alzheimer’s disease pathophysiology and risk factors with ... · ii. acknowledgements ariel...
TRANSCRIPT
Alzheimer’s Disease Pathophysiology and
Risk Factors with Amyloid Positron
Emission Tomography, an Open Science
Approach, and the Consideration of
Environmental Exposures
Eric E. Brown
A thesis submitted in conformity with
the requirements for the degree of Master of Science
Institute of Medical Science
University of Toronto
Copyright Eric E. Brown 2019
Alzheimer’s Disease Pathophysiology and Risk Factors with Amyloid
Positron Emission Tomography, an Open Science Approach, and the
Consideration of Environmental Exposures
Eric E. Brown
Master of Science
Institute of Medical ScienceUniversity of Toronto
2019
Abstract
Alzheimer’s disease (AD) is the most common underlying cause of dementia but is in-
completely understood. The pathophysiology of AD involves amyloid-beta plaques, neu-
rofibrillary tangles, and cerebrovascular changes involving white matter. Risk factors
including lead (Pb) exposure influence these processes. This thesis has four components
related to improving the understanding of AD pathophysiology. First, amyloid positron
emission tomography (PET) tracer delivery was hypothesized to be associated with white
matter integrity and was demonstrated to be correlated with established biomarkers in
mild cognitive impairment. Second, an open source software package for PET analy-
sis was created to improve transparency in AD research. Third, a systematic review of
case-control studies of Pb measurement in AD is presented, which highlights the possible
connection but identifies a need for studies that address early-life Pb exposure. And
fourth, a hypothesis that environmental microdose lithium may mitigate Pb toxicity
including cognitive impact is outlined with several literature reviews.
ii
Acknowledgements
Ariel Graff-Guerrero has been a supervisor and mentor not just during this Master’s
work but also during my years in the Clinician Scientist Program during my residency. I
thank him for his helpful guidance, teaching and direction. In particular, I thank him for
encouraging me to explore new ideas. My program advisory committee, including Philip
Gerretsen and Bruce Pollock, has likewise provided wonderful support and guidance,
both about the process and content of this research career I have started.
I would also like to thank Tiffany Chow, who mentored me as a medical student,
and later introduced me to the world of PET imaging, taking a chance on me to analyze
the data for a fascinating project that lead to my first scientific publication, a process
through which she supported me with kind mentorship.
Many colleagues in the Multimodal Imaging Group have been helpful teachers and
collaborators along the way. In particular, Fernando Caravaggio and Jun Chung helped
me learn PET analysis techniques.
I completed this Master’s thesis during the end of my psychiatric residency and
during a clinical fellowship, which caps off an academic journey that began over a decade
ago. Any achievements or success in this journey are due to the kind, loving, and generous
support of my extended family and friends. I thank my parents, Rosalie and Rob, for
their unwavering support for my interests and goals throughout my life. I thank my wife,
Jamie, for her loving support and encouragement through good times and hard times. I
am fortunate to have a family that is with me every step of the way, worth more than
anything else.
iii
Contributions
In Chapter 1, Figure 1.1 was reproduced from the figure 5 in the article “Tracking patho-
physiological processes in Alzheimer’s disease: an updated hypothetical model of dynamic
biomarkers” by Jack et al. (2013) and was reproduced with permission granted via Copy-
right Clearance Centre (Order License Id: 4601431467283).
Chapter 3, Amyloid Positron Emission Tomography as a Biomarkers of White
Matter Integrity, was reproduced, including figures and tables, with minor modifications
from the published article in Journal of Neuroimaging (E. E. Brown, Rashidi-Ranjbar,
et al. 2019), which is a permitted use as outlined in the Copyright Transfer Agreement.
Eric Brown conceived of the chapter’s hypothesis and statistical design in discussion
and under supervision and discussion with Ariel Graff-Guerrero, performed the PET
and MRI (except DTI) analysis, statistical analysis, and drafted the manuscript. Neda
Rashidi-Ranjbar performed the DTI analysis. The paper’s co-authors (Neda Rashidi-
Ranjbar, Fernando Caravaggio, Philip Gerretsen, Bruce Pollock, Benoit Mulsant, Tarek
Rajji, Corinne Fischer, Alastair Flint, Linda Mah, Nathan Herrmann, Christopher Bowie,
Aristotle Voineskos, and Ariel Graff-Guerrero) contributed to the conception, design and
implementation of the PACt-MD study. The co-authors provided valuable discussion
and feedback for this paper, which was incorporated into a revised manuscript. Research
assistants and staff aided in study implementation. Anonymous peer reviewers and the
editor made valuable feedback which was incorporated into the manuscript.
The data for this analysis is from a larger project, the PACt-MD randomized con-
trolled trial. This Project has been made possible by Brain Canada through the Canada
iv
Brain Research Fund, with the financial support of Health Canada and the Chagnon
Family.
The co-authors on the PACt-MD project have the following financial support and
disclosures: Neda Rashidi-Ranjbar has no potential conflicts of interest to declare and
has received a 3-year doctoral award from the Alzheimer Society Research Program–
Alzheimer Society of Canada. Fernando Caravaggio has no potential conflicts of interest
to declare and has received support and funding from CIHR, OGS, and the U of T Depart-
ment of Psychiatry Research Fellows Program. Philip Gerretsen has received research
support from CIHR, OMHF, Alternative Funding Program through the Ontario Ministry
of Health and Long-Term Care, and Centre for Addiction and Mental Health (CAMH).
Benoit H. Mulsant has received: research funding from Brain Canada, the CAMH Foun-
dation, the Canadian Institutes of Health Research, and the US National Institutes of
Health (NIH); research support from Bristol-Myers Squibb (medications for a NIH-funded
clinical trial), Eli-Lilly (medications for a NIH-funded clinical trial), Pfizer (medications
for a NIH-funded clinical trial), Capital Solution Design LLC (software used in a study
funded by CAMH Foundation), and HAPPYneuron (software used in a study funded by
Brain Canada). He directly own stocks of General Electric (less than $5,000). Tarek Rajji
has received research support from Brain Canada, Brain and Behavior Research Foun-
dation, BrightFocus Foundation, Canada Foundation for Innovation, Canada Research
Chair, Canadian Institutes of Health Research, Centre for Aging and Brain Health Inno-
vation, National Institutes of Health, Ontario Ministry of Health and Long-Term Care,
Ontario Ministry of Research and Innovation, and the Weston Brain Institute. Corinne
Fischer received grant funding from Roche pharmaceuticals and is the North American
PI for a device trial sponsored by Vielight Inc. Alastair Flint has received grant support
from the U.S. National Institutes of Health, the Patient-Centered Outcomes Research
Institute, the Canadian Institutes of Health Research, Brain Canada, the Ontario Brain
Institute, and Alzheimer’s Association. Christopher Bowie receives grant support from
Lundbeck, Takeda, and Pfizer; has been a speaker or on a board for Lundbeck and
Boehringer Ingelheim; and has received in-kind research materials from Scientific Brain
v
Training Pro. Aristotle Voineskos currently receives funding from the National Institute
of Mental Health, Canadian Institutes of Health Research, Canada Foundation for Inno-
vation, CAMH Foundation, and the University of Toronto. Linda Mah, Nathan Hermann,
Bruce Pollock, and Ariel Graff-Guerrero have no potential conflicts of interest to declare.
Chapter 4, Open Source PET Analysis in R, was based on the vignette included
with tacmagic, an R (R Core Team 2018) package and on an article summarizing
the package published in the Journal of Open Source Software (Brown 2019). In
both cases, Eric Brown was the primary author and retains the copyright, and both
the package and article are licenced with permissive licences. Eric Brown authored
the software and documentation. The package underwent open peer review by
rOpenSci (https://ropensci.org/software-review/) and two reviewers, Brandon Hurr
and Jon Clayden, provided detailed feedback (https://github.com/ropensci/software-
review/issues/280) on the package code which was incorporated into a revised version.
Chapter 5, Lead in Alzheimer’s Disease, is slightly modified from an article that was
published in Current Alzheimer’s Research (E. E. Brown, Shah, et al. 2019) and is repro-
duced in this thesis with written permission from the publisher Bentham Science. Eric
Brown is the primary author who conceived of and performed the review in consultation
with supervisor Ariel Graff-Guerrero, designed the systematic review and analysis, per-
formed the statistical analysis, and drafted the manuscript. Parita Shah independently
reviewed abstracts for inclusion. All co-authors (Eric Brown, Parita Shah, Bruce Pol-
lock, Philip Gerretsen, Ariel Graff-Guerrero) participated in intellectual discussion and
feedback, contributing to the intellectual content, and reviewed the manuscript. Anony-
mous peer reviewers made recommendations which were incorporated into the manuscript.
Dr. David Chettle provided input on a draft of the manuscript, and Dr. Patrick Parsons
and colleagues for shared raw data from a study that met inclusion criteria in the meta-
analysis.
Chapter 6, Water-supply Lithium, Environmental Lead Exposure, and Illness, is
modified from an article published in Medical Hypotheses (Brown et al. 2018) and is
included here in accordance with the policies of publisher Elsevier, which allows inclusion
vi
in a thesis without seeking permission. The article was derived from an essay written
for course work for this degree. Eric Brown conceived of the hypothesis and design
of the reviews and manuscript, performed the reviews and analyses and drafted the
manuscript. Dr. Tony George, Dr. Lena Quilty, and Karolina Kozak are acknowledged
for their insightful suggestions and feedback on the original proposal for the essay. All co-
authors (Eric Brown, Bruce Pollock, Philip Gerretsen, Ariel Graff-Guerrero) participated
in discussion and feedback, contributing to the intellectual content, and reviewed the
manuscript. Anonymous peer reviewers made recommendations which were incorporated
into the manuscript.
The formatting and typesetting template of this thesis was derived from a free and
open-source template by Pollard et al. (2016).
Financial support for this Masters and related work includes the Clinician Scientist
Program of the University of Toronto’s of the Department of Psychiatry, an Ontario
Graduate Scholarship, and the Canadian Institutes of Health Research Canada Graduate
Scholarships.
vii
Table of Contents
Abstract ii
Acknowledgements iii
Contributions iv
List of Tables xii
List of Figures xiii
Abbreviations xiv
1 Literature Review 1
1.1 Alzheimer’s Disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Pathophysiology of Alzheimer’s Disease . . . . . . . . . . . . . . . . . . 4
1.2.1 Amyloid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Tau . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.3 Neurodegeneration . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.4 Vascular Pathology . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Etiology and Risk Factors . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 Genetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.2 Major Depressive Disorder . . . . . . . . . . . . . . . . . . . . . 10
1.3.3 Hypertension and Vascular Factors . . . . . . . . . . . . . . . . 12
1.3.4 Lead . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4 Biomarkers of Alzheimer’s Disease Pathophysiology . . . . . . . . . . . 17
1.4.1 Cerebrospinal Fluid . . . . . . . . . . . . . . . . . . . . . . . . 18
viii
1.4.2 Positron Emission Tomography . . . . . . . . . . . . . . . . . . 19
1.4.3 Structural Magnetic Resonance Imaging . . . . . . . . . . . . . . 25
1.4.4 Reproducibility and Open Science in Neuroimaging . . . . . . . 30
1.4.5 Lead Measurement . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.5 A Modern Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2 Hypotheses and Research Aims 35
3 Amyloid Positron Emission Tomography as a Biomarkers of White
Matter Integrity 38
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.1 Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.2 Clinical characterization and image acquisition . . . . . . . . . . 43
3.2.3 Image analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2.4 Statistical tests . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3.1 Characterization of the sample . . . . . . . . . . . . . . . . . . . 48
3.3.2 White matter segmentation techniques . . . . . . . . . . . . . . 49
3.3.3 White matter hyperintensities, [11C]-PIB PET, and diffusion-
weighted imaging . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3.4 Potential confounding variables . . . . . . . . . . . . . . . . . . 51
3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.4.1 Sample Characteristics . . . . . . . . . . . . . . . . . . . . . . . 53
3.4.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.4.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4 Open Source PET Analysis in R 60
4.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.1.1 PET Analysis in R . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
ix
4.3 The tacmagic R package . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.1 Time-activity curve operations . . . . . . . . . . . . . . . . . . . 65
4.3.2 Model calculation . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3.3 Batch analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3.4 Cutoff calculations . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5 Lead and Alzheimer’s Disease 76
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.2.1 Scoping search and targeted searches . . . . . . . . . . . . . . . 79
5.2.2 Meta-analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.3.1 Scoping review . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.3.2 Targeted searches . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.4.1 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.5.1 Next steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.6 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6 Water-supply Lithium, Environmental Lead Exposure, and Illness 95
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.1.1 Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.2.1 Part 1: Systematic review of the health impacts of exposure to
environmental lithium . . . . . . . . . . . . . . . . . . . . . . . 98
6.2.2 Part 2: What are the psychiatric impacts of environmental lead
exposure? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.2.3 Part 3: Does lithium exposure protect against lead toxicity? . . . 100
6.2.4 Part 4: Biological mechanisms . . . . . . . . . . . . . . . . . . . 100
x
6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.3.1 Part 1: Health effects of environmental lithium exposure . . . . . 101
6.3.2 Part 2: Health effects of environmental lead exposure . . . . . . . 108
6.3.3 Part 3: Link between lithium and lead . . . . . . . . . . . . . . . 111
6.3.4 Part 4: Biological mechanisms . . . . . . . . . . . . . . . . . . . 113
6.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.4.1 Lithium in drinking water (Part 1) . . . . . . . . . . . . . . . . . 114
6.4.2 Environmental lead exposure (Part 2) . . . . . . . . . . . . . . . 115
6.4.3 The link between lithium and lead (Part 3) . . . . . . . . . . . . 116
6.4.4 Biological plausibility (Part 4) . . . . . . . . . . . . . . . . . . . 117
6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7 General Discussion 120
8 Conclusions 125
8.1 Amyloid PET tracer delivery in white matter may be a marker of white
matter integrity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
8.1.1 Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8.2 Open source PET analysis improves transparency and reproducibility . . 127
8.2.1 Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
8.3 Available case-control AD studies do not adequately address role of Pb in
AD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
8.3.1 Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
8.4 Drinking water lithium may mitigate harms of lead exposure and the
hypothesis requires testing . . . . . . . . . . . . . . . . . . . . . . . . . 130
8.4.1 Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.5 General Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
9 Future Directions 132
References 136
xi
List of Tables
3.1 Characterization of the sample. . . . . . . . . . . . . . . . . . . . . . . 49
3.2 Pearson correlations of the primary hypotheses (WM SUVmax and R1 with
WMH and FA) and of the cortical measures. For the 4 primary hypotheses,
p-values adjusted for false discovery rate (q) are also reported. . . . . . . 50
3.3 Pearson correlations between potential confounding variables and
biomarkers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4 Partial Pearson correlations of for primary hypotheses with age at PET
scan included as a controlled third variable. . . . . . . . . . . . . . . . . 52
5.1 Inclusion and exclusion criteria for the individual targeted searches. . . . 80
5.2 Summary of targeted searches. The search terms unique to each method
are shown, and all searches included terms for Alzheimer’s and lead as
described in the methods. . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.3 Summary of blood lead studies. . . . . . . . . . . . . . . . . . . . . . . 84
6.1 Studies that report on the association between suicide rates and environ-
mental lithium. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
xii
List of Figures
1.1 Conceptual model of biomarker changes in AD from Jack et al. (2013) . 33
3.1 Quality control output image (cropped) of LPA algorithm showing T2-
weighted FLAIR MRI image next to image with masked white matter
hyperintensities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.2 Relationships among primary white matter measures with least-square
regression line. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.1 Example output from tacmagic plotting function, as described in 4.3.1.3 69
4.2 Example output from tacmagic plotting function of a Logan plot. . . . 71
5.1 Meta-analysis of studies that compared whole blood lead in Alzheimer’s
disease compared to controls, and where group means and standard devi-
ations were available. . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.2 Meta-analysis of studies that compared serum lead in Alzheimer’s disease
compared to controls, and where group means and standard deviations
were available. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.1 Relative size comparisons of clinical Li doses (Liaugaudaite et al. 2017)
compared to estimated mean environmental daily intake (Schrauzer 2002),
represented with dose proportional to circle areas. . . . . . . . . . . . . 96
6.2 Results of systematic review of health impacts of environmental lithium. 102
xiii
Abbreviations
Aβ Amyloid beta peptide
AD Alzheimer’s Disease
ADNI Alzheimer’s Disease Neuroimaging Initiative
APP Amyloid Precursor Protein
ApoE Apolipoprotein E
AT(N) Amyloid, Tau, (Neurodegeneration)
BIDS Brain Imaging Data Structure
CAA Cerebral Amyloid Angiopathy
CBF Cerebral Blood Flow
CI Confidence Interval
CSF Cerebrospinal Fluid
CRAN Comprehensive R Archive Network
CT Computed Tomography
DSM 5 Diagnostic Stastistical Manual 5
DTI Diffusion Tensor Imaging
DVR Distribution Volume Rratio
DWI Diffusion-Weighted Imaging
FA Fractional Anisotropy
FDDNP 2-(1-6-[(2-[18F]fluoroethyl)(methyl)amino]-2-naphthylethylidene)malononitrile
FDG Fludeoxyglucose
FLAIR Fluid-Attenuated Inversion Recovery
GSK Glycogen Synthase Kinase
xiv
HC Healthy Controls
Li Lithium
MCI Mild Cognitive Impairment
MD Mean Diffusivity
MDD Major Depressive Disorder
MND Major Neurocognitive Disorder
MRI Magnetic Resonance Imaging
NFT Neurofibrillary Tangle
NHANES National Health and Nutrition Examination Survey
p-tau phosphorylated tau protein
PACt-MD Prevention of Alzheimer’s dementia with Cognitive remediation plus
Transcranial direct current stimulation in Mild cognitive impairment and Depression
PET Positron Emission Tomography
PIB Pittsburg Compound B
Pb Lead
ROI Region Of Iinterest
SRTM Simplified Reference Tissue Model
SUV Standardized Uptake Value
SUVR Standardized Uptake Value Ratio
TAC Time Activity Curve
t-tau total tau protein
WARM Washout Allometric Reference Method
WMH White Matter Hyperintensity
XRF X-ray Fluorescence
xv
Chapter 1
Literature Review
Dementia is a major, growing public health concern, which has been described as “the
greatest global challenge for health and social care in the 21st century” (Livingston et al.
2017). In Canada, 564,000 individuals were estimated to have dementia in 2016, which is
expected to grow to 937,000 people by 2031 (Chambers et al. 2016). By that time, there
will be around 66 million individuals with dementia worldwide (Livingston et al. 2017).
The burden of dementia to individuals, caregivers, and society is high. In Canada,
the total cost of caring for people with dementia was estimated to be $10.4 billion in
2016 (Chambers et al. 2016). An aging population and increasing longevity are driving
the increased prevalence of dementia, but it is noted that although age is a risk factor
for dementia, dementia is not considered an expected or inevitable part of normal aging
(Livingston et al. 2017). Death rates due to dementia have been increasing significantly,
with the age-adjusted death rate due to dementia in the United States increasing from
30.5 to 66.7 per 100,000 deaths from 2000 to 2017 (Kramarow & Tejada-Vera 2019).
The definition of dementia has been operationalized in the form of clinical criteria.
The American Psychiatric Association updated their clinical criteria in 2013, with the
publication of the Diagnostic and Statistical Manual 5 (DSM 5) (American Psychiatric
Association 2013). In DSM 5, the term dementia has been updated to major neurocogni-
tive disorder (MND), although dementia remains widely used in the literature (American
1
Psychiatric Association 2013). At its essence, the DSM 5 definition of MND involves a
significant decline in one or more cognitive domains which interferes with independence
in daily activities, but which are not due to delirium or another mental disorder (Amer-
ican Psychiatric Association 2013). When an individual has a more modest cognitive
decline which does not interfere with activities of daily living, they may have a mild neu-
rocognitive disorder (American Psychiatric Association 2013), commonly known as mild
cognitive impairment (MCI), which is considered a risk state for dementia (Livingston et
al. 2017).
Dementia is not a single disease, but a clinical syndrome with various underlying
causes. Alzheimer’s disease (AD) is the most common form of dementia (Barker et al.
2002). Additional criteria for specific forms of dementia are listed as specifiers within the
DSM 5, and alternative criteria for specific dementias are published by other groups, such
as the National Institute on Aging-Alzheimer’s Association (NIAA)’s clinical criteria for
AD (McKhann et al. 2011).
In the face of the rising burden of dementia, the underlying diseases, including
AD, do not yet have a cure, and current interventions are focused on prevention and
management of symptoms (Livingston et al. 2017). It is hoped that an improved un-
derstanding of the pathophysiology of AD will lead to the development of a curative
treatment (Canter et al. 2016), thus reducing the burden of this illness on individuals,
caregivers, families and society. This thesis adds to a large body of work, for example,
as summarized by Jack et al. (2013), which hopes to clarify the pathophysiology of AD.
1.1 Alzheimer’s Disease
The definition of AD is also operationalized within the DSM 5 as clinical criteria for
specifiers of major and mild neurocognitive disorder (American Psychiatric Association
2013). The AD specifier can be either Probable or Possible AD. For example, for major
neurocognitive disorder due to Probable AD, the onset of impairment is insidious and
2
impairment gradually progresses, and there must be either a personal or family history
of a “causative” genetic mutation (see 1.3.1 below), or all three of memory decline in
addition to another cognitive domain, a lack of plateaus in the steady progression of
cognitive decline, and an absence of evidence for a mixed etiology. The disorder must
also not be better explained by another illness or substance use (American Psychiatric
Association 2013).
Currently, AD is conceptualized to occur on a continuum of severity ranging from
pre-clinical, to mild cognitive impairment, and finally to mild, moderate and severe
dementia (Dubois et al. 2016). The stages of AD are outlined in a review by Holtzman
et al. Holtzman et al. (2011). A typical time course of dementia due to AD may be 7-10
years. Functioning becomes gradually more impaired from the mild to severe dementia,
and each stage may last several years (Holtzman et al. 2011). Often memory deficits are
an early symptom, but other cognitive domains can be involved early on, with worsening
deficits over time (Holtzman et al. 2011). Independence in daily living can be relatively
preserved in the mild stage, but is lost in the moderate stage, with total dependence
occurring by the severe stage until the individual dies (Holtzman et al. 2011).
While probable or possible AD can be diagnosed on the basis of its clinical presen-
tation without the use of biomarkers, a specific underlying pathology is presumed as the
cause of AD. The characteristic neuropathologic changes include (a) the accumulation of
amyloid beta-peptide (Aβ) plaques, and (b) neurodegeneration including neurofibrillary
tangles of tau proteins. These two elements of histopathology–now known as amyloid
plaques and neurofibrillary tangles–were described by Alois Alzheimer in 1906 in an au-
topsy of a 51-year-old woman who had progressive cognitive impairment, and the disease
was subsequently named after him by Kraeplin in 1910 (Hippius & Neundörfer 2003;
Berrios 1990). More recently, it has been contested whether this first case would meet
current criteria for AD, due to the presence of significant atherosclerosis (O’Brien 1996).
However, amyloid plaques and neurofibrillary tangles are central in modern models of
AD (Jack et al. 2010; Jack et al. 2013).
There has been a recent call to redefine AD, for research settings, from a clinical
3
(based on signs and symptoms) to a biological definition, as biomarkers of pathology
are measurable in vivo (Jack et al. 2018). The definition would be based on measures
of amyloid, tau and neurodegeneration, labelled AT(N) (Jack et al. 2018). The pro-
posal suggests that amyloid biomarkers could be used to determine if an individual is
in an “Alzheimer’s continuum”, tau biomarkers could then confirm that someone on the
continuum has AD, and neurodegeneration and cognitive symptoms are not part of the
definition, but can be used for staging (Jack et al. 2018).
1.2 Pathophysiology of Alzheimer’s Disease
1.2.1 Amyloid
Since the identification of senile plaques in the early case reports by Alois Alzheimer at
the beginning of the 20th century (Berrios 1990), amyloid has been implicated in AD.
Amyloid has been described as the “primary influence driving AD pathogenesis” in what
is known as the amyloid hypothesis or amyloid cascade hypothesis of AD (Hardy & Allsop
1991; Hardy 2002).
The amyloid hypothesis proposed that AD occurs due to a sequence of pathogenic
events beginning with increased production and accumulation of Aβ42, in some cases
as a result of genetic mutations of the amyloid precursor protein (APP) and presenilin
(PS) genes (Hardy 2002). A membrane protein, the physiologic role of APP is not fully
understood (Müller et al. 2017). Aβ is cleaved from APP by β- and γ-secretase, two
intramembrane cleaving enzymes. The specific site of the cleavage determines the form
of Aβ (with the number that follows Aβ indicating the length of the peptide). Some
forms of Aβ more easily self-aggregate into plaques, and are implicated in AD, including
Aβ42 (Selkoe & Hardy 2016).
Since it was proposed, much research has gone into amyloid and AD to explore
the hypothesis (Selkoe & Hardy 2016). There has been doubt and objection surrounding
4
the amyloid hypothesis (Makin 2018; Kametani & Hasegawa 2018). The low correlation
between extent of amyloid deposition and cognitive symptoms was an early concern
(Hardy 2002), and more recently, the failure of randomized controlled trials of amyloid-
reducing medications, have contributed to recent criticism of the hypothesis (Makin 2018).
However, the objection to amyloid plaques having a causal role in idiopathic AD may be
diminishing as a result of growing evidence for the sequence of pathological changes and
causal experiments in rodents (Selkoe et al. 2012).
There is a broad literature supporting a central role of amyloid in AD, which sum-
marized by Selkoe and Hardy Selkoe & Hardy (2016). In support of the hypothesis, the
genes which increase the risk of AD are known to affect Aβ: the PS genes, of which
mutations are rare causes of early-onset AD, encode γ-secretase, and lead to a relative
elevation of pathogenic, aggregating forms of Aβ, Aβ42/43 (Selkoe & Hardy 2016). A
more common genetic variation, the apoplipoprotein E (ApoE) ϵ4 allele, markedly in-
creases the risk of late-onset AD in a dose-dependent fashion and is associated with
decreased clearance of Aβ (Selkoe & Hardy 2016).
In longitudinal human studies, cerebral amyloid accumulation based on amyloid
positron emission tomography (PET) shows that significant accumulation of amyloid
occurs prior to other pathological features seen in AD, including increased cerebrospinal
fluid (CSF) tau, decreased cerebral metabolism, cortical atrophy, and the cognitive and
functional symptoms of dementia (Selkoe & Hardy 2016) (see below for a more detailed
overview of the evolution of biomarkers in AD). Rodent studies also support a causal
impact of amyloid. For example, human Aβ42 induces other AD-related pathology in
rats when injected (Selkoe & Hardy 2016).
1.2.2 Tau
In addition to the extracellular senile plaques now known to be aggregations of Aβ,
intracellular neurofibrillary tangles (NFTs) were the other hallmark finding in Alois
Alzheimer’s initial report. Consisting of accumulations of tau proteins, NFTs remain
5
a core pathological feature AD (Hippius & Neundörfer 2003; Jack et al. 2018). Tau is a
microtubule-associated protein, which has the physiologic role of stabilizing microtubules
(Guo et al. 2017). Tau protein abnormalities are not unique to AD; rather they occur in
a group of disorders called tauopathies, which includes Lewy Body and frontotemporal
dementia, which feature different forms of abnormal tau (Guo et al. 2017).
Tau proteins undergo post-translational modification, most frequently by phospho-
rylation (Guo et al. 2017). Phosphorylation of tau affects its physiologic function and is
increased in pathological states (Guo et al. 2017). For example, phosphorylation of tau
can decrease its affinity for microtubules and increase its self-aggregation. Regulation of
phosphorylation occurs via kinases and phosphatases. In AD, glycogen synthase kinase
(GSK) 3β has an important role, as it phosphorylates tau at threonine-231, which may
lead to tau aggregation (Guo et al. 2017).
In AD, tau neurofibrillary tangles have a characteristic pattern of progression. In
a landmark pathological study, Braak & Braak (1991) examined a series of 83 brains of
individuals with and without diagnosed dementia, across a range of ages. They reported
a sequence of stages in which tau pathology can be identified in progressively more brain
regions in a predictable order in AD (Braak & Braak 1991). Whereas amyloid accumula-
tion can occur many years prior to the onset of clinical symptoms, the progression of NFT
pathology is correlated with clinical symptoms and the progression of AD (Holtzman et
al. 2011).
1.2.3 Neurodegeneration
In addition to the microscopic findings of AD, significant atrophy can be noted on gross
examination (Holtzman et al. 2011). Various markers of neurodegeneration are present
in AD, and while they are not specific to AD, can be useful for staging the illness (Jack
et al. 2018). Neurodegenerative changes in AD include loss of synapses and populations
of neurons (Holtzman et al. 2011). Commonly reported regions of neuronal loss include
areas within the entorhinal cortex, the hippocampus, as well as within the temporal,
6
parietal and frontal cortex (Holtzman et al. 2011). Additionally, there is disruption of
specific networks and neurotransmitter systems, including cholinergic dysfunction (Holtz-
man et al. 2011). White matter disease is also observed in AD, with around half of cases
reported to be associated with angiopathy-related white matter disease in both deep and
periventricular regions (Englund 1998).
1.2.4 Vascular Pathology
While AD is the most common single cause of dementia, in most cases of dementia
(Schneider et al. 2007), and in about half of cases of AD, there is mixed pathology, of
which vascular factors are the most common (Snyder et al. 2015). However, vascular
pathology and dysfunction are not core defining features of AD. In fact, the DSM 5
clinical criteria specifier for AD requires that the illness is “not better explained by
cerebrovascular disease” (American Psychiatric Association 2013). Further, unless there
a “causative” genetic mutation for AD is present, then the presence of evidence for mixed
etiology would exclude the DSM 5 diagnosis of Probable AD (American Psychiatric
Association 2013).
In cases of a neurocognitive disorder where there is evidence of multiple etiologic
processes, DSM 5 assigns the diagnosis of Neurocognitive Disorder Due to Multiple Eti-
ologies (American Psychiatric Association 2013), which is more commonly called mixed
dementia. The most common pathologies of mixed dementia are vascular and AD, such
that the term often refers specifically to this combination (Langa et al. 2004).
Cerebrovascular pathology is nonetheless common and important in AD, and it
contributes to the pathogenesis and development of cognitive impairment (Jack et al.
2013). As described below in 1.3.3, vascular risk factors contribute to the development
of AD.
One way in which the pathological changes of AD directly contribute to vascular
pathology is via cerebral amyloid angiopathy (CAA), which is the deposition of Aβ in
7
blood vessel walls (Holtzman et al. 2011). The consequences of CAA include ischemia,
lobar hemorrhage, and in rare cases, vasculitis (Holtzman et al. 2011).
1.3 Etiology and Risk Factors
As a complex disease, AD does not have a single cause. Multiple risk factors contribute
to the development of AD and dementia. Further, the development of dementia, as men-
tioned, often involves multiple underlying pathologies, which commonly includes AD.
While genetics plays a role, environmental, demographic and medical comorbidities can
account for a significant proportion of cases of dementia. Livingston et al. (2017) used
published relative risks from systematic reviews to calculate the population attributable
fraction of known potentially modifiable risk factors for dementia. They found that ap-
proximately 35% of dementia cases can be attributed to nine risk factors: education,
midlife hypertension, midlife obesity, hearing loss, late-life depression, diabetes mellitus,
low physical activity, smoking and social isolation (Livingston et al. 2017). While this
report relies on mixed-quality evidence and is not specific to AD, it illustrates the com-
plexity of AD risk factors and the importance of taking a broach approach in prevention,
but also in research into the pathophysiology of dementia and AD. In this section, we
explore the major risk factors and causes of AD.
1.3.1 Genetics
As outlined amyloid above, the discovery of genetic mutations that increases the risk of
AD was instrumental in the development of the amyloid hypothesis (Hardy 2002). Ge-
netic heritability plays a major role in AD, and it has been studied extensively (Bertram
& Tanzi 2008; Cuyvers & Sleegers 2016). Having a single first-degree relative with AD
confers a relative risk of 3.5, which increases up to 7.5 in cases of multiple affected
first-degree relatives (Cuyvers & Sleegers 2016; van Duijn et al. 1991).
A brief overview of the genetics of AD could be dichotomized into the rare
8
“causative” mutations that dramatically increase the risk of AD and are associated
with an early onset of illness, and secondly the more common genetic mutations which
increase the risk of AD, but which are not necessary or sufficient to cause the illness.
The former category is included in the DSM 5 diagnostic criteria (American Psychiatric
Association 2013).
1.3.1.1 Causative Genetic Mutations
Though the early-onset form of AD is relatively rare, accounting for 1-2% of cases (Brouw-
ers et al. 2008), it has been found to be attributable to highly penetrant single-gene
mutations, which have helped to understand the pathophysiology of AD (Brouwers et al.
2008). The APP gene, on chromosome 21, was discovered to be implicated in AD after
AD-pathology was observed in Down’s syndrome, the syndrome associated with trisomy
21. Various types of mutations of the APP gene, including missense, duplication, and
promotor region mutations can lead to increased production and/or aggregation of Aβ
(Brouwers et al. 2008).
Mutations of the presenilin genes, PSEN1 and PSEN2 located on chromosome 14,
are another cause of early-onset AD, and were discovered for their role in AD (Brouwers et
al. 2008). Mutations of these genes have also been observed in late-onset AD (Brouwers
et al. 2008). The function of PSEN genes relates their γ-secretase activity, which cleaves
APP in Aβ. Mutations of PSEN affect the ratio of Aβ42 to Aβ40, with a relative
abundance of the amyloidogenic Aβ42 (Brouwers et al. 2008).
Taken together, mutations in APP, PSEN1, and PSEN2 explain 5-10% of early-
onset cases (Cuyvers & Sleegers 2016), but can also occur in late-onset AD. Other genetic
mutations have been identified but are less common (Brouwers et al. 2008).
9
1.3.1.2 Risk-associated Genes
Genome-wide association studies have helped to identify over 20 risk loci implicated in
AD, which may explain 28% of the heritability of liability of AD (Cuyvers & Sleegers
2016). Outside of the rare autosomal dominant mutations that can lead to early-onset
AD, the genetic factor that has the most significant impact on the risk of AD is the ϵ4polymorphism in the APOE gene.
Apolipoprotein E (ApoE) is a glycoprotein with physiologic roles in cholesterol
transport, the major brain cholesterol carrier protein (Puglielli et al. 2003). It is produced
in the periphery as well as within astrocytes in the central nervous system (Zhao et al.
2018). ApoE has three common isoforms encoded by the APOE ϵ2, ϵ3, and ϵ4 alleles
(Puglielli et al. 2003). The APOE ϵ4 allele is said to increase the risk of AD in a
dose-dependent manner, with the relative risk of AD highest with two ϵ4 alleles, with an
estimated relative risk around 8 (Puglielli et al. 2003). However, the risk may be different
in different populations, with the risk of AD conferred by the ϵ4 allele much lower in
individuals with African as compared to European ancestry, and highest in East-Asian
population (Rajabli et al. 2018). Further, ApoE status may interact with non-genetic
factors including blood pressure (Rajan et al. 2018), elevated blood glucose (Bangen et al.
2016), depression (da Silva et al. 2013), and head injury (Nicoll et al. 1996). Thus, ApoE
status is an important factor to ascertain in studies investigating the pathophysiology of
AD.
Over 20 other risk-conferring genetic loci have been identified, though none confer-
ring risk as strongly as ApoE ϵ4 (Cuyvers & Sleegers 2016).
1.3.2 Major Depressive Disorder
Depressed mood in late life can be an early symptom of AD. A lifetime history of major
depressive disorder (MDD) is associated with an increased risk of dementia, including
AD. Given that MDD is a common condition, for example which affects around 11% of
10
Canadians (Pearson et al. 2013), and one that is potentially treatable, its potential role
as a risk factor for AD has been widely investigated.
A meta-analysis including 20 case-control and cohort studies with a total of 102,172
individuals demonstrated that a history of depression is associated with an odds ratio
(OR) for AD of approximately 2 (Ownby et al. 2006). This study also found that the
interval between onset of MDD and AD was positively associated with the OR, suggesting
that MDD is a risk factor rather than merely a prodrome for AD. An analysis of 949
participants from the original Framingham Heart Study cohort found an increased risk
of AD with a hazard ratio of 1.76 over a 17-year follow up period (Saczynski et al. 2010).
A later meta-analysis similarly found similar overall results and highlighted a number of
contradictory studies suggesting either a risk of only early-onset or late-onset MDD on
dementia (da Silva et al. 2013). Weighing in on the conflicting literature, a more recent
28-year longitudinal study exploring the impact of MDD on later onset of dementia in
10,308 individuals found that early life MDD including recurrent MDD starting in early
life was not significantly associated with a higher risk of dementia, but that depressive
symptoms in the later 11-year phase of the study were associated with a significantly
increased risk of dementia. This study supported the possibility that MDD and dementia
share an underlying cause, rather than MDD being a causal factor in developing dementia
(Singh-Manoux et al. 2017).
While the specific nature of the relationship between AD and MDD is not yet
confirmed, the overlapping neurobiology has led to numerous biological mechanisms being
proposed to explain the link, including vascular factors, inflammation, Aβ, glucocorticoid-
mediated hippocampal atrophy, and neuronal growth factors (Byers & Yaffe 2011).
We previously reviewed the literature on the specific relationship between MDD
and tau and found that there was limited evidence available to conclude whether there
is a potential role of tau in MDD (E. E. Brown, Y. Iwata, et al. 2016).
Harrington et al. (2015) reviewed the literature exploring a role of Aβ in MDD.
Their systematic review identified studies that used CSF, plasma, and serum measure-
11
ment to compare Aβ40, Aβ42, and/or the Aβ40:Aβ42 ratio levels in individuals with
MDD and those without depression. A further five studies explored the role with PET
imaging. While there was heterogeneity and bias among the included studies, there was
evidence for a role of Aβ in MDD. All six studies that reported the Aβ40:Aβ42 ratio
found that it was higher in MDD than individuals without depression, as it is known
to be in AD. Notably, it was found in early- and late-life depression. No studies found
MDD to be associated with differences in Aβ40, but most studies that measured Aβ42found it to be lower in MDD as compared to non-depressed individuals. The PET studies
in the review also had heterogeneous results, with some identifying global and regional
increases in amyloid in MDD.
Our group analyzed data from the Alzheimer’s Disease Neuroimaging Initiative
(ADNI), comparing individuals with amnestic MCI with and without a lifetime history
of MDD (Chung et al. 2015). The groups were matched on age, gender, race, ethnicity,
education, cognition, level of functioning, ApoE status, and marital status, and the
groups did not significantly differ in current depressive symptoms. We found that in
amnestic MCI, a history of depression was associated with increased amyloid burden in
the frontal cortex.
Thus, while MDD is an established risk factor associated with AD, the specific
nature of the relationship and biological mechanisms requires further research.
1.3.3 Hypertension and Vascular Factors
The Framingham Heart Study is a major longitudinal study of multiple cohorts that
began in 1948, which, among many influential medical insights, has helped to identify the
major cardiovascular risk factors (Wilson et al. 1998). The cardiovascular risk factors,
now widely accepted, include age, hypertension, dyslipidemia, and cigarette smoking.
These risk factors have been shown to be associated with the progression from MCI to
AD (Viticchi et al. 2017) and advancing cognitive decline in AD (Viticchi et al. 2015).
12
As noted in 1.2.4, the presence of cerebrovascular pathology in AD is typical. Vascu-
lar pathology such as microinfarcts may contribute to the development of AD-pathology,
including Aβ deposition, which has been demonstrated experimentally in animal studies
(Garcia-Alloza et al. 2011). Thus, vascular factors may contribute to the development
of AD-related pathology and contribute to the progression of the illness. Even without
directly impacting AD pathology, concomitant cerebrovascular disease can reduce the
cognitive reserve and lead to earlier presentations of dementia (Livingston et al. 2017).
There is debate about the extent to which vascular risk factors contribute to AD-specific
pathology (plaques and tangles), rather than merely co-occuring vascular pathology that
contributes to cognitive decline and making dementia (including due to AD) more likely
to be diagnosed (Chui et al. 2011).
Hypertension is a major risk factor for cerebrovascular disease and therefore for de-
mentia and has been identified as a treatment target for dementia prevention (Livingston
et al. 2017). Midlife hypertension has been associated with an increased risk of dementia,
including AD specifically, later in life by several large longitudinal observational studies
(Launer et al. 2000; Kivipelto 2001; Whitmer et al. 2005; Abell et al. 2018). Other
cardiovascular risk factors in midlife have been associated with dementia and AD, includ-
ing smoking, elevated cholesterol and diabetes mellitus (Kivipelto 2001; Whitmer et al.
2005).
In addition to guiding clinical preventive strategies, the randomized controlled trial
design is best able to investigate a potential causal role of hypertension in AD. Despite
calls to treat hypertension to reduce the risk of dementia, the evidence directly supporting
such practice is mixed. For example, a Cochrane review from 2009, data from random-
ized controlled trials including 15,936 participants without cerebrovascular disease were
included in a meta-analysis, which did not demonstrate a benefit of antihypertensive
treatment on cognitive impairment or dementia in late-life (McGuinness et al. 2009).
The mean age of the included in that review was 75, which may be too late for a benefit
of blood pressure reduction. Late-life hypertension may not be associated with an in-
creased risk of dementia, and there is evidence suggesting that the onset of hypertension
13
after age 80 may be associated with a reduced risk of dementia (Corrada et al. 2017) even
compared to those without hypertension. Low blood pressure may be more problematic
than hypertension in the very old.
A recent large randomized controlled trial of 9361 individuals with hypertension
comparing intensive blood pressure control (target systolic blood pressure < 120) to
the usual target (< 140) did not find a statistically significant impact of the intensive
control, with a hazard ratio for probably dementia of 0.83; 95% CI, 0.67-1.04. The trial
was stopped early due to other favourable cardiovascular outcomes and may have been
underpowered for this endpoint. A secondary endpoint, MCI, was significantly lower in
the intensive control group (hazard ratio 0.81; 95% CI, 0.69-0.95) (The SPRINT MIND
Investigators for the SPRINT Research Group et al. 2019). The mean age of participants
was 68.
Given the association between vascular risk factors and dementia, the improved
treatment of vascular risk factors in certain populations has been proposed as a par-
tial explanation for an observed reduction in the age-specific reduction in incidence of
dementia (Satizabal et al. 2016; Langa 2015).
Thus, while hypertension in midlife has been repeatedly demonstrated to be asso-
ciated with the onset of dementia and AD later in life, there is limited evidence from
prospective randomized controlled trials to demonstrate a benefit of antihypertensive
treatment, which may be partly due to the age of the participants in the studies that
have been done.
1.3.4 Lead
While not discussed in the Livingston et al. (2017) paper on risk factors for dementia,
lead (Pb) exposure has long been proposed as a potential contributing cause of AD. It
was originally identified as a potential causative agent on the basis of epidemiological
data suggesting the role of an unidentified environmental agent, which Pb would be a
14
possible candidate (Prince 1998). Since then, there has been evidence connecting Pb
exposure to cognitive impairment and dementia risk factors supporting the possible role
of Pb in AD.
Due to its current and historical commercial uses, most notoriously as an anti-
knocking additive to gasoline used ubiquitously from the 1920s up until its phasing out
began in the 1970s (Nriagu 1990), human exposure to Pb has been a natural experiment.
Exposure to Pb is ongoing, particularly in the developing world, but also to a lesser extent
in developed countries. Routes of current exposure include leaded pipes and dust from
Pb-based paint, as well as in soil (Advisory Committee on Childhood Lead Poisoning
Prevention of the Centers for Disease Control and Prevention 2012).
The toxicity of Pb is well-established, with serious and widespread effects on multi-
ple organ systems. By substituting for calcium ions, Pb can cross the blood brain barrier
(Flora et al. 2012), where it has diverse effects including disruptions to cell signaling,
neurotransmitter and ion transport, and disruption to protein processing and enzyme
function (Garza et al. 2006), in addition to the generation of reactive oxygen species
(Flora et al. 2012). Acute poisoning, typically associated with industrial exposure, oc-
curs at a blood level around 100-120 �g/dl, while chronic poisoning is described at blood
levels around 40-60 �g/dl (Flora et al. 2012). More insidiously, impacts of lower levels of
exposure, even below 10 �g/dl, continue to be identified and include cognitive, cardiovas-
cular, immunological, and endocrine effects. Accordingly, the Centre for Disease Control
has stated that there is no safe level of Pb exposure (Advisory Committee on Childhood
Lead Poisoning Prevention of the Centers for Disease Control and Prevention 2012).
In principle, the impact of Pb on AD could occur by different mechanisms. Firstly,
it could affect cognition by non-AD related mechanisms, decreasing cognitive reserve;
secondly, by causing or interacting with other AD risk factors; and thirdly, by directly
cause or inducing AD-related pathological changes. There is evidence for each of these
mechanisms.
Firstly, as low-level Pb exposure impacts cognition, the risk of future AD could
15
be increased via a reduction in cognitive reserve. Several longitudinal studies of early
life exposure have confirmed the relationship between Pb exposure and IQ at the low-
est detectable levels, and this has been interpreted as a causal relationship (Advisory
Committee on Childhood Lead Poisoning Prevention of the Centers for Disease Control
and Prevention 2012). For example, a cohort study of 4,853 children ages 6-16 years
demonstrated that after accounting for potential confounding variables, Pb exposure
was inversely associated with intelligence and cognition even at levels below 10 �g/dl.
Another prospective longitudinal cohort study of blood Pb levels in 1,037 11-year-old
children found an inverse association between Pb and both cognitive measures and fu-
ture socioeconomic status at age 38 (Reuben et al. 2017). In adults, Pb exposure has
also been associated with cognitive impairment. There have been numerous longitudinal
studies that have measured bone Pb, a measure of chronic exposure (see 1.4.5) with
cognitive decline after adjusting for co-variates, as reviewed by Shih et al. (2007).
Secondly, there is also evidence implicating Pb exposure with risk factors for AD.
Specifically, Pb has known deleterious effects on the cardiovascular system (Flora et al.
2012). The effects of environmental Pb exposure on hypertension may be particularly
pronounced. For example, in a prospective cohort study with a sample of 14,289 adults
weighted to reflect population, the impact of blood Pb levels on health outcomes was
determined over a median follow up period of 19 years. In analyses adjusted for cardio-
vascular risk factors, the authors determined that higher Pb levels were associated with
an increased all-cause mortality, mortality from cardiovascular disease and hypertension.
They estimated that the number of deaths in the United States that can be attributed
to Pb exposure via cardiovascular disease was 250,000 annually (Lanphear et al. 2018).
This important study is in the context of a body of prospective trials linking Pb exposure
to hypertension, as reviewed by Navas-Acien et al. Navas-Acien et al. (2007).
Thirdly, Pb has also been implicated directly affecting the pathophysiologic process
of AD. Most of the evidence supporting this direct connection comes from experimental
studies in animals and in vitro cell studies. Pb(IV) ions have been demonstrated to have a
strong, specific binding to Aβ-40, which has been proposed as a potential mechanism con-
16
tributing to AD pathology (Wallin et al. 2017). Tau hyperphosphorylation in response to
Pb exposure has been demonstrated in differentiated human neuroblastoma cells (Bihaqi
et al. 2017). Animal studies strongly implicated early-life Pb exposure in inducing later
AD pathology. Primates exposed to Pb as infants have increased tau phosphorylation
in adulthood (Bihaqi & Zawia 2013), increased APP expression and increased brain Aβ
intracellularly and deposited as plaques (Wu et al. 2008). Similarly, rodents exposed to
Pb as neonates have increased APP expression and cerebral Aβ as adults (Basha 2005).
The observation that early life exposures can affect later-life genetic regulation and ex-
pression, ultimately leading to pathology as seen with Pb and AD-related changes, has
been conceptualized in a model termed Latent Early-life Associated Regulation (LEARn)
(Lahiri et al. 2009).
Thus even in the absence of reports of powered longitudinal studies that report the
association of the incidence of later-life AD (and not simply cognitive deficits more gener-
ally) with early-life Pb exposure, and without direct experimental support in humans for
an AD-specific impact, there is significant evidence to implicate Pb as a possible cause or
contributor to AD in humans via either a direct impact on amyloid and tau pathological
pathways, indirectly via cardiovascular or metabolic effects, or both.
1.4 Biomarkers of Alzheimer’s Disease Pathophysiology
Historically, diagnostic confirmation of AD has only been possible by post-mortem
histopathological exam, and where to clinicians, the term AD may refer to a clinical
entity with characteristic symptoms and course, to researchers, the term may refer to
the characteristic pathological process (Dubois et al. 2010). Since the availability of
reliable, validated biomarkers that allow the in vivo detection of the neuropathological
processes of AD, there have been proposals to redefine the disease on this basis (Dubois
et al. 2010; Jack et al. 2018). Irrespective of definitions, such biomarkers have proven
immensely valuable in advancing the understanding of the biology of AD, and they
may play an important role in the development of disease-modifying drugs (Blennow
17
et al. 2010; Jack et al. 2018). The most important biomarkers in AD include CSF
levels of Aβ and tau, amyloid and glucose PET, and magnetic resonance imaging (MRI)
(Dubois et al. 2010). These biomarkers have been categorized as “pathophysiological”–
reflecting the core pathophysiological changes that define AD, i.e. amyloid and tau, and
“topographical”–reflecting downstream brain changes, i.e. atrophy and reduced blood
flow and metabolism (Dubois et al. 2010).
1.4.1 Cerebrospinal Fluid
Surrounding the nervous system and within the brain’s ventricles, the CSF is in close
communication with the brain and its extracellular space. An important function of
the CSF is its role as a sink for the solutes from the nervous system’s extracellular
space (Simon & Iliff 2016). In AD, some of the normal clearance function of the CSF is
impaired, and the clearance of Aβ is specifically reduced, which may be a contributing
factor to the pathogenesis of AD (Simon & Iliff 2016). Due to the close communication
of the CSF with the extracellular space and the sink function, measurement of CSF
provides important information about the biochemistry of the brain, although unlike
neuroimaging, it cannot provide localized information about specific brain regions. The
modality has been extensively studied in AD and is established as an important biomarker
(Olsson et al. 2016; Blennow et al. 2010). The core CSF biomarkers in AD are Aβ42,total tau (t-tau), and phosphorylated tau (p-tau181 and p-tau231) (Blennow et al. 2010).
A recent meta-analysis included hundreds of studies of AD CSF biomarkers to assess
their diagnostic validity (Olsson et al. 2016).
The core amyloid CSF biomarker is Aβ42, which is lower in AD compared to
controls and is suspected to result from aggregation in plaques in the brain (Blennow et
al. 2010) and/or decreased clearance from the brain (Selkoe & Hardy 2016). There is
an inverse relationship between CSF Aβ42 and amyloid plaques (Tapiola et al. 2009).
The meta-analysis by Olsson et al. (2016) included 131 studies that included 9,949
participants with AD and 6,841 controls. The average ratio of CSF in AD to controls
18
was 0.56 (95% CI 0.55-0.58) (Olsson et al. 2016).
Total tau in CSF is hypothesized to reflect damage to neurons and axons in the
brain (Blennow et al. 2010), and it correlates with neurofibrillary tangle load (Tapiola et
al. 2009). The meta-analysis by Olsson et al. (2016) included 151 studies with a total of
11341 participants with AD and 7086 controls. The ratio of levels in AD to controls was
2.54 (95% CI 2.44–2.64) (Olsson et al. 2016). Phosphorylated tau in the CSF reflects
the formation of neurofibrillary tangles, and the state of tau phosphorylation, and is
commonly measured for phosphorylation at threonine-181 (p-tau181) or threonine-231 (p-
tau231) (Blennow et al. 2010). The meta-analysis by Olsson et al. (2016) included 89
studies with 7498 participants with AD and 5126 controls and the average ratio was 1.88
(95% CI 1.79–1.97).
For diagnostic purposes, combining CSF Aβ42 and tau improves accuracy. For
example, in individuals with MCI who went on to develop AD within 3 years, CSF
Aβ42/t-tau as a diagnostic test performed well with 97% sensitivity and 83% specificity
(Palmqvist et al. 2015).
1.4.2 Positron Emission Tomography
Positron emission tomography is an imaging modality that uses radiotracers that bind to
targets of interest. The radiotracers are labelled with a radioactive ligand, such as carbon-
11 (11C) or fluorine-18 (18F), which undergo radioactive decay resulting in the emission
of a positron. A PET scanner detects the location of the radioactive decay and the total
radioactivity by voxel is measured over time. Kinetic analysis of the time-activity data
can reveal estimates of radiotracer binding in regions of interest, among other measures
of the tracer kinetics. In AD, the PET radiotracers of interest target amyloid and tau,
to identify the accumulated abnormal proteins that occur in AD, and glucose, to identify
changes in cerebral metabolism and blood flow. Tracers are used in small doses and the
total radioactivity exposure is low and comparable to other radiologic tests (Scheinin et
al. 2007).
19
1.4.2.1 Fludeoxyglucose PET
A chronological account of the major PET radiotracers in AD would begin with
fludeoxyglucose (18F-FDG). By virtue of its resemblance to glucose, 18F-FDG enables
functional imaging of cerebral glucose metabolism (Bohnen et al. 2012). It was first
synthesized in 1976 and is a widely used PET radiotracer today (Portnow & Vaillancourt
2013). As reviewed by Bohnen et al. (2012), AD is associated with significant changes
in glucose metabolism that can be identified by 18F-FDG PET, including reductions in
temporoparietal regions, the posterior cingulate cortex and the frontal cortex, reflecting
hypometabolism associated with neurodegeneration.
As discussed above in 1.2.3, neurodegeneration including loss of synaptic function
and density, is a feature of AD, which is more closely related to the onset of clinical
symptoms than the presence of Aβ. However, a meta-analysis demonstrated that 18F-
FDG PET is useful in predicting progression from MCI to AD, with a sensitivity of 89%
and specificity of 85% (Yuan et al. 2009). While non-specific to AD, other forms of
dementia may be associated with different patterns of hypometabolism, which is another
reason why 18F-FDG PET can be useful clinically (Bohnen et al. 2012).
1.4.2.2 Amyloid PET
The first Aβ-specific tracer used in humans in vivo was Pittsburgh Compound B (PIB),
which was derived from thioflavin-T, an amyloid histological dye (Klunk et al. 2004).
The properties of this tracer enable in vivo identification and quantification of cerebral
Aβ deposition: it binds amyloid at sufficiently low concentrations, crosses the blood
brain barrier, and clears rapidly from brain tissue without amyloid (Klunk et al. 2004).
The radioligand used by PIB is 11C, which has a half-life of approximately 20 minutes,
whereas 18F has a longer half-life of 110 minutes, and is more practical for distributing
to sites without a cyclotron. Amyloid radiotracers using 18F have been developed and
include 18F-florbetapir (Blennow et al. 2015). Amyloid radiotracers have been validated
by comparing the analyzed images with postmortem neuropathological assessment. For
20
example, in a prospective study of 59 terminally ill patients, 18F-florbetapir was been
demonstrated to have high sensitivity and specificity for the detection of amyloid plaques
identified on neuropathological exam 1 and 2 years after image acquisition (Clark et al.
2012).
Amyloid PET has had a “transformative impact on AD research” Rabinovici (2015),
enabling in vivo identification, localization and quantification of amyloid, which was
previously only possible post-mortem. Though not required for diagnosis of AD, nor
recommended for routine clinical use, amyloid PET has also proven valuable in certain
clinical settings involving atypical presentations and diagnostic uncertainty (Johnson et
al. 2013; Laforce et al. 2016). Our group has previously shown that in cases of dementia
where there can often be diagnostic uncertainty, such as in late-onset frontotemporal
dementia, PIB-PET can differentiate such cases from AD (E. E. Brown, Graff-Guerrero,
et al. 2016).
1.4.2.2.1 Estimating amyloid burden from amyloid PET data Both categor-
ical and quantitative measures of amyloid plaques can be derived from amyloid PET
scans. In PET imaging analysis, to interpret and compare the activity values (e.g. in
kBq) from the time activity curves, it is adjusted for factors that directly impact the
measured activity: the participant’s body weight and the tracer dose injected. The con-
verted value is termed the standardized uptake value (SUV). A commonly used method
in amyloid PET for both quantitative and categorical analyses is the calculation of a
SUV ratio (SUVR), which is the ratio of the SUV in a region of interest for a specified
time window to the that of a reference region in another time window. As a ratio, it au-
tomatically accounts for factors common to both the numerator and denominator, such
as the dose of the radiotracer administered and the patient’s weight (McNamee et al.
2009). Thus, weight and dose are not needed to calculate SUVR, so with C representing
activity concentration, and t representing the time window, SUVR can be expressed as:
SUV R = SUVtarget
SUVreference
= Cttarget
Ctreference
21
Using the SUVR in place of kinetic analyses has major advantages: it does not
require arterial blood sampling, an invasive process, during PET image acquisition; it
only requires PET image acquisition during a relatively brief time frame after a delay
from tracer injection (e.g. a 15- to 20-minute window) rather than a full dynamic scan
starting from the moment of tracer injection (e.g. a 90-minute window), which is less
demanding on the participant who must be still for the duration of the scan. As a ratio,
using the SUVR relies on having a valid reference region in the brain in which little
amyloid is expected to be present. The use of SUVR with the cerebellum as reference
region has been validated with amyloid PET (e.g. Klunk et al. 2004; Clark et al. 2012).
An “optimal” post-injection time window for SUVR calculation depends on the
specific tracer and dose. For PIB, the most commonly used window is reported to be 50-
70 minutes (Klunk et al. 2015), which has been demonstrated to most reliably correlate
with measures using the full dynamic data (McNamee et al. 2009). However, windows
of 20- to 30-minute durations starting from between 40-70 minutes post-injection have
been used. The 60- to 90-minute window has been shown to be affected by changes in
cerebral blood flow (CBF), unlike kinetic methods calculated from the full dynamic data
(van Berckel et al. 2013). As flow is known to decrease in AD, kinetic models may be
preferred over SUVR in longitudinal studies (van Berckel et al. 2013), although SUVR,
with a relatively large effect size, has been recommended and used for this purpose (e.g
Lopresti et al. (2005); Villemagne et al. (2011)).
When the full dynamic data are obtained (time activity curve starting at the point
of injection), a kinetic model can be used that may more accurately amyloid burden, as
SUVR has the potential to overestimate binding, has greater variability, and as men-
tioned above, can be sensitive to changes in blood flow as seen in AD (van Berckel et
al. 2013). Simultaneous measurement of arterial tracer concentration is considered the
gold standard approach (McNamee et al. 2009), but non-invasive methods using the
cerebellum reference region are also accurate (Lopresti et al. 2005). A commonly used
method of estimating amyloid binding burden from dynamic data is the Logan graphical
analysis method of calculating a distribution volume ratio (DVR) (Logan et al. 1996),
22
see 4.3.2.2 for more details.
An alternative to the SUVR method of quantifying Aβ burden that accounts for
cerbebral blood flow is the Washout Allometric Reference Method (WARM), which the
authors showed had better discriminatory power with respect to diagnostic groups, al-
though the model has not been neuropathologically confirmed (Rodell et al. 2013; Rodell
et al. 2017).
Efforts have also been made to standardize the reporting of PIB PET amyloid
burden, to theoretically improve the comparison of PIB PET SUVR values across sites
using different scanners. The Centiloid project proposes scaling PIB PET SUVR by
anchoring prototypical amyloid-negative scans and amyloid-positive scans on a fixed scale
(Klunk et al. 2015).
1.4.2.2.2 Dual Biomarker As, in principle, PIB binds to Aβ and washes out of
other brain tissue over time, measures of amyloid burden including SUVR and the non-
invasive Logan model rely on late-scan activty data. As noted above, while PIB SUVR
measures in the late 40- to 90-minute range reflect cerebral Aβ burden, SUVR has been
shown in some cases to be impacted by changes in relative tracer delivery R1 (van Berckel
et al. 2013). However, the influence of CBF on PIB PET uptake has been proposed as
an advantage, as decreased CBF is an important marker of neuronal activity associated
with AD progression (see 1.4.2.1). In contrast to late scan SUVR, the tracer uptake
reflected in the early-scan activity curve, is not associated with Aβ binding, but has
been demonstrated to reflect CBF. As a result, PIB PET has been described as a “dual
biomarker” (Meyer et al. 2011).
With arterial monitoring in a study of PIB in rhesus monkeys, the unidirectional
influx constant of the tracer (K1), was shown to be associated with CBF as measured with15O-water PET, as the lipophilic tracer can passively diffuse across the blood brain barrier
(Blomquist et al. 2008). This finding was subsequently demonstrated in humans, where
it was also shown that permeability-surface area products are similar in AD and healthy
23
controls (HC), supporting the hypothesis that PIB influx may reflect CBF (Gjedde et al.
2013).
Whereas determination of K1 requires arterial sampling, the relative influx constant
R1, equal to the ratio of the influx K1 of a target region to a reference region, can
be determined from dynamic PET data alone, using methods such as the Simplified
Reference Tissue Model (SRTM) (Lammertsma & Hume 1996) and a related constrained
version, SRTM2 (Wu & Carson 2002). The relative delivery constant R1 is reduced in
AD compared to MCI and HC (van Berckel et al. 2013), and like K1, is correlated with
CBF, having been shown to strongly correlate with FDG PET (Meyer et al. 2011). Using15O-water PET as a marker of CBF, Chen et al. (2015) demonstrated that PIB R1 is
strongly associated with the relative uptake of 15O-water.
Early PIB activity relative the cerebellum was shown to be comparable to FDG
PET data (Rostomian et al. 2011), and another small study demonstrated an association
of early SUVR (the ratio of the first 6 minutes of activity in the cortical regions of interest
to the same period in the cerebellum reference, which they authors named ePIB) with
relative glucose delivery using FDG PET, and with PIB PET K1 (Forsberg et al. 2012).
Similarly, early PIB activity, in combination with late PIB activity reflecting amyloid
load, improved the diagnostic classification of HC, MCI and AD, supporting the dual-
biomarker role of PIB (Fu et al. 2014).
Rodriguez-Vietez et al. replicated the finding that early PIB activity and PIB SRTM
R1 are associated with glucose metabolism as measured by FDG PET (Rodriguez-Vieitez
et al. 2016; Rodriguez-Vieitez et al. 2017), as did Peretti et al. more recently (Peretti
et al. 2019). Ponto et al. replicated the finding that PIB R1 is associated with FDG
glucose metabolism, and also showed that the peak PIB SUV value (which occurs early
in the time-activity curve) is as well (Ponto et al. 2019).
Multiple groups have thus confirmed that PIB PET data can be analyzed to obtain
a surrogate marker of CBF, a marker of brain metabolism that is important in AD and
related research.
24
1.4.2.3 Tau PET
Even prior to the use of PIB in humans, FDDNP was developed, a radiotracer which binds
both Aβ plaques and tau neurofibrillary tangles (Agdeppa et al. 2001). Unlike PIB, the
binding of FDDNP correlates with tau as measured in the CSF (Tolboom et al. 2009).
As reviewed by Villemagne & Okamura (2014), relative to imaging Aβ placques, tau
PET radiotracer development is complicated by most tau aggregates being intracellular,
the occurence of multiple isoforms, a wide variety of posttranslational modifications that
occur to tau, overlap in the form of tau aggregates and Aβ sheets, and the relatively
lower concentration of tau compared to Aβ in AD. However, more recently, tau-specific
radiotracers have been developed. For example, flortaucipir has affinity for tau aggregates
seen in AD and is able to distinguish AD from clinical controls (Ossenkoppele et al. 2018).
Thus, there are now options for in vivo tau imaging, which may be a boon to research,
adding information about disease staging which compliments amyloid PET, which as
described above, reflects amyloid accumulation that is known to occur well before the
onset of clinical symptoms. For example, another tau radiotracer, T807, has been used
with amyloid imaging to evaluate distinct pathological topographies in MCI and AD
(Brier et al. 2016).
1.4.3 Structural Magnetic Resonance Imaging
Magnetic resonance imaging is an imaging modality that uses a strong magnetic field to
align protons in the body, a gradient magnetic field which affects the resonant frequency
of protons depending on their location within the gradient, pulses of radiofrequencies
to influence the nuclear spin of the protons and cause them to emit a radiofrequency
signal, and receiver coils to measure the emitted signal’s location (Berger 2002). Dif-
ferent pulse sequences can be used to measure different aspects of biological tissues, in
particular on the basis of their fat and water content (Berger 2002). Therefore, MRI can
be used to identify brain structures including grey matter, white matter, ventricles, and
pathological features including tumours. The common sequences for the evaluation of
25
brain structure include T1-weighted, T2-weighed, and fluid-attenuated inversion recovery
(FLAIR) (Wardlaw et al. 2013).
In AD, structural MRI can be used to visualize the atrophy reflective of neurodegen-
eration. Severe atrophy can be visually identified. More subtle changes can be analyzed
quantitatively with image analysis pipelines to extract volumetric information by region
of interest or globally. For example, reductions in hippocampal volume and the thickness
of entorhinal cortex and supramarginal gyrus are observed in MCI and even more so in
AD, and these measures together can discriminate MCI from HC (area under the curve
0.91-0.95) and AD from HC (area under the curve 1.0) (Desikan et al. 2009). While
total hippocampal volume is a useful AD biomarker, more granular measurements of
subregions of the hippocampus and related structures with high-resolution T2-MRI have
revealed patterns similar to the progression of tau described by Braak & Braak (1995)
and that regions CA1 and BA35 may be best able to discriminate MCI from HC (Wolk
et al. 2017).
1.4.3.1 Vascular-related Features on Magnetic Resonance Imag-
ing
In addition to quantifying global and regional brain volume, structural MRI has been used
to identify pathological changes associated with small vessel disease, which, as reviewed in
1.2.4, plays an important role in the pathophysiology of dementia including AD. Wardlaw
et al. (2013) have reviewed the MRI features of cerebral small vessel disease in the
context of the neurodegeneration of dementia and highlight the role of these features in
AD, noting that “the clinical differentiation of Alzheimer’s disease from vascular cognitive
impairment or vascular dementia is increasingly recognized to be blurred” (Wardlaw et al.
2013). The consensus group identified the major categories of neuroimaging features of
small vessel disease: recent small subcortical infarct, lacune of presumed vascular origin,
white matter hyperintensity (WMH) of presumed vascular origin, perivascular space, and
cerebral microbleed. Among these, WMH, associated with aging, cerebrovascular risk
26
factors common to AD, and cognition, are of particular interest (Wardlaw et al. 2013).
So-called WMH are named for their characteristic appearance with increased
(i.e. hyperintense) signal on T2 and FLAIR MRI sequences, but they may also appear
hypointense on T1 sequences (Wardlaw et al. 2013). Several other terms have been used
in the literature,1 with overlap from the white matter changes that appear hypodense
on computed tomography (CT) that were described even before MRI that were termed
leukoaraiosis (Prins & Scheltens 2015). While the risk factors for WMH on MRI have
been characterized and include age, hypertension, smoking, diabetes, and cardiac disease
(Fazekas et al. 1988; Habes et al. 2016), the underlying pathophysiology is still debated
(Wardlaw et al. 2013; Duering et al. 2013).
Early pathologic studies identified ischemic tissue damage of varying severity and
etiology (Fazekas et al. 1993). Possible etiologies include chronic hypoperfusion and
ischemic strokes similar to lacunar infarcts (Potter Gillian M. et al. 2010; Duering et
al. 2013). Using arterial spin labelling, an MRI method that is used to measure CBF,
brain regions with WMH were shown to be associated with lower CBF, suggesting a
relationship between WMH and brain perfusion among healthy adults (Brickman et al.
2009; Bahrani et al. 2017).
Distinctintion is often made between deep and periventricular WMH, although the
two phenomena are strongly correlated (DeCarli Charles et al. 2005), and white matter
disease is observed in both regions on neuropathological examination in AD (Englund
1998).
The clinical importance of WMH has been confirmed by a systematic review of
46 longitudinal studies that explored the association of WMH with the onset of disease,
using meta-analytic techniques, which demonstrated that WMH are associated with an
increased risk of stroke, cognitive decline, dementia, and death (Debette & Markus 2010).
Of the reviewed studies, three specifically looked at the risk of AD and could be included
in a meta-analysis, which demonstrated an increased risk (Debette & Markus 2010). In1Other names for WMH include leukoaraiosis, white matter lesions, white matter changes, leukoen-
cephalopathy, (ischemic) white matter disease, and white matter damage (Wardlaw et al. 2013).
27
this study, the effect was driven by a large population study but was not observed in the
two smaller studies of high-risk groups, and the authors hypothesized that WMH may
increase the risk of MCI but not progression to AD. However, subsequently it has been
reported that WMH in the context of MCI predicts both the onset of AD (Provenzano et
al. 2013) and rapid cognitive decline (Tosto et al. 2014). The cognitive profile associated
with WMH appears distinct from that associated with Aβ (Hedden et al. 2012).
Further, there has been additional evidence supporting an increased risk of AD
with WMH from a prospective longitudinal study (Brickman et al. 2012), and again
more recently in the prospective Framingham Offspring cohort study, baseline WMH were
demonstrated to be associated with MCI compared to cognitively intact controls cross-
sectionally, and also prospectively with the development of MCI from normal cognition
in a 6.5 year follow up period (Bangen et al. 2018).
The association of WMH with deficits in specific cognitive domains has been clar-
ified by a recent meta-analysis of cross-sectional studies in MCI and AD, which demon-
strated associations with overall cognition, attention, executive function, and processing
speed, but negative associations with memory, visuoconstruction, and language were also
observed (van den Berg et al. 2018). Among 142 participants with AD in another study,
70% had some degree of WMH on MRI, and the presence of WMH was associated with
worse cognitive and functional impairment (Heo et al. 2009).
Further complicating the relationship between WMH and AD is the potential for
moderating variables. For example, MDD is associated with WMH and cognitive impair-
ment (Park et al. 2018), which highlights the importance of assessing known risk factors
as potential moderators in studies of AD pathophysiology.
28
1.4.3.2 White Matter Integrity on Magnetic Resonance Imag-
ing
Diffusion-weighted MRI (DWI), or diffusion tensor imaging, is an MRI sequence that
measures the diffusion of water molecules, which can be used to image and assess the
orientation and integrity white matter (Sexton et al. 2011; Amlien & Fjell 2014). Com-
monly used measures calculated from DWI data include fractional anisotropy (FA), which
reflects the directionality of diffusion within a tissue or region, and mean diffusivity (MD),
a measure of total diffusion in a voxel (Amlien & Fjell 2014).
As the role of white matter disease in AD is increasingly recognized, DWI has been
used as an in vivo biomarker of white matter integrity (Amlien & Fjell 2014). Amlien &
Fjell (2014) reviewed the literature on diffusion tensor imaging in AD, including studies
that found DWI added predictive value to CSF biomarkers in the progression of MCI
to AD. In a meta-analysis of case-control studies of AD and MCI that used DWI by
Sexton et al. (2011), pooled data from 617 participants with AD and 915 controls
revealed reductions in FA in most regions studied, with parietal and internal capsule
white matter excepted, and MD was reduced in all regions (Sexton et al. 2011). In a
regression analysis with cognition, a greater effect size of FA in parietal white matter was
associated with reductions in cognition, and for MD, the regression was not significant
(Sexton et al. 2011).
Measures of white matter integrity from DWI are associated with WMH (Chao et
al. 2013), but may detect more subtle changes than WMH on FLAIR, which may reflect
more severe white matter damage (Prins & Scheltens 2015). For example, both DWI FA
and FLAIR WMH predict further development of WMH (Maillard et al. 2013). Thus,
DWI measures such as FA and MD may add to FLAIR measures of WMH in assessing
the spectrum and severity of white matter integrity (Maillard et al. 2013).
29
1.4.4 Reproducibility and Open Science in Neuroimaging
In a methodological discussion of neuroimaging biomarkers, it has become important to
discuss issues of reproducibility. A current “reproducibility crisis”–the idea that a signif-
icant proportion of the scientific literature is not reproducible, i.e. false–is increasingly
discussed and debated (Fanelli 2018). The psychological sciences, and neuroimaging re-
search, have specifically been criticized for low rates of reproducibility for reasons includ-
ing low statistical power, flexible data analysis protocols, and software errors (Poldrack
et al. 2017).
With respect to software errors, several solutions were suggested, including to use
large established software projects where available, to “learn and use good programming
practices, including the judicious use of software testing and validation”, and to make
any code publicly available.
Although the review by Poldrack et al. (2017) focuses on functional MRI, the same
issues are faced by PET/MRI methods applicable to AD research discussed here. Anal-
ysis pipelines for PET/MRI studies require many software steps including image format
conversion, orientation and cropping, motion correction and quality control, MRI segmen-
tation, normalization, atlas parcellation and region of interest delineation, PET–MRI
coregistration, and partial volume correction. Closed-source commercial pipelines are
widely used, and recently fully open-source automated pipelines have become available,
which incorporate more established open-source projects (Funck et al. 2018; Karjalainen
et al. 2019).
The result of PET image analysis is an averaged time-activity curve for a region of
interest. Subsequently, statistics such as the SUVR or kinetic models are calculated (for
example, as described in 1.4.2.2.1 above). These subsequent analyses may be included in
an analysis pipeline, manually calculated, or calculated with existing analysis software
libraries. Often, methods are not fully described and certain aspects, even relatively
simple steps such as averaging data from multiple regions of interest, could be impacted
by how or whether weighting was applied, for example.
30
Recent concerted efforts to address reproducibility and open protocols in neuroimag-
ing include the development of a standardized data structure for neuroimaging experi-
ments, the brain imaging data structure (BIDS) (Gorgolewski et al. 2016). The BIDS
initiative has been adopted by major image repositories and is supported by MRI analysis
pipelines. An initiative bringing formal peer review to software and stastical packages is
rOpenSci, whose mission statement is to “[foster] a culture that values open and repro-
ducible research using shared data and reusable software” (rOpenSci 2019).
1.4.5 Lead Measurement
There are different methods to measure Pb exposure and accumulation in humans as Pb
is distributed in blood, soft tissue and bone, and the relative merits of Pb biomarkers
have been reviewed by Barbosa et al. (2005).
The most widely used method is direct blood Pb measurement, usually by mass
spectrometry, which reflects recent exposure and which is influenced by other factors
including age, sex, and nutritional status, but which may also represent past exposure,
with a portion of blood Pb resulting from release from longer term bone stores, which
can be higher in children or individuals with low exposure (Barbosa et al. 2005). The
use of blood Pb as a biomarker of Pb exposure is problematic for these reasons. Serum,
rather than blood Pb may be preferred but is prone to contamination and is therefore
technically challenging to obtain and validate (Barbosa et al. 2005).
Direct measures of bone Pb are reflective of past and chronic exposure, given the
long half-life of Pb in the bone on the order of decades (with wide range estimates
published) (McNeill et al. 2017). Bone Pb can also be measured in vivo, with calcaneal
and tibial levels estimated with x-ray fluorescence (XRF) (McNeill et al. 2017).
As reviewed by Barbosa et al. (2005), other forms of Pb measurement have been
used but are problematic. Saliva levels have been used but may be low and unreliable.
Hair levels are vulnerable to external contamination and confounding factors. Nail levels
31
may reflect chronic exposure, but high variability even among different nails from the
same subject has been reported. Tooth Pb can be useful given the long-term accumulation
of Pb in teeth, and is promising in that it may provide details about timing of past
exposure across a lifespan, though confounds and methodologic limitations are present
(Barbosa et al. 2005).
With bone (and to a smaller and less predictable extent, blood) Pb measurement
reflecting chronic exposure, cross-sectional studies to assess the association of past Pb
exposure on present symptoms and diagnoses are possible. Blood Pb may be more useful
when multiple measurements are obtained in prospective studies, to allow the onset of
symptoms or disease to be correlated with past exposure. In the case of AD, where clinical
symptoms emerge long after the initiation and appearance of biological changes, early
life exposure or chronic lifetime exposure may be essential to evaluate any meaningful
association with Pb.
1.5 A Modern Model
This literature review does cover all aspects and theories about the pathophysiology of
AD, and is necessarily focused on the main theories of AD pathogenesis and aspects
relevant to this thesis work. Additional important theories under investigation have
been reviewed elsewhere, and include the shared pathophysiology with insulin resistance
and diabetes (AD as “type 3 diabetes”) (Kandimalla et al. 2017), a potential more
central role for GSK-3 (Takashima 2006), the role of prion-like spreading of amyloid and
tau (Goedert 2015), and a central role of the immune system (Heppner et al. 2015).
Similarly, a coverage of established and proposed medication interventions, including
the cholinesterase inhibitors, lithium, and anti-amyloid therapies (Panza et al. 2019) is
outside the scope of this thesis.
Nonetheless, with the summary of pathological features of AD, the etiological fac-
tors involved, and the methods of in vivo biomarker investigation above, we can turn to
32
the efforts to incorporate these diverse fields of investigation. While the pathophysiology
of AD likely cannot be reduced to a linear model, as simultaneous processes and feed-
back loops are likely involved (Selkoe et al. 2012), a model by Jack et al. incorporates
recent biomarker studies reflects longitudinal observations of the presumed underlying
pathological changes (Jack et al. 2010; Jack et al. 2013).
By carefully examining the longitudinal change biomarkers and the relationship
among different biomarkers cross-sectionally, a temporal sequence of biological changes
can be described, providing a framework upon which the known risk factors of AD can
be evaluated to clarify their role and impact.
In the conceptual model, changes in biomarkers from normal to abnormal are plot-
ted over time, reflecting disease progression. Jack et al. (2013) illustrate the model in a
figure reproduced here as Figure 1.1.
www.thelancet.com/neurology Vol 12 February 2013 211
Personal View
incident clinical diagnosis (of either dementia or mild cognitive impairment) as the anchoring event and then compared biomarker and clinical trajectories as a function of time relative to this event. The DIAN study78 has used the age of onset of dementia in the aff ected parent as the temporal anchor for mutation carriers, thus permitting estimates of the longitudinal behaviour of clinical and biomarker metrics from cross-sectional data.
Donohue and colleagues79 modelled long-term bio marker trends from short-term within-subject data in the ADNI using shape invariant modelling that places time on the horizontal axis. These methods can model subject-specifi c rescaling and shifting of time in datasets with no specifi c common anchoring event such as incident dementia.
Jedynak and colleagues80 and Mungas and colleagues81
focused on novel composite horizontal axes that capture the latent trait descriptors of underlying AD patho-physiological processes by combining several biomarkers in a non-linear way; the horizontal axis represents the entire disease range with one horizontal-axis metric that can be thought of as a latent trait.81 Each biomarker contributes to this single latent trait metric, with greater weighting where it is most dynamic in the disease range.
Model revisionFigure 5 is a revised version of our original 2010 model (fi gure 1) that incorporates new fi ndings and also addresses some of the shortcomings described in the preceding paragraphs. Although our revised model has many similarities with our 2010 model, diff erences do exist. Firstly, the horizontal axis in our revised model is expressed as time, not clinical disease stage. The absolute time in years needed to traverse the disease pathway from left to right and the specifi c age at which a person enters the disease pathway will vary among individuals.
A range of possible cognitive outcomes is shown at given positions along the horizontal axis. This refl ects the fact that each individual responds to AD pathophysiological changes uniquely.73,74 People with a high risk of cognitive impairment due to AD patho physiological processes have a cognitive impairment curve that is shifted to the left in time. Such high-risk individuals might harbour more genetic risk alleles, have low cognitive reserve, pursue lifestyles that increase the likelihood for cognitive impair-ment, or have other comorbid brain pathological changes. By contrast, low-risk individuals with a protective genetic profi le, high cognitive reserve, no comorbid brain pathological changes, and low lifestyle risks for dementia can coexist with substantial AD pathophysiology and still maintain normal cognitive function. Thus cognitive impairment in fi gure 5 is shown as a zone with low-risk and high-risk borders.
The revised model includes modifi cations of the specifi c ordering of some biomarkers on the basis of reports described above. CSF Aβ42 is now positioned slightly before amyloid PET, which is followed by CSF tau. FDG PET and MRI are drawn coincidentally as the last biomarkers to
become abnormal, but those that track most closely with progressive cognitive impairment.
All biomarkers are still confi gured as sigmoids, but the shapes of the sigmoid curves are no longer identical. The curves have a progressively steeper slope in the right-hand tail for later-changing biomarkers. Finally, the biomarker curves are drawn closer together in the revised model, indicating less distinct temporal separation.
Autopsy evidence that tau pathophysiology can precede Aβ depositionOurs is a model of the temporal evolution of AD bio-markers in relation to each other and to the progression of clinical symptoms. Although biomarkers do refl ect
Max
Min
Biom
arke
r abn
orm
ality
CSF Aβ42
Amyloid PETCSF tauMRI + FDG PETCognitive impairment
MCI
DementiaLo
w riskHigh ris
k
Normal
Max
Min
Biom
arke
r abn
orm
ality
Time
MCI
Dementia
Low ris
kHigh risk
Normal
T
A
B
Figure 5: Revised model of dynamic biomarkers of the Alzheimer’s disease pathological cascade(A and B) Neurodegeneration is measured by FDG PET and structural MRI, which are drawn concordantly (dark blue). By defi nition, all curves converge at the top right-hand corner of the plot, the point of maximum abnormality. Cognitive impairment is illustrated as a zone (light green-fi lled area) with low-risk and high-risk borders. (B) Operational use of the model. The vertical black line denotes a given time (T). Projection of the intersection of time T with the biomarker curves to the left vertical axis (horizontal dashed arrows) gives values of each biomarker at time T, with the lead biomarker (CSF Aβ42) being most abnormal at any given time in the progression of the disease. People who are at high risk of cognitive impairment due to Alzheimer’s disease pathophysiology are shown with a cognitive impairment curve that is shifted to the left. By contrast, the cognitive impairment curve is shifted to the right in people with a protective genetic profi le, high cognitive reserve, and the absence of comorbid pathological changes in the brain, showing that two patients with the same biomarker profi le (at time T) can have diff erent cognitive outcomes (denoted by grey circles at the intersection of time T). Aβ=amyloid β. FDG=fl uorodeoxyglucose. MCI=mild cognitive impairment.
Figure 1.1: Conceptual model of biomarker changes in AD from Jack et al. (2013)
The earliest detectable abnormalities, occurring up to decades prior to the onset
of clinical symptoms, are CSF Aβ42 and cerebral Aβ detected by amyloid PET. Tau
changes, detected by CSF (with tau PET not yet incorporated into the model), are
intermediate between amyloid and clinical symptoms. Biomarkers that reflect neurode-
generation, such as structural MRI and FDG PET are most closely associated with the
emergence of clinical symptoms, which follow. The specific delay until the emergence of
33
clinical symptoms is thought to be influenced by an individual’s risk profile. All along,
each biomarker progresses in severity, in a presumed sigmoidal manner from undetectable
to maximally abnormal.
As a model based on in vivo biomarkers rather than histopathology, it may not
capture more subtle changes that in vivo methods do not capture, such as the observation
that tau deposition occurs very early in life and may reflect an initiation of an AD-related
cascade, though alternatively may not be related to AD, or may reflect independent but
related processes (Jack et al. 2013).
The model by Jack et al. (2013) is useful to place current and future research into
risk factors, biomarkers and pathophysiological mechanisms into context. For example,
and to relate it to the themes of this thesis, in considering the role of vascular risk factors,
changes in CBF, and white matter integrity it is helpful to interpret in the context of
the known evolving biomarkers of AD, to hypothesize on how each process that the
biomarkers represent may impact each other. Turning to Pb, which has biological insults
throughout the lifespan, hypotheses on its potential impact can be conceptualized to
involve alterations in the onset, slope, or relative timing of the various biomarker curves.
34
Chapter 2
Hypotheses and Research Aims
The overarching focus of this thesis is to explore novel approaches in understanding the
complex interplay of pathophysiologic processes in AD. Having outlined the definition,
pathophysiology, modern biomarkers and research methodology, and a model of AD in
the preceding chapter, the following four chapters report recent research that adds to
this vast literature.
Chapter 3 presents a methodologic study using MRI and PIB PET. As reviewed
above in section 1.4.3.1 and 1.4.3.2, WMH and FA are clinically important biomarkers
of white matter integrity and have a role in AD specifically, in that they can predict AD
onset and be associated with severity. In section 1.4.2.2.2, the association of PIB PET
uptake measures with cerebral blood flow is reviewed. Given that cerebral blood flow and
white matter integrity are associated, with chronic hypoperfusion one hypothesized mech-
anism that leads to WMH, we hypothesize that PIB markers that reflect cerebral blood
flow, when evaluated in white matter, would be associated with established biomarkers of
white matter integrity, namely WMH volume and reductions in FA. Such a finding could
add value to PIB PET, an expensive but widely used technique in AD research. Mark-
ers of cerebral blood flow, vascular changes and white matter integrity are particularly
useful in the exploration of AD risk factors that may contribute to AD pathophysiology
via these effects, such as MDD, and also Pb, as reviewed in this thesis.
35
In developing the analysis used for Chapter 3, a gap in available open source tools
became apparent, with respect to the analysis of PET time activity curves. With the
principles proposed by Poldrack et al. (2017) as reviewed in 1.4.4, i.e. publishing analy-
sis code openly, using good programming practices including inclusion of software tests,
Chapter 4 presents an open source R (R Core Team 2018) package that was used for the
analysis in Chapter 3, but which also aims to be useful more broadly to others. The pack-
age provides tools to load time-activity curve data from multiple formats and pipelines
for further analysis, including weighted merging of regions of interest, calculation of
SUVR, the Logan non-invasive graphical analysis method of DVR calculation, plotting
of time-activity curves, and a method cut-off calculation used to establish PIB-positivity.
While Chapters 3 and 4 aim to present methodological advancements useful in AD
pathophysiology research, Chapters 5 and 6 use systematic reviews and meta-analytic
techniques to pool and analyze available literature and generate hypotheses in the area
of the role of Pb in AD.
Chapter 5 presents systematic reviews and meta-analytic methods to analyze case-
control studies of AD in which Pb was compared. As reviewed in section 1.3.4, Pb has
been identified as a known neurotoxic agent that has widely contaminated the environ-
ment in the past decades, with animal studies directly linking it to AD-related patho-
logical changes, the aim of the study presented in Chapter 5 was to identify the current
evidence in humans linking Pb to AD (or not) and to identify what next steps may be
useful in clarifying this relationship, in the context of the available Pb- and AD-related
biomarkers outlined in 1.4.
Chapter 6, unlike Chapters 3-4, draws on several disparate areas of literature to
lay out a hypothesis with respect to the impact of environmental exposures of the de-
velopment of disease. As reviewed in Chapters 1 and 5, environmental Pb exposure is
associated with problems associated with dementia and AD including cognitive deficits
and cerebrovascular disease. Lithium, which has been shown, when used as a medication,
to effectively reduce cognitive decline and the onset of AD in amnestic MCI (Forlenza et
al. 2019), has also been associated with lower rates of dementia in areas where drinking
36
water lithium concentrations are higher (Kessing et al. 2017a). Thus, on the face of it,
environmental lithium has opposite associations as environmental Pb. Chapter 6 pur-
sues this idea by summarizing the health impacts of Pb and systematically reviewing the
health conditions reported to be associated with drinking water Pb. By also reviewing
the known biological mechanisms of Pb and lithium, as well as available animal studies
that have experimentally tested for protective effects of lithium on Pb-mediated toxicity,
the chapter aims to explore the hypothesis that benefits of lithium in drinking water may
be due to a protection against the damaging effects of Pb exposure.
By aiming to maximize the useful information available in amyloid PET scans,
improving the openness of PET analysis via a well-tested R package, summarizing and
synthesizing the literature on the potential role of Pb in AD, and proposing a novel
mechanism by which environmental factors may interact to alter the risk of AD, the
broad aim of this thesis is to advance the understanding of AD, in concert with the work
of many others, hoping to one day improve the lives of individuals and families facing
this disabling and lethal disease.
37
Chapter 3
Amyloid Positron Emission
Tomography as a Biomarkers of
White Matter Integrity
This chapter is modified and reproduced from an article published in the Journal of
Neuroimaging (E. E. Brown, Rashidi-Ranjbar, et al. 2019). See the Contributions section
of this thesis for details.
3.1 Introduction
Alzheimer’s disease (AD) is a growing public health concern, with a high burden to
patients, their caregivers and society (Livingston et al. 2017). The pathophysiology
of AD involves a complex interplay between various factors including cerebral amyloid
plaques, neurofibrillary tangle accumulation, and vascular dysfunction (Jack et al. 2013).
Pittsburgh Compound B ([11C]-PIB), introduced in 2004, was the first positron
emission tomography (PET) radiotracer that enabled specific in vivo quantification of
cerebral amyloid (Klunk et al. 2004). The majority of individuals with AD are [11C]-PIB-
38
positive, having elevated cortical distribution volume ratios (DVR). In mild cognitive
impairment (MCI), [11C]-PIB-positivity is a significant predictor of progression to AD
(Villemagne et al. 2011). The presence of amyloid is also noted in approximately one
quarter of elderly HC, in whom it may also predict eventual progression to MCI and AD
(Villemagne et al. 2011). The ability to identify and quantify amyloid deposition in vivo
has been valuable in AD research, has been used in large coordinated research initiatives
(Chung et al. 2015), and also has diagnostic utility in clinical settings (Laforce et al.
2016).
In addition to quantifying amyloid burden, models of [11C]-PIB delivery into cere-
bral tissue may reflect individual differences in blood flow or glucose metabolism. For
example, early kinetic parameters, such as the unidirectional influx constant (K1), have
been shown to be correlated with CBF and glucose metabolism (Rodriguez-Vieitez et
al. 2016). Moreover, dynamic data fitted using the simplified reference tissue model
(SRTM) have shown that the kinetic parameter R1–the ratio of K1 in target versus a
reference tissue is also highly correlated with CBF and glucose metabolism (Blomquist
et al. 2008; Rodriguez-Vieitez et al. 2017; Meyer et al. 2011; Rodell et al. 2013; Chen
et al. 2015; Forsberg et al. 2012; Gjedde et al. 2013; Sojkova et al. 2015; Rodriguez-
Vieitez et al. 2016; Rodell et al. 2017). Associations of early [11C]-PIB standardized
uptake value (SUV) and R1 have recently been validated both with a measure of CBF,
[15O]-water PET and cerebral glucose metabolism measured with fludeoxyglucose PET
(Ponto et al. 2019; Peretti et al. 2019). Thus, [11C]-PIB-PET has been described as
a “dual biomarker”, with late post-injection data predominantly providing information
about amyloid deposition and early data reflecting CBF.
Magnetic resonance imaging (MRI) white matter hyperintensities (WMH) reflect
small-vessel cerebrovascular disease and are associated with an increased risk of stroke,
mild cognitive impairment, dementia and death (Bangen et al. 2018; Debette & Markus
2010). In MCI, WMH are associated with cognitive changes, and may be associated
with deficits in specific cognitive domains including attention, executive function and
processing speed (Bangen et al. 2018; van den Berg et al. 2018). Deep WMH have been
39
associated with an increased risk of progression from MCI to AD (Campbell et al. 2013).
Other dementia risk factors such as depression may be mediated by WMH and are thus
important to consider in pathophysiologic studies (Park et al. 2018).
Diffusion-weighted MRI, specifically with measures of fractional anisotropy (FA),
is used to evaluate white matter structural integrity. Mean diffusivity (MD) also reflects
white matter integrity but with less specificity than FA (Bosch et al. 2012). Decreases in
tract-based and global FA are associated with aging and AD (Zhang et al. 2009; Bennett
et al. 2009).
More generally, cardiovascular risk factors are known to have an important role in
the complex pathophysiology of AD and may be an important target for intervention
(Livingston et al. 2017). For example, the Framingham cardiovascular risk factors (age,
hypertension, dyslipidemia, cigarette smoking) are associated with progression from MCI
to AD (Viticchi et al. 2017), and progression of AD (Viticchi et al. 2015). The cardio-
vascular risk factors are also associated with WMH (Habes et al. 2016). A recent large
randomized controlled trial suggests that intensive hypertension treatment may reduce
the onset of MCI, although no effect was detected for reducing the risk of AD (The
SPRINT MIND Investigators for the SPRINT Research Group et al. 2019).
Given the potential importance of WMH in the pathophysiology of dementia, it
would be valuable to know whether [11C]-PIB-PET, in addition to providing information
about amyloid deposition and CBF, is associated with white matter structural impair-
ment. Given the strong association of WMH with vascular risk factors and specifically the
association of CBF with WMH (Brickman et al. 2009; Bahrani et al. 2017), we hypoth-
esized that [11C]-PIB-PET, which can reflect CBF, may also provide information about
white matter integrity and pathology. In this exploratory study, we tested whether [11C]-
PIB markers of CBF in white matter regions of interest were associated with established
MRI measures of vascular and white matter structural impairment.
40
3.2 Methods
3.2.1 Participants
Participants for this study are a subset of the participants with MCI and/or remitted
major depressive disorder (MDD) in an ongoing randomized controlled trial that started
in 2015 called PACt-MD (Prevention of Alzheimer’s dementia with Cognitive remedi-
ation plus Transcranial direct current stimulation in Mild cognitive impairment and
Depression; ClinicalTrials.gov Identifier: NCT02386670). The study was approved by
the institutional review board, and all participants provided written informed consent.
The PET imaging was obtained in a subset of participants from this trial with separate
protocol and written informed consent.
As part of this randomized controlled trial, participants underwent a comprehensive
baseline clinical assessment including neurocognitive testing and neuroimaging. At base-
line, 173 participants completed a [11C]-PIB PET scan and an MRI for co-registration.
One participant was excluded due to a technical error with the PET scan. The PACt-MD
protocol specified the T2-weighted fluid-attenuated inversion recovery (FLAIR) sequence
to be optional, so it was only obtained on a subset of the participants who received PET
scans, depending on time availability on the day of the scan. We included all 34 partici-
pants who had a valid full dynamic [11C]-PIB PET, and all of T1-weighted, T2-weighted,
T2-weighted FLAIR, and diffusion-weighted MRI sequences.
For the purpose of this study assessing neuroimaging methods, we combined two
groups of participants from the larger trial: those with MCI alone, and those with MCI
as well as a history of MDD, as we aimed to test for an association between different
brain imaging measures within participants. Including both groups increases the external
validity of any finding with more confidence that the two measures are associated in het-
erogeneous disorders, which would increase the likelihood that any identified biomarkers
are useful. The larger group also maximizes statistical power. The combined group is
equivalent to recruiting a group of individuals with mild cognitive impairment where a
41
current major depressive episode is an exclusion criterion, but a history of major depres-
sive disorder is not.
The inclusion criteria for all participants included a willingness to provide informed
consent, the availability of a study partner in contact with the participant, the ability to
read and write in English, and a Montgomery–Åsberg Depression Rating Scale (MADRS)
score <= 10 (Montgomery & Asberg 1979). Specific to the “MCI alone” group, inclusion
criteria included being age 60 or older and DSM 5 criteria for mild neurocognitive disorder
(American Psychiatric Association 2013). Specific to the MDD group, inclusion criteria
included being age 65 or older; DSM 5 criteria for one or more major depressive episode
with an offset between 2 months and 5 years or an offset of 5 years or longer if at least
one episode was during the participant’s adult life and it received medical attention
(e.g., hospitalization; saw a psychiatrist or primary care physician; or treatment with an
antidepressant).
Exclusion criteria included a DSM 5 diagnosis of alcohol or other substance use dis-
order within 12 months or a positive urine drug screen for drugs of abuse; a lifetime DSM
5 diagnosis of schizophrenia, bipolar disorder, or obsessive compulsive disorder; a signifi-
cant neurological condition (e.g., stroke, seizure disorder, multiple sclerosis); a high risk
for suicide; taking anticonvulsants, and other psychotropic medication that cannot be
safely discontinued, except for the following if they have been prescribed at a stable dose
for at least 4 weeks: antidepressants; zopiclone; trazodone; a benzodiazepine; gabapentin
or pregabalin (if prescribed for chronic pain); the use of a cognitive enhancer (acetyl-
cholinesterase inhibitor or memantine) within 6 weeks; an unstable medical illness (e.g.,
uncontrolled diabetes mellitus or hypertension); a pacemaker or other metal implants
precluding safe use of transcranial direct current stimulation; if completing one PET
scan would exceed allowed annual radiation exposure (20 mSv); being pregnant or lactat-
ing or a positive urine pregnancy test before PET scan; unable or unlikely to follow the
PET study protocol; and claustrophobia. Exclusion criteria specific to the MCI group in-
cluded DSM 5 criteria for MDD in the past 10 yrs. Specific to the MDD group, exclusion
criteria included DSM 5 criteria for Major Neurocognitive Disorder (“dementia”), and
42
having had electroconvulsive therapy within 6 months of the baseline neuropsychological
testing.
3.2.2 Clinical characterization and image acquisition
All PACt-MD participants had detailed clinical assessment including measurement of
vital signs and neuropsychiatric evaluation. The baseline diagnoses were confirmed in
a clinical consensus conference, and details of the clinical assessments and consensus
diagnosis procedures have been reported previously (Fischer et al. 2019). While cognitive
measures were obtained for the larger trial, the purpose of this study was to test for an
association among neuroimaging markers, rather than any association with diagnosis or
cognitive deficits.
We used a high-resolution research tomograph (HRRT) PET scanner to acquire
full dynamic [11C]-PIB PET images from time of injection to 90 minutes in frames of the
following duration: 5 x 1 minute, 20 x 2 minutes and 9 x 5 minutes (FORE & 2D Filtered
Back Projection reonstruction; with correction for attenuation and scatter). The mean
tracer dose was 9.8 (+/- 0.7) mCi delivered as a bolus. On a separate visit, we used
a 3-Tesla General Electric Echospeed MRI scanner to obtain T1-weighted, T2-weighted,
T2-weighted FLAIR, and diffusion-weighted sequences. We used the same scanners for
all participants.
3.2.3 Image analysis
3.2.3.1 Amyloid PET
To analyze the dynamic [11C]-PIB-PET image, we used PMOD version 3.807 with
PNEURO and used the Hammers-N30R83 atlas (Hammers et al. 2003). We loaded the
dynamic [11C]-PIB PET scan in atlas orientation, removed the first frame of noise while
retaining all post-injection data frames, and cropped the image while retaining data for
43
the whole brain. We used the structural T1-weighted MRI image for co-registration and
atlas parcellation.
We analyzed each participant individually with visual inspection for accuracy at
each step and using parameters based on the developer’s recommendation for amyloid
studies. Exceptions to the protocol were recorded as needed to improve accuracy. The
anatomical scans were segmented, normalized and co-registered to the PET image. The
normalized and matched PET-MR was segmented and transformed into atlas space.
We applied partial volume correction (PVC) to the PET data using a standard
MRI-based approach implemented in PMOD using the white-grey matter segmentation
(Müller-Gärtner et al. 1992). We used the PVC-corrected data for subsequent analyses
given our particular interest in the tracer delivery in the white matter over grey matter
region of interest, as without PVC, spillover effects could theoretically confound the
results.
For each region of the Hammers-N30R83 atlas, we saved regional average time-
activity curve data. We used R (v 3.5; R Core Team; Vienna, Austria) to analyze
time-activity curve data and perform subsequent statistical tests. For processing of time-
activity curve data, we used the open R package tacmagic version 0.1.9 (Brown 2019),
which combines atlas regions of interest (ROIs) into larger ROIs weighted by volume. We
created two ROIs: a cortical composite region, which included frontal, temporal, parietal,
cingulate, precuneus cortical regions; and a total white matter ROI.
3.2.3.1.1 Primary PET outcome measures To measure amyloid burden, we cal-
culated the SUV ratio (SUVR) by taking the mean activity of 4 frames spanning the
window from 50-70 minutes and dividing by the uptake in the cerebellar cortex as refer-
ence region. As surrogates for CBF, we used previously reported measures: the relative
delivery value R1, from the simplified reference tissue model (SRTM) model using the
cerebellum as reference region, and the early peak SUV (SUVmax or ePIB). We used the
tpcclib (bfmsrtm v. 0.6.17; Vesa Oikonen; Turku PET Centre, Finland) implementation
44
of the SRTM model. To calculate SUVmax, we identified the greatest activity value (kBq)
from the regional time-activity curve and divided by the product of the tracer dose (MBq)
and participant’s weight (kg). We calculated models on the white matter ROI to test
our primary hypothesis and on the cortical composite ROI for an exploratory analysis.
3.2.3.2 White matter hyperintensities
Measurement of WMH can be reported by total volume, total number of contiguous
lesions, or categorically by presence or absence. Quantification and classification can
be done manually or by semi- or fully-automated techniques. While manual delineation
of WMH by an expert neuroradiologist remains the gold standard, validated automatic
segmentation techniques increase objectivity and reproducibility. Visual rating scales are
less sensitive and less reliable than volumetric assessment (van Straaten et al. 2006).
We used four different published automated WMH segmentation algorithms that
each quantify total WMH volume (see below). We used the mean z-score from the four
algorithms as our measure of total WMH volume.
3.2.3.2.1 Lesion Segmentation Tool (LST) The first two segmentation algo-
rithms were from the LST toolbox 2.0.15 (Schmidt et al. 2012) in SPM 12 (v 7486; FIL
Methods Group, United Kingdom) running in MATLAB R2018b (Mathworks; Natick,
MA). We used both unsupervised algorithms available with the LST toolbox: “lesion
growth algorithm” (LGA), which uses a T1-weighted and T2-weighted FLAIR image,
and “lesion prediction algorithm” (LPA), which only uses the T2-weighted FLAIR image.
The LST toolbox algorithms output estimates of total lesion volume and number of
lesions and have been validated in different populations (Egger et al. 2017). The default
parameters were used for both algorithms.
The LGA algorithm automatically segments the T1-weighted image into cere-
brospinal fluid, grey matter and white matter and combines the co-registered T2-weighted
FLAIR image with the segmented information, generating lesion “belief maps” (Schmidt
45
et al. 2012). The algorithm uses a pre-specified threshold to generate a binary lesion
map, which then iteratively grows along hyperintense voxels from the T2-weighted
FLAIR image to create a lesion probability map (Schmidt et al. 2012). We estimated
the optimal initial threshold by repeating the analysis with a range of threshold values
(0.1, 0.2, 0.3, 0.4, 0.5) and visually comparing the results, which is the recommended
method when training data is not used (Schmidt et al. 2012).
3.2.3.2.2 Wisconsin White Matter Hyperintensities Segmentation Toolbox
(W2MHS) The third WMH algorithm was the W2MHS, version 2.1 (Ithapu et al.
2014), a MATLAB toolkit that uses machine learning, specifically support vector ma-
chines and random forest methods. The default parameters were used. This is an un-
supervised method that uses T1-weighted and T2-weighted FLAIR MRI sequences to
calculate effective volume for deep, periventricular, and total WMH. The toolbox has
been validated in other samples (Dadar, Pascoal, et al. 2017) and in a head-to-head
analysis validating 10 methods of automated white matter segmentation, the random
forest method performed best (Dadar, Maranzano, et al. 2017).
3.2.3.2.3 Automated monospectral optimal threshold intensity method The
fourth algorithm used was a fully automated technique using variable threshold intensity
method on T2-weighted FLAIR images and has been validated across different scanners
in a geriatric population (Yoo et al. 2014). The MATLAB script was provided by the
authors and depends on SPM8. The default parameters were used.
3.2.3.3 Diffusion tensor imaging (DTI)
We denoised the diffusion weighted images (DWI; 60 gradient directions with b = 1000,
5 baseline scans with b = 0) with the MRtrix3 denoise command (Tournier et al. 2012;
Veraart, Fieremans, et al. 2016; Veraart, Novikov, et al. 2016). We created a skull-
stripped magnitude image from the first b0 volume from each participant’s DWI dataset
46
(Jenkinson et al. 2012). Next, we used flirt (FSL) to register the previously skull-stripped
magnitude image to this b0 and generated a transformation matrix from the fieldmap
space to diffusion space (Jenkinson et al. 2002). This transformation matrix was applied
to the fieldmap to register it to the diffusion image. We corrected the DWI data for
motion and eddy current distortion with eddy (FSL) (Andersson & Sotiropoulos 2016),
and then used dtifit (FSL) to fit a tensor model and produce FA, mean diffusivity (MD),
V1 and sum of square errors (SSE) images (Behrens et al. 2003; Behrens et al. 2007).
We visually verified the corrected data and brain extraction and the colour encoding of
the primary eigenvector (V1) and the SSE map to ensure no imaging artefacts remained
following correction.
To calculate the mean white matter FA for each subject, we used fslmaths (-ero)
to erode the FA map (removing voxels around the exterior of the brain). Next, we
thresholded this eroded image between 0.2-0.8 to exclude grey matter and cerebrospinal
fluid and applied a binary threshold to produce a white matter mask from the FA image.
We calculated the total white matter mean FA value for each subject as the average
FA within this mask (fslstats -M). The mask was similarly applied to the MD image to
calculate the mean white matter MD.
3.2.4 Statistical tests
Our primary hypothesis was that measures of [11C]-PIB delivery thought to reflect CBF,
namely R1 and SUVmax, when applied to white matter regions of interest, are correlated
with measures of white matter pathology. Specifically, we hypothesized that [11C]-PIB R1
and SUVmax are negatively correlated with mean WMH volume and positively correlated
with mean global FA. To test these 4 relationships, we calculated the Pearson correlation
coefficient. We used an alpha level of 0.05 with two-sided tails to test for statistical
significance.
To address possibility of a false-positive result due to multiple comparisons, we ad-
justed the p-values of the 4 correlations of the primary hypothesis using the Benjamini-
47
Hochberg false discovery rate method as implemented in the base R stats package (Ben-
jamini & Hochberg 1995), which can be used in cases where there is positive regression
dependency.
As secondary analyses, to clarify whether the hypothesized relationship of [11C]-
PIB measures with white matter measures was limited to the white matter PET ROI,
we also tested for correlations using the cortical ROI. We tested for correlations of the
PET delivery markers with MD. Finally, to test whether the relationships were unique
to the CBF-related PET measures, we also tested for relationships with cortical SUVR,
a measure of global amyloid burden.
To explore the influence of potential confounding variables, we calculated the cor-
relation coefficient for the relationship between the PET and MRI measures, age, and
systolic and diastolic blood pressure. For potential confounding variables with signifi-
cant correlations with FA or WMH volume, we included the potential confound as a
third variable in partial correlations.
3.3 Results
3.3.1 Characterization of the sample
We included 34 elderly participants (19 male, 15 female) with mild cognitive impairment,
of which 9 also had remitted MDD. The mean age of the participants was 72.6 years.
Further characteristics of the sample are summarized in Table 3.1. The subset of partici-
pants who received T2-weighted FLAIR scans did not differ significantly in age, gender,
MMSE, MoCA or MADRS scores compared to the other participants. In determining
SUVmax, the frame with the maximal SUV was the 2nd (n = 9), 3rd (n = 11), 4th (n =
9), or 5th (n = 5) frame.
48
Table 3.1: Characterization of the sample.
Sample (n = 34)
Age (at PET scan) 72.6 +/- 6.8 years (60-88)
Sex (M:F) 19:15
History of MDD n = 9
CDR 0, n = 4; 0.5, n = 30
MMSE (n = 33) 27.5 ± 2.1 (21-30)
MoCA (n = 32) 24.2 ± 2.9 (16-29)
Days between PET and MRI 22.9 ± 18.1 (3-80)
Mean Total FA 0.367 ± 0.008
Mean Total MD 0.000809 ± 0.000028
Mean WM SUVmax 4.66 ± 1.03
Mean SRTM R1 1.08 ± 0.1
Values expressed as mean ± standard deviation (range) unless specified. FA = Frac-
tional Anisotropy; M/F = Male/Female; MDD = Major Depressive Disorder; MMSE =
Mini-Mental State Examination; MoCA = Montreal Cognitive Assessment; n = number
of participants; SRTM = simplified tissue reference model; SUV = standardized uptake
value.
3.3.2 White matter segmentation techniques
The total WMH volumes calculated by each of the four automated segmentation algo-
rithms were highly correlated, with pairwise Pearson correlations ranging from r(32) =
0.69 to 0.91, p < 0.0001 for all tests. A sample WMH mask produced by the LPA
algorithm is shown in Figure 3.1.
49
3.3.3 White matter hyperintensities, [11C]-PIB PET, and
diffusion-weighted imaging
Correlation tests for our primary hypothesis are summarized in Table 3.2 and Figure
3.2. Both white matter SUVmax and white matter SRTM R1 were negatively correlated
with the mean z-score WMH volume, and positively correlated with FA. In turn, FA was
negatively correlated with WMH volume (r(32) = -0.51, p = 0.002). Additionally, MD
was negatively correlated with SUVmax (r(32) = -0.43, p = 0.011) but not with white
matter SRTM R1 to a degree that reached statistical significance (r(32) = -0.28, p =
0.111).
Table 3.2: Pearson correlations of the primary hypotheses (WM SUVmax and R1 with WMH and FA) and ofthe cortical measures. For the 4 primary hypotheses, p-values adjusted for false discovery rate (q) are alsoreported.
WMH Volume DTI FA
WM SUVmax r = -0.4, p = 0.02, q = 0.027 r = 0.44, p = 0.01, q = 0.024
WM SRTM R1 r = -0.43, p = 0.012, q = 0.024 r = 0.36, p = 0.039, q = 0.039
Cortical SUVmax r = -0.34, p = 0.051 r = 0.34, p = 0.047
Cortical R1 r = 0.13, p = 0.448 r = -0.11, p = 0.525
Cortical SUVR50-70 r = 0.05, p = 0.787 r = 0.03, p = 0.863
Reported p-values are two-sided. DTI = diffusion tensor imaging; FA = fractional
anisotropy; MD = mean diffusivity; SRTM = simplified tissue reference model; SUV
= standardized uptake value; SUVR = standardized uptake value ratio; WM = white
matter; WMH = white matter hyperintensity.
The cortical composite [11C]-PIB SUVR50-70 (a measure of amyloid deposition) was
not significantly correlated with WMH volume, white matter SUVmax or R1, FA (Table
3.2) or MD. Cortical SUVmax, like white matter SUVmax, was negatively correlated with
WMH volume and positively correlated with FA (Table 3.2).
50
3.3.4 Potential confounding variables
We tested for relationships between the primary variables and the potential confounds
of age and blood pressure. Systolic and diastolic blood pressure were not significantly
correlated with WMH volume, white matter SUVmax or R1, DTI FA or MD, or SUVR
(Table 3.3). However, age was significantly associated with the biomarkers in the expected
directions: advanced age was associated with greater WMH volume and lower FA, white
matter SUVmax and R1 (Table 3.3). We therefore calculated partial correlations with age
as a third variable for the correlations of the primary hypothesis. The directions of the
relationships were unchanged but the magnitude of the correlations decreased and the
p-values increased for all tests, which are reported in Table 3.4.
Table 3.3: Pearson correlations between potential confounding variables and biomarkers.
Age (df = 32)
Systolic BP (df =
31)
Diastolic BP (df =
31)
WMH Volume r = 0.65, p =
<0.001
r = 0.07, p = 0.705 r = 0.01, p = 0.976
WM SUVmax r = -0.50, p =
0.003
r = -0.21, p =
0.249
r = -0.16, p =
0.360
WM SRTM R1 r = -0.33, p =
0.058
r = -0.12, p =
0.511
r = -0.10, p =
0.584
DTI FA r = -0.44, p =
0.01
r = -0.17, p =
0.333
r = -0.14, p =
0.453
DTI MD r = 0.63, p <
0.001
r = 0.31, p = 0.08 r = 0.25, p = 0.153
Cortical SUVR50-70 r = 0.35, p =
0.044
r = 0.06, p = 0.734 r = 0.07, p = 0.694
Reported p-values are two-sided. BP = blood pressure; df = degrees of freedom;
DTI = diffusion tensor imaging; FA = fractional anisotropy; MD = mean diffusivity;
51
SRTM = simplified tissue reference model; SUV = standardized uptake value; SUVR =
standardized uptake value ratio; WM = white matter; WMH = white matter hyperin-
tensity.
Table 3.4: Partial Pearson correlations of for primary hypotheses with age at PET scan included as a controlledthird variable.
WMH Volume DTI FA
WM SUVmax r = -0.11, p = 0.547 r = 0.28, p = 0.114
WM SRTM R1 r = -0.30, p = 0.095 r = 0.25, p = 0.161
DTI = diffusion tensor imaging; FA = fractional anisotropy; SRTM = simplified
tissue reference model; SUV = standardized uptake value; WM = white matter; WMH
= white matter hyperintensity.
3.4 Discussion
With studies confirming a relationship between measures of [11C]-PIB delivery and CBF,
we sought to test whether such markers specifically in white matter regions were associ-
ated with known markers of white matter pathology. We calculated a measure of WMH
volume and DTI FA. The selection of WMH and DTI FA was motivated by their clinical
significance and association with CBF.
With respect to our primary hypothesis that [11C]-PIB delivery in white matter is
associated with MRI measures of white matter integrity, we found correlations in the
expected directions. The peak [11C]-PIB uptake in white matter and the relative delivery
parameter R1 were moderately correlated with WMH volume and global white matter
FA.
To explore whether the primary hypotheses are specific to models previously asso-
ciated with CBF, rather than with amyloid, we also tested whether the cortical SUVR
52
(over a 50-70 minute window) was associated with either FA or WMH volume, and as
expected, it was not (see Table 2). Further, the association of [11C]-PIB delivery with
white matter integrity appears specific to the white matter ROI, which suggests that we
are not simply identifying a correlation between global cortical CBF and white matter
pathology.
Recently, the association of late-scan SUVR in white matter regions has been shown
to be associated with DTI markers of white matter integrity including FA and MD in
a larger sample (n = 537) (Zeydan et al. 2019), however, early-scan information was
not available, so the association with R1 or SUVmax could not be tested. Tracer delivery
estimates such as R1 and SUVmax may better reflect white matter integrity, and our
result warrants replication in a larger sample to confirm this.
3.4.1 Sample Characteristics
The hypothesis was tested in a sample of participants with MCI, and some of which
also had a history of MDD. By not excluding the participants with a history of MDD,
we were able to test the hypothesis in a larger sample. However, this also increased
the heterogeneity of the sample. A risk of using a heterogenous sample is that the
hypothesized relationship may only be true in a subset of the sample. In such a case, an
effect could be washed out or missed. Alternatively, in an effect is observed, there is a
risk of falsely concluding that it applies to all sub-populations within the group, when it
may only apply to a subset.
In this case, the hypothesized relationship is between two imaging modalities pre-
sumed to reflect brain pathology, i.e. white matter integrity, which is not a finding specific
to diagnostic categories. We are not correlating the markers with clinical or demographic
features, only among brain markers. Even if, in a subgroup, there is lower likelihood of
a specific pathology, the relationship between the imaging and (absence of) pathology
would hold true. Further, in identifying relationships generally, it is important to have a
broad range of values, as a restricted range can obfuscate a true correlation. The partic-
53
ipant sample used, of MCI with or without MDD, is also reflective of the population for
which such a biomarker could be useful. In the context of these benefits and limitations,
and interpreting any results in this context, for a first exploratory study such as this, the
selected sample is appropriate.
3.4.2 Limitations
Age was associated with all PET delivery measures, as reported with other tracers (Cum-
ming et al. 2013) and with FA (Maillard et al. 2012); it was therefore a potential con-
founding variable. CBF is known to decrease with age (Chen et al. 2011). Previous work
that explored associations between [11C]-PIB delivery measures and CBF measures did
not report inclusion or effect of age in the regression models, although some account for
age in age-matched group comparisons (Rodriguez-Vieitez et al. 2017; Meyer et al. 2011;
Chen et al. 2015; Forsberg et al. 2012; Gjedde et al. 2013; Rodriguez-Vieitez et al. 2016;
Peretti et al. 2019).
Our study is underpowered to explore tests beyond the primary hypotheses, but we
included partial correlations including age to highlight the potential confound and need
for further replication. The partial correlation tests do not reach statistical significance,
but the identified correlations are in the same direction as the correlations that do not
adjust for age. This highlights the need for replication with a larger sample size, in
addition to attention to age in studies investigating the relationship between CBF and
PET delivery measures.
We used a single white matter ROI, which may risk washing out local variations of
clinical importance. However, this approach reduces the number of statistical compar-
isons which decreases the risk of spurious associations. Further, this approach reflects a
practical analysis which could be implemented with existing PET analysis pipelines. A
global measure of FA, while sensitive for general white matter pathology, is relatively non-
specific and significant variation in the average measure reflects crossing fibres (Alexander
et al. 2007). A future study with a larger sample size could explore regional associations.
54
Our approach to WMH quantification attempts to balance the drawbacks of all
possible approaches: by using automated segmentation algorithms, we avoid the inaccu-
racy and variability associated with visual rating scales, and the cost and subjectivity
of manual delineation by neuroradiologist. We used automatic segmentation methods
that did not require additional training, which means the methods when applied to our
dataset may have less accuracy than when tested on the datasets on which they were
trained, although previous replication studies have shown good accuracy on new datasets
(Caligiuri et al. 2015).
No single optimal segmentation algorithm or technique has been identified (Caligiuri
et al. 2015). By using multiple methods with distinct approaches and taking the mean of
the z-scores, we attempted to reduce the error from any single algorithm, on the premise
that the methods would be more similar in their measurement of WMH and differ in error,
thus the z-score approach may more closely reflect WMH volume. Despite using different
approaches, we demonstrated that the 4 algorithms are highly correlated as is expected
with validated techniques. We also avoid the potential bias that could be introduced in
the selection of a method that has not been pre-specified, in the absence of evidence for
the superiority of any particular algorithm. That is, by using the standardized mean
of multiple methods, we mitigate the possibility that the PET measures are spuriously
correlated with error in the automated WMH segmentation.
3.4.3 Applications
The use of white matter [11C]-PIB delivery measures will not replace DTI for assess-
ment of white matter integrity or T2-weighted FLAIR in the characterization of vascular
related-WMH. However, white matter [11C]-PIB maximum SUV or R1 may provide im-
portant additional information in studies that use [11C]-PIB-PET. These measures may
reflect white matter structural or vascular changes that are distinct from, but associated
with, the pathologies that are reflected by WMH and FA, which, once clarified, may
prove useful.
55
The [11C]-PIB delivery measures are easily derived from the regional time-activity
curves produced from existing standard PET analysis pipelines. Thus, if confirmed, the
knowledge that white matter R1 reflects white matter pathology improves the cost-benefit
profile of [11C]-PIB-PET scans. This may be particularly valuable in large multisite stud-
ies that use amyloid imaging longitudinally to investigate the clinical and pathological
course of Alzheimer’s and related conditions, particularly when FLAIR or DWI MRI
sequences are not available.
This paper’s findings are unlikely to directly impact clinical care at this time. How-
ever, the analysis of amyloid PET data without the use of MRI is possible (Bourgeat
et al. 2018), so it would be theoretically possible to create a fully automated analysis
pipeline could to extract important clinical values from a dynamic amyloid PET scan
such as amyloid burden, cortical and white matter CBF, which could contribute to a risk
score, for example. While such a direct clinical impact is not possible today, this specula-
tive example highlights the potential future benefit of exploring the potential biomarkers
within an amyloid PET scan. Deriving multiple PET biomarkers from a single scan may
prove clinically useful when amyloid imaging is required.
Similar to with [11C]-PIB, CBF has been associated with the early uptake of flor-
betapir, another amyloid PET tracer (Devous et al. 2014). Thus, it is possible that our
techniques with [11C]-PIB could be applied to other amyloid tracers which may broaden
the research and clinical implications of these findings once replicated and validated.
In conclusion, in this exploratory analysis, [11C]-PIB R1, a marker of local white
matter CBF, was correlated with MRI markers of white matter pathology, WMH and
FA. Further exploration and replication are warranted to establish whether [11C]-PIB
may be a “triple biomarker”, with dynamic [11C]-PIB scans providing information about
amyloid deposition (late-scan cortical SUVR), cortical CBF (early scan cortical SUV
and delivery measures), and white matter pathology (early scan white matter SUV and
R1). Future studies are required to clarify the specific pathology that these white matter
delivery measures reflect, and to determine whether white matter SUVmax or SRTM R1
are associated with clinical outcomes such as cognitive decline and the risk of a major
56
neurocognitive disorder.
57
Figure 3.1: Quality control output image (cropped) of LPA algorithm showing T2-weighted FLAIR MRI imagenext to image with masked white matter hyperintensities.
58
Figure 3.2: Relationships among primary white matter measures with least-square regression line.
59
Chapter 4
Open Source PET Analysis in R
This chapter is partly modified and reproduced from work published as a vignette, as part
of the software package tacmagic (author Eric Brown) and summarized in publication
(Brown 2019). See the Contributions section of this thesis for details.
4.1 Background
Reproducibility in science is the ability for an independent scientist to be able to use the
data and methods of another scientist to reproduce the same results (Laine et al. 2007).
Poor transparency and lack of reproducibility is common in neuroimaging research, which
are contirbuting factors to the concern of a high rate of false scientific reports (Poldrack
et al. 2017).
Neuroimaging research often relies on complex analysis and statistical software,
with many steps and high flexibility (Poldrack et al. 2017). This flexibility can be can
extremely powerful to answer important research questions, but too much flexibility can
lead to spurious findings, and too much complexity can lead to software errors (Poldrack
et al. 2017). Further, when source code is kept private, even if the results are valid,
reproducing the results may not be feasible. Private code is also vulnerable to software
60
errors that may not be detected as readily as open source projects shared with many
users (Poldrack et al. 2017). While producing an additional tool, even if open source,
can add to the problem of excessive analysis flexibility by being yet another option, a
well-designed, open-source, freely available tool also has the potential to become widely
used and replace less transparent in-house, commercial, or ad hoc options.
For fully reproducible research, both the original data and a fully transparent ana-
lytic protocol are required (Laine et al. 2007). Addressing the former, large collaborative
efforts (such as ADNI, or the Open Access Series of Imaging Studies) are providing
thousands of images for researchers to analyze to answer important questions. These
collaborative efforts also address the frequent issue of insufficient power in neuroimaging
studies (Poldrack et al. 2017). To address the second element of reproducible research,
having a transparent analysis protocol, some groups are including full analysis scripts
along with the data and results at time of publication (Nichols et al. 2017).
Providing analysis code is more transparent than simply listing an overview of
analytic steps, and ideally code is provided as a “literate program”–a program that
includes documentation and code within a document that can be run to produce a human-
readable document (Peng et al. 2006). The R statistical programming language (R Core
Team 2018), itself an open-source project, has been recommended for this purpose (Peng
et al. 2006), and is widely used with many tools available (Gandrud 2015).
For code sharing, R is also extensible by “packages”, which are collections of func-
tions, tests, and data that are documented and bundled in a uniform way prescribed by
the maintainers of the R language (R Core Team 2019). rOpenSci is an organization
dedicated to supporting the use of vetted R packages for reproducible science (rOpenSci
2019). The organization faciliates peer-review of contributed packages and promotes the
use and further development of these vetted open source R packages. Sharing R packages
is facilitated by Comprehensive R Archive Network (CRAN), a repository from which
end users can easily download vetted packages.
61
4.1.1 PET Analysis in R
Positron emission tomography (PET) is a research and clinical imaging modality that uses
radioactive tracers that bind to target molecules of interest. A PET scanner identifies
the tracer location by virtue of the tracer’s radioactive decay, providing information
about the distribution of target in the body. Analysis pipelines are used to calculate
radiotracer activity over time within a spatial region of interest (ROI). The resulting
time-activity curves (TAC) are subsequently analyzed, often with kinetic modelling, to
answer important clinical and research questions (Dierckx et al. 2014).
As the spatial resolution of PET is relatively poor, analysis is frequently combined
with higher resolution imaging such as magnetic resonance imaging (MRI), which can
be spatially co-registered to the PET image. Identification of regions of interest (ROIs)
on the MRI is done with automated segmentation and parcellation techniques, which
subsequently allows radiotracer activity (over time) to be identified by ROI.
An image analysis pipeline is required to extract regional time activity curves
(TACs) from a dynamic PET image. There is no single standard analysis pipeline, and
the concerns about flexibility raised by Poldrack et al. (2017) are applicable. Multiple
approaches and pipelines exist including commercial solutions (e.g. PMOD Inc., which
is widely used) and in-house options (e.g. Moriguchi et al. (2019)), but more recently
fully open source pipelines have also become available, for example magia (Karjalainen
et al. 2019) and APPIAN (Funck et al. 2018). Pipelines generally implement the following
steps:
• Dynamic PET pre-processing (e.g. motion correction, decay-correction)
• PET image co-registration with structural MRI
• MRI segmentation, normalization and parcellation
Analysis piplelines may include the kinetic modelling steps, but may also save vol-
ume and related data for subsequent kinetic and statistical modelling. Different pipelines
save the data in different formats. Common steps including merging ROIs, transforming
62
the data, kinetic analysis, and calculation of cutoffs scores are often done in ad hoc ways
or with other software.
Previous options for PET analysis in R are limited. There are no available packages,
of the over 14,000 packages on CRAN, that provide basic time-activity curve loading and
processing or the analytic techniques most frequently used in AD-related research such
as SUVR, Logan DVR or PIB-positive cutoff calculation. Thus, a package that provides
such functionality could help improve transparency, reproducibility and efficiency in PET
analysis, including for AD and related research.
4.2 Methods
The software package, title tacmagic, was created to support the analysis of PET TAC
data. The major features of tacmagic include:
1. loading TAC and volume data for analysis
2. merging regional TAC data into larger ROIs weighted by volume,
3. basic TAC plotting,
4. calculation of standardized uptake value ratio (SUVR) (Lopresti et al. 2005; Dier-
ckx et al. 2014),
5. calculation and plotting of the non-invasive reference region Logan DVR model
(Logan et al. 1996; Oikonen 2018) and
6. calculation of cut-off values for dichotomizing data (Aizenstein et al. 2008).
To create a software package supporting basic PET and TAC data analysis for AD
and related research, we used the open source statistical programming language, R (R
Core Team 2018). We followed the principles for package style, structure, documentation
as outlined by the R core team (R Core Team 2019). For transparency, the source code
with a full history of modifications was tracked using git and hosted on the leading
software development platform github.com, so the source code can be accessed online and
63
to enable community collaboration. To reduce the likelihood of software errors (Poldrack
et al. 2017), software unit testing was used to cover all major functions, using the testthat
R package, a testing framework for R (Wickham 2015).
tacmagic was submitted to rOpenSci, where it underwent an open soft-
ware review process by an editor and 2 reviewers, which is archived online
(https://github.com/ropensci/software-review/issues/280). The reviewed version
was released and the source code was independently archived where it can be accessed
online (https://doi.org/10.5281/zenodo.2577946). The package was submitted and
archived to CRAN to enable wide distribution.
4.3 The tacmagic R package
While the source code for tacmagic can be accessed online, here we provide an overview
of the package’s features and their implementation. The sample data for this chapter
uses anonymized PIB PET and T1-weighted MRI scans of participants with AD, data
from http://www.gaain.org, which was made available for unrestricted use (Klunk et al.
2015).
There are two approaches to using the tacmagic package to analyze PET time-
activity curve data: either by loading participant data individually and using the various
functions to analyze it, or via the batch functions to list and analyze data from multiple
participants. Here, we illustrate the main features of tacmagic, by walking through the
analysis of a single participant. We provide an explanation of the batch mode at the end.
The R base software must be installed and is the base from which all R packages
are run. As tacmagic is found within the primary package repository, CRAN, tacmagic
can be installed with a single command, install.packages("tacmagic"), and once
installed it can be loaded for use with library(tacmagic).
64
4.3.1 Time-activity curve operations
4.3.1.1 Data loading
Time-activity curve (TAC) data is loaded via load_tac(), which is a wrapper for format-
specific functions. To specify which file format the TAC data is stored as, use the format
parameter. Supported formats can be viewed in help(load_tac).
The minimal amount of information required is the TAC data for one or more ROI,
including the start and stop times of each frame, the time units and the activity units.
This information may be in one or more files depending on the format and software that
created it.
For example, PMOD’s .tac files contain all of the information, but the TAC .voistat
files do not contain start and stop times, but this information could be specified using a
.acqtimes file. Support is also available for DFT format, which contains both TAC and
volume data.
We processed the PIB PET and T1MRI data with the PMOD (https://www.pmod.com/)
PNEURO software suite to produce a .tac file with TACs for all ROIs in the Hammer’s
atlas. The .tac file can be loaded with load_tac(). The filename object below will
be a path to the local file, and in the case of this example, we use the package’s built-in
file path:
> filename <- system.file("extdata", "AD06.tac", package="tacmagic")
> AD06_tac <- load_tac(filename, format="PMOD")
A TAC object is an R data.frame object with extra attributes including time and
activity units. A summary can be printed with the generic summary() function. For
example, the object we just created would display:
> summary(AD06_tac)
65
tac object
Activity unit: kBq/cc
Time unit: seconds
Number of ROIs: 172
Number of frames: 34
Time span: 0 - 5400 seconds
Unique frame durations: 15 30 60 180 300 600 seconds
PMOD’s suite also produces .voistat and .acqtimes formats, which together can
be used to produce the same data when .tac files are not available:
> f_acq <- system.file("extdata", "AD06.acqtimes", package="tacmagic")
> f_vs <- system.file("extdata", "AD06_TAC.voistat", package="tacmagic")
> tac2 <- load_tac(f_vs, format="voistat", acqtimes=f_acq)
The magia pipeline saves TAC data in a different format within a MATLAB .mat
file. Such files can be loaded similarly, though the units must be entered, because the
information is not encoded in the .mat file:
> f <- system.file("extdata", "AD06_tac_magia.mat", package="tacmagic")
> AD06_magia <- load_tac(f, format="magia", time_unit="seconds",
+ activity_unit="kBq/cc")
For other data sources, tacmagic TAC objects can be created from data.frame
objects with as.tac(). The time and activity units must be specified as arguments if
not already set as attributes in the data.frame. The columns of the data.frame are the
regional TACs, with the ROI names stored as column names.
> manual <- data.frame(start=c(0:4), end=c(2:6),
+ ROI1=c(10.1:14.2), ROI2=c(11:15))
66
> manual_tac <- as.tac(manual, time_unit="minutes", activity_unit="kBq/cc")
> summary(manual_tac)
tac object
Activity unit: kBq/cc
Time unit: minutes
Number of ROIs: 2
Number of frames: 5
Time span: 0 - 6 minutes
Unique frame durations: 2 minutes
4.3.1.2 ROI merging
Often it is desirable to combine TAC ROIs into larger ROIs. For example, if the PET
analysis pipeline created TACs for each atlas ROI, your analysis may call for merging
these “atomic” ROIs into larger regions, for example, to merging ROIs into a single ROI
for the frontal lobe. If this is done, the mean TAC data should be weighted by the
volumes of the atomic ROIs. If volume information is available, tac_roi() provides this
merging functionality.
In PMOD’s software, volume information is available in .voistat files. Units do
not matter because it is the relative volume information that is needed. In addition to
TAC and volume information, we must specify which atomic ROIs make up the merged
ROI. This is done in tacmagic by providing a named list, where the names are the
merged ROIs and the list items are themselves lists of the atomic ROIs that make up
each merged ROI. For the Hammer’s atlas, and as an example, typical data is built into
the package with roi_ham_stand(), roi_ham_full(), or roi_ham_pib().
> AD06_vol <- load_vol(f_vs, format="voistat")
> roi_ham_pib()[1] # The first definitions of merged ROIs, for example.
$leftfrontal
[1] "FL_mid_fr_G_l" "FL_precen_G_l" "FL_strai_G_l"
67
[4] "FL_OFC_AOG_l" "FL_inf_fr_G_l" "FL_sup_fr_G_l"
[7] "FL_OFC_MOG_l" "FL_OFC_LOG_l" "FL_OFC_POG_l"
[10] "Subgen_antCing_l" "Subcall_area_l" "Presubgen_antCing_l"
> AD06 <- tac_roi(tac=AD06_tac,
+ volumes=AD06_vol,
+ ROI_def=roi_ham_pib(),
+ merge=F, # T to also return atomic ROIs
+ PVC=T) # to use _C ROIs (PMOD convention)
A new TAC object has been created, AD06, which contains ROIs that were
merged with volume weighting from the original Hammer’s atlas ROIs as defined by
roi_ham_pib().
4.3.1.3 Plotting
Basic TAC plotting can be done by calling plot, which accepts one or two TAC objects,
e.g. from two participants or group means. The user specifies which ROIs to plot by
providing a vector of ROI names as they appear in the TAC object. As the TAC object
contains time unit information, the plot can convert to desired units, which can be
specified with the time argument. The following code produces the output shown in
Figure 4.1.
> plot(AD06,
+ ROIs=c("frontal", "parietal", "cerebellum"),
+ time="minutes",
+ title="PIB time activity curves for AD06")
68
0 20 40 60 80
05
1015
2025
PIB time activity curves for AD06
Time (minutes)
Act
ivity
(kB
q/cc
)
●
●
●
●
●●●
●●●
●●
●
●●
●●
●
●●
●
●
●
●
●
●●
●●
● ●●
●●
●
●
●
●
●●●●●●
●●●
●
●
●
●●
●●
●
●
●
●
●●
●●
●●
●●
● ●
●
●
●
●
●●●
●●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●● ● ●
● ● ●
●
●
●
frontalparietalcerebellum
Figure 4.1: Example output from tacmagic plotting function, as described in 4.3.1.3
4.3.2 Model calculation
4.3.2.1 SUVR
Once data is loaded as a TAC object, further analysis is possible. The SUVR is a simple
quantification of PET activity that is commonly used from many tracers including PIB.
It is the ratio of the tracer activity over a specified time period (Ct) in a target ROI to a
reference region. Using a ratio allows factors that are normally required to calculate an
SUV to cancel out, namely the tracer dose and patient body weight, therefore SUV R
can be calculated from TAC data alone:
SUV R = SUVT ARGET
SUVREF
= CtT ARGET
CtREF
In the literature, SUVR is variably described and calculated using the mean of
activity for the frames of the specified time period, or the area under the curve. For
PIB, the mean/summed activity has been used, and the time windows have varied from
starting at 40-50 minutes and ending at 60-90 minutes (Lopresti et al. 2005).
69
The suvr() function calculates SUVR for all regions in a TAC file based on the
provided time information (as a vector of frame start times) and the specified reference
region (a string). If the frames used are of different durations, the weighted mean is
used. In this example, we provide three start times, in seconds, as the SUVR window
definition: frames starting at 3000, 3300, and 3600, which is the equivalent of a 50- to
70-minute window.
AD06_SUVR <- suvr(AD06, SUVR_def=c(3000,3300,3600), ref="cerebellum")
An alternative method, using the area under the curve with the mid-frame times
as the x-axis is available with suvr_auc() and should provide very similar results.
AD06_SUVR2 <- suvr_auc(AD06, SUVR_def=c(3000,3300,3600), ref="cerebellum")
4.3.2.2 DVR
The Distribution Volume Ratio (DVR) is a method of quantifying tracer uptake that is
used as an alternative to the SUVR in PIB studies, for example. Like SUVR, it can be
calculated from TAC data without the need for arterial blood sampling, by making use
of a reference region. In this case, it is called the non-invasive Logan plot method. It
is calculated with a graphical analysis technique described by Logan et al (Logan et al.
1996).
In addition to the TAC data, depending on the tracer, a value for k2’ may need to
be specified. For PIB, this has a limited effect on the result, but can be specified, and a
fixed value of 0.2 has been recommended.1
The non-invasive Logan method of DVR calculation works by calculating the slope
of the line of the following equation after the time, t∗, where linearity has been reached:1http://www.turkupetcentre.net/petanalysis/analysis_11c-pib.html
70
∫ T
0 Croi(t)dt
Croi(t)= DV R[
∫ T
0 Ccer(t)dt + Ccer(t)/k2′
Croi(T )] + int
The time, t∗ (t_star), after which the relationship is linear can be found by test-
ing the point after which the error is below a certain threshold (the default is 10%). If
t_star=0, then tacmagic tries to find the suitable value. The following example demon-
strates how to invoke the Logan method for a single ROI, the frontal lobe region, using
the cerebellum as reference, and with a fixed k2’ of 0.2, with the optimal t∗:
AD06_DVR_fr <- DVR_ref_Logan(AD06, target="frontal", ref="cerebellum",
+ k2prime=0.2, t_star=0)
To visually confirm that the model behaved as expected with linearity, the Logan
model we just created can be plotted: plot(AD06_DVR_fr), as shown in Figure 4.2. The
right plot shows the Logan model, with the vertical line representing the identified t∗,and the linear model fitted to the points after that time. In this case, the line after t∗can be seen to fit well. The slope of that line is the DVR.
0 20 40 60 80
05
1015
2025
Time−activity curves
Time (minutes)
Act
ivity
(kB
q/cc
)
●
●
●
●
●●●●●●●●
●
●●
●●
●
●●
●
●
●
●
●
●●
●●
● ●●
●●
●
●
●
●
●●●●●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●● ● ●
● ● ●
●
●
frontalcerebellum
●●●●●●●●●●●●●●●
●●●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
0 10 30 50 70
020
4060
8010
012
0
Logan plot
x
y
Figure 4.2: Example output from tacmagic plotting function of a Logan plot.
Similarly, DVR can be calculated for all ROIs, either by setting t_star manually
71
or to 0 as before. If 0, a different value will be identified for each ROI.
AD06_DVR <- DVR_all_ref_Logan(AD06, ref="cerebellum",
+ k2prime=0.2, t_star=23)
For this data, the DVR calculation has been shown to produce equivalent results as
an existing tool (Oikonen 2018). A wrapper function dvr() is available to conveniently
calculate DVR for a target ROI or all ROIs, and currently defaults to using the Logan
reference method.
4.3.3 Batch analysis
In most cases, a project will involve the analysis of multiple participants. The above
workflow can be used to test and visualize an analysis, but a batch workflow will likely
be preferred to analyze multiple participants. All analyses can be run using two steps: a
batch data loading step and a batch analysis step.
Data loading is done by batch_load(). See help(batch_load) for the required
arguments. The first argument is a vector of participant IDs that corresponds to file
names, e.g. participants <- c("participant01", "participant02") if the files are
located e.g. /mypath/participant01.tac and /mypath/participant01_TAC.voistat.
In this case, the function call might look like the following:
my_data <- batch_load(participants, dir="/mypath/", tac_format="PMOD",
+ roi_m=T, vol_file_suffix="_TAC.voistat",
+ vol_format="voistat", ROI_def=roi_ham_stand(),
+ merge=F)
The above would load the appropriate TAC and voistat files, perform the
ROI merging specified by ROI_def, because roi_m = TRUE, and would return a list
72
where each element represents a participants, e.g. the first participant would be
my_data$participant1.
Once the TAC data is loaded, all analyses can be run using batch_tm(). The output
from batch_load() is the first argument for batch_tm(). The models implemented
in tacmagic can be specified using the models argument, e.g. models = c("SUVR",
"Logan") to calculate both SUVR and Logan DVR. The relevant model parameters will
also need to be specified, and all possible arguments are shown in help(batch_tm).
4.3.4 Cutoff calculations
In the analysis of PIB PET data, often researchers want to dichotomize patients into PIB+
vs. PIB-, i.e. to identify those with significant AD-related amyloid pathology (PIB+).
There are a number of approaches to this depending on the available data. We have
implemented a method described by Aizenstein et al. Aizenstein et al. (2008), which
uses a group of participants with normal cognition to establish a cutoff value above which
participants are unlikely to have minimal amyloid pathology.
The method identifies a group of participants out of the normal cognition group
with higher-PIB outliers removed. An outlier is a participant with any ROI with a DVR
higher than the upper inner fence, from a set of ROIs known to be associated with
amyloid deposition. Such participants are removed from the group, and this process is
done iteratively until no more outliers are removed. Then, cutoff values are determined
from this new group for each ROI, set again as the upper inner fence. Then these cutoff
values are applied to all participants, and a participant is deemed PIB+ if they have at
least one ROI above its cutoff. To calculate the cutoff values using this iterative method,
cutoff_aiz() takes 2 arguments: the DVR data, and the names of the variables of the
ROI DVRs to use (and there must be at least 2 for this method). The final step is to
apply the cutoffs to the full set of participants. With sample data, fake_DVR, generated
to demonstrate the package’s features, calculating cutoff scores can be done with the
following calls:
73
> cutoffs <- cutoff_aiz(fake_DVR, c("ROI1_DVR", "ROI2_DVR",
+ "ROI3_DVR", "ROI4_DVR"))
Iteration: 1 Removed: 10
Iteration: 2 Removed: 1
Iteration: 3 Removed: 0
> positivity_table <- pos_anyroi(fake_DVR, cutoffs)
> summary(positivity_table)
Mode FALSE TRUE
logical 39 11
The algorithm identified 11 PIB+ participants. In the generation of the sample
data, the DVRs from the first 10 participants were drawn from a normal distribution
with mean 1.9, sd 0.6 and for the latter 40 participants, from mean 1.3, sd 0.3; thus this
pattern is in line with what we would expect: all 10 of the first participants are PIB+,
and just 1 of the latter 40 was (by chance).
4.4 Conclusions
In the context of increasing recognition of the factors that contribute to invalid findings in
neuroimaging research and science more broadly, including but not limited to low sample
size and power, overly flexible data analysis workflows, and software error, there has been
a push for fully reproducible analysis pipelines, for example, from organizations such as
rOpenSci. In the course of planning the data analysis for a PIB PET project outlined in
Chapter 3, we identified a need for a safe, transparent way to process and analyze PET
TAC data, with no suitable existing software available. In this chapter, we outlined the
creation of an R package to serve this purpose.
Our package, tacmagic is an open source solution to loading, manipulating, and
plotting PET TAC data, in addition to featuring basic TAC analysis techniques. By using
standardized platforms to produce this package, including R, git, github, rOpenSci, and
74
CRAN, we created software that has been reviewed and tested, and has the potential to be
reviewed, improved, and expanded in a fully transparent manner. If other research groups
find the package useful, it has the potential to improve replicability and transparency in
PET research.
75
Chapter 5
Lead and Alzheimer’s Disease
This chapter is modified and reproduced from published work copyright by Bentham
Science and reproduced with permission (E. E. Brown, Shah, et al. 2019). See the
Contributions section of this thesis for details.
5.1 Introduction
Untangling the pathophysiology of AD is a research priority, given the disease’s rising
public health burden coinciding with an aging population, and the hope that a better
understanding will improve prevention strategies and clinical management. Like pieces
of a puzzle, diverse risk factors across the lifespan each confer a small and significant
risk of developing AD and its characteristic neuropathological changes, including tau
neurofibrillary tangles, amyloid plaques, and cerebrovascular pathology (Livingston et al.
2017). The more pieces that can be identified, the clearer the picture will become. Lead
(Pb) exposure, an ongoing public health concern and known cause of dementia, may be
another piece of the puzzle toward understanding the pathophysiology of AD (Bakulski
et al. 2012). Yet, the potential role of Pb in AD is uncertain.
Lead is a heavy metal that contaminates the environment due to current and histor-
76
ical uses across the globe and is a significant cause of morbidity and mortality in middle-
and low-income countries (Organization 2009). While Pb remains ubiquitously present
in the environment, there has been a reduction in industrialized regions, attributable to
the discontinuation of leaded gasoline, with a measurable decrease in human accumula-
tion (Organization 2009; McNeill et al. 2017). At the same time, despite an increasing
overall prevalence of dementia in many locations due to an aging population, a decreasing
incidence of dementia has been reported, which has not been fully explained (Satizabal
et al. 2016). If Pb were a cause of AD, such a trend might be expected.
The neurotoxicity of Pb to humans is well-established, both acutely at high doses,
and at the lowest levels found in the environment (Organization 2009). Rare case reports
have connected high levels of Pb exposure with AD pathology (Niklowitz & Mandybur
1975). Effects of chronic exposure include neuropsychiatric and cognitive effects, in-
cluding IQ deficits in children, which persist throughout life and impact socioeconomic
potential (Reuben et al. 2017). Additionally, chronic Pb exposure has major effects
on hypertension, an important contributor to cognitive decline, and is associated with
increased mortality (Lanphear et al. 2018).
Experimental studies in animals have established a causal link between Pb exposure
and amyloid, tau and GSK-3 beta abnormalities in rodents (Bihaqi et al. 2018), but
such designs are not ethically possible in humans. To ascertain the impact of chronic Pb
exposure in humans, exposure must be estimated by history or biological measurement.
Longitudinal community studies using bone Pb measurement have established an
association between environmental Pb exposure and cognitive decline (Shih et al. 2007),
with several large well-designed studies, including the Veteran’s Affairs Normative Aging
Study (Farooqui et al. 2017), the Baltimore Memory Study (Bandeen-Roche et al. 2009),
and the Nurses’ Health Study (Power et al. 2014; Weuve et al. 2009). While these
community studies do not report on the development of AD specifically, cognitive decline
is a major risk factor for AD, for example, with more than 1 in 5 individuals with mild
cognitive impairment going on to develop AD within 3 years (Livingston et al. 2017).
77
Longitudinal studies that recruit individuals in early life must include sufficient
numbers of participants to compare the subset of individuals that go on to develop AD
to those who do not. Alternatively, cross-sectional case-control designs with carefully
selected controls along with biological measures that reflect past exposure could clarify
the potential role of chronic Pb exposure in AD.
To clarify the role of Pb in AD, and to identify gaps in the current literature for
future investigation, we sought to perform a systematic review of human case-control
studies that compare AD to matched controls with any measure of Pb exposure. We
performed meta-analyses of the data where possible.
5.2 Methods
As Pb exposure can be estimated by different methods, we began with a scoping review
to identify categories of human Pb studies by method of Pb measurement (Arksey &
O’Malley 2005). Next, we conducted a series of systematic reviews, one for each category,
to clarify the role of Pb exposure in AD. The objective was to identify case-control studies
that included AD and control participants, comparing Pb levels by any method, and
where possible, to combine the study results in meta-analyses to help draw conclusions
about the potential role of Pb in AD.
To reduce risk of bias and to understand the current literature inclusively, all forms
of Pb measurement were included, regardless of the degree to which they may reflect Pb
exposure during the window in which such exposure may confer risk of developing AD.
An iterative search process was used also because searching for papers about Pb is
complicated by “lead” having a common homonym and “PB” a common acronym. Thus,
the scoping review was restricted to “Lead” as a medical subheading term (a strategy
other reviews have used) (Esteban-Vasallo et al. 2012) or the combination of the terms
“Pb” and “metal”. Subsequent searches with more specific methods specified can use
broader terms to identify Pb, without identifying more records than can be feasibly re-
78
viewed. The combination of search strategies reduces the risk of missing relevant articles.
For the same reason, we chose the MEDLINE database, which is focused on biomedicine.
5.2.1 Scoping search and targeted searches
We began with a scoping review to identify human studies indexed in MEDLINE with
the Pubmed interface, in which a measure of Pb exposure was obtained, as well as clinical
or biological manifestations of dementia. We used the following search query on August
13, 2018:
(Alzheimer* OR dement* OR amyloid* OR tau OR neurofib* OR NFT OR neu-
rocogniti* OR cognit*) AND (lead[MeSH Terms] OR (Pb AND metal)).
Search records were reviewed, and articles were retrieved first based on the scoping
inclusion and exclusion criteria. The scoping search inclusion criteria included articles
that were: in peer-reviewed academic journals; human studies; primary literature (exper-
iment or observational study); reported a measure of cognition, neurocognitive diagnosis
or biomarker; reported a neurocognitive disorder. The scoping search exluded articles
that: reported only groups including children or individuals <= 19 years old; had no hu-
man participants/subjects; reported no primary data; reported only acute Pb exposure
(e.g. acute overdose); did not report any measure of congition, biomarker, or neurocogni-
tive disorder diagnosis; were non English.
The retrieved articles were read to identify categories of Pb measurement, and
although the systematic review focused on case-control studies comparing AD and HC,
longitudinal studies were retrieved and reviewed for background and discussion. The
methods of Pb quantification were identified and informed subsequent targeted searches
with stricter inclusion and exclusion criteria (table 5.1).
79
Table 5.1: Inclusion and exclusion criteria for the individual targeted searches.
Inclusion Criteria Exclusion Criteria
� all inclusion criteria from scoping search � all exclusion criteria from scoping
search
� case-control design � group comparison on the basis of
exposed vs. unexposed rather than
diagnosis
� a group of Alzheimer’s disease
� a group of healthy controls
� lead measured by specified method
Targeted searches were used to identify additional articles that used each method of
Pb measurement. The searches included a unique term focused on the method, as listed
in table 5.2, as well as the following term in common: (Alzheimer* OR dementia OR
“neurocognitive disorder”) AND ((lead AND (Pb OR metal OR element)) OR (Pb AND
(lead OR metal or element))) NOT “may lead” NOT “lead compound”. Two researchers
independently reviewed the abstracts (EB, PS) for inclusion and exclusion criteria for
consensus. Where it was not possible to exclude on the basis of the abstract, the article
was retrieved. Where it could be determined that multiple papers reported the same
data, the paper with the larger sample size was included in the analysis.
5.2.2 Meta-analyses
For each category of Pb measurement method, where more than one study provided
means and standard deviations, and a meaningful meta-analysis was possible, we per-
formed a fixed and random effects meta-analysis to pool the data. Where studies met
inclusion criteria but did not present means and standard deviations, the corresponding
author was contacted to request the missing data. Meta-analysis was performed in R
version 3.5 using the package “meta” version 4.9-2 (Schwarzer 2007). Heterogeneity was
80
tested by calculation of I2 and tau statistics, and results were presented in forest plots.
5.3 Results
5.3.1 Scoping review
The scoping review was performed on August 13, 2018, and produced 800 results, which
were reviewed to identify methods of quantification of human Pb exposure to inform
targeted searches by method. Human studies of Pb exposure and cognitive impairment
included Pb measurement by the following methods: Pb accumulation in bone, as mea-
sured by K-shell X-ray fluorescence, either tibial, calcaneal or patellar; blood Pb levels,
in whole blood, serum, plasma or erythrocytes; in urine; in nails and hair; and in brain
tissue postmortem.
5.3.2 Targeted searches
The targeted searches were completed on October 12, 2018. Table 5.2 summarizes the
search results from each of the 6 targeted searches.
Table 5.2: Summary of targeted searches. The search terms unique to each method are shown, and allsearches included terms for Alzheimer’s and lead as described in the methods.
Measurement method Total Included Unique term
Blood 129 10 + 5 AND (blood
OR serum
OR
erythrocyt*
OR plasma
OR circulat*
OR
hemoglobin*)
81
Measurement method Total Included Unique term
Cerebrospinal fluid 33 5 AND
(cerebrospin*
OR CSF OR
spinal OR
fluid OR
lumbar OR
ventric*)
Hair and nail 19 1 AND
(toenail OR
fingernail
OR nail OR
nails OR
unguis OR
ungal OR
hair OR skin
OR dermis
OR epiderm*
OR
epitheli*)
Bone 15 0 AND (bone
OR KXRF
OR K-XRF
OR XRF
OR shell OR
tibia* OR
calcane* OR
patell*)
82
Measurement method Total Included Unique term
Urine 11 1 AND (urine
OR
urinalysis
OR urinary)
Postmortem 21 1 AND
(postmortem
OR mortem
OR autopsy)
5.3.2.1 Whole blood, serum and erythrocytes
Blood studies included measurements in serum, whole blood and in one study, erythro-
cytes. The 6 studies found in the scoping search were also identified by the targeted
search, as well as 4 additional case-control blood Pb studies. Additionally, a systematic
review of circulatory levels of various metals was identified that included Pb (Xu et al.
2018). By not specifically targeting Pb, this review identified 5 additional studies (3
from the same lab) which met our inclusion criteria, but where Pb was not mentioned
in the title or abstract, and one article from a journal not indexed in PubMED (Bocca
et al. 2005; Bocca et al. 2006; Alimonti et al. 2007; González-Domínguez et al. 2014;
Paglia et al. 2016). Most studies used inductively-coupled mass spectroscopy to measure
Pb, with one using energy dispersive x-ray fluorescence (Basun et al. 1991) and another
using atomic-absorption spectroscopy (Fathabadi et al. 2018).
The 15 identified blood Pb studies are summarized in Table 5.3. Of the 14 studies
that reported group differences, 12 reported no significant difference between AD and
HC. A single study reported greater blood Pb in AD vs. HC (Fathabadi et al. 2018),
while a single study reported significantly a lower level in AD vs. HC (Giacoppo et al.
2014). No study found significant group differences in serum Pb, and the single study
83
that reported on erythrocyte Pb did not find a group-wise difference by diagnosis (Hare
et al. 2016).
Table 5.3: Summary of blood lead studies.
Citation
AD
n
Age
(SD)
HC
n
Age
(SD)
Diagnostic
criteria
Pb
measurement Result
1991
Basun
24 75 (8) 28 78 (3) DSM-III-R Serum NS
2005
Pino
60 NR 109 NR NR Serum, whole
blood
Group
data
NR
2005
Bocca
60 74.6
(6.39)
44 > 45 NINCDS-
ADRDA
Serum, whole
blood
NS
2006
Bocca
28 72.8
(7.2)
30 62.5
(6.1)
NINCDS-
ADRDA
Serum, whole
blood
NS
2007
Alimonti
53 74.5
(6.5)
124
(57 >
45 y)
44.8
(12.7)
NINCDS-
ADRDA
Serum NS;
means
NR
2008
Gerhardsson
173 75, 76 54 73 DSM-IV;
NINCDS-
ADRDA
Plasma NS;
means
NR
2012 Lee 80 NR 130 NR NR Serum, whole
blood
NS
2012
McIntosh
19 77 (7) 24 73 (6) NR Serum, whole
blood
NS
2014
Giacoppo
15 73.27
(10.05)
10 72.40
(8.796)
NR Whole blood HC
>
AD
84
Citation
AD
n
Age
(SD)
HC
n
Age
(SD)
Diagnostic
criteria
Pb
measurement Result
2014
Park
64 74.97
(5.37)
67 73.36
(5.29)
DSM-IV;
NINCDS-
ADRDA
Serum NS
2014
González-
Domínguez
30 80.9
(4.5)
30 74.0
(5.7)
NINCDS-
ADRDA
Serum NS
2016
Hare
206 78.0
(8.6)
758 70.0
(7.0)
NINCDS-
ADRDA
Serum;
erythrocytes
(subset)
NS
2016
Paglia 34
72.44
(7.48)
40 65.53
(6.37)
NINCDS-
ADRDA
Serum NS
2018
Yang
170 76.48
(7.41)
264 71.82
(8.61)
DSM-IV Whole blood NS
2018
Fathabadi
27 70.85
(8.54)
54 67.55
(6.54)
NIAAA Whole blood AD
>
HC
Abbreviations: DSM Diagnostic and Statistical Manual (American Psychiatric
Association); NIAAA National Institute on Aging and the Alzheimer’s Association;
NINCDS-ADRDA National Institute of Neurological and Communicative Disorders
and Stroke and the Alzheimer’s Disease and Related Disorders Association; NR Not
reported; NS No significant group difference.
One study did not report group-wise mean Pb levels or group differences (Pino et al.
2005). Two of the studies which otherwise met inclusion criteria did not include means
and standard deviations and so could not be included in the meta-anaylses (Alimonti et
al. 2007; Gerhardsson et al. 2008). The raw data from one study was made available
by the authors so the mean and standard deviations could be calculated (McIntosh et al.
85
2012).
Two separate meta-analyses were possible, as some studies reported serum and
whole blood Pb levels separately. The 7 studies that reported whole blood means were in-
cluded in a meta-analysis, and the 8 studies that reported serum Pb means were included
in a separate meta-analysis, with Pb levels converted to common units for comparison
of standardized mean differences.
The meta-analyses are summarized in Figures 5.1 and 5.2. No significant difference
in Pb levels was identified between the groups by either meta-analysis. Both fixed and
random effects produced similar results, and due to differences in the methods of Pb
measurement, a random effects model is preferred.
Study
Fixed effect modelRandom effects modelHeterogeneity: I2 = 78%, τ2 = 0.1294, p < 0.01
Bocca 2005Bocca 2006Lee 2012McIntosh 2012Giacoppo 2014Fathabadi 2018Yang 2018
Total
399
60 28 80 19 15 27170
Mean
4.734.551.901.671.40
22.222.50
SD
2.25002.53000.97000.41001.1090
28.57001.3500
ExperimentalTotal
556
44 30
130 24 10 54
264
Mean
5.095.652.191.902.467.882.36
SD
2.01002.23000.94000.94000.91096.63001.0200
Control
−1.5 −1 −0.5 0 0.5 1 1.5
Standardised MeanDifference SMD
−0.03−0.12
−0.17−0.46−0.30−0.30−0.99
0.820.12
95%−CI
[−0.17; 0.10][−0.44; 0.20]
[−0.56; 0.22][−0.98; 0.07]
[−0.58; −0.02][−0.90; 0.31]
[−1.84; −0.13][ 0.34; 1.30]
[−0.07; 0.31]
(fixed)
100.0%−−
11.3%6.3%
21.9%4.7%2.4%7.4%
46.1%
Weight(random)
−−100.0%
15.7%13.3%17.7%11.8%
8.3%14.0%19.1%
Weight
Figure 5.1: Meta-analysis of studies that compared whole blood lead in Alzheimer’s disease compared tocontrols, and where group means and standard deviations were available.
There was significant heterogeneity in the whole blood analysis (I2 = 78%, tau =
0.1294, p < 0.01). To attempt to reduce heterogeneity, the study that used a different
method of Pb quantification, as well being the only study with significantly higher Pb in
AD, was removed (Fathabadi et al. 2018). The analysis was repeated but there was no
86
Study
Fixed effect modelRandom effects modelHeterogeneity: I2 = 10%, τ2 = 0.0033, p = 0.36
Basun 1991Bocca 2005Bocca 2006Lee 2012Park 2014Gonzalez 2014Hare 2016Paglia 2016
Total
514
12 60 28 80 64 30206 34
Mean
4.410.440.582.001.520.050.260.13
SD
3.73000.27000.24002.70001.41000.03530.60400.0900
ExperimentalTotal
1124
25 44 30 130 67 30 758 40
Mean
4.680.520.551.701.960.050.300.16
SD
3.19000.22000.19001.20001.31000.03080.93800.1700
Control
−0.6 −0.2 0 0.2 0.4 0.6
Standardised MeanDifference SMD
−0.05−0.06
−0.08−0.32
0.140.16
−0.320.13
−0.04−0.21
95%−CI
[−0.16; 0.05][−0.18; 0.06]
[−0.77; 0.61][−0.71; 0.07][−0.38; 0.65][−0.12; 0.44][−0.67; 0.02][−0.38; 0.63][−0.19; 0.11][−0.67; 0.25]
(fixed)
100.0%−−
2.5%7.7%4.5%
15.2%10.0%
4.6%49.9%
5.6%
Weight(random)
−−100.0%
3.1%9.1%5.4%
16.7%11.5%
5.6%41.7%
6.8%
Weight
Figure 5.2: Meta-analysis of studies that compared serum lead in Alzheimer’s disease compared to controls,and where group means and standard deviations were available.
difference in the main result, and heterogeneity remained significant (I2 = 63%, tau2 =
0.0608, p = 0.02).
5.3.2.2 Cerebrospinal fluid
The scoping search identified 2 case-control studies that measured Pb in CSF, which
presented data from the same participants, and which were also identified in the tar-
geted search (Gerhardsson et al. 2008; Gerhardsson et al. 2009). The targeted search
found a third paper from this group (Gerhardsson et al. 2008; Gerhardsson et al. 2009;
Gerhardsson et al. 2011). The series of papers reported on the Malmo Alzheimer Study,
and provided group means for CSF Pb, serum Pb and the quotient of CSF/serum.
The authors described 174 participants with AD, 90 participants with AD and
minor vascular changes, and 54 control participants. The groups were similar in age.
Cerebrospinal fluid was analyzed for metal concentrations including Pb, and medians
and ranges were reported. Both AD groups had significantly lower CSF Pb than the
87
control group (Gerhardsson et al. 2008; Gerhardsson et al. 2009), and the quotient of
Pb in CSF/serum was lower in AD (Gerhardsson et al. 2011).
Two additional studies were identified in the targeted search, for 5 in total. Basun et
al. (Basun et al. 1991) reported CSF Pb in a subset of a larger study, with measurements
in 13 participants with AD and 11 HC. There was no significant group difference found.
The oldest study from 1983 did not detect significant levels of CSF Pb in either group
(Hershey et al. 1983). Without multiple studies reporting means and standard deviations,
a quantitative meta-analysis of was not possible.
5.3.2.3 Hair and nail
The scoping review identified a single study that reported group differences in hair and
nail Pb in AD compared to controls. The specific search identified 2 additional older
studies which compared dementia groups to control groups, but the dementia groups
were of mixed neurodegenerative conditions and not specific to AD. The single included
study found a significantly lower level of Pb in hair but not nail Pb (Koseoglu et al.
2017).
5.3.2.4 Urine
A single identified study reported measuring urinary Pb in AD compared to controls and
did not report values or a group difference (McIntosh et al. 2012).
5.3.2.5 Postmortem
The 3 identified postmortem studies did not find associations of Pb in AD. Szabo et
al. measured Pb, among other metals, deposited in the frontal cortex and in ventricular
fluid of deceased individuals, comparing a group of 17 with AD to a control group of 15.
There was no significant difference in Pb levels between the groups (Szabo et al. 2016).
88
Haraguchi et al. reported a postmortem study that included 4 cases of AD, 6 cases of
diffuse neurofibrillary tangles with calcification (DNTC) and 9 elderly controls. This
small study did not identify a significant difference between AD and controls, though
DNTC was associated with elevated Pb, especially in the temporal cortex. In an older
study primarily about Parkinson’s, Pb was measured in the brain of 2 subjects with AD
but an association, if any, was not reported (Uitti et al. 1989).
5.3.2.6 Bone
Whereas longitudinal studies of aging that were identified in the scoping review included
bone Pb measurement, there were no case-control studies of AD that used the bone Pb
method, nor were any identified in the targeted search.
5.4 Discussion
Case-control studies can be well-suited to compare physiological differences associated
with disease when groups are well-matched and appropriate co-variates are selected. How-
ever, the onset of AD may begin decades prior to the emergence of the clinical symptoms
on which the diagnosis is based. In the case of Pb exposure as a potential contributor
to AD, this time difference may be critical, as early-life exposure to Pb may have la-
tent effects via epigenetic changes, as conceptualized in the Latent Early-life Associated
Regulation (LEARn) model (Lahiri et al. 2009; Maloney & Lahiri 2016). The model is
supported by biochemical evidence in animal experiments demonstrating that early-life
exposure to Pb induces amyloid precursor protein (APP) expression transiently and also
when the animals were older in the absence of ongoing Pb exposure, in combination with
accumulation of amyloid plaques (Basha 2005). In contrast, Pb exposure in old age did
not induce similar changes.
Therefore, any marker of Pb exposure that does not reflect earlier life exposure may
be of limited utility in clarifying its potential role in the etiology of AD in cross-sectional
89
designs. For this reason the preferred measure of exposure is K-shell X-ray fluorescence
measurements of bone Pb, which reflects chronic exposure, with a half-life of bone Pb
in the order of decades (McNeill et al. 2017). By comparison, although markedly more
accessible, blood Pb is unlikely to be a useful biomarker for this purpose, given the half-
life around 1 month. While to some extent blood Pb may reflect Pb released from longer
term bone stores, it is likely more closely related to recent exposure (McNeill et al. 2017).
With blood Pb reflecting an unknown combination of recent exposure, remote ex-
posure and other individual factors, it is difficult to know how to interpret the blood
Pb studies of AD. The mixed and uncertain origin of blood Pb may help explain the
mixed and mostly non-significant results of the blood Pb studies. Even more limited,
serum Pb levels, given their extremely low levels, are particularly vulnerable to error by
contamination, and may explain the higher levels observed in some studies (Rodushkin
& Ödman 2001). Despite blood Pb being the most studied Pb measurement in direct
AD research, on its own, it is unlikely to be a useful marker of chronic Pb exposure.
Our hypothesis is based on the directionality that Pb exposure increases the risk
of developing AD. However, blood Pb, in addition to reflecting acute exposure, is also
affected by physiological differences, and physiological changes of AD could affect Pb
blood levels, given that metabolic changes have been described in AD for other metals
(Ayton et al. 2015).
Our meta-analysis contradicts results from another meta-analysis of blood Pb levels
in AD, which concluded that blood Pb is lower in AD. Our meta-analysis incorporated
several additional studies, including a recent study with the largest sample of AD par-
ticipants among the whole blood studies (Yang et al. 2018). We excluded data from a
thesis and data that did not include published or author-provided means and standard
deviations. Further, in the previous meta-analysis, the authors combined whole blood
and serum in a single meta-analysis of 10 studies. To address this issue of heterogeneity,
we reported 2 separate whole blood and serum meta-analyses with 7 and 8 studies, re-
spectively, which included data from studies where both serum and whole blood levels
were reported. Both analyses suggest, that if there is an association of blood or serum
90
Pb with AD, it is likely to be a very small difference and unrelated to chronic or earlier
life exposure.
Few studies were available that employed Pb methods of measurement other than
blood. Cerebrospinal fluid Pb may be lower in AD, which may relate to lower acute
exposure or physiologic differences in AD. Lead in human nails may reflect exposure
approximately 1 year prior to collection, depending on individual growth rates, and are
not correlated with longer term exposure (Grashow et al. 2015), as would be required to
ascertain the role of Pb in the development of AD. The single study that measured Pb in
nails found no difference in AD. The same study did find lower levels of hair Pb among the
AD group (Koseoglu et al. 2017). The timing of exposure for hair Pb is highly dependent
on hair length, with hair growing approximately 1 cm per month (Keil et al. 2011). The
included study did not report on hair length. In summary, hair and nail measurements,
like blood, are indicative of current or recent Pb exposure and person-specific physiologic
factors rather than reflective of Pb exposure over the time course in which AD develops.
Intriguingly, the lack of significant Pb deposition in AD brains postmortem may suggest
that if Pb exposure does increase the risk of AD, it is not via long term deposition in the
brain, which is consistent with proposed models of how Pb may increase the risk of AD
(Lahiri et al. 2009).
5.4.1 Limitations
This is the first systematic review that attempts to summarize all case-control studies that
measure Pb in AD. With respect to the ability to draw conclusions about the role of Pb
in the development or pathophysiology of Alzheimer’s, this review is significantly limited
by the available literature. The methods used to measure Pb exposure predominantly
reflect recent exposure and physiologic factors rather than capture exposure during the
period in which the AD-related pathophysiological changes began. Our search strategy,
which explicitly identified articles about Pb, has the potential to miss articles where
Pb analysis was a secondary or peripheral aspect of the research, such that it was not
91
mentioned in the abstract. This limitation, in combination with publication bias, could
increase the risk of falsely concluding that there is a significant association with Pb even
if there is none.
5.5 Conclusions
Lead has been proposed as a potential cause or contributor to AD for at least 20 years
(Prince 1998), and while there is significant support for the hypothesis from animal stud-
ies, the evidence drawn from human studies including case-control designs, is method-
ologically lacking. Most case-control studies compared Pb in blood, which may only min-
imally, if at all, reflect exposure during the time in which the individual developed AD.
There were no bone Pb case-control studies, which may be the best available biomarker
of chronic Pb exposure.
5.5.1 Next steps
Given the time course of AD onset and development, occurring decades prior to the onset
of symptoms, bone Pb measurement is likely the most useful biomarker of Pb exposure.
A case-control study comparing participants with AD, particularly if they are early in
the disease course, using bone Pb K-XRF measurement may help clarify the role of Pb
exposure as a risk factor for AD. A similar case-control design with individuals at high risk
of AD, such as those with mild cognitive impairment, could also help to clarify the role
of Pb by increasing the chance that the bone Pb measurement and disease onset reflect
the same time period. Including other known risk factors, particularly measurements of
hypertension and vascular risk factors, would be important to include as covariates, and
to assess for interactions and as mediating variables. Confirming the presence of AD-
related pathological changes, such as amyloid and/or tau positron emission tomography,
could identify individuals with AD-pathology early enough in the course of illness that
bone Pb results become more meaningful.
92
Future longitudinal studies that have early-life Pb measurement or bone-Pb mea-
surement reflecting chronic exposure could plan to test the association between Pb expo-
sure and incidence of AD when the population is of sufficient age. Likewise, investigators
from previous longitudinal epidemiology studies with existing early-life or bone Pb mea-
surements could consider whether their study populations have advanced enough age to
warrant either reassessment with AD-related biomarkers or screening for incidence of AD
diagnosis.
Cohort study designs that compare groups of individuals with higher Pb exposure,
such as those working in Pb battery factories could compare the incidence of AD in
these groups in comparison to a control group matched for other known AD risk factors,
though the complexity in both prospective and retrospective designs, due to the time
interval between exposure and outcome is likely why cognition, rather than incidence of
AD is reported in the literature (Shih et al. 2007).
Despite the longstanding hypothesis that Pb is implicated in AD, there is little
direct evidence from human studies. Evidence supporting a link between Pb and AD in
humans includes the known general neurotoxicity of Pb, the association of chronic Pb
exposure with cognitive decline, and the possible decreasing incidence of AD (Satizabal et
al. 2016) with coincidental decreasing environmental contamination by Pb. Further, an-
imal studies strongly implicate Pb in the development of pathological changes associated
with AD (Bihaqi et al. 2018; Lahiri et al. 2009; Zawia & Basha 2005). The relation-
ship is important to clarify, as it would inform public health prevention and intervention
strategies, as well as our understanding of the pathophysiology of AD.
5.6 Acknowledgements
The authors thank Dr. David Chettle for his input on a draft of the manuscript, and
Dr. Patrick Parsons and colleagues for generously sharing raw data from a study that
met inclusion criteria in the meta-analysis. We also thank the anonymous reviewers for
93
their helpful feedback, which improved this manuscript.
94
Chapter 6
Water-supply Lithium,
Environmental Lead Exposure, and
Illness
This chapter is modified and reproduced from published work copyrighted by Elsevier
(Brown et al. 2018). See the Contributions section of this thesis for details.
6.1 Introduction
Lithium, the third element on the periodic table, is an established medication. It is a
mood stabilizer: a first-line treatment for bipolar disorder (Yatham et al. 2013), and
it is also used in unipolar depression. Treatment with Li reduces the risk of suicide
across psychiatric diagnoses, with high quality evidence in bipolar disorder and unipolar
depression (Cipriani et al. 2013). Clinically, typical doses may range from 600 mg to 1800
mg per day and individual doses are based on clinical effect, side effects, and blood levels
(Kessing et al. 2017b; Liaugaudaite et al. 2017). At lower doses, Li may also reduce the
progression to dementia from mild cognitive impairment (Kessing et al. 2017b).
95
Not only found in the medicine cabinet, Li is abundant in the Earth’s crust. Present
in minerals in varying amounts in different regions, Li dissolves into groundwater and is
commonly found in drinking water. In turn, Li is consumed by humans who drink the
water and eat the grains and vegetables that take it up (Schrauzer 2002). Daily intake of
Li therefore ranges by location and diet, with estimates of mean daily intake in the range
of 348–1560 �g/day (0.348–1.560 mg/day) (Schrauzer 2002), i.e. two to three orders of
magnitude lower than effective clinical doses (see Figure 6.1 for visualization).
High clinical doseLow clinical doseHigh environmental exposureLow environmental exposure
Figure 6.1: Relative size comparisons of clinical Li doses (Liaugaudaite et al. 2017) compared to estimatedmean environmental daily intake (Schrauzer 2002), represented with dose proportional to circle areas.
Nonetheless, mounting evidence from epidemiological studies that correlate
drinking-water Li levels with health outcomes suggests that exposure to microdose
environmental Li may be beneficial. For example, higher drinking water concentrations
are shown to correlate with lower rates of suicide, homicide, and dementia (Mauer et al.
2014; Vita et al. 2015). Despite its status as an established effective medication, the
mechanism underlying the clinical benefits of Li is uncertain (Alda 2015). Lithium has
broad effects on cellular signalling pathways in the brain involving glycogen synthase
96
kinase 3 (GSK-3), cyclic adenosine monophosphatase response element binding protein
(CREB), and Na+-K+ adenosine triphosphatase (ATPase), with influences on calcium
homeostasis (Alda 2015). The beneficial mechanism of action may differ in different
populations [8]. As with bipolar disorder, the mechanism of the purported beneficial
effects of microdose Li is uncertain (Kessing et al. 2017b). Meanwhile, Pb is described
as ubiquitous in the environment in varying amounts, detectable even in regions of the
arctic (World Health Organization 2011). Lead is noted to be a “cumulative general
poison”, and is neurotoxic (World Health Organization 2011). The Centre for Disease
Control reports that for children, no safe lower limit of Pb blood level has been found
(Advisory Committee on Childhood Lead Poisoning Prevention of the Centers for
Disease Control and Prevention 2012). Lead toxicity is a global public health problem:
0.2% of deaths and 0.6% of disability-adjusted life years are attributed to Pb exposure,
surpassing urban outdoor air pollution and climate change (Organization 2009).
6.1.1 Hypothesis
A parsimonious mechanistic explanation of the effects of microdose Li would account for
the breadth of its apparent effects. A possible clue to that mechanism may be in the
broad and complementary effects of Pb, another environmental element. We hypothesize
that if the harms of Pb exposure are opposite to the benefits of Li exposure, then the
benefits of Li may be due to mitigation of the toxicity of Pb. This possibility is important
to clarify, as it would have implications on any recommendations to supplement Li.
The hypothesis would be supported if: the harms of Pb are opposite to the benefits
of Li; Pb and Li co-occur in the environment where such harms and benefits are observed;
Pb and Li have effects on shared biological processes; there is experimental evidence
demonstrating Li mitigates the neurotoxicity of Pb. The aim of this paper is to review
the literature for evidence supporting or refuting these empirical possibilities in order to
clarify the relationship and determine the next steps that may be required.
97
6.2 Methods
This paper aims to explore the hypothesis that Li mitigates the negative health impacts
of Pb. First, in Part 1, a systematic review was done to identify the health outcomes asso-
ciated with environmental Li exposure. Second, in Part 2 the health risks of Pb exposure
were reviewed. Third, potential causal connections between Li and Pb were identified
by reviewing experimental studies (Part 3). And finally, potential areas of overlap in
the mechanisms of action in Li and Pb were identified by highlighting physiologic and
biologic studies (Part 4).
6.2.1 Part 1: Systematic review of the health impacts of
exposure to environmental lithium
6.2.1.1 Literature search
A systematic search was completed using MEDLINE with the PubMED interface on
November 16, 2017 with the following query: ((“lithium”[MeSH Terms] OR “lithium”[All
Fields]) AND (“water”[MeSH Terms] OR “water”[All Fields] OR “drinking water”[MeSH
Terms] OR (“drinking”[All Fields] AND “water”[All Fields]) OR “drinking water”[All
Fields])). The human filter was applied. By using the human filter, articles that have
not yet been indexed with MeSH subheadings are excluded, so there is a risk of missing
the most recently published articles. Therefore, the search was repeated without the
filter to identify potential recent articles from January 1, 2017 to November 16, 2017.
These articles were combined with the first search and duplicates removed.
Entries were included if they described a peer-reviewed, primary literature study
that reported on a direct or indirect measure of drinking water Li and a psychiatric or
non-psychiatric medical outcome. The rationale for considering all health outcomes is
the arbitrary distinction between mental and medical illness, as well as the potential
impact of medical problems on illnesses categorized as psychiatric (for example the im-
98
pact of thyroid diseases on mood disorders and vascular disease on cognitive disorders).
Entries were excluded if they did not contain an abstract; were not in English; or were
reviews (which were separately retrieved for background and to identify any missed pri-
mary articles), commentaries, letters, or hypotheses if they did not contain original data.
Interventional studies that administered Li as a treatment were excluded.
6.2.2 Part 2: What are the psychiatric impacts of environ-
mental lead exposure?
Compared to the potential benefits of exposure to environmental Li, Pb toxicity is an
established fact in the medical and scientific literature and a topic of great importance
to public health. Therefore, for efficiency and accuracy, Part 2 relied on previously-
published, recent high-quality reviews. A non-systematic search for recent scoping re-
views from governmental and non-governmental bodies such as the World Health Orga-
nization (WHO) and the government of Canada was completed. The reviews were read
and summarized to answer specific questions relating to the hypothesis. When there was
insufficient information in the identified reviews, primary source literature was used to
answer the question and critically appraise the articles individually.
6.2.2.1 Is lead present in the environment where benefits of
lithium have been suspected?
If exposure to Li through drinking water reduces the risk of mental and physical health
conditions, and if the mechanism is by mitigating harm from environmental Pb exposure,
then it is necessary but not sufficient that environmental Pb exposure be associated with
conditions in which drinking-water Li is purported to have benefits. To answer this
question, the reviews of environmental Pb toxicity were summarized, and articles found
in the systematic search from Part 1 were reviewed to identify whether they also included
a measure of Pb exposure.
99
6.2.2.2 Is exposure to lead associated with effects opposite to
the effects associated with lithium in humans?
If Li mitigates the toxic effects of Pb, then the effects of Pb should be opposite to those
associated with Li. As Pb toxicity cannot be evaluated by randomized controlled trial,
large, well-designed, well-controlled observational studies are best suited to answer these
questions. Subsequent to identifying the associated effects of Li, the effects of Pb were
noted on the basis of the available reviews, and where information on a specific effect was
missing from the identified reviews on Pb, a targeted literature search was completed,
and the identified articles were critically appraised.
6.2.3 Part 3: Does lithium exposure protect against lead
toxicity?
Experimental methods such as randomized controlled trials are the gold-standard de-
sign to ascertain causality. A literature search identified experimental studies (including
animal studies), which demonstrate causality, where Li has been used to mitigate Pb
exposure. A PubMed search was conducted using the following query: (“lithium”[MeSH
Terms] OR “lithium”[All Fields]) AND (“lead”[MeSH Terms] OR “Pb”[All Fields]). Ad-
ditional articles were retrieved by checking the citations of previously identified articles
and with additional specific searches.
6.2.4 Part 4: Biological mechanisms
Systematically reviewing the basic science literature that explores pharmacological and
physiological mechanisms of Li and Pb is beyond the scope of this report. Therefore,
recent reviews and targeted searches were used to identify and illustrate possible mecha-
nisms by which Li and Pb may interact, which would be consistent with the hypothesis
that Li alters the effects of Pb.
100
6.3 Results
6.3.1 Part 1: Health effects of environmental lithium expo-
sure
The results of the systematic review are summarized in a PRISMA (Moher et al. 2009)
diagram (Figure 6.2). The initial search produced 3195 entries, which was reduced to 822
by application of the human filter. An additional 194 recent abstracts were identified, for
a combined 1006 unique entries to review. Based on the above inclusion and exclusion
criteria following review of the titles and abstracts, 60 abstracts were included and 946
excluded. Of the 60 abstracts, 56 full articles were retrievable. Full text versions for 4
older studies (1970–1988) were not retrievable and were therefore excluded. Following
retrieval, articles were read, and 32 articles met inclusion criteria.
A systematic review completed in August 2013 identified 11 epidemiologic studies
of trace Li doses (Mauer et al. 2014). Ten of these were identified by the above search.
A systematic review of environmental Li and suicide identified 9 studies in a search
completed in December 2013 (Vita et al. 2015). All 9 were captured by the above
search.
The 32 primary literature studies identified were categorized based on the health
outcome: suicide, homicide, dementia, other psychiatric effects, and non-psychiatric med-
ical effects.
6.3.1.1 Suicide
Of the 32 articles, 15 publications tested the association between suicide rates and Li
levels measured in drinking water using 12 unique samples, as 4 publications reported
different analyses of the same data set (Helbich et al. 2015; Helbich et al. 2013; Helbich
et al. 2012; Kapusta et al. 2011). The articles are summarized in Table 6.1. All studies
reported on rates of suicide to test for an effect of levels of lithium in drinking water. The
101
Figure 6.2: Results of systematic review of health impacts of environmental lithium.
studies cover countries from 3 continents: Austria, England, Greece, Denmark, Lithuania,
Italy, Japan, and the United States.
Table 6.1: Studies that report on the association between suicide rates and environmental lithium.
Citation
Location
Years
Samples
Li Mean (SD);
Range (µg/L) Covariates
Association of Li
with incidence of
suicide
Helbich
2015
Austria
2005–2009
6460
10 (10); NR Li prescription Inverse in M; not in F; no impactof Li prescription
Helbich
2013
As above 10 (11); NR Altitude, populationdensity, income,unemployment,religion, care providersper capita
Inverse in low altitudes; positiveassociation in high altitudes
102
Citation
Location
Years
Samples
Li Mean (SD);
Range (µg/L) Covariates
Association of Li
with incidence of
suicide
Helbich
2012
As above 11.3 (27); NR Inverse
Kapusta
2011
As above 11.3 (27); NR Inverse
Bluml 2013 Texas
1999–2007
Unemployment,population density,income, ethnicity
Inverse
Giotakos
2013
Greece
1999–2010
3123
11.1 (21.16);
2.8–219.0
Inverse
Ishii 2015 Japan
(Kyushu)
2010–2013
149
4.2 (9.3); NR Demographics,temperature
Inverse overall; inverse among M;no association among F
Kabacs
2011
England
2006–2007
434
NR; < 1–21 No association
Knudsen
2017
Denmark
1991–2012
47
11.6 (6.8); NR No association
Liaugaudaite
2017
Lithuania
2009–2013
151
10.9 (9.1); NR Inverse among men; noassociation among F
Ohgami
2009
Japan (Oita)
2002–2006
22
NR; 0.7–59 Inverse overall; marginal in F
103
Citation
Location
Years
Samples
Li Mean (SD);
Range (µg/L) Covariates
Association of Li
with incidence of
suicide
Pompili
2015
Italy
1980–2011
18
5.28; 0.11–60.8 Inverse association in 1980–1989but not later decades
Schrauzer
1990
Texas
1978–1987
157
NR; 0–160 Inverse
Shiotsuki
2016
Japan
(Hokkaido,
Kyushu)
2010–2011
27
3.8 (5.3);
0.1–43
Meteorological
data
Inverse in M, not F
Sugawara
2013
Japan
(Aomori)
Unreported
153
NR; 0–12.9 Medical institutions,unemployment
Inverse in F, not M; NS withco-variates
F = Female; M = Male; NR = Not Reproted; NS = Not Significant; SD = Standard
Deviation.
Only 2 of the 15 studies did not identify any relationship between Li level and
suicide. The two negative studies were from England (Kabacs et al. 2011) and Denmark
(Knudsen et al. 2017) where the range of drinking water Li was found to be low (0–21
�g/L and 0.6–30.7 �g/L respectively).
Six studies used covariates to account for potential confounds that may impact
suicide rates. Controlling for sociodemographic covariates did not account for the effect
of Li in the studies identified. However, one study used altitude as a covariate (Helbich
104
et al. 2013) and found that overall higher Li is associated with lower suicide, but that
the relationship was reversed in areas of high altitude, and cited previous research that
had associated altitude with suicide rates. In their previous systematic review, Vita et
al. (Vita et al. 2015) concluded that drinking water Li may be associated with a reduced
risk of suicide. Since that publication, there have been an additional 6 studies of suicide
risk, of which only one did not find an association (Knudsen et al. 2017).
Sex differences in the association of drinking water Li and suicide were described. In
subgroup analyses, several papers reported a negative association in men and a weaker
or no association in women (Liaugaudaite et al. 2017; Helbich et al. 2015; Ishii et
al. 2015; Ohgami et al. 2009; Shiotsuki et al. 2016), whereas only one study saw a
possible association in women and no association in men (Sugawara et al. 2013). More
frequent use violent means of suicide among men has been proposed as an explanation
for the observed sex differences (Liaugaudaite et al. 2017; Ishii et al. 2015; Shiotsuki et
al. 2016), with Li suspected to reduce suicide via reduction of violence and aggression
(Smith & Cipriani 2017).
6.3.1.2 Homicide
Two studies reported on the association of drinking water Li and homicide, one of which
also investigated suicide and also appears in Table 6.1 (Schrauzer & Shrestha 1990). In
this study, 27 Texas counties were grouped based on levels of Li in the drinking water
(0–12, 13–60, and 70–160 �g/L) and group differences in suicide and crime rates were
compared with t-tests. With respect to homicide, the counties with the highest levels of
Li had the lowest homicide rates (7.5 per 100,000) as compared to the medium (13.4 per
100,000) and low (16.9 per 100,000) Li counties which was statistically significant.
More recently, Giotakos et al. (Giotakos et al. 2015) reported higher drinking
water Li levels to be associated with lower homicide rates when looking at the same 34
Greek prefectures as in the group’s earlier study reporting an association with suicide
(Giotakos et al. 2013). Using linear regression, there was a statistically significant neg-
105
ative relationship between homicide and drinking water Li. However, when the analysis
was weighted by population, the direction of the relationship remained, but was no longer
statistically significant.
6.3.1.3 Dementia
Two studies reported on rates of dementia. The retrospective population-based study by
Kessing et al. is notable for the large sample size (73,731 cases with dementia and 733,653
controls) and long study period (1995–2013) (Kessing et al. 2017a). The group also linked
each person’s address through the study period to a nationwide map of Li levels computed
from 151 measurements. The incidence rate ratio was calculated for individuals in 4
groups based on mean Li exposure over the range of 0–27.0 �g/L. The lowest rates of both
vascular and AD were found in the groups with the highest Li exposure. However, there
was a small but statistically significantly higher rate of dementia in the second-lowest
exposure group. The authors acknowledged that unmeasured social or environmental
factors could be confounding the results.
Fajardo and colleagues also sought to identify the relationship between drinking
water Li and changes in mortality rates from AD (Fajardo et al. 2017). In their 2017
study, they calculated changes in age-adjusted mortality rates from AD between 2000–
2006 and 2009–2015 for 155 Texas counties. They included extensive covariates that
have been linked to dementia including sociodemographic factors as well as pollution,
physical activity, obesity and presence of type II diabetes. They had access to 6180
water samples. As expected, they found an overall increase in mortality from Alzheimer’s
dementia over time: 27%. They also found that higher levels of Li were negatively
correlated with increases in mortality rates. The association remained when controlling
for sociodemographic factors but not when physical activity, obesity, and type II diabetes
were included in the analysis. They also found a negative association between Li levels
and obesity and type 2 diabetes. A post hoc analysis also showed that in counties with
fewer than 20 Alzheimer’s-related deaths over the time period, Li levels were higher.
106
6.3.1.4 Other psychiatric effects
In a study of similar design as the group’s dementia paper, Kessing et al. also reported on
the association of bipolar disorder with drinking water Li levels and found no association
(Kessing et al. 2017b).
In the only identified study to specifically look at adolescent mental health, Ando et
al. correlated questionnaire responses of 3040 adolescents in Kochi, Japan with drinking
water Li and found a significant negative association with depressive symptoms and
violence, but no association with suicidal ideation or self-harm. They used a multivariate
regression analysis and included age, sex, size of school, and whether they were living
with parents as covariates.
Two studies found that drinking water Li levels were associated with psychological
traits. Matsuzaki and colleagues found that Li levels were positively correlated with
hyperthermic temperament scores and negatively correlated with depressive temperament
scores (Matsuzaki et al. 2017). In the only identified study that measured endogenous Li
levels as a surrogate measure rather than drinking water Li, Norra et al. found Li levels
to be negatively correlated with emotional lability (Norra et al. 2010).
In the oldest identified study looking at psychiatric outcomes and drinking water Li,
higher levels of drinking water Li were associated with lower rates of psychiatric hospital
admissions as well as the diagnosis of psychosis, neurosis, and personality disorders in
Texas (Dawson 1970).
6.3.1.5 Non-psychiatric medical effects
Zarse and colleagues looked at the connection between Li in drinking water Li and all-
cause mortality in Oita, Japan. In 1,206,174 people from 18 municipalities, there was a
strong negative correlation between water Li levels and all-cause mortality (Zarse et al.
2011). This relationship persisted even after controlling for suicide.
107
A negative correlation between municipal water Li levels and atherosclerotic heart
disease in Caucasians in American cities was reported (Voors, A.W. 1970).
A group from Argentina published 4 studies on the health correlates of environ-
mental Li exposure (Broberg et al. 2011; Harari et al. 2016; Harari, Bottai, et al. 2015;
Harari, Langeén, et al. 2015). They noted that Argentinian Andes mountains are as-
sociated with very high levels of drinking water Li (e.g. up to 1005 �g/L) (Broberg et
al. 2011). They used urinary excretion of Li as a surrogate measure of environmental
Li exposure and found an association with thyroid function (T4) within a normal range
in healthy individuals (Broberg et al. 2011). In a sample of pregnant women, blood Li
levels were positively correlated with thyroid stimulating hormone (Harari, Bottai, et al.
2015), negatively correlated with fetal measures and birth weight (Harari, Langeén, et
al. 2015), and associated with levels of vitamin D3, calcium and magnesium (Harari et
al. 2016).
Budd and Rossof were unable to replicate an association of leukemia with drinking
water Li (Budd & Rossof 1980). No association was found between drinking water Li
and the incidence of anencephaly (Elwood 1977).
6.3.2 Part 2: Health effects of environmental lead expo-
sure
The WHO has published extensive background documents and guidelines about Pb toxi-
city, including specifically for drinking water quality (World Health Organization 2011).
In the WHO report on global health risks, 0.2% of deaths and 0.6% of disability-adjusted
life years are attributed to Pb exposure, surpassing urban outdoor air pollution and cli-
mate change (Organization 2009). Given the global scope and severity of the problem,
individual countries, including Canada in a detailed 2017 report (Health Canada 2017),
have published extensive reviews for consultations and guidelines.
108
6.3.2.1 Presence of lead in the environment
All of the locations identified in Part 1, in Europe, America and Asia are affected by
environmental Pb contamination (World Health Organization 2011).
A major source of Pb in the environment is the global use of leaded gasolines
introduced in the 1920s, with consumption peaking in the 1970s, and currently phased
out in most countries (World Health Organization 2011; Nriagu 1990). As a result,
concentrations of Pb in the air and environment vary depending on location and year
(World Health Organization 2011). For example, in Canada, mean air concentration at
measuring stations decline from 0.74 to 0.10 �g/m3 from 1973 to 1989 (World Health
Organization 2011). The WHO notes that water has become the “largest controllable
source of Pb exposure in the USA”, as a result of mineral dissolution and plumbing
systems (World Health Organization 2011). In addition to air and water contamination,
humans are significantly exposed to Pb through paint, dust, soil and food (World Health
Organization 2011).
As the toxic effects of Pb are both acute and cumulative, present as well as past
Pb contamination is relevant for present health concerns. For example, older Canadians
had significantly higher blood Pb levels than younger Canadians, and there has been a
significant reduction in blood levels over time (Health Canada 2017).
6.3.2.2 Impacts of chronic environmental lead exposure on
health outcomes identified in Part 1
Both the WHO and Health Canada reports highlight the neurological impacts of chronic
exposure (World Health Organization 2011; Health Canada 2017). Chronic Pb exposure
has been shown to impact multiple systems and is associated with neurologic, renal and
hematologic disorders. High levels of Pb exposure have been associated with low IQ in
children, and this is noted to be the most consistent finding among children (World Health
Organization 2011). This association was also recently reported in the well-characterized,
109
prospective, longitudinal Dunedin cohort, where childhood Pb exposure was associated
with declining IQ, and decreasing socioeconomic status, after adjusting for confounds
(Reuben et al. 2017).
6.3.2.2.1 Dementia Epidemiologic studies implicate Pb in cognitive deficits includ-
ing memory loss, impaired reaction time, and impaired verbal concept formation (Health
Canada 2017). Exposure to environmental Pb is associated with cognitive decline in
older adults, and the relationship appears to be causal (Shih et al. 2007). Lead has been
proposed as an etiologic agent in Alzheimer’s dementia, the most common form of demen-
tia (Prince 1998), and the possible nature of the connection has recently been reviewed
(Bakulski et al. 2012). In support of the relationship between Pb and the development of
dementia, Pb exposure has been associated with Alzheimer-related pathological changes
in animal studies (Health Canada 2017).
6.3.2.2.2 Suicide, homicide, and crime Suicide, homicide and crime were not
specifically mentioned in the WHO or Health Canada reports on the impacts of environ-
mental Pb (World Health Organization 2011; Health Canada 2017).
Whether Pb exposure is associated with an increased risk of suicide was directly
tested in a large cohort study of 81,067 Pb-exposed workers in South Korea. The cohort
was divided based on blood Pb levels into a highand low-Pb groups (> 20 �g/dl and <
10 �g/dl, respectively), and analyses were adjusted for age and exposure to other metals.
The causes of death were compared between the groups. All-cause mortality was higher
in the high-Pb group for both sexes. The high-Pb group had significantly higher risk
of suicide in men. Given the low incidence of suicide, there were only 13 deaths in the
male population, nonetheless a significant association with Pb was identified (Kim et al.
2015).
Environmental Pb exposure through air pollution has been associated with in-
creased rates of homicide in the United States after accounting for other environmental
pollutants and sociodemographic factors (Stretesky & Lynch 2001).
110
As part of an ongoing project to measure blood Pb levels in St. Louis City, Mis-
souri, a recent ecological study assessed the relationship between areas where high blood
Pb levels are found and rates of violent crimes, on the basis that Pb exposure risk is
concentrated geographically (Boutwell et al. 2017). The authors used blood Pb level
data from 59,645 children. This indirect study found that homicide and violent crimes
were significantly associated with areas with higher blood Pb levels.
6.3.2.2.3 Non-psychiatric medical effects In a prospective study of 14,289 adults
from the National Health and Nutrition Examination Survey (NHANES-III), Lanphear
et al. (2018) correlated blood Pb level with cardiovascular outcomes over a mean follow
up of 19.3 years. Despite generally low blood levels and only a single measurement, the
authors identified large and statistically-significant associations with mortality, reporting
population attributable fractions of blood Pb to be 18.0% for all-cause mortality, 28.7%
for cardiovascular disease mortality, and 37.4% for ischemic heart disease mortality (Lan-
phear et al. 2018).
6.3.3 Part 3: Link between lithium and lead
6.3.3.1 Epidemiologic and cross-sectional human studies
Very few studies identified examined both Li and Pb on health outcomes in humans. On
the basis of the systematic review from Part 1, no abstract reviewed indicated drinking
water Li and health outcomes included a measure of Pb. A study investigating the
impact of trace elements on ischemic stroke found higher serum Li and Pb in stroke
patients compared to HC and did not test for interaction (Skalny et al. 2017). A Turkish
study investigated the impact of Li and Pb in drinking water on body composition and
found very low levels of both in the water supplies and no significant association with
either (Cetin et al. 2017).
111
6.3.3.2 Animal studies
On the basis of two translated abstracts, a group from China reported that administering
Li to rats mitigated Pb-induced hippocampal damage and memory impairment (Li et al.
2006), and neuronal damage (Yang et al. 2004).
Wang et al. sought to test whether pre-treatment with Li can mitigate the toxic
effects of Pb on the liver, spleen, kidney, and brain, and performed a number of experi-
ments in vitro and with mice (Wang et al. 2015). The main mouse experiment compared
four groups, with each group receiving different daily injections over 2 weeks: saline (con-
trol), Pb, Li, and a group with injection of Li then Pb 2 h later. On gross and microscopic
examination, they found that Li pre-treatment significantly mitigated the damage to the
spleen, liver and kidney seen in the Pb-only group. The Pb-only group performed signif-
icantly worse on a memory test than the Li pretreated group, which performed nearly as
well as controls.
In an experiment to determine the effects of Pb, Li and the combination on thyroid
function in rats, Singh and Dhawan found that, compared to controls, Pb treatment
increased thyroid 131iodine uptake, Li treatment decreased 131iodine uptake, and the
combination of Li and Pb caused the largest increase in 131iodine uptake (Singh & Dhawan
2000), demonstrating that Li and Pb have the potential to interact.
Another rat experiment showed that Li and Pb both alter the uptake of other trace
ions in rats, including an interaction between Li and Pb with respect to arsenic uptake:
Pb increased arsenic uptake significantly; Li alone did not affect arsenic uptake, but in
combination, Li mitigated the uptake of arsenic (Singh et al. 1994).
6.3.3.3 In-vitro studies
Banijamali and colleagues sought to test whether Li could mitigate the toxic effects of
Pb on non-adherent mouse bone marrow stem cells, a known target of Pb toxicity. They
found that Li administration significantly reduced cell apoptosis and necrosis induced by
112
Pb (Banijamali et al. 2016).
6.3.4 Part 4: Biological mechanisms
Recent reviews synthesize the vast research into the respective physiological impacts of
Pb and Li (Alda 2015; Mason et al. 2014). The mechanisms of each are broad and an
exhaustive review is beyond the scope of this report, but the processes are affected by
both elements and can be highlighted.
GSK-3 is implicated in the pathophysiology of dementia and involved in phosphory-
lation of tau, a hallmark neurolopathological feature of Alzheimer’s dementia. An exper-
imental study in rats showed that perinatal Pb exposure stimulated GSK-3, increasing
GSK-3 dependent tau hyperphosphorylation (Gassowska et al. 2016). Conversely, Li
inhibits GSK-3 activity (Alda 2015). In the study by Wang et al. described above, pre-
treatment with Li mitigated Pb-associated reduction of phosphorylated-GSK-3� (Wang
et al. 2015).
An additional mechanism that appears to be oppositely affected by Li and Pb is
Na+ /K+ ATPase. Lithium increases Na+ /K+ ATPase while Pb inhibits it (Alda
2015; Mason et al. 2014). Lithium inhibits the calmodulin-related activities of adenylate
cyclase and protein kinase A, while Pb activates protein kinase C and induces calmodulin-
dependent processes (Alda 2015; Mason et al. 2014). Both Li and Pb affect N-methyl-d-
aspartate (NMDA) receptors, influencing calcium and other ion transport in the brain.
Calcium homeostasis across the blood-brain barrier is tightly regulated, with cal-
cium adenosine triphosphatase (ATPase) being the main transporter. Lithium may re-
duce the activity of this transporter (Cho 1995). Additionally, Li stabilizes intracellu-
lar calcium via NMDA receptor antagonism and inhibition of inositol monophosphatase
(Wallace 2014). By comparison, Pb crosses the blood-brain barrier by substituting for
calcium, both being divalent cations.
113
6.4 Discussion
6.4.1 Lithium in drinking water (Part 1)
The systematic review in Part 1 is the most extensive and most recent conducted to date
and is a major strength of this report. This review highlights the growing evidence base
supporting a beneficial association of water-supply Li with various outcomes.
While some studies included covariates of possible confounds, many did not, which
is a limitation of this review. However, this elevates the importance of looking at the
consistency of the association across different countries, as certain geographic confounds
(e.g. a city high in poverty by chance also having low levels of drinking water Li) may
account for a spurious association in one study, but not in all, and can therefore be
accounted for in a systematic review. The most concerning potential confounds are those
where Li levels are systematically associated with psychiatric risk factors, but not in a
biological sense. For example, one study used altitude as a covariate and found that the
association between suicide and Li depended on altitude, but no other study included
this as a covariate. A meta-analysis, which would quantitatively combine the different
studies, may add further value, and may be feasible given the common reporting of Li
levels and suicide incidence ratios.
The evidence is strongest for suicide, where higher drinking water Li has been
associated with lower suicide rates across different continents and time periods in the
majority of studies. Fewer studies address the relationship with homicide, crime, but
there is support for this relationship as well. Two studies address dementia, with the
recent study by Kessing notable for its large sample size and robustness of measurements
of both Li and clinical variables, suggesting a negative correlation between drinking water
Li and dementia in Denmark. The findings would have been strengthened if the authors
had controlled for potential demographic, social and geographical confounding variables.
On the whole, the epidemiologic studies support the negative association between
drinking water Li and a reduction in psychiatric outcomes, most strongly for suicide
114
and dementia. The non-psychiatric medical effects are less studied, with only one in-
vestigating all-cause mortality and another for ischemic heart disease; though both were
positive studies. Future studies are warranted and should include social and geophysical
covariates, especially those that may vary with Li in geographically distinct regions, such
as altitude. An additional covariate that has not been considered could be the major
sources of food supply, as food is also a source of Li, and if the food is not made locally,
then it may reflect Li levels where it is grown.
6.4.2 Environmental lead exposure (Part 2)
Unlike in Part 1, the review of the effects of Pb exposure did not include an inclusive
systematic search of the primary literature and is a limitation of this report. However,
compared to drinking-water Li, the effects of environmental Pb are generally better stud-
ied and reviewed, given the known public health concerns.
Unlike with Li, where water is a main source of exposure, and where levels are
measurable and consistent within a large area served by a single water supply, estimating
Pb exposure is more complicated. Lead exposure by water is more associated with local
plumbing. Additional routes of exposure are also more varied within an area, and more
challenging to measure, such as air (e.g. proximity to a roadway), soil, and dust (e.g. from
Pb paint). Despite Pb accumulating in the body, measures that reflect long half-life are
invasive or costly (bone measurement by biopsy or X-ray fluorescence) as compared to
serum tests. Therefore, population studies of Pb are challenging and more susceptible to
ascertainment bias.
Lead is present in the environment, though levels in general are declining. Nonethe-
less, current generations have been exposed to its effects and there is sufficient evidence
to conclude that Pb could have affected individuals where a beneficial effect of Li was
noted. Likewise, since the pioneering work of Needleman (Roehr 2017), the general neu-
rotoxicity of chronic Pb exposure has been well established, especially in children, where
the relationship with IQ has been replicated in multiple longitudinal studies. Recent
115
reviews combining longitudinal, cross-sectional and animal studies suggest that Pb is
associated with cognitive impairment and specifically dementia in late life. Interestingly,
a decline in the incidence of dementia over time has been noted (Satizabal et al. 2016),
but it does not appear that the role of Pb (which is declining simultaneously) has been
proposed or tested as a factor, a separate hypothesis that warrants further study.
6.4.3 The link between lithium and lead (Part 3)
The paucity of epidemiological studies that measure both Li and Pb is a major limitation.
This question has not been investigated adequately. A future study could include mea-
sures of environmental Li and environmental Pb levels. We therefore turned to studies
that separately investigated the associated health effects of environmental Li and Pb to
see whether each element had opposing, complementary associated health effects.
On the basis of our review, drinking water Li may be negatively correlated with rates
of suicide, homicide, dementia, general psychiatric and behavioural concerns, all-cause
mortality and ischemic heart disease, with evidence of varying quality. In contrast, there
is no evidence supporting a protective association of drinking water Li on bipolar disorder,
with one large negative study. Environmental Pb exposure appears to be positively
correlated with suicide (in men), homicide, cognitive impairment and dementia, lower
IQ, lower socioeconomic status, all-cause mortality, cardiovascular and ischemic heart
disease, with evidence of varying quality. Therefore, environmental Li and Pb exposure
appear to have complementary associated health effects, supporting our hypothesis.
A lack of experimental studies in humans to test the possibility that Li mitigates
Pb is an unsurprising but significant limitation. However, animal studies support a
protective role of Li in Pb toxicity. The study by Wang et al. strongly supports the
role of Li in preventing Pb-induced toxicity including impacts on brain and behaviour
(Wang et al. 2015). This important study warrants replication. Future animal studies
could test whether very low doses of Li also have a protective effect on cognition and
neuropathology in chronic Pb exposure.
116
In vivo studies in humans could theoretically be used to clarify the role of Li in
protecting from Pb-related toxicity. For example, there are large databases of participants
who have undergone structural magnetic resonance imaging, cerebrospinal fluid analyses
and positron emission tomography to quantify the amount of amyloid, an Alzheimer’s-
related neuropathological finding. It would be possible, with considerable cost and effort,
to obtain biomarkers of Pb exposure from bone, in addition to estimating Li exposure
based on place of residence. This would enable testing to see whether Pb correlates
with Alzheimer’s related neuropathology, and whether environmental Li was negatively
correlated with neuropathology. A first step may be to establish a relationship between
bone Pb accumulation and Alzheimer’s pathological changes in vivo.
6.4.4 Biological plausibility (Part 4)
There is significant overlap in the pathophysiology of Pb toxicity and the pharmacody-
namics of Li. While this is circumstantial evidence and does not imply that Li protects
against the Pb toxicity, an absence of overlap would have challenged the hypothesis that
Li and Pb interact. These data are correlational, and it is a challenge to ascertain whether
observed phenomena are direct or downstream effects of Li and Pb. However, these pos-
sibly overlapping mechanisms may also provide a place to look in future animal studies
as mentioned above, for example, in comparing groups exposed to Pb, Li and both, to
assess group differences in GSK-3, intracellular calcium, NMDA receptor binding, and
Na+/K+ ATPase activity.
6.5 Conclusion
This report sought to review several disparate bodies of evidence to answer several ques-
tions to clarify whether the environmental exposure to Li may mitigate the harms of
chronic environmental Pb exposure. It does not appear that this question has been
asked before, but despite the novelty of the question, there is significant circumstantial
117
evidence supporting the hypothesis.
On the basis of this review, environmental exposure to low levels of Li is nega-
tively associated with suicide, homicide, and possibly with dementia, criminality, other
psychiatric disturbances, ischemic heart disease and all-cause mortality. In contrast, low
levels Pb exposure is positively correlated with similar outcomes. Lead is present in
the areas where Li has been associated with better outcomes. Additional evidence sup-
porting the possibility that Li reduces the neurotoxic effects of Pb comes from animal
studies. There is significant overlap in the neurophysiologic mechanisms affected by Li
and Pb, especially as they both impact calcium signalling pathways and GSK-3. Further
studies are required to clarify each of these areas, but the current literature is consistent
with the hypothesis. Ultimately, a randomized controlled trial using chronic low dose Li
would be required to confirm a causal relationship in humans. However, our hypothesis
is well-supported by the current literature.
On the other hand, there may also be multiple mechanisms at play: the mitigation
of Pb toxicity might explain part of the benefit of Li in some individuals. An observation
that could call the Pb hypothesis into question is the overlap in the benefits of Li observed
at clinical doses and microdoses found in drinking water (i.e. reduction in suicide, pre-
vention of cognitive decline). This may suggest a common mechanism between clinically
dosed and microdose Li exposure. The notable exception is that microdose Li has not
been identified to be associated with a lower risk of bipolar disorder. Therefore, the par-
tial overlap in Li’s clinical effects with the associated outcomes of drinking water Li could
merely reflect confirmation bias across the literature. Recent research into correlates of
drinking-water Li is likely in large part motivated by the known clinical benefits of Li,
with the most recent studies investigating suicide, bipolar disorder, and dementia. There
may therefore be systematic or confirmation bias reflected in the available literature. It
is notable that there is no evidence to support a benefit of drinking water Li on bipolar
disorder. That is, the effects associated with drinking water Li appear to map better to
the opposite effects of Pb exposure than they do to the effects of Li at clinical doses.
While it is currently premature for this hypothesis to have clinical consequences,
118
there is currently sufficient evidence to support the consideration of Pb exposure in
epidemiologic studies investigating the impact of drinking water Li. In these future
studies, outcome measures should include the effects of chronic Pb exposure. Given the
potential public health implications, these are important studies to be done.
Some investigators have concluded that Li is an essential nutrient on the basis of the
observed benefits, including the purported psychiatric benefits reviewed here (Schrauzer
2002). This may be premature without a mechanistic understanding or robust evidence
from a randomized controlled trial and without safety trials. For example, if this report’s
hypothesis is true, the majority of the benefit of Li may be seen in those exposed to
the highest levels of Pb. A small benefit observed across the general population may
be a watered-down average, where a small subset of the population has a much greater
effect (e.g. those with specific mental disorders, genetic vulnerabilities, and/or specific
environmental exposure such as to Pb). Further, even if Li is beneficial on average, as with
any treatment, the possibility of harm must be considered, and adds to the importance
of identifying the subgroups who stand to benefit. Given the dramatically smaller dose
range as compared to current clinical practice that may be beneficial in reducing Pb
toxicity, the safety profile including therapeutic window would be favourable, and the
blood level monitoring that is required in current clinical practice would likely not be
required, but this would have to be confirmed in safety trials.
Additionally, if the hypothesis of this report is proven to be true, microdose Li
may one day be considered in areas where environmental Pb remains high, and where
reduction of Pb contamination has failed or is infeasible. If the benefit of microdose Li
can established in reducing Pb toxicity in humans, microdose Li supplementation or even
addition of Li to the water supply may have benefit, analogous to fluoride for tooth decay.
Another area of investigation may be the possible role of microdose Li in localized Pb
contaminations as in the Flint water crisis, analogous to the use of potassium iodide after
exposure to radioactivity. With further investigation into these areas, microdose Li may
be confirmed to have significant public health benefits.
119
Chapter 7
General Discussion
This thesis began with an overview of AD, outlining a complex and heterogeneous illness
that is incompletely understood, and for which a treatment or cure would have far-
reaching benefits given the high morbidity and prevalence of the disease. Despite intense
research into the pathophysiology of AD, and despite many highly funded trials of course-
modifying treatments, the latter have not been successful and there are calls to consider
alternatives to the amyloid hypothesis and to investigate the pathophysiology of AD more
broadly (Panza et al. 2019).
This thesis presents several iterative contributions to the field of clinical AD patho-
physiological research: empirical evidence supporting a proposed novel PET biomarker
of white matter integrity in Chapter 3, an open source PET analysis tool in Chapter 4,
a systematic review clarifying the potential role of Pb in AD in Chapter 5, and in Chap-
ter 6, a series of systematic reviews and meta-analyses exploring a hypothesis about
environmental Pb and Li and its role in AD and disease more broadly.
Within the overview of AD, the role of cerebrovascular changes was highlighted
as particularly important, given cerebrovascular pathology is commonly observed in AD,
the risk factors–some modifiable–for vascular disease are also risk factors for AD, cere-
brovascular factors may contribute to the pathophysiology of AD, and cerebrovascular
disease can impact cognition and function independently of AD. Further, neuroimaging
120
biomarkers can identify cerebrovascular-related changes in vivo and can be correlated
with important clinical features. The overview of vascular factors also set the stage for
the work we presented in the rest of the thesis, as each project relates to the cerebrovas-
cular changes that accompany AD.
Chapter 3 presented an exploratory analysis empirically testing for an association
between established biomarkers of white matter integrity: white matter hyperintensities
on T2-weighted FLAIR and reduced global FA on DWI, with novel PET biomarkers:
maximum SUV in a white matter ROI and relative tracer delivery, R1, in a white matter
ROI. The premise of the hypothesis rested on the previously demonstrated associations
between grey matter PET R1 and early SUV measures with CBF, with CBF reductions in
turn previously being associated with WMH and reduced FA. Indeed, an etiologic theory
of WMH is chronic hypoperfusion, a cerebrovascular phenomenon. Moderate correlations
were observed between the white matter PIB PET delivery measures with established
markers of white matter integrity.
The exploratory nature of this study highlighted a need and benefit of the R package,
which was in turn created for the analysis of time-activity curve data and presented in
Chapter 4. By specifying the version of open-source software in an article, scientists
looking to audit or replicate a study can refer to the archived source code that is available
online in a long-term repository. The exact code can be examined. Further exploration
and modifications can be contributed to the package, and all are transparent given the
infrastructure used (Git, GitHub, Zenodo, CRAN). As the R package in Chapter 4 was
developed simultaneously to meet the needs of the PET study in Chapter 3, it was not
specified in advance. However, in future studies, the intent to use the package (and thus
its fully specified analytic methods) as part of the proposed analysis can be included in a
registered report. Registered reports address many “questionable research practices” and
can greatly improve the confidence in research findings in neuroscience (D. Chambers et
al. 2014).
The proposed PET biomarkers from Chapter 3 rely on the early post-injection
frames: SUVmax occurs within the first minutes post-injection, and calculation of R1 via
121
SRTM uses the full dynamic data. This required participants to remain still in a scanner
for the full 90 minutes. Studies that have used amyloid PET for the specific purpose of
quantifying amyloid have sometimes used simplified protocols for comfort and efficiency,
and only acquired PET data in the late window that is required for amyloid quantification
with SUVR (e.g. starting acquisition 50-70 minutes post injection, for example, as with
the ADNI amyloid PET protocols).
With the increasing evidence confirming the utility of PIB PET as a surrogate mea-
sure of CBF and potential alternative to the use of FDG for certain applications, and if
the use of such surrogates in white matter regions provides additional useful information,
future studies may prefer to obtain the full dynamic scan. A potential compromise for
comfort may be to obtain early- and late-scan data in two separate windows, allowing
the participant a break between the two acquisitions. The decision will likely hinge on
the purpose for obtaining the amyloid PET scan. For simple categorization of amyloid
status, i.e. PIB-positive status, late-scan SUVR may be sufficient, however, for studies
investigating AD pathology, or planning on contributing to a public repository for sub-
sequent studies, measuring during the full dynamic window would maximize the benefit
of the PET scan.
Chapters 5 and 6 are examples of research questions that may particularly benefit
from amyloid PET, CBF and white matter biomarkers. Specifically, Chapter 5 identified
a lack of human studies evaluating markers of chronic Pb exposure and the onset of AD.
The potentially long period of time between Pb exposure and development of clinical
AD, as suggested by animal models (Bihaqi et al. 2014), poses a challenge to the field.
A potential solution is to use measures of Pb that most reflect long-term exposure as in
the bone, and biomarkers of AD-related changes that occur early, as in amyloid imaging.
With bone Pb KXRF reflecting Pb exposure over 1-2 decades, and amyloid imaging
positivity occurring up to 1-2 decades prior to the onset of clinical AD, a cross-sectional
study with these biomarkers has the potential to bridge long periods of time.
The major impact of low-level environmental Pb exposure on hypertension, cardio-
vascular disease and mortality (Lanphear et al. 2018), as reviewed in section 6.3.2.2.3,
122
strongly implicates vascular factors as a pathophysiological link between Pb exposure
and cognitive changes. Meanwhile, experiments have demonstrated that Pb exposure in-
duces amyloid accumulation and tau phosphorylation. The extent to which such factors
are responsible for a link between Pb and AD is unknown. The use of neuroimaging to
characterize amyloid, tau and vascular changes associated with early-life or long-term Pb
exposure could bring clarity to this question.
There are several potential benefits to clarifying the relationship between chronic
Pb exposure and AD. Firstly, while the harmful impacts of even low-level Pb exposure are
undisputed, fully characterizing the impacts adds weight to public health interventions
to reduce ongoing exposure. Such evidence brings justification to costly programs to
remove Pb from the environment (e.g. water distribution pipes, old paint, and industrial
exposure), which is higher in developing countries (Organization 2009).
Secondly, understanding the mechanisms by which Pb contributes to AD and re-
lated pathology can help to understand AD more generally, which in principle may help
the development of future interventions to prevent or treat AD due to that improved
understanding, even where Pb may not have had a significant role in a given case. The
mechanism by which Pb may increase the risk of AD could be via pathological processes
that occur in AD without Pb exposure which are simply exacerbated or modulated by Pb,
e.g. tau phosphorlation via GSK-3, as has been demonstrated to occur in rats exposed
to Pb (Gassowska et al. 2016).
However, thirdly, and most directly related to the work presented in Chapter 6,
is the possibility for an intervention that mitigates the harms of Pb exposure, which
could be targeted to locations where there is known ongoing chronic exposure, or where
exposure is known to be elevated, for example due to an infrastructure-related emergency
as occurred in Flint, Michigan. Recently in a randomized controlled trial in MCI with
follow up over several years, low-dose Li was demonstrated to prevent cognitive decline
compared to placebo (Forlenza et al. 2019), which is purportedly due to its inhibition
of GSK-3 and downstream reduction of tau phosphorylation. Thus, Li may modify the
pathophysiology of AD, and it may do so by inhibiting mechanisms that occur in cases
123
unrelated to Pb but which also occur to a greater extent in Pb exposure.
A major challenge to studying the impact of environmental exposures on health are
potential co-variates. For example, as discussed, in Chapter 6, the occurrence of lithium
in drinking water is impacted by geographical features such as altitude, which one study
found could account for the differences associated with Li levels. In determining the
impact of Pb exposure, there are a wide variety of potential confounding factors including
other toxic exposures, and other social factors that may be associated with Pb exposure
such as income, housing, and other social determinants of health. Such confounds are
important to understand, but hard to test with correlational studies. Even if accounted
for statistically, the directionality of the association remains unclear, particularly with
cross-sectional studies, and causation cannot be established definitively. For example, if
it were argued that a relationship between high Pb exposure and AD could be accounted
for by the association of both with certain social determinants of health, it would be
important to clarify whether Pb mediates that risk.
The reported decline in incidence of AD in developed countries has been hypothe-
sized to be due to improvements in education and the treatment of vascular risk factors
including hypertension (Langa 2015). Notably, low level chronic Pb exposure has de-
clined over the same time period (McNeill et al. 2017). Thus, it may be an alternative
or additional reason for the possible decline in AD. Indeed, if Pb were a significant cause
of AD, a decline in the regions where Pb levels have decreased would be expected. As
chronic low-level Pb exposure has been identified as a major factor in hypertension and
cardiovascular mortality (Lanphear et al. 2018), the health improvements attributable to
reductions in environmental Pb could be misattributed to other factors thought to have
improved hypertension on a population scale, such as improved hypertension diagnosis
and treatment.
124
Chapter 8
Conclusions
With the overall goal to advance the understanding of the pathophysiology of AD and
the tools available for such research, this thesis presented projects with several separate
but related aims. In this section we will revisit the hypotheses and draw conclusions.
8.1 Amyloid PET tracer delivery in white matter may be a
marker of white matter integrity
The work in Chapter 3 proposed a novel biomarker using amyloid PET to model a
surrogate for white matter integrity. The use of the peak SUV and the relative delivery
R1 values from PIB PET scans was based on the previous association of these markers
in cortical regions with CBF, but we were the first to test for a specific association of
these markers in a white matter ROI with more established biomarkers of white matter
integrity. With the caveat that our analysis was exploratory in the sense that we devised
the hypothesis and method after the data were collected, we were able to demonstrate a
moderate correlation as hypothesized.
White matter peak SUV and R1 were positively correlated with global FA and
inversely correlated with WMH volume. Thus, PIB PET white matter SUV and R1 may
125
be markers of white matter integrity. We aimed to test whether this was the case, and
while these findings support the hypothesis, there are alternative explanations that have
not yet been ruled out.
With our study, we cannot rule out the contribution of several potential confounds.
Associations of the novel measures with blood pressure were not significant, but they were
associated with age. We used partial correlation to account for age, and the resulting
correlations were no longer statistically significant. Given the small sample, we were
unable to confidently conclude whether the observed correlation is due to an independent
association with white matter integrity or merely due to a common association with age.
8.1.1 Implications
It remains to be seen whether PIB PET white matter R1 and SUVmax are independent
markers of white matter integrity, and whether they independently predict clinical find-
ings similar to other markers of white matter integrity. The specific biological correlates
of these measures are not yet known.
The direct implications of our study are that given the association identified in
support of our hypothesis, further investigation is warranted to confirm the relationship
and determine the clinical utility. In the following Chapter (9), the next steps required
to determine the clinical and research utility of these markers are outlined.
If it is subsequently determined that PIB tracer deliver in white matter regions
reflects white matter integrity independent of CBF, age and other co-variates, it may be
a convenient biomarker, trivial to calculate when a dynamic PIB PET scan is obtained.
In principle, similar analyses of other amyloid or tau PET tracers could potentially be
used in a similar way, as the hypothesized connection to white matter integrity is via
blood flow related changes reflected by tracer delivery rather than the specific tracer
binding.
Likely, T2-weighted FLAIR and DTI are more direct measures of white matter
126
integrity, and PET-derived surrogates are unlikely to replace such measures when assess-
ment of white matter integrity is a primary goal. However, T2-weighted FLAIR and
DTI scans do come at an extra time and cost, and the analysis of the images to extract
meaningful measures of white matter integrity is complex (e.g., section 3.2.3.2). While
structural MRI (with a T1-weighted sequence) is often obtained in studies using amy-
loid PET for co-registration, T2-weighted FLAIR sequences are not always obtained. In
longitudinal studies multiple PET scans may be obtained and a single T1-weighted MRI
scan used for co-registration. In both cases, white matter integrity information from the
PET scan could have value.
Further, the analysis of amyloid PET data without the use of MRI is possible
(Bourgeat et al. 2018; Bourgeat et al. 2015), and integration of white matter analyses
into such pipelines may be possible as well. Thus, such surrogate biomarkers may prove
to be a cost-effective way to derive multiple data points from a single scan, when amyloid
imaging is required.
As a further example, if the proposed biomarkers are proven to be associated with
clinical outcomes, a risk score could theoretically be calculated from an amyloid scan that
includes information about amyloid deposition, brain volume, CBF and white matter
integrity.
8.2 Open source PET analysis improves transparency and repro-
ducibility
The work presented in Chapter 4, a tool to automate the analysis of PET time-activity
curve data, had aims for the current work–to transparently analyze the data presented
in Chapter 3, but also more broadly: to be a tool that other researchers may find useful
and efficient in their own analyses, to improve transparency and reproducibility in PET
and AD research.
127
The first goal, of creating a transparent, automated pipeline to analyze the time
activity curve data in the project in Chapter 4 was achieved by implementing it in R,
which has a central repository (CRAN) and community with a well-documented and
widely used structure, and by publishing and archiving the entire source code.
The second broader aim can only be achieved once other researchers determine
whether the tool meets their needs. To improve the package’s code in addition to its ac-
cessibility and credibility, it was submitted to and underwent peer review in the rOpenSci
process and is now a part of the rOpenSci repository.
8.2.1 Implications
The presented R package, in addition to meeting the immediate needs of the present anal-
yses, may prove to be a useful tool for other researchers in PET analysis and AD research,
improving transparency, efficiency and accuracy. Future development and collaboration
may expand its integration with other open source PET analysis tools, contributing to
a fully transparent, robust and free option for PET analysis.
8.3 Available case-control AD studies do not adequately address
role of Pb in AD
The aim of Chapter 5 was to systematically identify, review, and where possible, meta-
analyze studies that used a case-control design to test for an association of Pb accumu-
lation and AD.
As Pb exposure has been associated with cognitive impairment in humans, and
has been associated with AD pathology in animal studies of early-life exposure, we hy-
pothesized that any Pb measure that may reflect exposure prior to disease onset may be
elevated in cases of AD, which would suggest that Pb is a cause or risk factor for AD.
128
Several methods of Pb measurement were identified by the scoping search, with
targeted searches identifying studies that applied the measurements in case-control stud-
ies of AD. The methods included Pb measurement in blood and serum, hair and nails,
CSF and postmortem pathological exam. We have that none of the measurements used
reflected earlier life exposure.
Most identified studies used blood (whole and/or serum) Pb measurements, and
when pooled in a meta-analysis, there was no difference in Pb levels in AD compared to
controls. There was some evidence to suggest AD was associated with lower levels of Pb
in hair and cerebrospinal fluid.
Bone Pb measurement was used in longitudinal studies of cognition, but none that
looked specifically at AD or in an AD case-control design, yet bone measurement may
be the most suitable method, as it partially accounts for Pb accumulation over decades.
Thus, the aim to review the available literature on case-control studies of Pb in AD
was successful, but this aspect of the literature did not provide any evidence to support
a role of Pb in AD, due to limitations of the Pb measurement methods.
8.3.1 Implications
The implications of the systematic review on Pb in AD relate both to what was found: no
elevation of Pb in individuals who have already been diagnosed with AD, and what was
notably absent from the literature: studies which include measures of Pb accumulation
reflecting remote exposure (e.g. bone) or longitudinal studies that assess incidence of AD
based on early life exposure. As animal models most strongly support the role of Pb in
AD as an early-life risk causal agent, human studies to test this possibility are required.
129
8.4 Drinking water lithium may mitigate harms of lead exposure
and the hypothesis requires testing
The paper presented in Chapter 6 aimed to lay out a hypothesis and systematically draw
on the literature to test the elements of the hypothesis.
The systematic review identified that drinking water lithium has been associated
with reductions in health conditions and symptoms in different locations, including sui-
cide, homicide, and possibly with dementia, criminality, other psychiatric disturbances,
ischemic heart disease and all-cause mortality. The strongest evidence was for a reduc-
tion in suicide rates. Only two studies investigated dementia, and both found significant
associations with drinking water lithium: a reduced incidence (Kessing et al. 2017a), and
reduced dementia-related mortality (Fajardo et al. 2017).
The identified health conditions negatively associated with lithium exposure are
also known to be positively associated with Pb exposure. The ubiquitous nature of Pb
in the environment supports the hypothesis.
Most directly testing the hypothesis mechanistically, are animal studies that demon-
strate mitigation of Pb toxicity. Further, in a review of the known biological processes
impacted by Pb and Li, overlap was identified, particularly in relation to GSK-3 and
calcium signaling.
Thus, while there is no direct human evidence for a protective role of Li against Pb
toxicity, the available literature supports the hypothesis.
A potential challenge to the hypothesis that the possible benefits to cognition of
lithium in drinking water are via mitigation of Pb toxicity is the growing evidence of
a cognitive and functional benefit of lithium treatment in mild cognitive impairment
(Forlenza et al. 2019). As we saw in Chapter 5, there is no association between the
present diagnosis of AD and acute blood Pb levels–if Pb causes AD, it is via earlier-life
(or in utero or epigenetic) exposure.
130
8.4.1 Implications
With the hypothesis supported but not demonstrated in any prospective human study, it
is premature to advocate for microdose lithium to prevent any disease. However, given
the potentially significant public health benefits–e.g. reduction in rates of dementia, crime
and suicide–the main implication is that such prospective testing is warranted.
8.5 General Conclusion
In this thesis, several contributions to field of AD research have been made, including:
• the proposal of a novel analysis technique that may provide additional information
about white matter integrity from existing PET scan information,
• the creation of an open source PET analysis software that implements the analysis
and standard analytic techniques with broad access and compatibility,
• a systematic review that highlights gaps in the literature on Pb, a potential cause
of AD,
• and the proposal of a novel hypothesis that appears well-supported by the literature
that, if confirmed by prospective experimental testing via a randomized controlled
trial, could have broad public health benefits including the reduction of cognitive
impairment and possibly dementia.
131
Chapter 9
Future Directions
A recent exploratory analysis has demonstrated that the early PIB uptake (in this case,
the SUVR with window 1-9 minutes after injection), which have been shown to reflect
CBF, is positively associated with cognition in a cohort including MCI and dementia
(Tiepolt et al. 2019). This study did not look at a white matter region of interest. The
pattern reported was similar to that observed for FDG PET. Thus, analogous to how
early PIB uptake and delivery measures have been shown to reflect CBF and metabolism,
and in turn have similar cognitive correlates as CBF, our study demonstrating the possi-
ble connection between white matter region PIB delivery with markers of white matter
integrity invites future work to establish whether white matter regions of PIB delivery
are associated with clinical features similar to WMH and FA.
The initial next step will be to replicate this finding in another sample to validate
the exploratory analysis. In this study, we used a subset of the participants of the larger
PACt-MD trial who had all of the biomarkers required to test our hypothesis, restricting
the sample to 34 participants. However, a larger subset is available to test the association
between the test the association between global FA and the novel PET biomarkers (those
who were excluded only because T2-weighted FLAIR, and thus a measure of WMH was
not available). This analysis has not yet been done for these participants and would
therefore be a partial replication, though would not be as convincing as a new prospec-
132
tively acquired sample. Further, the larger subset may be suitably powered to test for
association with cognitive test performance.
In the exploratory study, the white matter ROI was averaged and treated as a
single region. However, local differences may be present and important. A future analysis
could use WMH masks to investigate for local tracer delivery differences within the white
matter, which could help clarify the underlying pathology.
Further exploration into the optimal use of partial volume correction techniques
is also important. Standard techniques rely on the assumption that white matter is
homogenous, which may not be the case. However, partial volume correction may be
particularly important when looking for effects specific to white matter regions.
Longitudinal studies of multimodal imaging and other biomarkers could clarify the
pathology reflected by PIB white matter R1 and SUVmax. Our initial cross-sectional
exploratory analysis demonstrated that these PET markers vary with the severity of
WMH and DTI FA. A longitudinal design could clarify the sequence of changes. For
example, decreases in white matter SUVmax and R1 may precede the onset or accumula-
tion of WMH as might be expected given WMH are hypothesized to result from chronic
hypoperfusion and/or microvascular infarcts.
Once confirmed, the PIB delivery measures that may reflect white matter integrity
could be particularly useful, in combination with other measures of white matter integrity,
cerebrovascular pathology and dysfunction, and clinical factors, in a study to test for
an association with biomarkers reflective of chronic Pb exposure (e.g. bone Pb). The
simplest design would be to simply assess for correlations between the AD-, vascular- and
Pb-related biomarkers in a cross-sectional study of individuals at risk for AD (e.g. with
amnestic MCI).
Such a design could also be feasible in the context of an add-on study, as the same
population is desirable in studies such as randomized controlled trials for the prevention
or treatment of dementia. This would maximize the value of such a study in which
biomarkers are already being obtained.
133
A more complex design, but one that would further clarify the role of Pb in AD,
could involve the longitudinal assessment of cerebrovascular, amyloid and tau biomarkers,
in combination with Pb monitoring. Given the significant reductions of environmental
Pb in certain locations, such a study may be best done in locations where chronic envi-
ronmental Pb exposure is relatively high. Otherwise, adding AD-related biomarkers such
as amyloid and tau imaging, to the follow up procedures of the ongoing cohort studies in
which Pb exposure has already been measured (as in the NHANES, Nurses Health Study,
Baltimore Memory Study) is a way to test for an association between earlier-life Pb expo-
sure and AD-related pathology, which could yield more accurate and more timely results
than diagnostic incidence. Given the costs of such biomarkers, it would be more feasible
to test a subset of the cohort studies. Longitudinally, the possible role of hypertension
as a mediating factor between Pb and onset of AD could be tested.
Future work to clarify the possible role of Li in mitigating the harms of Pb exposure
was outlined in Chapter 6. Most importantly, a randomized controlled trial of microdose
Li could help to establish whether there is any clinical benefit among Pb-exposed in-
dividuals. However, the design of such an RCT would be complicated. The current
evidence relies on large population-based epidemiologic studies, which could be detecting
an important effect at a population level, but the effect size for an individual may be
small. The animal evidence that demonstrated a benefit of Li pre-exposure relies on de-
signs that would be unethical in humans, because toxic Pb is administered. Large-scale
administration of microdose Li analagous to water supply fluoridation may be safe but
once can speculate that it would politically challenging in the absence of strong human
evidence. The most feasible approach may be to select a population in which there is
high and variable Pb exposure and conduct a large RCT of microdose Li.
In the context of emerging evidence for a benefit of low dose Li in the prevention
of cognitive decline in MCI (Forlenza et al. 2019), Li may prove to be helpful in the
prevention of AD outside of mitigating effects of early and chronic Pb exposure. If
confirmed, this finding on its own could have significant clinical implications. It would be
important to know whether certain subgroups of individuals with MCI benefit most from
134
Li. It is possible that Pb-exposed individuals with MCI benefit most from Li, for example
if the treatment effects are by disrupting a process in which Pb plays a role such as GSK-
mediated tau phosphorylation. Measuring Pb accumulation in a study replicating the
treatment effects of low-dose Li in MCI could clarify whether there are greater benefits
in individuals who have a higher Pb load. Forlenza et al. (2019) used low-dose lithium
(serum target 0.25–0.5 mEq/L) to reduce the risk of adverse effects compared to standard
doses, and on the basis of preliminary work showing 50% inhibition of (GSK-3 $beta$)
activity in peripheral blood platelets at this level. Lower doses of Li have not been tested
in an RCT of MCI and may also have benefit given the epidemiologic data. Using even
lower doses would further reduce the risk of adverse effects, as the most serious adverse
effects of lithium are dose-dependent.
The future directions of the PET analysis R package presented in Chapter 4 are
open-ended. The package functions by converting time-activity curve data in different
file formats into a common internal representation as an R object. This allows for future
support of any file format by implementing loading and conversion functions, which would
make all of the package’s features available for the new format. Similarly, new features
such as kinetic models can be implemented on the basis of the internal R time-activity
curve object, without regard for external file formats.
The open source model of the package and its use of popular software development
infrastructure (i.e. Git, GitHub, Travis CI) enables collaboration, inclusion in the rOpen-
Sci repository facilitates discovery and quality, and integration of an extensive test-suite
helps to test for accuracy as the package’s features are extended.
135
References
Abell, J.G. et al., 2018. Association between systolic blood pressure and dementia in the Whitehall II
cohort study: Role of age, duration, and threshold used to define hypertension. European Heart Journal,
39(33), pp.3119–3125.
Advisory Committee on Childhood Lead Poisoning Prevention of the Centers for Disease Con-
trol and Prevention, 2012. Low Level Lead Exposure Harms Children: A Renewed Call for Primary
Prevention.
Agdeppa, E.D. et al., 2001. Binding Characteristics of Radiofluorinated 6-Dialkylamino-2- Naph-
thylethylidene Derivatives as Positron Emission Tomography Imaging Probes for nl-Amyloid Plaques in
Alzheimer’s Disease. Journal of Neuroscience, 21, p.5.
Aizenstein, H.J. et al., 2008. Frequent amyloid deposition without significant cognitive impairment
among the elderly. Arch Neurol, 65, pp.1509–17.
Alda, M., 2015. Lithium in the treatment of bipolar disorder: Pharmacology and pharmacogenet-
ics. Molecular psychiatry, 20(6), pp.661–670.
Alexander, A.L. et al., 2007. Diffusion tensor imaging of the brain. Neurotherapeutics, 4(3),
pp.316–329.
Alimonti, A. et al., 2007. Serum chemical elements and oxidative status in Alzheimer’s disease,
Parkinson disease and multiple sclerosis. NeuroToxicology, 28(3), pp.450–456.
American Psychiatric Association, 2013. Diagnostic and statistical manual of mental disorders:
DSM-5, 5th ed., Arlington, VA, US: American Psychiatric Publishing, Inc.
Amlien, I. & Fjell, A., 2014. Diffusion tensor imaging of white matter degeneration in Alzheimer’s
disease and mild cognitive impairment. Neuroscience, 276, pp.206–215.
Andersson, J.L. & Sotiropoulos, S.N., 2016. An integrated approach to correction for off-resonance
136
effects and subject movement in diffusion MR imaging. NeuroImage, 125, pp.1063–1078.
Arksey, H. & O’Malley, L., 2005. Scoping studies: Towards a methodological framework. Inter-
national Journal of Social Research Methodology, 8(1), pp.19–32.
Ayton, S., Lei, P. & Bush, A.I., 2015. Biometals and Their Therapeutic Implications in
Alzheimer’s Disease. Neurotherapeutics, 12(1), pp.109–120.
Bahrani, A.A. et al., 2017. White Matter Hyperintensity Associations with Cerebral Blood Flow
in Elderly Subjects Stratified by Cerebrovascular Risk. Journal of Stroke and Cerebrovascular Diseases,
26(4), pp.779–786.
Bakulski, K.M. et al., 2012. Alzheimer’s disease and environmental exposure to lead: The epi-
demiologic evidence and potential role of epigenetics. Current Alzheimer research, 9(5), pp.563–573.
Bandeen-Roche, K. et al., 2009. Cumulative lead dose and cognitive function in older adults.
Epidemiology (Cambridge, Mass.), 20(6), pp.831–839.
Bangen, K.J. et al., 2016. Interaction Between Midlife Blood Glucose and APOE Genotype
Predicts Later Alzheimer’s Disease Pathology J. Brandt, ed. Journal of Alzheimer’s Disease, 53(4),
pp.1553–1562.
Bangen, K.J. et al., 2018. Baseline White Matter Hyperintensities and Hippocampal Volume are
Associated With Conversion From Normal Cognition to Mild Cognitive Impairment in the Framingham
Offspring Study. Alzheimer Disease & Associated Disorders, 32(1), pp.50–56.
Banijamali, M., Rabbani-Chadegani, A. & Shahhoseini, M., 2016. Lithium attenuates lead in-
duced toxicity on mouse non-adherent bone marrow cells. J Trace Elem Med Biol, 36, pp.7–15.
Barbosa, F. et al., 2005. A Critical Review of Biomarkers Used for Monitoring Human Exposure
to Lead: Advantages, Limitations, and Future Needs. Environmental Health Perspectives, 113(12),
pp.1669–1674.
Barker, W.W. et al., 2002. Relative Frequencies of Alzheimer Disease, Lewy Body, Vascular and
Frontotemporal Dementia, and Hippocampal Sclerosis in the State of Florida Brain Bank: Alzheimer
Disease & Associated Disorders, 16(4), pp.203–212.
Basha, M.R., 2005. The Fetal Basis of Amyloidogenesis: Exposure to Lead and Latent Overex-
pression of Amyloid Precursor Protein and -Amyloid in the Aging Brain. Journal of Neuroscience, 25(4),
pp.823–829.
Basun, H. et al., 1991. Metals and trace elements in plasma and cerebrospinal fluid in normal
137
aging and Alzheimer’s disease. Journal of Neural Transmission. Parkinson’s Disease and Dementia
Section, 3(4), pp.231–258.
Behrens, T. et al., 2007. Probabilistic diffusion tractography with multiple fibre orientations:
What can we gain? NeuroImage, 34(1), pp.144–155.
Behrens, T. et al., 2003. Characterization and propagation of uncertainty in diffusion-weighted
MR imaging. Magnetic Resonance in Medicine, 50(5), pp.1077–1088.
Benjamini, Y. & Hochberg, Y., 1995. Controlling the False Discovery Rate: A Practical and Pow-
erful Approach to Multiple Testing. Journal of the Royal Statistical Society: Series B (Methodological),
57(1), pp.289–300.
Bennett, I.J. et al., 2009. Age-related differences in multiple measures of white matter integrity:
A diffusion tensor imaging study of healthy aging. Human Brain Mapping, 31, pp.378–390.
Berger, A., 2002. How does it work?: Magnetic resonance imaging. BMJ, 324(7328), pp.35–35.
Berrios, G.E., 1990. Alzheimer’s disease: A conceptual history. International Journal of Geriatric
Psychiatry, 5(6), pp.355–365.
Bertram, L. & Tanzi, R.E., 2008. Thirty years of Alzheimer’s disease genetics: The implications
of systematic meta-analyses. Nature Reviews Neuroscience, 9(10), pp.768–778.
Bihaqi, S.W. et al., 2018. Influence of Early Life Lead (Pb) exposure on alpha-Synuclein, GSK-
3beta and Caspase-3 mediated Tauopathy: Implications on Alzheimer’s disease. Current Alzheimer
research.
Bihaqi, S.W. et al., 2014. Infantile exposure to lead and late-age cognitive decline: Relevance to
AD. Alzheimer’s & Dementia, 10(2), pp.187–195.
Bihaqi, S.W., Eid, A. & Zawia, N.H., 2017. Lead exposure and tau hyperphosphorylation: An in
vitro study. Neurotoxicology, 62, pp.218–223.
Bihaqi, S.W. & Zawia, N.H., 2013. Enhanced taupathy and AD-like pathology in aged primate
brains decades after infantile exposure to Lead (Pb). Neurotoxicology, 0.
Blennow, K. et al., 2010. Cerebrospinal fluid and plasma biomarkers in Alzheimer disease. Nat
Rev Neurol, 6, pp.131–44.
Blennow, K. et al., 2015. Amyloid biomarkers in Alzheimer’s disease. Trends in Pharmacological
Sciences, 36(5), pp.297–309.
138
Blomquist, G. et al., 2008. Unidirectional influx and net accumulation of PIB. The open neu-
roimaging journal, 2, p.114.
Bocca, B. et al., 2006. Alterations in the level of trace metals in Alzheimer’s disease. Trace
Elements & Electrolytes, 23(4), pp.270–276.
Bocca, B., Marchione, F. & Giubilei, F., 2005. Monitoring of chemical elements and oxidative
damage in patients affected by Alzheimer’s disease., p.8.
Bohnen, N.I. et al., 2012. Effectiveness and Safety of 18F-FDG PET in the Evaluation of Dementia:
A Review of the Recent Literature. Journal of Nuclear Medicine, 53(1), pp.59–71.
Bosch, B. et al., 2012. Multiple DTI index analysis in normal aging, amnestic MCI and AD.
Relationship with neuropsychological performance. Neurobiology of Aging, 33(1), pp.61–74.
Bourgeat, P. et al., 2018. Implementing the centiloid transformation for 11C-PiB and β-amyloid
18F-PET tracers using CapAIBL. NeuroImage, 183, pp.387–393.
Bourgeat, P. et al., 2015. Computational Analysis of PET by AIBL (CapAIBL): A cloud-based
processing pipeline for the quantification of PET images. Journal of Nuclear Medicine, 56(supplement
3), pp.149–149.
Boutwell, B.B. et al., 2017. Aggregate-level lead exposure, gun violence, homicide, and rape J.
Meliker, ed. PLOS ONE, 12(11), p.e0187953.
Braak, H. & Braak, E., 1991. Neuropathological stageing of Alzheimer-related changes. Acta
Neuropathologica, 82(4), pp.239–259.
Braak, H. & Braak, E., 1995. Staging of Alzheimer’s disease-related neurofibrillary changes.
Neurobiology of Aging, 16, pp.271–8; discussion 278–84.
Brickman, A.M. et al., 2012. Regional White Matter Hyperintensity Volume, Not Hippocampal
Atrophy, Predicts Incident Alzheimer Disease in the Community. Archives of Neurology, 69(12), p.1621.
Brickman, A.M. et al., 2009. Reduction in cerebral blood flow in areas appearing as white matter
hyperintensities on magnetic resonance imaging. Psychiatry Research: Neuroimaging, 172(2), pp.117–
120.
Brier, M.R. et al., 2016. Tau and Ab imaging, CSF measures, and cognition in Alzheimer’s disease.
Science Translational Medicine, 8(338), pp.1–9.
Broberg, K. et al., 2011. Lithium in Drinking Water and Thyroid Function. Environmental Health
139
Perspectives, 119(6), pp.827–830.
Brouwers, N., Sleegers, K. & Van Broeckhoven, C., 2008. Molecular genetics of Alzheimer’s
disease: An update. Annals of Medicine, 40(8), pp.562–583.
Brown, E.E., 2019. Tacmagic: PET Analysis in R. Journal of Open Source Software, 4(34),
p.1281.
Brown, E.E. et al., 2018. Psychiatric benefits of lithium in water supplies may be due to protection
from the neurotoxicity of lead exposure. Medical Hypotheses, 115, pp.94–102.
Brown, E.E. et al., 2016. Amyloid deposition in semantic dementia: A positron emission tomog-
raphy study: Amyloid-β in SD. International Journal of Geriatric Psychiatry, 31(9), pp.1064–1074.
Brown, E.E. et al., 2016. Tau in Late-Life Depression: A Systematic Review and Meta-Analysis
Y. Motoi, ed. Journal of Alzheimer’s Disease, 54(2), pp.615–633.
Brown, E.E. et al., 2019. Brain Amyloid PET Tracer Delivery is Related to White Matter Integrity
in Patients with Mild Cognitive Impairment. Journal of Neuroimaging.
Brown, E.E. et al., 2019. Lead (Pb) in Alzheimer’s Dementia: A Systematic Review of Human
Case-control Studies. Current Alzheimer Research, 16, pp.353–361.
Budd, J.L. & Rossof, A.H., 1980. Drinking water lithium levels fail to predict for the incidences
of acute or chronic granulocytic leukemia. In Lithium Effects on Granulopoiesis and Immune Function.
Springer, pp. 411–416.
Byers, A.L. & Yaffe, K., 2011. Depression and risk of developing dementia. Nature Reviews
Neurology, 7(6), pp.323–331.
Caligiuri, M.E. et al., 2015. Automatic Detection of White Matter Hyperintensities in Healthy
Aging and Pathology Using Magnetic Resonance Imaging: A Review. Neuroinformatics, 13(3), pp.261–
276.
Campbell, N.L. et al., 2013. Risk Factors for the Progression of Mild Cognitive Impairment to
Dementia. Clinics in geriatric medicine, 29(4), pp.873–893.
Canter, R.G., Penney, J. & Tsai, L.-H., 2016. The road to restoring neural circuits for the
treatment of Alzheimer’s disease. Nature, 539(7628), pp.187–196.
Cetin, I. et al., 2017. How Trace Element Levels of Public Drinking Water Affect Body Compo-
sition in Turkey. Biological Trace Element Research, 175(2), pp.263–270.
140
Chambers, L.W., Bancej, C. & McDowell, I., 2016. Prevalence and Monetary Costs of Dementia
in Canada: Population Health Expert Panel, Alzheimer Society Canada.
Chao, L.L. et al., 2013. Associations between White Matter Hyperintensities and β Amyloid on
Integrity of Projection, Association, and Limbic Fiber Tracts Measured with Diffusion Tensor MRI K.
Herholz, ed. PLoS ONE, 8(6), p.e65175.
Chen, J.J., Rosas, H.D. & Salat, D.H., 2011. Age-Associated Reductions in Cerebral Blood Flow
Are Independent from Regional Atrophy. NeuroImage, 55(2), pp.468–478.
Chen, Y.J. et al., 2015. Relative 11C-PiB Delivery as a Proxy of Relative CBF: Quantitative
Evaluation Using Single-Session 15O-Water and 11C-PiB PET. Journal of Nuclear Medicine, 56(8),
pp.1199–1205.
Cho, Y.W., 1995. Lithium-induced inhibition of Na-K ATPase and Ca ATPase activities in rat
brain synaptosome. Journal of Korean Medical Science, 10(1), pp.7–13.
Chui, H.C. et al., 2011. Vascular risk factors and Alzheimer’s disease: Are these risk factors for
plaques and tangles or for concomitant vascular pathology that increases the likelihood of dementia? An
evidence-based review. Alzheimer’s Research & Therapy, 3(6), p.36.
Chung, J.K. et al., 2015. Lifetime History of Depression Predicts Increased Amyloid-Accumulation
in Patients with Mild Cognitive Impairment. Journal of Alzheimer’s Disease, 45, pp.907–919.
Cipriani, A. et al., 2013. Lithium in the prevention of suicide in mood disorders: Updated
systematic review and meta-analysis. BMJ, 346(jun27 4), pp.f3646–f3646.
Clark, C.M. et al., 2012. Cerebral PET with florbetapir compared with neuropathology at autopsy
for detection of neuritic amyloid-β plaques: A prospective cohort study. The Lancet Neurology, 11(8),
pp.669–678.
Corrada, M.M. et al., 2017. Age of onset of hypertension and risk of dementia in the oldest-old:
The 90+ Study. Alzheimer’s & Dementia, 13(2), pp.103–110.
Cumming, P. et al., 2013. Surrogate markers for cerebral blood flow correlate with [ 18 F]-fallypride
binding potential at dopamine D 2/3 receptors in human striatum. Synapse, 67(4), pp.199–203.
Cuyvers, E. & Sleegers, K., 2016. Genetic variations underlying Alzheimer’s disease: Evidence
from genome-wide association studies and beyond. The Lancet Neurology, 15(8), pp.857–868.
Dadar, M. et al., 2017. Performance comparison of 10 different classification techniques in seg-
menting white matter hyperintensities in aging. NeuroImage, 157, pp.233–249.
141
Dadar, M. et al., 2017. Validation of a Regression Technique for Segmentation of White Matter
Hyperintensities in Alzheimer’s Disease. IEEE Transactions on Medical Imaging, 36(8), pp.1758–1768.
da Silva, J. et al., 2013. Affective disorders and risk of developing dementia: Systematic review.
British Journal of Psychiatry, 202(03), pp.177–186.
Dawson, 1970. The Mathematical Relationship of Drinking Water Lithium and Rainfall to Mental
Hospital Admission.
D. Chambers, C. et al., 2014. Instead of “playing the game” it is time to change the rules:
Registered Reports at AIMS Neuroscience and beyond. AIMS Neuroscience, 1(1), pp.4–17.
Debette, S. & Markus, H.S., 2010. The clinical importance of white matter hyperintensities on
brain magnetic resonance imaging: Systematic review and meta-analysis. BMJ, 341(jul26 1), pp.c3666–
c3666.
DeCarli Charles et al., 2005. Anatomical Mapping of White Matter Hyperintensities (WMH).
Stroke, 36(1), pp.50–55.
Desikan, R.S. et al., 2009. Automated MRI measures identify individuals with mild cognitive
impairment and Alzheimer’s disease. Brain, 132(8), pp.2048–2057.
Devous, M. et al., 2014. The first 5 minutes of F18-Florbetapir scans as an estimate of regional
cerebral blood flow and comparison to F18-FDG. Journal of Nuclear Medicine, 55(supplement 1), pp.247–
247.
Dierckx, R.A.J.O. et al. eds., 2014. PET and SPECT in Psychiatry, Berlin Heidelberg: Springer-
Verlag.
Dubois, B. et al., 2010. Revising the definition of Alzheimer’s disease: A new lexicon. Lancet
Neurology, 9, pp.1118–27.
Dubois, B. et al., 2016. Preclinical Alzheimer’s disease: Definition, natural history, and diagnostic
criteria. Alzheimer’s & Dementia, 12(3), pp.292–323.
Duering, M. et al., 2013. Incident lacunes preferentially localize to the edge of white matter hy-
perintensities: Insights into the pathophysiology of cerebral small vessel disease. Brain, 136(9), pp.2717–
2726.
Egger, C. et al., 2017. MRI FLAIR lesion segmentation in multiple sclerosis: Does automated
segmentation hold up with manual annotation? NeuroImage: Clinical, 13, pp.264–270.
142
Elwood, J.M., 1977. Anencephalus and drinking water composition. American journal of epi-
demiology, 105(5), pp.460–468.
Englund, E., 1998. Neuropathology of White Matter Changes in Alzheimer’s Disease and Vascular
Dementia. Dementia and Geriatric Cognitive Disorders, 9(1), pp.6–12.
Esteban-Vasallo, M.D. et al., 2012. Mercury, Cadmium, and Lead Levels in Human Placenta: A
Systematic Review. Environmental Health Perspectives, 120(10), pp.1369–1377.
Fajardo, V.A. et al., 2017. Examining the Relationship between Trace Lithium in Drinking Water
and the Rising Rates of Age-Adjusted Alzheimer’s Disease Mortality in Texas G. William, ed. Journal
of Alzheimer’s Disease, pp.1–10.
Fanelli, D., 2018. Opinion: Is science really facing a reproducibility crisis, and do we need it to?
Proceedings of the National Academy of Sciences, 115(11), pp.2628–2631.
Farooqui, Z. et al., 2017. Associations of cumulative Pb exposure and longitudinal changes in
Mini-Mental Status Exam scores, global cognition and domains of cognition: The VA Normative Aging
Study. Environmental Research, 152, pp.102–108.
Fathabadi, B. et al., 2018. Comparison of Blood Lead Levels in Patients With Alzheimer’s Disease
and Healthy People. American Journal of Alzheimer’s Disease & Other Dementias, p.153331751879403.
Fazekas, F. et al., 1993. Pathologic correlates of incidental MRI white matter signal hyperinten-
sities. Neurology, 43(9), pp.1683–1689.
Fazekas, F. et al., 1988. White matter signal abnormalities in normal individuals: Correlation
with carotid ultrasonography, cerebral blood flow measurements, and cerebrovascular risk factors. Stroke,
19(10), pp.1285–1288.
Fischer, C.E. et al., 2019. Examining the Link Between Cardiovascular Risk Factors and Neu-
ropsychiatric Symptoms in Mild Cognitive Impairment and Major Depressive Disorder in Remission.
Journal of Alzheimer’s Disease, 67(4), pp.1305–1311.
Flora, G., Gupta, D. & Tiwari, A., 2012. Toxicity of lead: A review with recent updates. Inter-
disciplinary Toxicology, 5(2), pp.47–58.
Forlenza, O.V. et al., 2019. Clinical and biological effects of long-term lithium treatment in older
adults with amnestic mild cognitive impairment: Randomised clinical trial. The British Journal of
Psychiatry, pp.1–7.
Forsberg, A. et al., 2012. The use of PIB-PET as a dual pathological and functional biomarker
143
in AD. Biochimica et Biophysica Acta (BBA) - Molecular Basis of Disease, 1822(3), pp.380–385.
Fu, L. et al., 2014. Comparison of dual-biomarker PIB-PET and dual-tracer PET in AD diagnosis.
European Radiology, 24(11), pp.2800–2809.
Funck, T. et al., 2018. APPIAN: Automated Pipeline for PET Image Analysis. Frontiers in
Neuroinformatics, 12.
Gandrud, C., 2015. Reproducible Research with R and R Studio 2nd ed., Chapman and Hall/CRC.
Garcia-Alloza, M. et al., 2011. Cerebrovascular lesions induce transient beta-amyloid deposition.
Brain, 134(12), pp.3697–3707.
Garza, A., Vega, R. & Soto, E., 2006. Cellular mechanisms of lead neurotoxicity. Medical science
monitor : international medical journal of experimental and clinical research, 12(3), pp.RA57–65.
Gassowska, M. et al., 2016. Perinatal exposure to lead (Pb) promotes Tau phosphorylation in the
rat brain in a GSK-3beta and CDK5 dependent manner: Relevance to neurological disorders. Toxicology,
347-349, pp.17–28.
Gerhardsson, L. et al., 2009. Concentrations of Metals, β-Amyloid and Tau-Markers in
Cerebrospinal Fluid in Patients with Alzheimer’s Disease. Dementia and Geriatric Cognitive
Disorders, 28(1), pp.88–94.
Gerhardsson, L. et al., 2011. Cerebrospinal fluid/plasma quotients of essential and non-essential
metals in patients with Alzheimer’s disease. Journal of Neural Transmission, 118(6), pp.957–962.
Gerhardsson, L. et al., 2008. Metal Concentrations in Plasma and Cerebrospinal Fluid in Patients
with Alzheimer’s Disease. Dementia and Geriatric Cognitive Disorders, 25(6), pp.508–515.
Giacoppo, S. et al., 2014. Heavy metals and neurodegenerative diseases: An observational study.
Biological trace element research, 161(2), pp.151–160.
Giotakos, O. et al., 2013. Lithium in the public water supply and suicide mortality in Greece.
Biological Trace Element Research, 156(1-3), pp.376–379.
Giotakos, O. et al., 2015. A Negative Association Between Lithium in Drinking Water and the
Incidences of Homicides, in Greece. Biological Trace Element Research, 164(2), pp.165–168.
Gjedde, A. et al., 2013. Blood-brain transfer of Pittsburgh compound B in humans. Frontiers in
Aging Neuroscience, 5, p.70.
Goedert, M., 2015. Alzheimer’s and Parkinson’s diseases: The prion concept in relation to assem-
144
bled A , tau, and -synuclein. Science, 349(6248), pp.1255555–1255555.
González-Domínguez, R., García-Barrera, T. & Gómez-Ariza, J.L., 2014. Characterization of
metal profiles in serum during the progression of Alzheimer’s disease. Metallomics, 6(2), pp.292–300.
Gorgolewski, K.J. et al., 2016. The brain imaging data structure, a format for organizing and
describing outputs of neuroimaging experiments. Scientific Data, 3, p.160044.
Grashow, R. et al., 2015. Cumulative lead exposure is associated with reduced olfactory recogni-
tion performance in elderly men: The Normative Aging Study. Neurotoxicology, 49, pp.158–164.
Guo, T., Noble, W. & Hanger, D.P., 2017. Roles of tau protein in health and disease. Acta
Neuropathologica, 133(5), pp.665–704.
Habes, M. et al., 2016. White matter hyperintensities and imaging patterns of brain ageing in
the general population. Brain, 139(4), pp.1164–1179.
Hammers, A. et al., 2003. Three-dimensional maximum probability atlas of the human brain,
with particular reference to the temporal lobe. Human Brain Mapping, 19(4), pp.224–247.
Harari, F. et al., 2016. Exposure to lithium through drinking water and calcium homeostasis
during pregnancy: A longitudinal study. Environmental Research, 147, pp.1–7.
Harari, F. et al., 2015. Exposure to Lithium and Cesium Through Drinking Water and Thyroid
Function During Pregnancy: A Prospective Cohort Study. Thyroid, 25(11), pp.1199–1208.
Harari, F. et al., 2015. Environmental exposure to lithium during pregnancy and fetal size: A
longitudinal study in the Argentinean Andes. Environment International, 77, pp.48–54.
Hardy, J., 2002. The Amyloid Hypothesis of Alzheimer’s Disease: Progress and Problems on the
Road to Therapeutics. Science, 297(5580), pp.353–356.
Hardy, J. & Allsop, D., 1991. Amyloid deposition as the central event in the aetiology of
Alzheimer’s disease. Trends Pharmacol. Sci., 12, pp.383–388.
Hare, D.J. et al., 2016. Lead and manganese levels in serum and erythrocytes in Alzheimer’s
disease and mild cognitive impairment: Results from the Australian Imaging, Biomarkers and Lifestyle
Flagship Study of Ageing. Metallomics : integrated biometal science, 8(6), pp.628–632.
Harrington, K.D. et al., 2015. Amyloid-beta and depression in healthy older adults: A systematic
review. Aust N Z J Psychiatry, 49, pp.36–46.
Health Canada, 2017. Lead in Drinking Water: Document for Public Consultation.
145
Hedden, T. et al., 2012. Cognitive profile of amyloid burden and white matter hyperintensities
in cognitively normal older adults. The Journal of neuroscience : the official journal of the Society for
Neuroscience, 32(46), pp.16233–16242.
Helbich, M. et al., 2013. Does altitude moderate the impact of lithium on suicide? A spatial
analysis of Austria. Geospatial health, 7(2), pp.209–218.
Helbich, M., Leitner, M. & Kapusta, N.D., 2012. Geospatial examination of lithium in drinking
water and suicide mortality. International journal of health geographics, 11(1), p.19.
Helbich, M., Leitner, M. & Kapusta, N.D., 2015. Lithium in drinking water and suicide mortality:
Interplay with lithium prescriptions. The British Journal of Psychiatry, 207(1), pp.64–71.
Heo, J.-H. et al., 2009. White Matter Hyperintensities and Cognitive Dysfunction in Alzheimer
Disease. Journal of Geriatric Psychiatry and Neurology, 22(3), pp.207–212.
Heppner, F.L., Ransohoff, R.M. & Becher, B., 2015. Immune attack: The role of inflammation in
Alzheimer disease. Nature Reviews Neuroscience, 16(6), pp.358–372.
Hershey, C.O. et al., 1983. Cerebrospinal fluid trace element content in dementia: Clinical, radi-
ologic, and pathologic correlations. Neurology, 33(10), pp.1350–1350.
Hippius, H. & Neundörfer, G., 2003. The discovery of Alzheimer’s disease. Clinical research, 5(1),
p.8.
Holtzman, D.M., Morris, J.C. & Goate, A.M., 2011. Alzheimer’s Disease: The Challenge of the
Second Century. Science Translational Medicine, 3(77), pp.77sr1–77sr1.
Ishii, N. et al., 2015. Low Risk of Male Suicide and Lithium in Drinking Water. The Journal of
Clinical Psychiatry, pp.319–326.
Ithapu, V. et al., 2014. Extracting and summarizing white matter hyperintensities using super-
vised segmentation methods in Alzheimer’s disease risk and aging studies: WMH in the Prediction of
AD Risk and Aging Studies. Human Brain Mapping, 35, pp.4219–4235.
Jack, C. et al., 2013. Tracking pathophysiological processes in Alzheimer’s disease: An updated
hypothetical model of dynamic biomarkers. Lancet Neurology, 12, pp.207–216.
Jack, C.R. et al., 2018. NIA-AA Research Framework: Toward a biological definition of
Alzheimer’s disease. Alzheimer’s & Dementia, 14(4), pp.535–562.
Jack, C.R. et al., 2010. Hypothetical model of dynamic biomarkers of the Alzheimer’s pathological
146
cascade. The Lancet Neurology, 9(1), pp.119–128.
Jenkinson, M. et al., 2002. Improved Optimization for the Robust and Accurate Linear Registra-
tion and Motion Correction of Brain Images. NeuroImage, 17(2), pp.825–841.
Jenkinson, M. et al., 2012. FSL. NeuroImage, 62, pp.782–790.
Johnson, K.A. et al., 2013. Appropriate Use Criteria for Amyloid PET. Alzheimer’s & dementia
: the journal of the Alzheimer’s Association, 9(1), pp.e–1–16.
Kabacs, N. et al., 2011. Lithium in drinking water and suicide rates across the East of England.
The British Journal of Psychiatry, 198(5), pp.406–407.
Kametani, F. & Hasegawa, M., 2018. Reconsideration of Amyloid Hypothesis and Tau Hypothesis
in Alzheimer’s Disease. Frontiers in Neuroscience, 12.
Kandimalla, R., Thirumala, V. & Reddy, P.H., 2017. Is Alzheimer’s disease a Type 3 Diabetes? A
critical appraisal. Biochimica et Biophysica Acta (BBA) - Molecular Basis of Disease, 1863(5), pp.1078–
1089.
Kapusta, N.D. et al., 2011. Lithium in drinking water and suicide mortality. Br J Psychiatry,
198, pp.346–50.
Karjalainen, T. et al., 2019. Magia: Robust automated modeling and image processing toolbox
for PET neuroinformatics. bioRxiv.
Keil, D.E., Berger-Ritchie, J. & McMillin, G.A., 2011. Testing for Toxic Elements: A Focus on
Arsenic, Cadmium, Lead, and Mercury. Laboratory Medicine, 42(12), pp.735–742.
Kessing, L.V. et al., 2017a. Association of Lithium in Drinking Water With the Incidence of
Dementia. JAMA Psychiatry.
Kessing, L.V. et al., 2017b. Lithium in drinking water and the incidence of bipolar disorder: A
nation-wide population-based study. Bipolar Disord.
Kim, M.-G. et al., 2015. Blood Lead Levels and Cause-Specific Mortality of Inorganic Lead-
Exposed Workers in South Korea M. Costa, ed. PLOS ONE, 10(10), p.e0140360.
Kivipelto, M., 2001. Midlife vascular risk factors and Alzheimer’s disease in later life: Longitudi-
nal, population based study. BMJ, 322(7300), pp.1447–1451.
Klunk, W.E. et al., 2004. Imaging brain amyloid in Alzheimer’s disease with Pittsburgh
Compound-B. Ann Neurol, 55, pp.306–319.
147
Klunk, W.E. et al., 2015. The Centiloid Project: Standardizing quantitative amyloid plaque
estimation by PET. Alzheimer’s & Dementia, 11(1), pp.1–15.e4.
Knudsen, N.N. et al., 2017. Lithium in Drinking Water and Incidence of Suicide: A Nationwide
Individual-Level Cohort Study with 22 Years of Follow-Up. Int J Environ Res Public Health, 14.
Koseoglu, E. et al., 2017. Trace metal concentrations in hair and nails from Alzheimer’s disease
patients: Relations with clinical severity. Journal of trace elements in medicine and biology : organ of
the Society for Minerals and Trace Elements (GMS), 39, pp.124–128.
Kramarow, E. & Tejada-Vera, B., 2019. Dementia Mortality in the United States, 20002017.
National Vital Statistics Reports, Centers for Disease Control and Prevention, 68(2), p.29.
Laforce, R. et al., 2016. Canadian Consensus Guidelines on Use of Amyloid Imaging in Canada:
Update and Future Directions from the Specialized Task Force on Amyloid imaging in Canada. Canadian
Journal of Neurological Sciences / Journal Canadien des Sciences Neurologiques, 43(04), pp.503–512.
Lahiri, D.K., Maloney, B. & Zawia, N.H., 2009. The LEARn model: An epigenetic explanation
for idiopathic neurobiological diseases. Molecular Psychiatry, 14(11), pp.992–1003.
Laine, C. et al., 2007. Reproducible Research: Moving toward Research the Public Can Really
Trust. Annals of Internal Medicine, 146(6), p.450.
Lammertsma, A.A. & Hume, S.P., 1996. Simplified Reference Tissue Model for PET Receptor
Studies. NeuroImage, 4(3), pp.153–158.
Langa, K.M., 2015. Is the risk of Alzheimer’s disease and dementia declining? Alzheimer’s
Research & Therapy, 7(1).
Langa, K.M., Foster, N.L. & Larson, E.B., 2004. Mixed Dementia: Emerging Concepts and
Therapeutic Implications. JAMA, 292(23), p.2901.
Lanphear, B.P. et al., 2018. Low-level lead exposure and mortality in US adults: A population-
based cohort study., p.8.
Launer, L.J. et al., 2000. Midlife blood pressure and dementia: The HonoluluAsia aging study73.
Neurobiology of Aging, 21(1), pp.49–55.
Li, J.S., Yang, F. & Zhao, X., 2006. [Protection of lithium on hippocampal cholecystokinin and
nitric oxide synthase neuron in lead exposed rats]. Zhonghua lao dong wei sheng zhi ye bing za zhi =
Zhonghua laodong weisheng zhiyebing zazhi = Chinese journal of industrial hygiene and occupational
diseases, 24(3), pp.147–150.
148
Liaugaudaite, V. et al., 2017. Lithium levels in the public drinking water supply and risk of
suicide: A pilot study. Journal of Trace Elements in Medicine and Biology, 43, pp.197–201.
Livingston, G. et al., 2017. Dementia prevention, intervention, and care. The Lancet, 390(10113),
pp.2673–2734.
Logan, J. et al., 1996. Distribution Volume Ratios without Blood Sampling from Graphical
Analysis of PET Data. Journal of Cerebral Blood Flow & Metabolism, 16(5), pp.834–840.
Lopresti, B.J. et al., 2005. Simplified quantification of Pittsburgh Compound B amyloid imaging
PET studies: A comparative analysis. J Nucl Med, 46, pp.1959–1972.
Maillard, P. et al., 2013. FLAIR and Diffusion MRI Signals Are Independent Predictors of White
Matter Hyperintensities. American Journal of Neuroradiology, 34(1), pp.54–61.
Maillard, P. et al., 2012. Effects of systolic blood pressure on white-matter integrity in young
adults in the Framingham Heart Study: A cross-sectional study. The Lancet Neurology, 11(12), pp.1039–
1047.
Makin, S., 2018. The amyloid hypothesis on trial. Nature, 559(7715), pp.S4–S7.
Maloney, B. & Lahiri, D.K., 2016. Epigenetics of dementia: Understanding the disease as a
transformation rather than a state. The Lancet Neurology, 15(7), pp.760–774.
Mason, L.H., Harp, J.P. & Han, D.Y., 2014. Pb Neurotoxicity: Neuropsychological Effects of
Lead Toxicity. BioMed Research International, 2014, pp.1–8.
Matsuzaki, H. et al., 2017. Re-analysis of the association of temperature or sunshine with hy-
perthymic temperament using lithium levels of drinking water. Journal of Affective Disorders, 223,
pp.126–129.
Mauer, S., Vergne, D. & Ghaemi, S.N., 2014. Standard and trace-dose lithium: A systematic
review of dementia prevention and other behavioral benefits. Australian & New Zealand Journal of
Psychiatry, 48(9), pp.809–818.
McGuinness, B. et al., 2009. Blood pressure lowering in patients without prior cerebrovascular
disease for prevention of cognitive impairment and dementia Cochrane Dementia and Cognitive Improve-
ment Group, ed. Cochrane Database of Systematic Reviews.
McIntosh, K.G. et al., 2012. Evaluation of a Prototype Point-of-care Instrument Based on
Monochromatic X-Ray Fluorescence Spectrometry: Potential for Monitoring Trace Element Status of
Subjects with Neurodegenerative Disease. Journal of Toxicology and Environmental Health, Part A,
149
75(21), pp.1253–1268.
McKhann, G.M. et al., 2011. The diagnosis of dementia due to Alzheimer’s disease: Recommenda-
tions from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines
for Alzheimer’s disease. Alzheimers Dement, 7, pp.263–9.
McNamee, R.L. et al., 2009. Consideration of optimal time window for Pittsburgh compound B
PET summed uptake measurements. J Nucl Med., 50, pp.348–355.
McNeill, F.E. et al., 2017. The decrease in population bone lead levels in Canada between 1993
and 2010 as assessed by in Vivo XRF. Physiological Measurement, 39(1), p.015005.
Meyer, P.T. et al., 2011. Dual-Biomarker Imaging of Regional Cerebral Amyloid Load and Neu-
ronal Activity in Dementia with PET and 11C-Labeled Pittsburgh Compound B. Journal of Nuclear
Medicine, 52(3), pp.393–400.
Moher, D. et al., 2009. Preferred reporting items for systematic reviews and meta-analyses: The
PRISMA statement. PLoS medicine, 6(7), p.e1000097.
Montgomery, S.A. & Asberg, M., 1979. A new depression scale designed to be sensitive to change.
The British Journal of Psychiatry, 134(4), pp.382–389.
Moriguchi, S. et al., 2019. Monoamine Oxidase B Total Distribution Volume in the Prefrontal
Cortex of Major Depressive Disorder: An [ 11 C]SL25.1188 Positron Emission Tomography Study. JAMA
Psychiatry.
Müller, U.C., Deller, T. & Korte, M., 2017. Not just amyloid: Physiological functions of the
amyloid precursor protein family. Nature Reviews Neuroscience, 18(5), pp.281–298.
Müller-Gärtner, H.W. et al., 1992. Measurement of Radiotracer Concentration in Brain Gray Mat-
ter Using Positron Emission Tomography: MRI-Based Correction for Partial Volume Effects. Journal
of Cerebral Blood Flow & Metabolism, 12(4), pp.571–583.
Navas-Acien, A. et al., 2007. Lead Exposure and Cardiovascular DiseaseA Systematic Review.
Environmental Health Perspectives, 115(3), pp.472–482.
Nichols, T.E. et al., 2017. Best practices in data analysis and sharing in neuroimaging using MRI.
Nature Neuroscience, 20(3), pp.299–303.
Nicoll, J.A.R., Roberts, G.W. & Graham, D.I., 1996. Amyloid β-Protein, APOE Genotype and
Head Injury. Annals of the New York Academy of Sciences, 777(1), pp.271–275.
150
Niklowitz, W.J. & Mandybur, T.I., 1975. Neurofibrillary Changes Following Childhood Lead
EncephalopathyCase Report. Journal of Neuropathology & Experimental Neurology, 34(5), pp.445–455.
Norra, C. et al., 2010. Differential effects of endogenous lithium on neurobehavioural functioning:
A study on auditory evoked potentials. Psychiatry Research, 178(1), pp.176–181.
Nriagu, J.O., 1990. The rise and fall of leaded gasoline. Science of the total environment, 92,
pp.13–28.
O’Brien, C., 1996. Science History: Auguste D. And Alzheimer’s Disease. Science, 273(5271),
pp.28–0.
Ohgami, H. et al., 2009. Lithium levels in drinking water and risk of suicide. The British Journal
of Psychiatry, 194(5), pp.464–465.
Oikonen, V., 2018. TPCCLIB.
Olsson, B. et al., 2016. CSF and blood biomarkers for the diagnosis of Alzheimer’s disease: A
systematic review and meta-analysis. The Lancet Neurology, 15(7), pp.673–684.
Organization, W.H. ed., 2009. Global health risks: Mortality and burden of disease attributable to
selected major risks, Geneva, Switzerland: World Health Organization.
Ossenkoppele, R. et al., 2018. Discriminative Accuracy of [18F]Flortaucipir Positron Emission
Tomography for Alzheimer Disease vs Other Neurodegenerative Disorders. JAMA, 320(11), p.1151.
Ownby, R.L. et al., 2006. Depression and risk for Alzheimer disease: Systematic review, meta-
analysis, and metaregression analysis. Archives of General Psychiatry, 63, pp.530–8.
Paglia, G. et al., 2016. Distinctive Pattern of Serum Elements During the Progression of
Alzheimer’s Disease. Scientific Reports, 6(1).
Palmqvist, S. et al., 2015. Detailed comparison of amyloid PET and CSF biomarkers for identi-
fying early Alzheimer disease. Neurology, 85(14), pp.1240–1249.
Panza, F. et al., 2019. Amyloid-β immunotherapy for alzheimer disease: Is it now a long shot?:
Aβ Immunotherapy Gamble. Annals of Neurology, 85(3), pp.303–315.
Park, J.H. et al., 2018. Depression Plays a Moderating Role in the Cognitive Decline Associated
With Changes of Brain White Matter Hyperintensities. The Journal of Clinical Psychiatry, 79(5), pp.e1–
e7.
Pearson, C., Janz, T. & Ali, J., 2013. Mental and substance use disorders in Canada. Health at
151
a Glance, p.10.
Peng, R.D., Dominici, F. & Zeger, S.L., 2006. Reproducible Epidemiologic Research. American
Journal of Epidemiology, 163(9), pp.783–789.
Peretti, D.E. et al., 2019. Relative cerebral flow from dynamic PIB scans as an alternative for
FDG scans in Alzheimer’s disease PET studies. PLoS ONE, 14(1), p.e0211000.
Pino, A. et al., 2005. Discriminant analysis to study trace elements in biomonitoring: An appli-
cation on neurodegenerative diseases. Ann Ist Super Sanità, 41(2), p.6.
Poldrack, R.A. et al., 2017. Scanning the horizon: Towards transparent and reproducible neu-
roimaging research. Nature Reviews Neuroscience, 18(2), pp.115–126.
Pollard, T. et al., 2016. Template for writing a PhD thesis in Markdown. Zenodo.
Ponto, L.L.B. et al., 2019. Early Phase PIB-PET as a Surrogate for Global and Regional Cerebral
Blood Flow Measures: Early PIB as Surrogate for CBF. Journal of Neuroimaging, 29(1), pp.85–96.
Portnow, L.H. & Vaillancourt, D.E., 2013. The history of cerebral PET scanning., p.5.
Potter Gillian M. et al., 2010. Counting Cavitating Lacunes Underestimates the Burden of Lacu-
nar Infarction. Stroke, 41(2), pp.267–272.
Power, M.C. et al., 2014. Lead exposure and rate of change in cognitive function in older women.
Environmental Research, 129, pp.69–75.
Prince, M., 1998. Is Chronic Low-Level Lead Exposure in Early Life an Etiologic Factor in
Alzheimer’s Disease? Epidemiology, 9(6), pp.618–621.
Prins, N.D. & Scheltens, P., 2015. White matter hyperintensities, cognitive impairment and
dementia: An update. Nature Reviews Neurology, 11(3), pp.157–165.
Provenzano, F.A. et al., 2013. White Matter Hyperintensities and Cerebral Amyloidosis: Neces-
sary and Sufficient for Clinical Expression of Alzheimer Disease? JAMA Neurology, 70(4), p.455.
Puglielli, L., Tanzi, R.E. & Kovacs, D.M., 2003. Alzheimer’s disease: The cholesterol connection.
Nature Neuroscience, 6(4), pp.345–351.
Rabinovici, G.D., 2015. The Translational Journey of Brain β-Amyloid Imaging: From Positron
Emission Tomography to Autopsy to Clinic. JAMA Neurology, 72(3), p.265.
Rajabli, F. et al., 2018. Ancestral origin of ApoE E4 Alzheimer disease risk in Puerto Rican and
152
African American populations G. S. Barsh, ed. PLOS Genetics, 14(12), p.e1007791.
Rajan, K.B. et al., 2018. Blood pressure and risk of incident Alzheimer’s disease dementia by
antihypertensive medications and APOE E4 allele: BP and Risk of Incident AD. Annals of Neurology,
83(5), pp.935–944.
R Core Team, 2018. R: A language and environment for statistical computing.
R Core Team, 2019. Writing R Extensions,
Reuben, A. et al., 2017. Association of Childhood Blood Lead Levels With Cognitive Function
and Socioeconomic Status at Age 38 Years and With IQ Change and Socioeconomic Mobility Between
Childhood and Adulthood. JAMA, 317(12), p.1244.
Rodell, A. et al., 2013. Washout allometric reference method (WARM) for parametric analysis of
[11C]PIB in human brains. Frontiers in Aging Neuroscience, 5, p.45.
Rodell, A.B. et al., 2017. Cerebral Blood Flow and Aβ-Amyloid Estimates by WARM Analysis
of [11C]PiB Uptake Distinguish among and between Neurodegenerative Disorders and Aging. Frontiers
in Aging Neuroscience, 8, p.321.
Rodriguez-Vieitez, E. et al., 2016. Comparison of Early-Phase 11C-Deuterium-L-Deprenyl and
11C-Pittsburgh Compound B PET for Assessing Brain Perfusion in Alzheimer Disease. Journal of
Nuclear Medicine, 57(7), pp.1071–1077.
Rodriguez-Vieitez, E. et al., 2017. Comparability of [18F]THK5317 and [11C]PIB blood flow
proxy images with [18F]FDG positron emission tomography in Alzheimer’s disease. Journal of Cerebral
Blood Flow & Metabolism, 37(2), pp.740–749.
Rodushkin, I. & Ödman, F., 2001. Assessment of the contamination from devices used for sam-
pling and storage of whole blood and serum for element analysis. Journal of Trace Elements in Medicine
and Biology, 15(1), pp.40–45.
Roehr, B., 2017. Herbert L Needleman. BMJ, p.j3684.
rOpenSci, 2019. About us.
Rostomian, A.H. et al., 2011. Early 11C-PIB Frames and 18F-FDG PET Measures Are Compa-
rable: A Study Validated in a Cohort of AD and FTLD Patients. Journal of Nuclear Medicine.
Saczynski, J.S. et al., 2010. Depressive symptoms and risk of dementia: The Framingham Heart
Study. Neurology, 75(1), pp.35–41.
153
Satizabal, C.L. et al., 2016. Incidence of Dementia over Three Decades in the Framingham Heart
Study. New England Journal of Medicine, 374(6), pp.523–532.
Scheinin, N.M. et al., 2007. Biodistribution and Radiation Dosimetry of the Amyloid Imaging
Agent 11C-PIB in Humans. Journal of Nuclear Medicine, 48(1), pp.128–133.
Schmidt, P. et al., 2012. An automated tool for detection of FLAIR-hyperintense white-matter
lesions in Multiple Sclerosis. NeuroImage, 59(4), pp.3774–3783.
Schneider, J.A. et al., 2007. Mixed brain pathologies account for most dementia cases in
community- dwelling older persons. Neurology, 69(24).
Schrauzer, G.N., 2002. Lithium: Occurrence, dietary intakes, nutritional essentiality. Journal of
the American College of Nutrition, 21(1), pp.14–21.
Schrauzer, G.N. & Shrestha, K.P., 1990. Lithium in Drinking-Water and the Incidences of Crimes,
Suicides, and Arrests Related to Drug Addictions. Biological Trace Element Research, 25, pp.105–113.
Schwarzer, G., 2007. Meta: An R package for meta-analysis. R News, 7(3), pp.40–45.
Selkoe, D.J. & Hardy, J., 2016. The amyloid hypothesis of Alzheimer’s disease at 25 years. EMBO
Molecular Medicine, 8(6), pp.595–608.
Selkoe, D., Mandelkow, E. & Holtzman, D., 2012. Deciphering Alzheimer Disease. Cold Spring
Harbor Perspectives in Medicine, 2(1), pp.a011460–a011460.
Sexton, C.E. et al., 2011. A meta-analysis of diffusion tensor imaging in mild cognitive impairment
and Alzheimer’s disease. Neurobiology of Aging, 32(12), pp.2322.e5–2322.e18.
Shih, R.A. et al., 2007. Cumulative lead dose and cognitive function in adults: A review of studies
that measured both blood lead and bone lead. Environ Health Perspect, 115, pp.483–92.
Shiotsuki, I. et al., 2016. Trace lithium is inversely associated with male suicide after adjustment
of climatic factors. Journal of Affective Disorders, 189, pp.282–286.
Simon, M.J. & Iliff, J.J., 2016. Regulation of cerebrospinal fluid (CSF) flow in neurodegenerative,
neurovascular and neuroinflammatory disease. Biochimica et Biophysica Acta (BBA) - Molecular Basis
of Disease, 1862(3), pp.442–451.
Singh, B. & Dhawan, D., 2000. Effect of lithium on thyroidal (131)Iodine uptake, its clearance,
and circulating levels of triiodothyronine and thyroxine in lead-treated rats,
Singh, B. et al., 1994. Combined action of lead and lithium on essential and nonessential elements
154
in rat blood. Biological trace element research, 46(1), pp.15–28.
Singh-Manoux, A. et al., 2017. Trajectories of Depressive Symptoms Before Diagnosis of Dementia:
A 28-Year Follow-up Study. JAMA Psychiatry, 74(7), p.712.
Skalny, A.V. et al., 2017. Serum trace elements are interrelated with hormonal imbalance in men
with acute ischemic stroke. Journal of Trace Elements in Medicine and Biology.
Smith, K.A. & Cipriani, A., 2017. Lithium and suicide in mood disorders: Updated meta-review
of the scientific literature. Bipolar Disorders, 19(7), pp.575–586.
Snyder, H.M. et al., 2015. Vascular contributions to cognitive impairment and dementia including
Alzheimer’s disease. Alzheimer’s & Dementia, 11(6), pp.710–717.
Sojkova, J. et al., 2015. Voxelwise Relationships Between Distribution Volume Ratio and Cerebral
Blood Flow: Implications for Analysis of -Amyloid Images. Journal of Nuclear Medicine, 56(7), pp.1042–
1047.
Stretesky, P.B. & Lynch, M.J., 2001. The Relationship Between Lead Exposure and Homicide.
Sugawara, N. et al., 2013. Lithium in Tap Water and Suicide Mortality in Japan. International
Journal of Environmental Research and Public Health, 10(11), pp.6044–6048.
Szabo, S.T. et al., 2016. Comparison of Metal Levels between Postmortem Brain and Ventricular
Fluid in Alzheimer’s Disease and Nondemented Elderly Controls. Toxicological sciences : an official
journal of the Society of Toxicology, 150(2), pp.292–300.
Takashima, A., 2006. GSK-3 is essential in the pathogenesis of Alzheimer’s disease. Journal of
Alzheimer’s Disease, 9(s3), pp.309–317.
Tapiola, T. et al., 2009. Cerebrospinal Fluid β-Amyloid 42 and Tau Proteins as Biomarkers of
Alzheimer-Type Pathologic Changes in the Brain. Archives of Neurology, 66(3).
The SPRINT MIND Investigators for the SPRINT Research Group et al., 2019. Effect of Intensive
vs Standard Blood Pressure Control on Probable Dementia: A Randomized Clinical Trial. JAMA, 321(6),
pp.553–561.
Tiepolt, S. et al., 2019. Early after Administration [11C]PiB PET Images Correlate with Cognitive
Dysfunction Measured by the CERAD Test Battery. Journal of Alzheimer’s Disease, 68(1), pp.65–76.
Tolboom, N. et al., 2009. Relationship of Cerebrospinal Fluid Markers to 11C-PiB and 18F-
FDDNP Binding. Journal of Nuclear Medicine, 50(9), pp.1464–1470.
155
Tosto, G. et al., 2014. Predicting Aggressive Decline in Mild Cognitive Impairment: The Impor-
tance of White Matter Hyperintensities. JAMA Neurology, 71(7), p.872.
Tournier, J.-D., Calamante, F. & Connelly, A., 2012. MRtrix: Diffusion tractography in crossing
fiber regions. International Journal of Imaging Systems and Technology, 22(1), pp.53–66.
Uitti, R.J. et al., 1989. Regional metal concentrations in Parkinson’s disease, other chronic
neurological diseases, and control brains. The Canadian journal of neurological sciences. Le journal
canadien des sciences neurologiques, 16(3), pp.310–314.
van Berckel, B.N.M. et al., 2013. Longitudinal Amyloid Imaging Using 11C-PiB: Methodologic
Considerations. Journal of Nuclear Medicine, 54(9), pp.1570–1576.
van den Berg, E. et al., 2018. White Matter Hyperintensities and Cognition in Mild Cogni-
tive Impairment and Alzheimer’s Disease: A Domain-Specific Meta-Analysis D. Libon, ed. Journal of
Alzheimer’s Disease, 63(2), pp.515–527.
van Duijn, C.M. et al., 1991. Familial aggregation of Alzheimer’s disease and related disorders:
A collaborative re-analysis of case-control studies. International Journal of Epidemiology, 20 Suppl 2,
pp.S13–20.
van Straaten, E.C. et al., 2006. Impact of White Matter Hyperintensities Scoring Method on
Correlations With Clinical Data: The LADIS Study. Stroke, 37(3), pp.836–840.
Veraart, J., Fieremans, E. & Novikov, D.S., 2016. Diffusion MRI noise mapping using random
matrix theory. Magnetic Resonance in Medicine, 76(5), pp.1582–1593.
Veraart, J. et al., 2016. Denoising of diffusion MRI using random matrix theory. NeuroImage,
142, pp.394–406.
Villemagne, V.L. & Okamura, N., 2014. In vivo tau imaging: Obstacles and progress. Alzheimer’s
& Dementia, 10(3), pp.S254–S264.
Villemagne, V.L. et al., 2011. Longitudinal assessment of Abeta and cognition in aging and
Alzheimer disease. Ann Neurol, 69, pp.181–92.
Vita, A., De Peri, L. & Sacchetti, E., 2015. Lithium in drinking water and suicide prevention: A
review of the evidence. International Clinical Psychopharmacology, 30(1), pp.1–5.
Viticchi, G. et al., 2015. Framingham risk score can predict cognitive decline progression in
Alzheimer’s disease. Neurobiology of Aging, 36(11), pp.2940–2945.
156
Viticchi, G. et al., 2017. Framingham Risk Score and the Risk of Progression from Mild Cognitive
Impairment to Dementia M. Tombini, ed. Journal of Alzheimer’s Disease, 59(1), pp.67–75.
Voors, A.W., 1970. Minerals in the municipal water and atherosclerotic heart death.
Wallace, J., 2014. Calcium dysregulation, and lithium treatment to forestall Alzheimer’s disease
a merging of hypotheses. Cell Calcium, 55(3), pp.175–181.
Wallin, C. et al., 2017. Alzheimer’s disease and cigarette smoke components: Effects of nicotine,
PAHs, and Cd(II), Cr(III), Pb(II), Pb(IV) ions on amyloid-beta peptide aggregation. Scientific reports,
7(1), p.14423.
Wang, J.T. et al., 2015. Protective effects of lithium against lead-induced toxicities in multiple
systems of adult mouse. Toxicology Research, 4, pp.1523–1534.
Wardlaw, J.M. et al., 2013. Neuroimaging standards for research into small vessel disease and its
contribution to ageing and neurodegeneration. The Lancet Neurology, 12(8), pp.822–838.
Weuve, J. et al., 2009. Cumulative exposure to lead in relation to cognitive function in older
women. Environmental health perspectives, 117(4), pp.574–580.
Whitmer, R.A. et al., 2005. Midlife cardiovascular risk factors and risk of dementia in late life.
Neurology, 64(2), pp.277–281.
Wickham, H., 2015. R Packages: Organize, Test, Document, and Share Your Code 1st ed.,
O’Reilly Media.
Wilson, P.W.F. et al., 1998. Prediction of Coronary Heart Disease Using Risk Factor Categories.
Circulation, 97(18), pp.1837–1847.
Wolk, D.A. et al., 2017. Medial temporal lobe subregional morphometry using high resolution
MRI in Alzheimer’s disease. Neurobiology of Aging, 49, pp.204–213.
World Health Organization, 2011. Lead in Drinking-water: Background document for developing
WHO Guidelines for Drinking-water Quality.
Wu, J. et al., 2008. Alzheimer’s Disease (AD)-Like Pathology in Aged Monkeys after Infantile
Exposure to Environmental Metal Lead (Pb): Evidence for a Developmental Origin and Environmental
Link for AD. Journal of Neuroscience, 28(1), pp.3–9.
Wu, Y. & Carson, R.E., 2002. Noise Reduction in the Simplified Reference Tissue Model for
Neuroreceptor Functional Imaging. Journal of Cerebral Blood Flow & Metabolism, 22(12), pp.1440–
157
1452.
Xu, L. et al., 2018. Circulatory Levels of Toxic Metals (Aluminum, Cadmium, Mercury, Lead) in
Patients with Alzheimer’s Disease: A Quantitative Meta-Analysis and Systematic Review. Journal of
Alzheimer’s Disease, 62(1), pp.361–372.
Yang, F., Li, J. & Yang, F., 2004. [Neuroprotection of lithium and the antagonism of lithium
to toxic effects of lead on primary culturing cerebrocortical neurons in vitro]. Wei Sheng Yan Jiu =
Journal of Hygiene Research, 33(6), pp.671–674.
Yang, Y.-W. et al., 2018. Risk of Alzheimer’s disease with metal concentrations in whole blood
and urine: A caseControl study using propensity score matching. Toxicology and Applied Pharmacology,
356, pp.8–14.
Yatham, L.N. et al., 2013. Canadian Network for Mood and Anxiety Treatments (CANMAT)
and International Society for Bipolar Disorders (ISBD) collaborative update of CANMAT guidelines for
the management of patients with bipolar disorder: Update 2013: CANMAT Guidelines for Bipolar
Disorder. Bipolar Disorders, 15(1), pp.1–44.
Yoo, B.I. et al., 2014. Application of variable threshold intensity to segmentation for white matter
hyperintensities in fluid attenuated inversion recovery magnetic resonance images. Neuroradiology, 56(4),
pp.265–281.
Yuan, Y., Gu, Z.-X. & Wei, W.-S., 2009. FluorodeoxyglucosePositron-Emission Tomography,
Single-Photon Emission Tomography, and Structural MR Imaging for Prediction of Rapid Conversion
to Alzheimer Disease in Patients with Mild Cognitive Impairment: A Meta-Analysis. American Journal
of Neuroradiology, 30(2), pp.404–410.
Zarse, K. et al., 2011. Low-dose lithium uptake promotes longevity in humans and metazoans.
European Journal of Nutrition, 50(5), pp.387–389.
Zawia, N.H. & Basha, M.R., 2005. Environmental risk factors and the developmental basis for
Alzheimer’s disease. Reviews in the Neurosciences, 16(4), pp.325–337.
Zeydan, B. et al., 2019. Investigation of white matter PiB uptake as a marker of white matter
integrity. Annals of Clinical and Translational Neurology, 6(4), pp.678–688.
Zhang, Y. et al., 2009. White matter damage in frontotemporal dementia and Alzheimer’s disease
measured by diffusion MRI. Brain, 132(9), pp.2579–2592.
Zhao, N. et al., 2018. Apolipoprotein E, Receptors, and Modulation of Alzheimer’s Disease.
158
Biological Psychiatry, 83(4), pp.347–357.
159