Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Vasoprotective functions of high-density lipoproteins on the cerebrovasculature and their relevance for… Button, Emily Beth 2020

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
24-ubc_2020_may_button_emily.pdf [ 21.28MB ]
Metadata
JSON: 24-1.0389964.json
JSON-LD: 24-1.0389964-ld.json
RDF/XML (Pretty): 24-1.0389964-rdf.xml
RDF/JSON: 24-1.0389964-rdf.json
Turtle: 24-1.0389964-turtle.txt
N-Triples: 24-1.0389964-rdf-ntriples.txt
Original Record: 24-1.0389964-source.json
Full Text
24-1.0389964-fulltext.txt
Citation
24-1.0389964.ris

Full Text

  VASOPROTECTIVE FUNCTIONS OF HIGH-DENSITY LIPOPROTEINS ON THE CEREBROVASCULATURE AND THEIR RELEVANCE FOR ALZHEIMER’S DISEASE  by   EMILY BETH BUTTON   B.Sc., The University of Waterloo, 2014       A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY  in  THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES  (Pathology and Laboratory Medicine)    THE UNIVERSITY OF BRITISH COLUMBIA  (Vancouver)         April 2020   © Emily Beth Button, 2020   ii  The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, the dissertation entitled: Vasoprotective Functions of High-Density Lipoproteins on the Cerebrovasculature and their Relevance for Alzheimer's Disease  submitted by Emily Button in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Pathology and Laboratory Medicine  Examining Committee: Dr. Cheryl Wellington, Professor, Department of Pathology and Laboratory Medicine, UBC Supervisor  Dr. Haakon Nygaard, Assistant Professor, Division of Neurology, UBC Supervisory Committee Member Dr. Lara Boyd, Professor, Department of Physical Therapy, UBC University Examiner Dr. Thalia Field, Associate Professor, Division of Neurology, UBC University Examiner  Additional Supervisory Committee Members: Dr. Mari DeMarco, Clinical Associate Professor, Department of Pathology and Laboratory Medicine, UBC Supervisory Committee Member Dr. Shernaz Bamji, Professor, Department of Cellular & Physiological Sciences, School of Biomedical Engineering, UBC  Supervisory Committee Member   iii Abstract  One in eleven Canadians over the age of 65 suffers from dementia, the most common form being Alzheimer’s disease (AD). AD has traditionally been characterized by the presence of amyloid beta (Aβ) plaques and neurofibrillary tangles and in more recent years also by neuroinflammation and cerebrovascular dysfunction. We hypothesize that circulating high-density lipoproteins (HDL) may protect against such cerebrovascular dysfunction and inflammation to reduce AD risk. HDL in healthy people is already known to have several vasoprotective functions on cells in peripheral arteries and higher levels of HDL cholesterol (HDL-C) or apolipoprotein A-I (apoA-I), its primary protein component, in blood are associated with reduced AD risk. Furthermore, studies in AD transgenic mouse models suggest that HDL protects against vascular Aβ deposition, memory deficits, and neuroinflammation. This thesis used mouse models and in vitro human cellular models to extend the current knowledge on the protective effects of HDL against AD.  First, transgenic AD mice were employed to further investigate the role of HDL on cerebrovascular-specific pathologies in AD using a genetic approach and a pharmacological approach. Genetic loss of apoA-I in AD transgenic mice exacerbated vascular Aβ deposition, activation of cerebrovascular endothelial cell and vessel-associated astrocytes, global amyloid burden, neuroinflammation, and cognitive deficits. Pharmacological modulation of cholesterol metabolism exclusively outside of the brain resulted in reduced neuroinflammation, activation of cerebrovascular endothelial cells, and cognitive deficits.  Next, 2-dimensional (2D) in vitro cell cultures of human brain-derived endothelial cells (EC) and 3D bioengineered human arteries were used to show that several of the known vasoprotective functions of HDL extend to brain cells and that HDL has novel protective functions against Aβ vascular deposition and Aβ-induced vascular inflammation in bioengineered tissues. Finally, efforts were made to develop high-throughput assays of these novel HDL functions in order to evaluate whether these functions are lost in people with AD or AD risk factors.     iv In summary, this thesis suggests novel pathways by which circulating HDL can promote cerebrovascular health using in vivo and in vitro models and suggests that improving the levels of functional HDL may be a valuable therapeutic or preventative strategy for AD.    v Lay Summary  Alzheimer’s disease (AD) is a devastating illness affecting half a million Canadians. People with AD have clumps of amyloid beta proteins in their brain and most have problems with blood vessel function in the brain. This thesis aimed to learn how high-density lipoproteins (HDL), particles in the blood made of fats and proteins, may protect brain blood vessels in AD. HDL already have known beneficial effects on blood vessels outside of the brain and there are links between high levels of HDL and lower risk of AD. We used mouse models and test tube models of human blood vessels to show that HDL prevents harmful deposits of amyloid beta in the walls of brain blood vessels and protects brain blood vessels against inflammation. Overall, this thesis suggests that raising the levels of functional HDL through drugs or healthy lifestyle changes may protect against detrimental brain blood vessels changes in AD.   vi Preface  All work in this thesis was performed at the Djavad Mowafaghian Centre for Brain Health and the Centre for Disease Modeling at the University of British Columbia or the Centre for Molecular Medicine and Therapeutics at the Child and Family Research Institute in Vancouver, BC unless otherwise noted. All projects and methods were approved by the University of British Columbia Research Ethics Board. • Biosafety approval: B14-0136 (Factors affecting the risk of Alzheimer’s Disease), B15-0066 (Biosafety for AD and TBI) • Animal care approval: A14-0003 (HDL, vascular biology and Alzheimer’s Disease), A14-0057 (ABCA1 and apoE in Alzheimer’s Disease) • Human ethics approval: H14-03357 (Lipoprotein function in cerebrovascular health), H17-01776 (Lipoprotein function and cerebrovascular health in diabetes). • Radiation safety: R17-0017 (Open Source Radiation Permit) • Personal animal care, safety, and ethics training: Rodent Biology and Husbandry (00000151), Canadian Council on Animal Care (6909-14), Rodent Anaesthesia (00000088), Animal Ethics (6906-14), Biological Safety (2019-nL4uo), Chemical Safety (2014-5oB3y), Radiation Safety (2015-RS1f), Tri-Council Policy Statement 2 Course on Research Ethics (Date of issue 16 June, 2015)  Details on the personal contributions and contribution of collaborators to the data in Chapter 2, 3, 4, and 5 are available in Appendix A.  Portions of Chapter 1 have been previously published in two review articles:  Boyce G, Button E, Soo S, Wellington C (2017) Pleiotropic vasoprotective functions of high-density lipoproteins (HDL). Journal of Biomedical Research. May 26. doi: 10.7555/JBR.31.20160103.   This review was designed and prepared by all co-authors. I was responsible for approximately 40% of the writing and led the design and creation of the figure and table in the published review   vii also appearing in Chapter 1 (Figure 1.2 and Table 1.1). I wrote sections focused on high-density lipoproteins (HDL) while Guilaine Boyce wrote sections on the cerebrovascular system, Sonja Soo wrote other minor sections, and Dr. Cheryl Wellington had major contributions to the concepts, structure, and revisions and wrote sections on Gaps and Opportunities. I wrote all published sections appearing within this thesis.  Button EB, Robert J, Caffrey TM, Fan J, Zhao W, Wellington CL (2019) HDL from an Alzheimer’s disease perspective. Current Opinion in Lipidology. Jun;30(3): 224-234.   All co-authors contributed to the concepts, structures, and content of this review. I wrote the sections on HDL and vascular resilience, epidemiological evidence linking HDL and Alzheimer’s disease (AD), HDL effects in AD animal models, considerations for evaluating HDL and an AD therapeutic, and considerations for HDL as an AD biomarker in addition to contributing to sections on the cerebrovasculature in AD in total comprising approximately 60% of the writing. I also led the design and creation of the table and figure in the published review that appear in Chapter 6 (Figure 6.1 and Table 6.1). Dr. Jerome Robert wrote the introduction and the section on the cerebrovasculature in AD, Dr. Tara Caffrey wrote the section on in vitro models, and Dr. Jianjia Fan and Wenchen Zhao wrote the paragraph on drugs targeting adenosine triphosphate binding cassette transporter A1 (ABCA1).  Sections from the above publications make up approximately 25% of Chapter 1, approximately 70% of that being originally written by myself. The remaining 75% of Chapter 1 is my own writing expanding the published sections or entirely new sections.   A version of Chapter 2 has been published in Button EB, Boyce GK, Wilkinson A, Stukas S, Hayat A, Fan J, Wadsworth BJ, Robert J, Martens KM, Wellington CL (2019) ApoA-I deficiency increases cortical amyloid deposition, cerebral amyloid angiopathy, cortical and hippocampal astrogliosis, and amyloid-association astrocyte reactivity in APP/PS1 mice. Alzheimer’s Research and Therapy. 11(4): 44.     viii This study was originally designed by Arooj Hayat, Guilaine Boyce, Sophie Stukas, and Cheryl Wellington. I took over the re-design and led the study after all mouse tissues were collected. Specifically, I performed plasma lipid analysis, ribonucleic acid (RNA) isolation, enzyme linked immunosorbent assay (ELISA) measurements, image processing, imaging analysis, and half of the real-time quantitative polymerase chain reaction (RT-qPCR) analysis. I led all of the data analysis and interpretation to re-orient the direction of the study and prepare it for publication including writing the manuscript and generating figures.   Contributions of co-authors: • Guilaine Boyce contributed the design to the original study, was responsible for the collection of experimental mouse tissues, and was responsible for generating the behavior data. • Anna Wilkinson was responsible for protein extraction, fixed brain sectioning and immunohistological staining including contributions to the developing of co-staining methods. • Dr. Sophie Stukas contributed to the design of the original study. • Arooj Hayat contributed to the design of the original study as well as breeding and tissue collection in a related pilot study. • Dr. Jianjia Fan contributed to RT-qPCR analysis. • Brennan Wadsworth made major contributions to the development of image analysis protocols.  • Dr. Jerome Robert contributed to interpretation of the data. • Dr. Kris Martens was responsible for the breeding of mouse cohorts. • Dr. Cheryl Wellington led the design of the original study and had major contributions to the re-design of the study, data interpretation, and manuscript revisions.  I designed the study in Chapter 3 along with Dr. Sophie Stukas and Dr. Cheryl Wellington. I led the treatments of the mice, tissue collection, data collection, data analysis and interpretation, writing, and figure generation. In terms of data collection, I was specifically responsible for   ix analysis of plasma lipids, RNA extraction, RT-qPCR analysis, ELISA measurements, image processing, and image analysis.   Contributions of others: • Dr. Sophie Stukas contributed to the design of the study and was responsible for the cytokine protein analysis. • Dr. Kris Martens was responsible for the behavior analysis. • Sabrina Wei and Harleen Cheema were responsible for the protein extraction.  • Anna Wilkinson was responsible for fixed brain sectioning and most immunohistological staining.  • Dr. Wai Hang Cheng was responsible for the Iba1 immunohistochemical staining. • Dr. Cheryl Wellington contributed to the design of the study and had major contributions to data interpretation. • Dr. Yuri Bukhtiyarov from Vitae Pharmaceuticals Inc. (Fort Washington, PA) supplied the VTP compound. • Joan Guo and Linghang Zhuang from Vitae Pharmaceuticals Inc. were responsible for the mass spectrometry analysis.  A version of Chapter 4 has been published in Robert J*, Button EB*, Stukas S, Boyce GK, Gibbs E, Cowan CM, Gilmour M, Cheng WH, Soo SK, Yuen B, Bahrabadi A, Kang K, Kulic I, Francis G, Cashman N, Wellington CL (2017) High-density lipoproteins suppress Aβ-induced PBMC adhesion to human endothelial cells in bioengineered vessels and in monoculture. Molecular Neurodegeneration. Aug 22;12(1): 60. *co-first author.   I contributed to the design of the study with Dr. Jerome Robert, Dr. Sophie Stukas, and Dr. Cheryl Wellington. Dr. Jerome Robert and I had equal efforts in collecting the majority of the data, specifically, I isolated HDL for the majority of experiments, performed approximately 40% of the experiments performed in monocultured endothelial cells, prepared and measured human brain homogenates by ELISA, and performed the analysis and interpretation of these experiments. I also   x assisted in the writing of the manuscript and was responsible for the generation of approximately 50% of the figures and lead the overall design of figure layout.   Co-author contributions: • Dr. Jerome Robert was co-first author on this manuscript and was responsible for the experiments in three dimensional (3D) bioengineered arteries and immunofluorescent staining experiments as well as approximately 50% of the monoculture experiments. Dr. Robert also undertook the majority of the manuscript writing in particular in the introduction and discussion sections and  provided partial funding for the study.   • Dr. Sophie Stukas contributed to the concepts of the manuscript and data interpretation as well as preliminary experiments in monoculture and HDL isolation. • Guilaine Boyce was responsible for the collection of blood for peripheral blood mononuclear cells (PBMC) isolation and assisted in PBMC isolation. • Dr. Ebrima Gibbs produced oligomeric and fibrillar amyloid beta (Aβ) from monomeric Aβ.  • Dr. Catherine Cowan provided transmission electron microscopic images of monomeric, oligomeric, and fibrillar Aβ.  • Megan Gilmour cultured 3D bioengineered arteries. • Dr. Wai Hang Cheng assisted in immunofluorescent staining and statistical analysis. • Sonja Soo, Brian Yuen, Arvin Bahrabadi, and Kevin Kang provided assistance in protein extraction, western blotting, immunofluorescence, and cell culture.  • Dr. Iva Kulic provided insight on data analysis in particular regarding the combination of data across experiments. • Dr. Gordon Francis provided isolated HDL for initial experiments and data interpretations • Dr. Neil Cashman provided assistance in producing oligomeric and fibrillar Aβ as well as images of Aβ species. • Dr. Cheryl Wellington had major contributions to the design of the study, data interpretation, manuscript writing, and providing funding and equipment.     xi Small portions of Chapter 4 (Figure 4.12 and Figure 4.13) were published in Robert J, Button EB, Yuen B, Gilmour M, Kang K, Bahrabadi A, Stukas S, Zhao W, Kulic I, Wellington CL (2017) Clearance of beta-amyloid is facilitated by apolipoprotein E and circulating high-density lipoproteins in bioengineered human vessels. Elife. Oct 10;6. pii: e29595.  This study was designed and led by Dr. Jerome Robert who also performed the majority of the experiments. I assisted in data interpretation and was responsible for all of the ELISA measurements. All other co-authors contributed as described for Robert, Button et al. (2017) above.  Portions of Chapter 5 were previously published in Button EB, Gilmour M, Cheema HK, Martin EM, Agbay A, Robert J, Wellington CL (2019) Vasoprotective functions of high-density lipoproteins relevant to Alzheimer’s disease are partially conserved in apolipoprotein B-depleted plasma. International Journal of Molecular Sciences. Jan 22;20(3). pii:E462.  I led the study design, most experimental procedures, data analysis, interpretation, writing, and figure generation for this manuscript.   Co-author contributions: • Meghan Gilmour and Andrew Agbay generated 3D bioengineered arteries and assisted in cell culture. • Harleen Cheema assisted in HDL isolation and apolipoprotein B-depletion of plasma as well as immunoblotting. • Emma Martin assisted in the quantification of monocytes adhered to 3D bioengineered arteries, immunoblotting, and cell culture. • Dr. Jerome Robert had major contributions to study design, data interpretation, and manuscript writing. Dr. Robert was also responsible for cholesterol efflux experiments and monocyte adhesion experiment in 3D bioengineered arteries. • Dr. Cheryl Wellington had major contributions to study design, data interpretation, and manuscript writing.    xii Unpublished sections of Chapter 5 include the comparison of HDL and serum in 3D bioengineered artery assays, assessment of serum assay variation, and a proof-of-concept study with plasma obtained from a clinical laboratory to assess the feasibility of these assays in investigating differences in HDL function on the basis of disease. I designed these studies with Dr. Jerome Robert and Dr. Cheryl Wellington. I performed ELISA measurements for all experiments, isolated HDL from serum, performed the assays for all experiments except those comparing HDL and serum, performed data analysis and interpretation, generated figures, and wrote the Chapter text. Megan Gilmour and Andrew Agbay produced the 3D bioengineered arteries. Dr. Robert performed assays comparing serum with HDL and analyzed monocyte adhesion. Dr. Mari DeMarco and Amy Nguyen provided plasma for the clinical proof-of-concept study working at the clinical laboratories at St. Paul’s Hospital in Vancouver, BC. Drs. Robert and Wellington had major contributions to data interpretation.  Sections of Chapter 6 were previously published in Button EB, Robert J, Caffrey TM, Fan J, Zhao W, Wellington CL (2019) HDL from an Alzheimer’s disease perspective. Current Opinion in Lipidology. Jun;30(3): 224-234.   As discussed above, all co-authors contributed to the concepts, structures, and content of this review. Overall, approximately 10% of Chapter 6 appears in this published review, 30% is composed of expansions of the published sections, and 60% is unrelated to the published review. I wrote all sections, published and unpublished, in Chapter 6.     xiii Table of Contents  Abstract ......................................................................................................................................... iii Lay Summary .................................................................................................................................v Preface ........................................................................................................................................... vi Table of Contents ....................................................................................................................... xiii List of Tables ............................................................................................................................. xxii List of Figures ........................................................................................................................... xxiii List of Abbreviations ............................................................................................................... xxvi Acknowledgements ................................................................................................................ xxxiii Dedication .................................................................................................................................xxxv Chapter 1: Introduction ................................................................................................................1 1.1 Summary ......................................................................................................................... 1 1.2 Alzheimer’s disease overview ........................................................................................ 1 1.2.1 Epidemiology and economic burden ...................................................................... 2 1.2.2 Familial versus sporadic AD ................................................................................... 3 1.3 Amyloid pathology ......................................................................................................... 4 1.4 Tau pathology ................................................................................................................. 5 1.5 Neurodegeneration .......................................................................................................... 7 1.6 Neuroinflammation ......................................................................................................... 7 1.7 Clinical criteria for diagnosing AD ................................................................................. 8 1.8 Blood-based biomarkers for AD ..................................................................................... 9 1.9 Risk factors for AD and dementia ................................................................................ 10 1.9.1 Non-modifiable risk factors .................................................................................. 10 1.9.2 Apolipoprotein E ................................................................................................... 11 1.9.2.1 Function of peripheral apoE .............................................................................. 11 1.9.2.2 Function of brain apoE ...................................................................................... 12   xiv 1.9.2.3 APOE as a risk factor for AD ........................................................................... 13 1.9.2.4 ApoE isoform effects on amyloid pathology .................................................... 14 1.9.2.5 ApoE isoform effects on tau pathology ............................................................ 14 1.9.2.6 ApoE isoform effects on blood-brain barrier function ..................................... 15 1.9.2.7 ApoE isoform effects on astrocytes .................................................................. 15 1.9.2.8 ApoE isoform effects on microglia ................................................................... 15 1.9.3 Modifiable risk factors .......................................................................................... 17 1.9.3.1 Physical activity ................................................................................................ 18 1.9.3.2 Diet .................................................................................................................... 19 1.9.3.3 Type 2 diabetes mellitus ................................................................................... 19 1.9.3.4 Hypertension ..................................................................................................... 20 1.9.3.5 Hypercholesterolemia ....................................................................................... 20 1.10 Therapeutics for AD ..................................................................................................... 21 1.10.1.1 Existing AD therapeutics .................................................................................. 21 1.10.1.2 AD therapeutics in ongoing clinical trials ........................................................ 21 1.11 The cerebrovascular system .......................................................................................... 22 1.11.1 The relationship between the cerebrovasculature and AD ................................... 23 1.11.2 Cerebrovascular clearance of Aβ from the brain .................................................. 24 1.11.3 Cerebral amyloid angiopathy ................................................................................ 25 1.11.4 Tau and the cerebrovasculature ............................................................................ 26 1.11.5 Cerebrovascular inflammation .............................................................................. 27 1.11.6 Vascular co-morbidities in AD ............................................................................. 29 1.12 In vitro models of the BBB ........................................................................................... 30 1.13 High-density lipoproteins (HDL) .................................................................................. 33 1.13.1 HDL protective associations with cardiovascular diseases .................................. 34 1.13.2 HDL function in health and disease ...................................................................... 35 1.13.2.1 HDL and cholesterol efflux capacity ................................................................ 35 1.13.2.2 HDL and inflammation ..................................................................................... 38 1.13.3 HDL composition in health and disease ............................................................... 39 1.14 The relationship between HDL and AD ....................................................................... 42   xv 1.14.1 Mixed genetic evidence on HDL and cardiovascular disease and AD risk. ......... 42 1.14.2 Epidemiological evidence for a protective effect of HDL on AD ........................ 43 1.14.3 Potential protective mechanisms of action of HDL against AD ........................... 44 1.14.4 Vasoprotective functions of HDL in AD animal models ...................................... 45 1.14.4.1 Mouse models of AD ........................................................................................ 45 1.14.4.2 Genetic alteration of HDL in AD animal models ............................................. 46 1.14.4.3 HDL-based therapeutics on AD-relevant outcomes in animal models ............. 47 1.15 Liver X receptor and its therapeutic potential for AD .................................................. 47 1.16 Summary, research hypothesis, and specific objectives ............................................... 51 1.16.1 Summary ............................................................................................................... 51 1.16.2 Research hypothesis .............................................................................................. 51 1.16.3 Specific objectives ................................................................................................ 52 Chapter 2: Amyloid pathology, neuroinflammation, cerebrovascular pathology, and cognition in apoA-I deficient, APP/PS1 mice ............................................................................53 2.1 Summary ....................................................................................................................... 53 2.2 Introduction ................................................................................................................... 54 2.3 Methods......................................................................................................................... 55 2.3.1 Animals ................................................................................................................. 55 2.3.2 Tissue collection ................................................................................................... 56 2.3.3 Plasma lipid measurements ................................................................................... 57 2.3.4 Histology ............................................................................................................... 57 2.3.5 Image analysis ....................................................................................................... 58 2.3.6 Protein extraction .................................................................................................. 59 2.3.7 Enzyme linked immunosorbent assay (ELISA) .................................................... 60 2.3.8 Isolation of RNA and real-time quantitative polymerase chain reaction (RT-qPCR) 60 2.3.9 Contextual and cued fear conditioning ................................................................. 61 2.3.9.1 Training ............................................................................................................. 61 2.3.9.2 Context testing .................................................................................................. 61 2.3.9.3 Cued testing ...................................................................................................... 61   xvi 2.3.10 Statistical analysis ................................................................................................. 61 2.4 Results ........................................................................................................................... 62 2.4.1 Loss of apoA-I significantly reduces plasma total cholesterol and HDL cholesterol concentrations ....................................................................................................................... 62 2.4.2 ApoA-I deficiency increases cortical parenchymal and vascular amyloid burden of APP/PS1 mice .................................................................................................................. 63 2.4.3 ApoA-I deficiency increases pro-inflammatory protein and mRNA levels in the cortices of APP/PS1 mice ..................................................................................................... 66 2.4.4 ApoA-I deficiency increases total ICAM-1 protein concentration in the brain, total ICAM-1 positive area in the hippocampus, and vascular ICAM-1 positive area in the cortex and hippocampus of APP/PS1 mice .......................................................................... 68 2.4.5 ApoA-I deficiency exacerbates Aβ-mediated increases in total and cerebrovascular GFAP levels in both cortex and hippocampus ........................................... 70 2.4.6 ApoA-I deficiency worsens fear memory deficits in APP/PS1 mice ................... 73 2.5 Discussion ..................................................................................................................... 75 2.6 Supplemental figures .................................................................................................... 80 Chapter 3: Amyloid pathology, neuroinflammation, cerebrovascular pathology, and cognition in APP/PS1 mice treated with a non-brain penetrant LXR agonist .......................84 3.1 Summary ....................................................................................................................... 84 3.2 Introduction ................................................................................................................... 84 3.3 Methods......................................................................................................................... 86 3.3.1 Animals ................................................................................................................. 86 3.3.2 Tissue collection ................................................................................................... 86 3.3.3 Plasma lipid measurements ................................................................................... 86 3.3.4 Histology ............................................................................................................... 86 3.3.5 Image analysis ....................................................................................................... 87 3.3.6 Protein extraction .................................................................................................. 87 3.3.7 Enzyme-linked immunosorbent assay (ELISA) ................................................... 87 3.3.8 Ribonucleic acid (RNA) isolation and real-time quantitative polymerase chain reaction (RT-qPCR) .............................................................................................................. 87   xvii 3.3.9 Contextual and cued fear conditioning ................................................................. 87 3.3.10 Measurement of plasma and brain drug concentration ......................................... 87 3.3.11 Statistical analysis ................................................................................................. 88 3.4 Results ........................................................................................................................... 88 3.4.1 VTP does not enter the brain in aged WT or APP/PS1 mice ................................ 88 3.4.2 VTP induces Abca1 messenger RNA (mRNA) expression in the small intestine but not in the brain ................................................................................................................ 89 3.4.3 Soluble Aβ42 concentrations tend to be reduced in APP/PS1 mice treated with the non-brain penetrant LXR agonist VTP in pilot study ........................................................... 90 3.4.4 VTP induces Abca1 messenger RNA (mRNA) expression in the small intestine but not in the brain ................................................................................................................ 91 3.4.5 VTP does not significantly alter Aβ or tau pathologies ........................................ 94 3.4.6 VTP treatment reduces some markers of neuroinflammation .............................. 95 3.4.7 VTP does not alter the number of cortical or hippocampal Iba1+ microglia ....... 97 3.4.8 VTP reduces total and vascular ICAM-1 expression in the hippocampus ........... 97 3.4.9 VTP does not alter total levels of GFAP+ astrocytes in the brain but may subtly reduce cerebrovascular astrogliosis .................................................................................... 100 3.4.10 VTP rescues deficits in cued fear memory in APP/PS1 mice ............................ 103 3.5 Discussion ................................................................................................................... 105 3.6 Supplemental Figures .................................................................................................. 111 Chapter 4: Investigation of high-density lipoprotein functions relevant to Alzheimer’s disease using brain-derived endothelial cells and 3-dimensional bioengineered arteries. ..113 4.1 Summary ..................................................................................................................... 113 4.2 Introduction ................................................................................................................. 113 4.3 Methods....................................................................................................................... 115 4.3.1 Fabrication of tissue engineered vessels ............................................................. 115 4.3.2 Characterization of engineered vascular tissue ................................................... 116 4.3.3 Preparation of HDL and PBMCs ........................................................................ 116 4.3.4 Preparation of Aβ peptides ................................................................................. 117 4.3.5 Monocyte adhesion in engineered vessels .......................................................... 117   xviii 4.3.6 Static monotypic cell culture .............................................................................. 118 4.3.7 Monolayer PBMC adhesion assay ...................................................................... 118 4.3.8 Ab oligomerization/fibrilization and electron microscopy confirmation ........... 119 4.3.9 Measurement of intracellular NO ....................................................................... 119 4.3.10 Cell surface biotinylation .................................................................................... 120 4.3.11 Monolayer Ab association, binding and uptake ................................................. 120 4.3.12 Molecular Biology .............................................................................................. 121 4.3.13 Immunoblot ......................................................................................................... 121 4.3.14 Beta-sheet formation assay ................................................................................. 122 4.3.15 Human brain protein extraction and ELISA ....................................................... 122 4.3.16 Vascular Aβ accumulation and clearance in bioengineered vessels ................... 123 4.3.17 Statistical analysis ............................................................................................... 123 4.4 Results ......................................................................................................................... 124 4.4.1 Bioengineering dynamic, human 3D vessels ...................................................... 124 4.4.2 HDL suppresses Ab-mediated monocyte adhesion to ECs in 3D dynamic engineered human vessels ................................................................................................... 125 4.4.3 HDL attenuates Ab-induced PBMC adhesion in monotypic ECs ...................... 126 4.4.4 HDL reduces the fibrillization rate of Ab42 and Ab40 in cell-free conditions .. 128 4.4.5 Aβ structure does not affect HDL’s ability to reduce PBMC adhesion to hCMEC/D3 ......................................................................................................................... 130 4.4.6 The ability of HDL to suppresses Aβ-induced PBMC adhesion to hCMEC/D3 is independent of NO production and Anx1 ........................................................................... 131 4.4.7 HDL-mediated suppression of Ab-induced PBMC adhesion to hCMEC/D3 is independent of miR-233 ..................................................................................................... 133 4.4.8 HDL-mediated suppression of Ab-induced PBMC adhesion to hCMEC/D3 is independent of ICAM-1 and VCAM-1 ............................................................................... 134 4.4.9 Suppressing Ab uptake into hCMEC/D3 blocks PBMC adherence ................... 138 4.4.10 SR-BI is necessary for HDL to suppress Ab-induced PBMC adhesion ............. 140 4.4.11 Aβ accumulates in the walls of 3D bioengineered vessels ................................. 142   xix 4.4.12 HDL suppresses the accumulation of Aβ in the vessel wall of bioengineered arteries and tends to increase Aβ transport into the circulation .......................................... 144 4.5 Discussion ................................................................................................................... 144 4.6 Supplemental Figures .................................................................................................. 148 Chapter 5: Development of assays for the evaluation of brain-relevant high-density lipoprotein functions in human blood specimens. ...................................................................152 5.1 Summary ..................................................................................................................... 152 5.2 Introduction ................................................................................................................. 152 5.3 Methods....................................................................................................................... 154 5.3.1 Isolation and comparison of HDL isolated by ultracentrifugation and apoB-depleted plasma ................................................................................................................... 154 5.3.1.1 Blood collection .............................................................................................. 154 5.3.1.2 HDL and apoB-depleted plasma isolation ...................................................... 155 5.3.1.3 Fabrication of 3D bioengineered vessels ........................................................ 156 5.3.1.4 Amyloid beta clearance, accumulation, and monocyte adhesion assays using engineered vessels ........................................................................................................... 156 5.3.1.5 Enzyme linked immunosorbent assay (ELISA) .............................................. 156 5.3.1.6 Static monotypic cell culture .......................................................................... 156 5.3.1.7 Intracellular NO production ............................................................................ 156 5.3.1.8 Preparation of Aβ monomers .......................................................................... 157 5.3.1.9 PBMC adhesion assays ................................................................................... 157 5.3.1.10 Cell-free assay of Aβ fibrillization ................................................................. 157 5.3.1.11 Characterization of HDL preparations ............................................................ 157 5.3.1.12 Gel electrophoresis .......................................................................................... 157 5.3.2 Comparison of HDL isolated by ultracentrifugation and serum in 3D bioengineered artery functional assays ............................................................................... 158 5.3.3 Evaluation of variability in Aβ accumulation assay in 3D bioengineered arteries with serum from healthy volunteers ................................................................................... 158 5.3.4 Proof-of-concept assays of plasma from T2DM subjects ................................... 158 5.3.4.1 Plasma collection ............................................................................................ 158   xx 5.3.4.2 Plasma defibrination ....................................................................................... 159 5.3.4.3 Aβ accumulation in 3D bioengineered arteries ............................................... 159 5.3.5 Statistical analysis ............................................................................................... 159 5.4 Results ......................................................................................................................... 160 5.4.1 ApoB-depleted plasma recapitulated the ability of HDL isolated by ultracentrifugation to reduce Ab accumulation in the walls of bioengineered vessels. ...... 160 5.4.2 ApoB-depleted plasma does not retain the ability of HDL to reduce Aβ-induced monocyte binding to the endothelium ................................................................................ 161 5.4.3 ApoB-depleted plasma is functionally equivalent to HDL with respect to delaying Aβ42 fibrillization ............................................................................................................... 163 5.4.4 ApoB-depleted plasma does not induce NO in human brain-derived ECs ......... 164 5.4.5 Non-HDL plasma components in apoB-depleted plasma interfere with some anti-inflammatory activities of purified HDL ............................................................................ 165 5.4.6 Human serum recapitulates the ability of HDL to suppress Aβ accumulation in 3D bioengineered arteries but not Aβ-induced monocyte adhesion ......................................... 167 5.4.7 Aβ accumulation assays in 3D bioengineered arteries with human serum have sufficiently low variation to allow for the analysis of clinical biospecimens ..................... 170 5.4.8 Plasma from individuals with T2DM does not differ from those without T2DM in its ability to suppress Aβ accumulation in 3D bioengineered arteries ................................ 172 5.5 Discussion ................................................................................................................... 175 5.6 Supplementary Figures and Tables ............................................................................. 182 Chapter 6: Discussion and concluding remarks .....................................................................186 6.1 Summary and significance of findings ....................................................................... 186 6.2 Limitations of these studies ........................................................................................ 189 6.2.1 In vivo mouse model ........................................................................................... 189 6.2.2 In vitro model of the BBB .................................................................................. 190 6.2.3 Mechanistic data on AD-relevant HDL functions .............................................. 191 6.2.4 Simplified view of inflammation in AD ............................................................. 193 6.3 Future directions ......................................................................................................... 194   xxi 6.3.1 Evaluation of HDL structure and subclasses important for AD-relevant vasoprotective functions ..................................................................................................... 194 6.3.2 HDL-based therapeutics in development for AD ............................................... 196 6.3.2.1 HDL-based therapeutics in development of the treatment of acute coronary syndrome and atherosclerosis ......................................................................................... 198 6.3.2.2 Lipid modifying therapeutics for dementia prevention or treatment .............. 199 6.3.2.3 HDL as a potential treatment for the adverse effects of therapeutics on the cerebrovasculature .......................................................................................................... 201 6.3.2.4 Considerations for the repurposing of HDL-based therapeutics for AD ........ 201 6.3.3 Lifestyle modifications affecting HDL levels and function ............................... 203 6.3.4 HDL as a biomarker for cerebrovascular dysfunction ........................................ 205 6.4 Conclusions ................................................................................................................. 208 References ...................................................................................................................................210 Appendices ..................................................................................................................................289 Appendix A ............................................................................................................................. 289    xxii List of Tables  Table 1.1 HDL compositional and functional heterogeneity in disease. ...................................... 37 Table 1.2 Summary of studies treating AD model mice with GW3965 investigating amyloid, neuroinflammation, and memory deficits. .................................................................................... 49 Table 1.3 Summary of studies treating AD model mice with TO901317 investigating amyloid, neuroinflammation, and memory deficits. .................................................................................... 50 Table 3.1 VTP brain penetrance in 12-month old APP/PS1 mice and wildtype littermates. ....... 89 Table 3.2 VTP brain penetrance in 12-month old APP/PS1 mice and wildtype littermates. ....... 92 Table 3.3 VTP and GW brain penetrance in 9-month old APP/PS1 mice and wildtype littermates...................................................................................................................................................... 111 Table 4.1 Demographic data, Aβ40 and Aβ42, and adhesion molecule quantification in AD patient and NCI control brains. ................................................................................................... 135 Table 5.1 Donor demographics, HDL-C concentration, and apolipoprotein concentrations in serum and isolated HDL. ............................................................................................................ 168 Table 5.2 Donor demographics for the analysis of assay variation using 25% serum in 3D bioengineered vessels. ................................................................................................................. 170 Table 5.3 Coefficients of variation for Aβ accumulation assay in 3D bioengineered vessels with 25% human serum. ...................................................................................................................... 171 Table 5.4 Characteristics of patient cohort. ................................................................................ 173 Table 6.1 HDL-based therapeutics under investigation in clinical trials. ................................... 196 Table 0.1 Specific contributions of candidate and collaborators in Chapter 2. .......................... 289 Table 0.2 Specific contributions of candidate and collaborators in Chapter 3. .......................... 289 Table 0.3 Specific contributions of candidate and collaborators in Chapter 4. .......................... 290 Table 0.4 Specific contributions of candidate and collaborators in Chapter 5. .......................... 292    xxiii List of Figures  Figure 1.1 Aβ clearance via the cerebrovasculature. .................................................................... 25 Figure 1.2 HDL composition and function ................................................................................... 33 Figure 2.1 APP/PS1 survival rates. ............................................................................................... 56 Figure 2.2 Plasma lipid and brain apolipoprotein concentrations in apoA-I deficient APP/PS1 mice. .............................................................................................................................................. 63 Figure 2.3 Amyloid plaques and vascular Aβ deposition in apoA-I deficient APP/PS1 mice. .... 64 Figure 2.4 Pro-inflammatory protein and mRNA markers in apoA-I deficient APP/PS1 mice. .. 67 Figure 2.5 Total and endothelial associated ICAM-1 in apoA-I deficient APP/PS1 mice. .......... 69 Figure 2.6 Reactivity of astrocytes associated with the cerebrovasculature in apoA-I deficient APP/PS1 mice. .............................................................................................................................. 71 Figure 2.7 Astrocyte reactivity to Aβ plaques and vascular Aβ deposits in apoA-I deficient APP/PS1 mice. .............................................................................................................................. 73 Figure 2.8 Cued and contextual fear memory in apoA-I deficient APP/PS1 mice. ...................... 74 Figure 2.9 Analysis strategy for histology. ................................................................................... 80 Figure 2.10 Morphological discrimination of vascular and parenchymal amyloid. ..................... 81 Figure 2.11 Total vascular area in apoA-I deficient APP/PS1 mice. ............................................ 82 Figure 2.12 GFAP staining area in the hypothalamus in apoA-I deficient APP/PS1 mice. ......... 83 Figure 3.1 Intestinal and brain Abca1 mRNA expression in VTP and GW treated APP/PS1 mice in pilot study. ................................................................................................................................ 89 Figure 3.2 Brain Aβ peptide concentrations in VTP and GW treated APP/PS1 mice in pilot study. ............................................................................................................................................. 90 Figure 3.3 Lipoprotein metabolism in VTP and GW treated APP/PS1 mice. .............................. 93 Figure 3.4 Amyloid and tau pathology in VTP and GW treated APP/PS1 mice. ......................... 94 Figure 3.5 Biochemical markers of neuroinflammation in VTP and GW treated APP/PS1 mice........................................................................................................................................................ 96 Figure 3.6 Iba+ microglia numbers in VTP and GW treated APP/PS1 mice. .............................. 97 Figure 3.7 Endothelial and parenchymal ICAM-1-positive staining area in VTP and GW treated APP/PS1 mice. ............................................................................................................................ 100   xxiv Figure 3.8 Total and vessel associated GFAP levels in VTP and GW treated APP/PS1 mice. . 102 Figure 3.9 Cued and contextual fear conditioning in VTP and GW treated APP/PS1 mice. ..... 104 Figure 3.10 Body weight comparison by housing facility. ......................................................... 112 Figure 3.11 Cortical and hippocampal vascular area in VTP and GW treated APP/PS1 mice. . 112 Figure 4.1 Bioengineering dynamic human 3-dimensional vessels. ........................................... 125 Figure 4.2 HDL suppresses Aβ-induced monocyte adhesion to 3D bioengineered vessels. ...... 126 Figure 4.3 Aβ40 and Aβ42 induce PBMC adhesion to ECs, which is suppressed by HDL. ..... 127 Figure 4.4 HDL delays beta-sheet formation and attenuates Aβ-induced PBMC adherence independent of Aβ structure. ....................................................................................................... 129 Figure 4.5 HDL suppression of Aβ-induced inflammation is independent of eNOS and S1P. .. 132 Figure 4.6 HDL does not signal through miR-223 to reduce Aβ-induced inflammation in hCMEC/D3. ................................................................................................................................ 134 Figure 4.7 Adhesion molecules are enhanced by TNFα but not Aβ. .......................................... 136 Figure 4.8 Aβ does not activate phosphorylation of multifunctional serine/threonine protein kinases. ........................................................................................................................................ 137 Figure 4.9 HDL reduces Aβ association, binding and uptake to hCMEC/D3 whereas blocking Aβ binding or uptake reduces PBMC adhesion to hCMEC/D3. ...................................................... 139 Figure 4.10 HDL suppression of Aβ-induced inflammation requires SR-BI. ............................ 141 Figure 4.11 HDL suppresses Aβ-induced monocyte adhesion in engineered vessels via SR-BI...................................................................................................................................................... 142 Figure 4.12 Aβ40 and Aβ42 accumulation within and transport through bipartite vessels. ....... 143 Figure 4.13 HDL facilitates Aβ transport and reduces accumulation in bioengineered bipartite vessels. ........................................................................................................................................ 144 Figure 4.14 Aβ induces dose-dependent PBMC adhesion to hCMEC/D3 and HDL attenuates the PBMC adhesion in a dose-dependent manner. ........................................................................... 148 Figure 4.15 HDL suppression of Aβ-induced inflammation is independent of eNOS and S1P in HUVEC. ...................................................................................................................................... 149 Figure 4.16 Cortical ICAM-1 expression is increased in AD. .................................................... 150 Figure 4.17 HDL treatment does not alter LRP1 or RAGE protein levels in hCMEC/D3. ........ 151 Figure 5.1 HDL isolation and depletion of apoB-containing lipoproteins from human plasma. 155   xxv Figure 5.2 Aβ accumulation in bioengineered vessels treated with HDL isolated by ultracentrifugation or apoB-depleted plasma. ............................................................................. 161 Figure 5.3 Monocyte binding to EC treated with HDL isolated by ultracentrifugation or apoB-depleted plasma. .......................................................................................................................... 162 Figure 5.4 Anti-fibrillary effects of HDL isolated by ultracentrifugation and apoB-depleted plasma. ........................................................................................................................................ 164 Figure 5.5 NO production in human brain-derived ECs treated with HDL isolated by ultracentrifugation or apoB-depleted plasma. ............................................................................. 165 Figure 5.6 Effect of non-HDL plasma components and PEG on PBMC adhesion and NO production assays. ....................................................................................................................... 166 Figure 5.7 Clots on 3D bioengineered vessels treated with plasma and serum. ......................... 167 Figure 5.8 Aβ accumulation in 3D bioengineered vessels treated with HDL and serum from young healthy donors. ................................................................................................................. 169 Figure 5.9 Intra- and inter-donor and experimental variability of the Aβ accumulation assay in 3D bioengineered vessels with unfractionated serum. ................................................................ 171 Figure 5.10 Power calculation for comparison of subject groups for the ability of 25% serum to reduce Aβ accumulation in 3D bioengineered vessels. .............................................................. 172 Figure 5.11 Aβ accumulation in 3D bioengineered vessels treated with plasma from individuals with T2DM and age and sex-matched controls .......................................................................... 174 Figure 5.12 Composition and cholesterol efflux function of HDL and apoB-depleted plasma. 183 Figure 5.13 HDL-C and apolipoprotein concentrations in HDL isolated by ultracentrifugation and unfractionated serum. ........................................................................................................... 183 Figure 5.14 Raw data for comparison of HDL and serum in Aβ accumulation and monocyte adhesion assays 3D bioengineered vessels. ................................................................................ 184 Figure 5.15 Raw data for intra- and inter-donor and experimental variability of the Aβ accumulation assay in 3D bioengineered vessels with unfractionated serum. ........................... 184 Figure 5.16 HDL-C and apolipoprotein concentrations in treatments for 3D bioengineered vessels. ........................................................................................................................................ 185 Figure 6.1 Vasoprotective functions of HDL relevant for Alzheimer’s disease. ........................ 208    xxvi List of Abbreviations  2D    two dimensional 3D    three dimensional ATN    amyloid/tau/neurodegeneration framework α-SMA   alpha smooth muscle actin Aβ    amyloid beta AAV    adeno-associated virus ABC  Aβ plaque score (A), Braak NFT score (B), CERAD staging for neuritic plaques ABCA1   ATP binding cassette transporter A1 ABCA7   ATP binding cassette transporter A7 ABCG1   ATP binding cassette transporter G1 AD    Alzheimer’s disease ADAS-Cog13  AD Assessment Scale-Cognitive Subscale ADCS-ADL-MCI AD Cooperative Study-Activity of Daily Living Inventory (MCI version) ADNI    Alzheimer’s Disease Neuroimaging Initiative AIBL    Australian Imaging Biomarkers and Lifestyle Study of Ageing ALS    amyotrophic lateral sclerosis ANOVA   analysis of variation Anx1   annexin A1 Apo    apolipoprotein APP    amyloid precursor protein APP/PS1   transgenic mice with the Swedish APP mutation and exon 9 deletion in PS1 ARIA    amyloid-related imaging abnormalities ARIC    Atherosclerosis Risk in Communities ASL-MRI   arterial spin labelling MRI ATP    adenosine triphosphate  BACE-1  beta-secretase 1 BBB    blood-brain barrier   xxvii BCA    bicinchoninic acid BLT-1   block lipid transport-1 BSA    bovine serum albumin CAA    cerebral amyloid angiopathy CAARI   CAA-related inflammation CAD    coronary artery disease CAIDE   Cardiovascular Risk Factors Aging and Dementia CCCDTD4  Canadian Consensus Conference on the Diagnosis and Treatment of Dementia CD31    cluster of differentiation 31 cDNA    complementary DNA CDR-SB   Clinical Dementia Rating-Sum of Boxes CEC    cholesterol efflux capacity CERAD   Consortium to Establish a Registry for AD CETP    cholesterol ester transfer protein CHD    coronary heart disease Clu   clusterin/apoJ CNS    central nervous system CR1    complement receptor 1 CSF1F   colony stimulating factor 1 receptor CVD    cardiovascular disease CSF    cerebrospinal fluid CTFβ    carboxy-terminal fragment beta DAB    3,3’Diaminobenzidine DAF-2-DA   4,5-diaminofluorescein diacetate DAPI    4’,6-diamidino-2-phenylindole DMEM   Dulbecco’s Modified Eagle Medium DMSO   dimethyl sulfoxide DNA    deoxyribonucleic acid EC    endothelial cell   xxviii ECL    enhanced chemiluminescence EDTA    ethylenediaminetetraacetic acid ELISA   enzyme linked immunosorbent assay eNOS    endothelial nitric oxide synthase EOAD   early-onset Alzheimer’s disease ETV2    ETS variant 2 FBS    fetal bovine serum FDG-PET   fluorodeoxyglucose positron emission tomography FINGER  Finnish Geriatric Intervention Study to Prevent Cognitive Impairment and Disability FITC    fluorescein isothiocyanate FTLD    frontotemporal lobar degeneration GAPDH   glyceraldehyde 3-phosphate dehydrogenase GFAP    glial fibrillary acidic protein GW    GW3965, classical LXR agonist GWAS   genome-wide association studies HbA1c   glycated hemoglobin hCMEC/D3  human cerebral microvascular endothelial cell line HDL    high-density lipoprotein HDL-C   HDL cholesterol HEM    hemizygous HEPES   4-(2-hydroxyethyl)-piperazineethanesulfonic acid HIF-1α  hypoxia-inducible factor alpha HFIP    hexafluoroisopropanol HMG-CoA   3-hydroxy-3-methyl-glutaryl-CoA HR    hazard ratio HSPG    heparin sulphate proteoglycans HUVEC   human umbilical vein endothelial cell Iba1    ionized calcium binding adaptor molecule 1 ICAM-1   intercellular adhesion molecule 1   xxix IL-1β    interleukin 1 beta iPSC    induced pluripotent stem cells JNK    jun amino-terminal kinase KO    knockout LCAT    lecithin cholesterol acyltransferase LDL    low-density lipoprotein LDLR    LDL receptor L-NAME   L–NG-nitroarginine methyl ester LOAD   late-onset Alzheimer’s disease LRP-1   LDL receptor related protein LXR    liver X receptor MAPK   mitogen activated protein kinase MAPT   microtubule associated protein tau MAPT   French Multidomain Alzheimer Prevention Trial MCI    mild cognitive impairment miRNA   micro RNA MMSE   mini-mental state examination MRI    magnetic resonance imaging mRNA   messenger RNA MS    multiple sclerosis NCI    non-cognitively impaired  NIA-AA   National Institute on Aging and Alzheimer’s Association NFκB    nuclear factor kappa beta NfL    neurofilament light chain NGS    normal goat serum NFT    neurofibrillary tangle NMDA   N-methyl-D-aspartate NO    nitric oxide NOS    nitric oxide synthase PBMC   peripheral blood mononuclear cell   xxx PBS    phosphate buffered saline PCL    polycaprolactone PDGFRβ   platelet derived growth factor beta PEG    polyethylene glycol PEG-P   apoB-depleted plasma isolated by PEG precipitation PEG-UC   HDL isolated by a combination of PEG precipitation and ultracentrifugation PGA    polyglycolic-acid PI3K    phosphoinositide 3-kinase PIB-PET   Pittsburgh compound B PET PICALM  phosphatidylinositol binding clathrin assembly protein PET    positron emission tomography PFA    paraformaldehyde PLA    polylactate PON-1   paraoxonase 1 PPARγ   peroxisome proliferator activating receptor gamma PREDIMED  Primary Prevention of Cardiovascular Disease with a Mediterranean Diet PreDIVA   Dutch Prevention of Dementia by Intensive Vascular Care PrP    prion protein PSEN1   presenilin 1 PSEN2  presenilin 2 p-tau    phosphorylated tau PVDF    polyvinylidene difluoride RAGE   receptor for advanced glycated endproducts RAP    receptor associated protein RCT    randomized controlled trial rHDL    reconstituted HDL RIPA    radioimmunoprecipitation assay RNA    ribonucleic acids ROI    region of interest RPMI    Roswell Park Memorial Institute (medium)   xxxi RR    relative risk RT-qPCR   real-time quantitative polymerase chain reaction RXR    retinoid X receptor S1P    sphingosine-1-phosphate SAA    serum amyloid A SAPK    stress activated protein kinase SD    standard deviation SDS    sodium dodecyl sulphate SDS-PAGE   SDS-polyacrylamide gel electrophoresis  Ser    serine SMC    smooth muscle cell sMRI    structural magnetic resonance imaging SORL1   sortilin related receptor 1 SREBP-1c  sterol regulatory element binding protein-1c SR-B1   scavenger receptor class B type I T2DM   Type 2 diabetes mellitus TBS    tris buffered saline TBS-X   TBS with triton X-100 TEM    transmission electron microscopy THP-1   human monocytic cell line Thr    threonine tRNA    transfer RNA TREM2   triggering receptor expressed on myeloid cells 2 Tyr    tyrosine UC-HDL   HDL isolated by ultracentrifugation UC-P    apoB-depleted plasma produced by ultracentrifugation VCAM-1   vascular cell adhesion molecule VEGF   vascular endothelial growth factor VLDL    very low-density lipoprotein VTP    VTP-38443, non-brain penetrant LXR agonist   xxxii WHO    World Health Organization WT    wildtype X-34    1,4-Bis(3-carboxy-4-hydroxyphenylethenyl)benzene ZO-1    zona occludens 1     xxxiii Acknowledgements  First and foremost, thank you to my supervisor Dr. Cheryl Wellington. I am very grateful that five years ago she saw a scientist within me and gave me the best support and mentorship to become that scientist. Her guidance in building skills of scientific thinking, her ability to challenge me to grow and to support me through the growing pains, her help in identifying my strengths, and her encouragement for me to own my work have made a world of difference throughout my studies.   The Wellington lab has been a special place to conduct by PhD studies largely because of all of its fantastic people with their scientific input, technical support, and friendship. I would like to thank the current and existing members for all their support, Dr. Jerome Robert, Dr. Sophie Stukas, Dr. Jianjia Fan, Dr. Tara Caffrey, Dr. Asma Bashir, Dr. Wai Hang Cheng, Guilaine Boyce, Andrew Agbay, Anna Wilkinson, Carlos Barron, Jasmine Gill, Jennifer Cooper, Wenchen Zhao, Dr. Iva Kulic, Megan Gilmour, Sonja Soo, Dr. Kris Martens, Dr. Dhananjay Namjoshi, and Jeniffer Chan and the many co-op students over the years. Thank you also to the volunteers, co-op students, and junior graduate students that I had the opportunity to mentor, Sabrina Wei, Harleen Cheema, Emma Martin, and Elyn Rowe. Their technical support was invaluable and each of them helped me to grow as a mentor. Another special thank you to my co-supervisor in everything but name, Dr. Jerome Robert. He provided critical technical and scientific mentorship to get me started as a doctoral student and has continued to enhance my work through his support and feedback.   Thank you also to both Pathology graduate student advisors that have existed during my studies, Drs. Haydn Pritchard and Dana Devine. In particular thank you to Dr. Haydn Pritchard who was instrumental in my decision to study at UBC.   My PhD project would not have been possible without the support and guidance of my supervisory committee who always made time for our meetings despite their busy schedules. Thank you to Drs. Shernaz Bamji, Mari DeMarco, Angela Devlin, and Haakon Nygaard for your time, your advice, and your feedback.    xxxiv I was very fortunate to receive funding for the first four years of my PhD, in particular I would like to acknowledge the Canadian Institutes of Health Research for the Doctoral Award I received that not only provided me with three years of salary funding but also with travel funding for conferences. The opportunity to present my work, build an international scientific network, and travel the world greatly enriched my PhD studies and was incredibly valuable to me personally.  I would also thank and acknowledge critical people in my academic journey that lead me to begin a career in research in the first place. First, Dr. Robin Duncan at the University of Waterloo who provided me with my first taste of research over two years of volunteering, working, and learning in her lab. Even earlier, I would like to thank my high school Biology and Chemistry teacher Mr. Campbell who was the first person who taught me how to think like a scientist.  Critical to my success has been my family and friend network. My parents, who have always been my most avid supporters, my friends both in Vancouver and elsewhere, who have provided well-needed balance to my life, and my husband, who has been the most supportive partner and greatest friend through this journey.                      xxxv Dedication    To my parents, whose hard work opened the doors for me to pursue this PhD and whose unending encouragement helped to push me through and complete it.  1 Chapter 1: Introduction  1.1 Summary Alzheimer’s Disease International estimates there to be over 50 million people around the world living with dementia in 2019 [1,2]. The most common form of dementia in people over the age of 65 is Alzheimer’s disease (AD), a disease characterized by amyloid-beta (Aβ) plaques, neurofibrillary tangles (NFT), and neurodegeneration [3]. In addition to these pathological hallmarks, cerebrovascular dysfunction has emerged as another leading contributor to AD [4]. Cerebral amyloid angiopathy (CAA), infarcts, vascular degeneration, or other forms of cerebrovascular dysfunction exist in 60-90% of AD autopsy cases [5] and changes to blood-brain barrier permeability have been observed in living people even at early stages of cognitive impairment [6]. The cerebrovasculature also provides two major routes of Aβ clearance from the brain [7]. Cerebral vessels are therefore an attractive target for future AD therapeutics or preventative strategies.   High-density lipoproteins (HDL) have well-established vasoprotective functions on peripheral vessels [8] and therefore may also protect against cerebrovascular dysfunction in AD. Human epidemiological studies and studies in mouse models also support a protective role for HDL against AD however the mechanisms behind this protective effect are poorly understood. In this thesis, mouse models and human in vitro models are used to define AD-relevant, cerebrovascular protective functions of HDL and efforts are made to develop in vitro assays of these functions capable of evaluating HDL from clinical populations. Together, work in this thesis provides a rationale for the exploration of HDL-based therapies for AD and HDL-based functional assays as biomarkers for cerebrovascular health.  1.2 Alzheimer’s disease overview Alzheimer’s disease is named for Alois Alzheimer, a German physician and researcher who documented one of the first cases. Dr. Alzheimer first examined the 51-year old August Deter in 1901 and noted reduced comprehension and memory, aphasia, disorientation, unpredictable behavior, paranoia, auditory hallucinations, and psychosocial impairments [9]. Upon autopsy,   2 Alzheimer found plaques, neurofibrillary tangles, and arteriosclerotic changes in this patient, which he reported in a lecture in 1906 and published in 1907 [10]. This case was followed by several other similar published reports and in 1910 dementia with plaques and tangles was first referred to as “Alzheimer’s disease” in a psychiatric textbook [9]. Arguably equally important yet less known in the characterization of AD is Dr. Oskar Fischer [11]. Dr. Fischer was the first to describe neuritic plaques and published his findings of this feature in 12 cases of senile dementia in 1907 [12], the same year as Dr. Alzheimer’s initial publication. Dr. Fischer continued his neuropathological study of dementia and published his findings of neuritic plaques and neurofibrillary tangles in 191 [13]0 and 1912 [14]. However, his contributions were forgotten for many decades and remain undervalued compared to that of Dr. Alzheimer due to political and anti-Semitic tensions in the Austro-Hungarian Empire where he performed his research [11]. Tragically, Dr. Fischer was forced out of his research institution during this tumultuous time and later imprisoned at a concentration camp until his death in 1942 [11].   1.2.1 Epidemiology and economic burden The most recent statistics on the global prevalence of AD and dementia from 2016 found that 43.8 million people suffer from dementia worldwide [15]. A systematic review and meta-analysis of 119 studies on dementia prevalence by Fiest et al. found an overall annual prevalence of dementia in people over 60 years of age of 30.4/1,000 people ranging from 11.7/1000 people in Asia to 103.6/1,000 in North America. Similarly, global annual incidence in people over 60 years of age overall was reported as 34.1/1,000 people but ranged from 11.5 in Nigeria to 97.8 in the USA. These regional differences, although not significant, are interesting and may be explained by differences in screening, reporting, life expectancy, competing risks, or dementia risk factor exposure  [16]. In Canada, over 500,000 people are reported to have dementia with an increasing prevalence with age and a higher prevalence in females at each age. Specifically, 1.9% of males and 2.8% of females between the ages of 65 and 74 are affected by dementia while 28.7% of males and 37.1% of females over the age of 85 are affected. The direct and indirect costs of dementia in Canada is estimated at up to $33 billion annually [17].     3 Changes in prevalence and incidence over time are of great interest to allow for prediction of disease burden in the future. The Global Disease Burden study on dementia evaluated the change in global prevalence from 1990 to 2016 and reported an increase from 20.2 million people worldwide to 43.8 million people. While this represents a 177% increase in total prevalence, the age adjusted increase was only 1.7%, indicating that most of the increase in dementia prevalence is due to a growing and aging population. They also found region specific differences including a reduced prevalence in high income countries [15]. Other have similarly seen decreases in dementia prevalence or incidence. The Framingham Study reported a decline in dementia incidence over their three decade study and suggested that better control of vascular risk factors may explain some of this shift [18]. The Cognitive Function and Ageing Study in England and Wales reported a 20% decrease in dementia incidence over two decades that was driven by reduced incidence in males [19].  1.2.2 Familial versus sporadic AD People with AD are subcategorized into those diagnosed at ages below 65, called early-onset AD (EOAD), and over 65, called late-onset AD (LOAD). It is estimated that approximately 3-5% of AD cases are EOAD [20,21]. Although it has historically been assumed that EOAD cases are primarily due to autosomal dominant inheritance of mutated genes in amyloid-related proteins, more recent research suggests that only around 10% of EOAD cases are inherited this way [22,23]. Of the small proportion of EOAD patients with an autosomal dominant inheritance pattern, only 12% had known mutations in amyloid related proteins [22]. This suggests that 90% of EOAD cases are likely to have some form of genetic causality due to autosomal recessive mutations, genes with variable heterozygous penetrance, a mixture of genetic risk factors, or not yet discovered mutations [23]. In addition to the well-known, autosomal dominant mutations in APP, PSEN1, and PSEN2, other mutations in MAPT, SORL1, TREM2, and several genes in the endolysosomal pathway have also been found to be associated with EOAD [24]. Several gene associations, such as TREM2, APOE, and SORL1, are common to EOAD and LOAD [24,25]. Although heritability of EOAD is estimated at over 90%, heritability of LOAD is estimated to be between 45% and 81% [23].    4 EOAD and LOAD are not only different based on their onset and heritability but also based on pathophysiology and clinical presentation. Non-memory impairment, specifically apraxia and visuospatial dysfunction may be higher in EOAD patients [26]. Although total neuropsychiatric index scores have been reported to be similar between EOAD and LOAD cases [27], others have reported more psychosocial problems in EOAD patients [25]. In terms of pathophysiology, EOAD patients show greater tau, NFT, and neuritic plaque burden, greater posterior hypometabolism, and reduced hippocampal volume loss [25]. Due to the reduced age of onset of EOAD, subjects exhibit less co-morbidities [25] including fewer vascular risk factors [28].  1.3 Amyloid pathology Aβ plaques are one of the neuropathological characteristics of AD first observed by Dr. Alzheimer [10] and the “amyloid hypothesis”, which  posits that Aβ aggregates cause neurodegeneration and memory loss in AD, is a leading theory for the pathobiology of AD [29]. Aβ peptides are formed when the membrane-bound amyloid precursor protein (APP) is cleaved first by β-secretase (BACE-1) to release a soluble APP fragment and a β-carboxyl terminal fragment (CTFβ). The CTFβ is then cleaved by γ-secretase to produce an APP intracellular domain and Aβ monomers. Cleavage of APP by BACE-1 and γ-secretases occurs predominantly in endosomes where the pH environment is optimal for BACE-1 function and γ-secretases are abundant relative to their expression at the plasma membrane [30–32]. The major components of γ-secretase are presenilin 1 and presenilin 2, whose encoding genes PSEN1 and PSEN2 are mutated in some EOAD cases [30].  Aβ is released as peptides ranging from 37 to 49 residues, depending on the final cleavage site by γ-secretase, although Aβ40 and Aβ42 are the most commonly found species in amyloid plaques [33]. Monomeric Aβ, in particular Aβ42, is prone to aggregation and forms oligomers then protofibrils then fibrils, which aggregate into plaques [33]. However, it is the oligomeric species of Aβ that is thought to be the most toxic to neurons [34]. The physiological functions of APP and Aβ are not fully understood, however, recent research has suggested that Aβ may function as an anti-microbial agent that traps infectious agents in the brain to be neutralized [35] and APP may function in regulating synaptic transmission by binding a γ-aminobutyric acid neurotransmitter receptor [36].     5 Importantly, Aβ plaques and vascular amyloid deposition are also commonly observed in the brains of people without cognitive impairment. For example, 23% of healthy 70-year olds in a Swedish cohort had elevated cerebrospinal fluid (CSF) Aβ42 levels [37] and 33% of healthy controls in the Australian Imaging Biomarkers and Lifestyle Study of Ageing (AIBL) had amyloid pathology observed with Pittsburgh compound B (PIB) positron emission tomography (PET) [38]. The presence of these pathologies in some cognitively normal people may occur because there is a delay between the onset of amyloid pathology and cognitive impairment. Indeed, the proportions of cognitively normal people with positive amyloid PET scans in the AIBL study reflected the proportions of people with cognitive impairment 15 to 20 years later [38]. Similarly, a recent study found the duration of amyloid positivity correlates better with cognitive decline than amyloid positivity on its own, suggesting a cumulative effect of amyloid on cognition [39]. Alternatively, some individuals may have more resilience to the presence of amyloid in their brains. Factors that contribute to this resilience are of great interest to the field and may include genetics, cognitive reserve, or sociodemographic characteristics [40].   1.4 Tau pathology The other major pathological hallmark of AD first observed by Dr. Alzheimer was NFTs [10], later found to be composed of hyperphosphorylated tau protein. Tau is most studied for its function in microtubule assembly and stability where it maintains neuronal projections to allow for transmission of signals along the axon to the synapse and reception of synaptic signals in post-synaptic dendrites [41]. Phosphorylation of tau reduces its binding to microtubules and results in self-assembly of tau into filaments that then form tangles [41]. It is not completely clear which structural forms of tau are the most toxic [41]. Pathological tau has also been shown to bind nucleoporins in neurons resulting in impaired nuclear import and export [42].  Tau is known to spread between neurons in a pattern throughout the brain that has been classically used for Braak staging of AD [43] and has been compared to prion propagation. Braak staging refers to the histological classification of abnormal tau neuropathology into stages that correspond to severity of cognitive deficits. Briefly, the distribution of tau tangles progresses from isolation in the transentorhinal layer to additional involvement of entorhinal cortex then finally to   6 involvement of the entire isocortext [44]. In human brains, tau seeds capable of misfolding monomeric tau are found not only in regions containing NFT but also in subsequent Braak staging regions without NFT. That these seeds are found mainly in the synaptic fraction of regions without NFT suggests they are transmitted synaptically [45]. Interestingly, when tau-containing brain homogenates are injected into the hippocampi of mice, tau first propagates to the hippocampus contralateral to the injection then spreads to synaptically connected regions and fails to spread to regions without synaptic connections, even when those regions are in close proximity [46]. How tau is spread between functionally connected neurons is less clear, although it has been demonstrated that neural activity is required [47] and some evidence suggests that exosomes may be responsible. Tau has been found in exosomes in the CSF [48] and blood [49] of AD patients in higher amounts than in controls even up to 10 years before diagnosis [49]. The feasibility of exosomal transmission of tau has been demonstrated in hippocampal slices and microfluidic systems [50] however this mechanism remains controversial.  PET imaging of tau pathology has allowed for great advances in the understanding of tau pathology progression and its relationship with cognitive function. From these studies it has been observed that tau pathology may in fact be a better predictor of cognitive decline than amyloid pathology or structural changes by magnetic resonance imaging (MRI). For example, in a retrospective longitudinal analysis of a cohort of 152 subjects aged 55 to 91 years old, only tau predicted decline in episodic memory and executive function [51]. Furthermore, the temporality of AD pathologies have been evaluated using PET imaging in cognitively normal people where it has been shown that first Aβ levels rise and then elevated tau levels follow, and that people with the greatest changes in tau were most likely to progress to mild cognitive impairment (MCI) [52].  Pathological tau is not unique to AD. Mutations in microtubule associated protein tau (MAPT), the gene encoding tau, are associated with frontotemporal lobar degeneration (FTLD), a group of disorders often affecting personality, behaviour, executive function and language, and with Parkinson’s disease [41]. Chronic traumatic encephalopathy is a neurodegenerative tauopathy pathologically characterized by aggregates of phosphorylated tau in neurons and astrocytes surrounding small blood vessels in the depths of cortical sulci [53].    7  1.5 Neurodegeneration Neurodegeneration is another pathological hallmark of AD that is obvious at autopsy based on gross atrophy of the brain. Ante-mortem, neurodegeneration can now be observed using MRI and suspected based on the result of fluid biomarkers. Early and progressive hippocampal atrophy is a key marker of AD on MRI, although it is not specific to AD [54]. Other areas of atrophy in AD include the neocortex, which reflects NFT pathology development, and subcortical structures [55]. CSF total-tau is used as a fluid biomarker for neurodegeneration in AD [37] and ongoing research is investigating additional biomarkers including  CSF and plasma neurofilament light chain (NfL). NfL is an axon cytoskeletal protein released from damaged neurons [56]. Longitudinal measurements show that high baseline or change in plasma or serum NfL levels associate with reductions in cortical thickness, hippocampal volume, brain glucose metabolism measured with fluorodeoxyglucose (FDG)-PET [57,58], and cognitive function [57–59] as well as increases to amyloid PET burden [57]. In people with familial AD, plasma NfL levels begin to increase 16.2 years before diagnosis and the rate of change can predict conversion to dementia [59]. Elevated plasma NfL levels are not specific to AD and are also observed in other neurodegenerative conditions including frontotemporal dementia, amyotrophic lateral sclerosis (ALS), Huntington’s disease, and multiple sclerosis (MS) [60], as well as after neurotrauma [61].  1.6 Neuroinflammation In recent years, neuroinflammation has emerged as another key component of AD pathophysiology. About 50% of genetic AD risk loci are related to microglia and innate immunity including genes for adenosine triphosphate (ATP)-binding cassette transporter A7 (ABCA7) and TREM2, which are expressed on microglia and myeloid cells, and complement receptor 1 (CR1) [62–64]. In vivo imaging studies show increased microglia in AD subjects in the temporal lobe and these increases correlate with amyloid load and inversely correlate with mini-mental state examination (MMSE) scores [65]. The role of microglia in AD is complex. In the early stages of AD, microglia may promote Aβ seeding in an attempt to trap infectious agents [35,62]. Mid-stage microglia may be more protective as they promote compaction of amyloid plaques to limit their toxicity to surrounding neurons. Late-stage microglia help to remove dead neurons but also   8 contribute to further neuronal damage as they activate tau kinases, produce reactive oxygen species and nitric oxide (NO), and promote astrocyte activation [62]. The switch from a homeostatic to a “neurodegenerative” or “disease associated” state is coupled with several transcriptional changes, including upregulation of APOE and TREM2 and downregulation of TGFβ, and is commonly observed in mouse models of AD, ALS, and MS and in human AD brains [66,67]. Indeed, several recent studies have found that depleting microglia with an inhibitor for colony stimulating factor 1 receptor (CSF1R) in AD mouse models can improve pathology. Administration of CSF1R inhibitor in 2-month old 5xFAD mice reduced neuritic plaque load and cognitive impairments [68] while treating 10-month old 5xFAD mice, 15-month old 3xTg-AD, or 6-month old APP/PS1 mice also improved memory but had no effect on amyloid load [69–71]. Similarly, CSF1R blockade in 8-week old P301S mice, expressing a mutant form of human MAPT, reduced tau pathology, neurodegeneration, and motor deficits [72] while microglia ablation in 6-month old P301S mice expressing human APOE-ε4, the greatest genetic risk factor for LOAD, suppressed neurodegeneration and had subtle effects on tau pathology [73]. A key role for microglia in AD pathogenesis is further supported by a recent analysis of cell-specific regulatory regions. As most AD GWAS hits correspond to noncoding regions, researchers defined enhancer-promoter regions specific to neurons, astrocytes, oligodendrocytes, and microglia in healthy brains then searched for AD risk genes in those regions. Variants associated with AD were mostly found in microglial enhancers suggesting the detrimental effects of these variants occurs through the actions of microglia [74].  1.7 Clinical criteria for diagnosing AD In 2011, a working group from the National Institute on Aging and Alzheimer’s Association (NIA-AA) published updated criteria for diagnosing AD and MCI as well as a research framework for classifying preclinical AD and, in 2012, updated neuropathological criteria for AD [54,75–78]. These guidelines have been widely adopted, including by Canada at the 4th Canadian Consensus Conference on the Diagnosis and Treatment of Dementia (CCCDTD4) [79].  According to the 2011 NIA-AA guidelines, the criteria for AD dementia is defined by cognitive or behavioural symptoms interfering with functional abilities, a decline in these functions from   9 previous observations, no alternative disorder existing to explain the symptoms, and diagnosis of cognitive impairment both from a patient history and an objective cognitive assessment. The cognitive assessment must include at least two of the following impairments: cannot acquire and remember new information, impaired reasoning and handling of complex tasks, poor judgement, impaired visuospatial abilities, impaired language abilities, or changes in personality or behaviour [54]. The core clinical criteria for MCI are much the same as those for AD, however, the impairment is in only one or more cognitive domains and does not interfere with the individual’s functional abilities. Tests for episodic memory are used to identify MCI patients likely to progress to dementia [80]. The presence of autosomal dominant AD mutations as well as imaging and fluid biomarkers increases the certainty of the AD or MCI diagnosis. Low CSF Aβ42, elevated CSF total tau or phosphorylated-tau (p-tau), elevated amyloid PET, decreased FDG-PET, or atrophy in particular regions on structural MRI (sMRI) were recommended as biomarkers in the 2011 guidelines [54,76].  The NIA-AA guidelines in 2011 recognized the importance of the preclinical stage of AD and the utility of biomarkers in identifying subjects in this stage, however, the working group determined that the existing evidence was only sufficient to develop a research framework, and not clinical criteria, for preclinical AD [77]. The most recent NIA-AA guidelines for preclinical AD, the ATN Framework, were proposed in 2018 [3]. In this framework, people are grouped according to their status in amyloid (A), tau (T), and neurodegeneration or neuronal injury (N) categories. Amyloid biomarkers are reduced CSF Aβ42, reduced CSF Aβ42/40 ratio, or increased amyloid PET. Positivity for tau is indicated by elevated CSF p-tau or tau PET. Reduction in FDG-PET, atrophy on sMRI, and increased CSF total tau constitute positivity for neuronal injury. Subjects are then classified as A-T-N- if they have no positive biomarkers, A+T+N+ if they have positive biomarkers in each category, or any other combination in between depending on their biomarker status.   1.8 Blood-based biomarkers for AD Although the importance of imaging and CSF biomarkers were recognized in the 2011 and 2018 NIA-AA clinical criteria and research frameworks for AD [3,75] and CSF biomarkers are entering   10 clinical practice for use in supporting diagnoses [81], these biomarkers remain expensive and invasive. Fortunately, less-invasive blood-based biomarkers have also seen substantial developments in recent years. Examples of the more promising candidates include Aβ42/Aβ40 ratio [82], total tau [83], p181-tau [84], NfL [59],  BACE-1 [85], and multi-protein panels [56,86].  Measurement of plasma Aβ42/Aβ40  in particular has made great strides towards clinical use with the mass spectrometry-based device from C2N Diagnostics receiving FDA breakthrough designation in 2019 [87,88] and several other platforms following close behind. Other plasma  Aβ42/Aβ40 measurement methods include the electrochemiluminescence-based Elecsys system by Roche [89] already in use for CSF, sandwich enzyme linked immunosorbent assays (ELISA) such as the ABtest system by Araclon Biotech [90], ultrasensitive paramagnetic ELISA platforms such as the Simoa (Single molecule array) system by Quanterix [82,91], immunomagnetic ELISAs [92], and assays that combine immunoprecipitation with mass spectrometry assays such as the Shimadzu assay [93,94]. These assays have had considerable success in predicting amyloid positivity by PIB PET imaging in patients, even in people without cognitive impairment [95], and especially when combined with APOE genotype data [89]. The use of plasma Aβ assays rather than amyloid PET in screening for clinical trials would also come with significant cost and time saving benefits [89].   1.9 Risk factors for AD and dementia 1.9.1 Non-modifiable risk factors Age is the single greatest risk factor for AD. Prevalence of AD increases dramatically with increasing age, although, importantly, AD is not a normal part of aging [15,21]. Sex is another non-modifiable risk factor for AD with an enormous burden. Women have twice the lifetime risk of AD and make up 2/3 of AD cases [15,21]. Although it was initially assumed that women’s longer life expectancy explained their greater AD risk, there is actually little difference in AD incidence at any given age [96], suggesting that other factors are at play.  Other potential explanations for the increased risk for women are hormonal effects, lower education levels, more sleep disorders, and survival bias due to the increased risk among men for cardiovascular disease (CVD) [97]. Many studies have focused on the potentially protective effect of estrogen including a recent report that demonstrated an inverse association between duration of reproductive status   11 (from menarche to menopause) and dementia risk as well as an increased risk for women with a late menarche, early menopause, or a hysterectomy [98].   Genetics makes up another large share of the non-modifiable risk for AD, the heritability of LOAD is estimated to be 45-81%  and EOAD up to 90% [23]. A small portion of the heritability of AD is attributed to rare autosomal dominant mutations in amyloid processing genes [99]. Mutations in APP, encoding the precursor to Aβ, as well as in PSEN1 and PSEN2, encoding components of the γ-secretase complex that cleaves APP into Aβ [30,34], have long been known to be associated with EOAD [100–102]. Genome-wide association studies (GWAS) have provided the most recent wave of discovery for the genetic contribution to AD. Most genes identified in GWAS have a relatively small effect on AD risk [99], with the exception of the recently discovered TREM2 risk alleles [103,104].  Importantly, a majority of these risk loci are in genes that regulate cholesterol and lipid metabolism, immune, endocytosis, APP metabolism, and tau binding protein pathways [63,64]. APOE is the greatest genetic risk factor for AD as it is relatively common and can exert a 2- to 33-fold increase in AD risk for people homozygous for the APOE-ε4 allele with greater AD risk for women compared to men [105,106].  1.9.2 Apolipoprotein E ApoE is the major apolipoprotein forming HDL-like particles in the brain [107,108] and is found on 5-10% of HDL in the plasma [109–111] and the majority of particles in the CSF [112]. The source of peripheral apoE is primarily the liver while CNS apoE is produce mainly by astrocytes and microglia  [107,108]. Observations in subjects after a liver transplant have suggested that these pools of apoE are separate from each other, as plasma apoE was found to phenoconvert to that of the liver transplant donor but CSF apoE remained the same [113].  1.9.2.1 Function of peripheral apoE ApoE is well understood to be a key regulator of peripheral lipid metabolism as it facilitates the uptake of lipoproteins by the low-density lipoprotein receptor (LDLR) and low-density lipoprotein receptor-related protein 1 (LRP1) [114]. Knockout of apoE in mice results in hypercholesterolemia   12 due to the inability to clear triglyceride-rich lipoproteins and these mice develop atherosclerosis, particularly when on a high-fat diet [115,116].   ApoE-containing HDL make up a small but important subclass of HDL formed through secretion of apoE from the liver after uptake, recycling of triglyceride-rich lipoproteins, and from apoE secreted from macrophages [114]. It is estimated that apoE-containing HDL accounts for approximately 10% of total HDL based on percent measurements by cholesterol mass [111] and by the percent of apoA-I mass [109,110] found on HDL particles containing apoE. ApoE-HDL also makes up close to 10% of HDL synthesis [109]. Recent focus on apoE-containing HDL has shown that after secretion into circulation, it quickly expands then is rapidly cleared [109] in a manner that can be affected by dietary fat consumption [117]. This rapid expansion and clearance suggests that apoE-HDL may contribute to reverse cholesterol transport, a role that may in part explain epidemiological reports of protective associations between apoE-HDL levels and coronary heart disease (CHD) risk [109,118,119]. ApoE-containing HDL has been shown to be cardioprotective through its actions reducing arterial stiffness, modifying extracellular matrix gene expression [120], and displacing hepatic lipase from heparin sulfate proteoglycans to allow for lipoprotein triglyceride hydrolysis [121]. Levels of plasma apoE-HDL are lower in adolescent males with T2DM compared to lean controls [122], in normolipidemic females compared to males [121], and during the postprandial state compared to during fasting in normolipidemic individuals [121].  Interestingly, evidence for a protective effect of plasma apoE-HDL on brain health is also emerging. In the Gingko Evaluation of Memory study, higher levels of apoE on plasma HDL was associated with reduced amyloid burden on PET scans [123] and higher levels of apoA-I on HDL particles containing apoE was associated with higher scores on a modified MMSE [124].  1.9.2.2 Function of brain apoE In contrast to the effects of apoE deficiency on peripheral lipids and vascular health, the effects within the mouse brain are subtle and not consistent across studies. Although some have observed no effect of apoE deficiency on spatial learning with Morris Water Maze or on synaptic density or plasticity [125,126], others have observed specific alterations to hippocampal synaptic plasticity [127] and deficits in spatial learning by Morris Water Maze [128–130]. Importantly, the studies   13 by Anderson et al. finding no effect were performed both in male and female mice of similar ages to the other studies and the genotype of the mice was confirmed on the basis of significantly elevated plasma cholesterol levels [125,126]. However, different transgenic lines were used across studies, which may lead to different phenotypes.   The function of apoE-containing lipoproteins in the brain is similar to the periphery in that they are responsible for transporting cholesterol and phospholipids however there are many other functions related to inflammation, amyloid pathology, tau pathology, and the cerebrovascular system, to be discussed below, all of which are affected by apoE isoform.   1.9.2.3 APOE as a risk factor for AD In humans, apoE is mainly found in three isoforms, although other rare variants exist [131,132]. These three isoforms differ in amino acids at two residues with Cys112 and Cys158 corresponding to apoE2, Cys112 and Arg158 corresponding to apoE3, and Arg112 and Arg158 corresponding to apoE4 [133]. The three most common APOE gene variants have varying associations with AD risk and age of onset where APOE-ε2 is protective, APOE-ε3 neutral, and APOE-ε4 detrimental. The effect of APOE-ε4 on AD risk ranges from approximately a 2 to 4-fold increase in risk for people carrying one allele and a 2 to 33-fold increase for people with two alleles [106]. The large range in APOE-ε4 risk effects depends in part on ethnic group as demonstrated in a meta-analysis where Japanese people were most affected and Hispanics least affected by the risk allele [106]. Age of onset also shows a gene-dose dependent effect of APOE-ε4 by approximately 10 years per allele, declining from an average age of onset of 84 years old with no APOE-ε4 alleles to 76 years old for APOE-ε4 heterozygotes and 68 years old for APOE-ε4 homozygotes [105]. The prevalence of one copy of the APOE-ε4 allele in the people without AD is approximately 15-30% and two copies less than 5% depending on the region  studied, while the rates for two copies in AD cases is approximately 10-15% [106]. Overall, the APOE-ε4 allele is the greatest genetic risk factor for late-onset AD [134] and the risk of cognitive decline associated with APOE-ε4 only increases when combined with other factors including vascular risk factors [135] and female sex [136].     14 1.9.2.4 ApoE isoform effects on amyloid pathology Given the substantial effect of APOE-ε4 on AD risk, many researchers have explored how apoE4 affects amyloid pathology as a possible explanation for the increased disease risk. A meta-analysis has shown that APOE-ε4 carriers have a 2 to 3-fold increase in amyloid pathology as observed with PET imaging [137] while other studies have observed greater amyloid plaque density and CAA in human APOE-ε4 carriers post-mortem [138]. Studies in mice have aimed to understand the mechanisms behind the increase in amyloid burden and have mainly converged on differences in the clearance of Aβ via the blood-brain barrier (BBB). Castellano et al. found that the concentration of Aβ in the interstitial fluid of transgenic mouse brains was higher in those expressing human APOE-ε3/ε4 or APOE-ε4/ε4 compared to APOE-ε4 noncarriers and the clearance of Aβ across the BBB was reduced [139]. Work in mice by the Zlokovic group demonstrated that apoE facilitates Aβ clearance across the BBB using specific receptors and that the preferred receptor and speed of transport via those receptors is dependent on apoE isoform. Specifically, exogenously administered radiolabelled apoE4 redirects Aβ to the slower VLDLR receptor while apoE2 and apoE3 more rapidly transport Aβ across the BBB using both VLDLR and LRP1 [140]. ApoE can also affect the proteolytic cleavage of LRP1 and LDLR transporters at the endothelial cell (EC) surface induced by Aβ [141]. ApoE2 and apoE3 suppress this Aβ-induced ectodomain shedding, thereby improving Aβ clearance, while apoE4 has no such beneficial effect [142].  A mouse study using cell-specific, inducible expression of human apoE isoforms suggests that the deleterious effect of astrocytic apoE4 on amyloid pathology is most pronounced during the Aβ seeding stage [143].  1.9.2.5 ApoE isoform effects on tau pathology The effect of apoE isoform on tau pathology has been understudied compared to its effect on amyloid however a few studies suggest a detrimental effect of apoE4. Among humans with frontotemporal dementia, a tauopathy, those who carry an APOE-ε4 allele show worse behavioural symptoms [144]. A relatively recent study crossed P301S mice expressing a human mutant form of tau with human APOE knock-in mice and observed worse neurodegeneration and more neuroinflammation in the APOE-ε4 knock-in mice compared to APOE-ε3 knock-in controls [145]. Building upon this work, this research group next depleted microglia in the same human mutant   15 tau expressing APOE-ε4 knock-in mice and found that these mice were protected against neurodegeneration [73]. Surprisingly, they observed that mice lacking microglia had increased astrocytic and neuronal apoE concentrations, yet improved tau pathology and neurodegeneration. Together this suggests that the cellular source of apoE or cell-specific apoE signaling affects the pathological consequences of apoE in tau mice.   1.9.2.6 ApoE isoform effects on blood-brain barrier function Human APOE-ε4 carriers with AD show more BBB breakdown than noncarriers [146] including a higher CSF/plasma albumin quotient and greater levels of pericyte degeneration [147]. Studies in mice suggest that apoE-isoform dependent BBB breakdown is caused by pro-inflammatory cyclophilin A-matrix metalloproteinase 9 signaling in pericytes [148]. Studies with isogenic human EC produced from iPSC show more pro-inflammatory and pro-thrombotic signaling in APOE-ε4 cells even if these cells do not secret apoE [149]. Interestingly, APOE-ε4 and isogenic APOE-ε3  iPSC derived BBB-EC show no differences [150].   1.9.2.7 ApoE isoform effects on astrocytes As astrocytes are the primary producer of apoE in the brain, it is of great interest to determine how astrocytes expressing various apoE isoforms may differ. Studies in APOE-ε4 knock-in mice [151] and human iPSC derived astrocytes expressing APOE-ε4 [152] both suggest that APOE-ε4 astrocytes have reduced phagocytic activity. Lipid metabolism is also disrupted in APOE-ε4 iPSCs, specifically, APOE-ε4 astrocytes show increased cholesterol accumulation [152] and secrete smaller, less lipidated particles [153] than APOE-ε3 astrocytes. Interestingly, these APOE-ε4 iPSC derived astrocytes did not support the growth and synaptogenesis of co-cultured iPSC neurons as well as APOE-ε3 iPSC derived astrocytes [153].  1.9.2.8 ApoE isoform effects on microglia Although astrocytes are the primary producers of apoE in the brain in a healthy, resting state, microglia have been shown to have increased apoE expression during aging and neurodegenerative or inflammatory conditions in mice [154,155]. It has been long been reported that Aβ-expressing apoE knockout mice have reduced amyloid burden compared to apoE expressing mice [156–158],   16 however, it has more recently been shown that amyloid plaques in apoE deficient mice are larger, less compact, and surrounded by more dystrophic neurons [159,160], suggesting instead that apoE may protect against amyloid. The ability of apoE to compact amyloid plaques is dependent on TREM2. Mice lacking TREM2 or expressing a mutated form show reduced microglial activation, more amyloid deposition during the amyloid seeding phase, less microglial coverage of plaques, and less apoE deposition in the plaques [159]. Similarly, human AD patients with TREM2 mutations exhibit less clustering of microglia around plaques, particularly in those with TREM2 mutations which are associated with increased AD risk [159]. TREM2-dependent apoE signaling has been shown to be a key regulator of the switch of microglia from a homeostatic state to a neurodegenerative state in aging mice and mouse models of neurodegenerative diseases [66].   Studies on apoE isoform effects in microglia suggest that APOE-ε4 microglia are also dysfunctional. Human APOE-ε4 knock-in AD transgenic mice have larger amyloid plaques [161,162], reminiscent of the increased plaque size in apoE knockout mice described above, as well as reduced microglia clustering around plaques and reduced TREM2 expression by microglia compared to APOE-ε3 knock-in mice [162]. In vitro, murine microglia expressing human APOE-ε4 have a reduced ability to phagocytose Aβ compared to microglia expressing APOE-ε3 [163] and iPSC microglia from APOE-ε4 donors were found to have markedly altered phagocytic activity, inflammatory function, cholesterol metabolism, Aβ production, and tau phosphorylation compared to those from APOE-ε3 donors or APOE-ε3 isogenic cells [150,152,164]. Interestingly, iPSC microglia from people with familial AD mutations showed little difference from wildtype controls, suggesting APOE is a particularly important genetic factor in microglia [164].   Finally, as discussed above, a recent study depleted microglia from APOE-ε4-knock-in mice expressing human mutant tau [73]. They found that the neurodegeneration observed in mice with tau and APOE-ε4 does not occur in the absence of microglia, suggesting a detrimental function specifically of apoE4 produced by microglia or of apoE4 signalling in microglia.    17 1.9.3 Modifiable risk factors Although non-modifiable risk factors, namely age, sex, and genetics, undeniably have a large contribution to AD risk, increasing evidence in recent years suggests that dementia can potentially be prevented through addressing modifiable risk factors. According to a 2017 report commissioned by The Lancet, 35% of dementia risk is potentially modifiable [165].  This report evaluated the population attributable fraction for various modifiable risk factors, specifically education as an early-life risk factor, midlife hearing loss, hypertension and obesity as mid-life risk factors, and smoking, depression, physical inactivity, social isolation, and diabetes as late-life risk factors [165].   The Finnish Geriatric Intervention Study to Prevent Cognitive Impairment and Disability (FINGER) study was a double-blind randomized controlled trial (RCT) of 1,260 individuals aged 60 to 77 and was the first RCT evaluating a multi-modal lifestyle intervention for cognitive decline. The two-year intervention consisted of diet, exercise, cognitive training, and vascular risk factor monitoring while the control group received general health advice. In this landmark study, the intervention group showed improvements in executive function, processing speed, and neuropsychological test battery score and reduced the risk of cognitive decline [166] even in APOE-ε4 carriers [167]. Although both the French Multidomain Alzheimer Prevention Trial (MAPT) and Dutch Prevention of Dementia by Intensive Vascular Care trial (PreDIVA) found no benefits of multidomain lifestyle interventions in their RCTs overall, subgroup and post-hoc analyses were more promising. The MAPT trial showed benefits of the lifestyle intervention to parts of the MMSE and improvements to individuals at increased risk of dementia [168] while the PreDIVA trial found that lifestyle interventions reduced dementia incidence for non-AD dementia and reduced dementia risk in people with baseline untreated hypertension [169,170]. Several trials are ongoing testing the successful FINGER trial multi-domain intervention in other regions [170].   In May 2019, the World Health Organization (WHO) formally recognized the importance of modifiable risk factor reduction for dementia prevention in a report recommending preventative interventions and outlining the strength of each intervention based on existing research. The WHO guidelines strongly recommend tobacco cessation, physical activity, a balanced diet, management   18 of hypertension, and management of type 2 diabetes mellitus (T2DM). Recommendations of conditional strength were given for management of alcohol abuse, cognitive interventions, weight management, and management of dyslipidemia [171]. Select modifiable risk factors with a vascular component are discussed below.  1.9.3.1 Physical activity Physical activity interventions were not only a component of the seminal, multi-domain FINGER study but have also been extensively evaluated as a single intervention for the prevention of cognitive decline. A meta-analysis of 15 prospective studies found that all levels of physical activity consistently protected against cognitive decline, with an overall hazard ratio (HR) of 0.62 for high physical activity and HR of 0.65 for low to moderate levels [172]. A recent 44-year longitudinal Swedish birth cohort study of 800 women aged 38 to 54 found that physical activity was associated with an HR of 0.43 for mixed dementia and 0.47 for dementia with cerebrovascular disease, however, no risk reduction was found specifically for AD [173].   Data from RCTs studying older adults (generally over 65 years of age) are less consistent. Although several meta-analyses of RCTs including cognitively normal subjects or subjects with MCI or dementia have found a beneficial effect of aerobic exercise on cognitive function [174–177], a Cochrane systematic review found no benefit. This Cochrane review analyzed 12 RCTs including cognitively normal subjects at least 55 years of age and found that aerobic exercise exerted no benefit on any cognitive domain regardless of the type of control group used or the benefits to cardiovascular fitness acquired during the trial [178]. Although the trials included in the Cochrane review were shorter in duration than most in the previously mentioned meta-analyses, a recent, longer RCT similarly found no effect of exercise on cognition in healthy adults. This multi-centre US-based RCT of 1,635 sedentary adults between ages 70 and 89 studied the effects of a 24-month moderate-intensity intervention and found no improvements to any cognitive outcomes compared to a health education intervention, except in those over 80 years old and those with the lowest baseline physical performance [179]. Differences in the cognitive assessments used, age of study participants, aerobic exercise intensity, duration, and frequency, and the type of control group used may account for some of these inconsistencies. Overall, while observational   19 studies suggest that physical activity during mid-life protects against cognitive impairment, RCTs do not consistently show a benefit of aerobic exercise on the cognitive function of older adults.  1.9.3.2 Diet Research on diet and dementia has mainly focused on the potential benefits of the Mediterranean diet. For example, in a 6.5-year RCT of 522 people with high vascular risk aged 75 on average, those on the Mediterranean diet supplemented with nuts or extra virgin olive oil performed better on the MMSE and clock drawing test than those on the low-fat control diet [180]. An observational study on a younger population of 70 people aged 30 to 60 similarly found benefits to the Mediterranean diet, those with higher reported adherence to the diet had improved FDG-PET and PIB-PET both at baseline and three years later [181]. A cross-sectional study also in 30- to 60-year old subjects found that greater Mediterranean diet adherence was associated with greater cortical thickness on MRI [182].  1.9.3.3 Type 2 diabetes mellitus Although T2DM is not entirely a modifiable risk factor, its association with dementia risk is strong and management of T2DM has been recommended to prevent dementia [171]. A meta-analysis of 17 studies with a total of 1,747,777 participants found that the relative risk (RR) for AD for people with diabetes is 1.52 [183]. In particular, people with untreated diabetes [184] and diabetics who are also APOE-ε4 carriers [185] are at an increased risk of dementia. Brain imaging abnormalities have been well-documented in subjects with T2DM including cortical [182,186] and hippocampal [187] atrophy and deep white matter lesions [188] on MRI. T2DM also has specific effects on the BBB, for example, increased BBB permeability that was also correlated with biomarkers of endothelialisation in the CSF [189]. Others have observed increased BBB permeability as well as pericyte degeneration, basement membrane thickening, and capillary density changes in in vitro and in vivo models of diabetes or hyperglycemia, which may result from oxidative stress and inflammation [190].    20 1.9.3.4 Hypertension The relationship between blood pressure and AD risk is complex with different associations occurring at different ages, indications of risks of both hypertension and hypotension, and mixed results in clinical trials. Several recent, large population studies have made great advances in understanding the relationship. For example, the Atherosclerosis Risk in Communities (ARIC) has suggested that mid-life, but not late-life hypertension is associated with more cognitive decline two decades later [191]. Furthermore, they found that mid-life hypertension followed by late-life hypotension actually increases the risk of dementia (HR 1.62) at a rate comparable to sustained hypertension during mid-life through to late-life (HR 1.49) [192]. The 1946 British birth cohort study (Insight 46) found that hypertension even earlier in life can have damaging effects on the brain. Specifically, high systolic and diastolic blood pressure at 36 and 53 years of age as well as an increase in blood pressure from 43 to 53 years of age in this cohort was associated with increased white matter hyperintensity volume at 69 to 71 years of age. Similarly, higher diastolic blood pressure at 43 years of age and increases from 36 to 43 years of age was associated with smaller whole brain volume at late-life follow-up while increases in systolic blood pressure from 36 to 43 years of age was associated with smaller hippocampal volume at follow-up [193]. It is possible that dementia risk could be reduced in hypertensive individuals if blood pressure is successfully lowered, although few studies attempting this preventative strategy have been published, most trials fail to find any benefits, and the benefits may depend on the age at which hypertension is treated and the severity of the hypertension at baseline [194–196].  1.9.3.5 Hypercholesterolemia The effect of blood cholesterol concentration on AD risk similarly seems to depend on age. Systematic reviews and meta-analyses suggest that mid-life total cholesterol concentrations positively correlate with AD risk but late-life total cholesterol has no such association or even an inverse association [197–199]. For example, serum total cholesterol concentration measured at 75 years of age had no association with incident AD risk in original cohort of the Framingham study [200] while positive associations were observed between mid-life cholesterol concentrations in both the ARIC study at follow-up 20 to 25 years later [201] and in the Three-City Study at follow-up 13 years later [202]. On the other hand, mid-life cholesterol measured in the Prospective   21 Population Study of Women in Sweden was not associated with AD 32 years later [203]. The effect of statins on AD risk is similarly unclear. Statins are drugs that inhibit a key enzyme in cholesterol synthesis and are widely administered to people with hypercholesterolemia. Although prospective studies consistently find that statin use reduces dementia and AD risk [204–206], RCTs find no such effect [204,207]. Mendelian randomization studies have attempted to discern whether the associations between hypercholesterolemia and AD are causal or a result of cofounding variables by studying the effects of genetically inherited polymorphisms that alter plasma cholesterol concentrations. Several such studies have found that genetic predisposition to high serum total cholesterol is not related to AD risk [165,208,209], however, this does not rule out an effect of hypercholesterolemia arising from non-genetic causes. High plasma triglyceride and low-density lipoprotein cholesterol (LDL-C) concentrations and low HDL-C concentrations are also found to be associated with increased AD risk in some studies, although the relationship is even less clear [197,202,210]. The association between plasma HDL-C and AD risk is more thoroughly discussed in section 1.14.2.  1.10 Therapeutics for AD 1.10.1.1 Existing AD therapeutics There are currently no disease altering AD therapeutics, only drugs for easing AD symptoms. The two classes of symptomatic drugs currently available include cholinesterase inhibitors (donepezil, rivastigmine, and galantamine) and an N-methyl-D-aspartate (NMDA) receptor antagonist (memantine). Meta-analyses suggest that donepezil can improve cognitive dysfunction in AD [211] and while memantine can also be beneficial in moderate-severe AD, its effects are subtle [212], inferior to donepezil [211], and best when in combination with donepezil [213].   1.10.1.2 AD therapeutics in ongoing clinical trials As amyloid is found in all AD subjects and the “amyloid hypothesis” suggests that Aβ accumulation is responsible for neurodegeneration and memory loss in AD, most therapeutics in development for AD target Aβ [214]. Amyloid antibodies, such as aducanumab, have progressed further in clinical trials than any other AD therapeutic. Although promising phase I trial results for aducanumab [215] were followed by a phase III trial termination due to futility [216], a recent re-  22 analysis of the phase III trials has revived excitement for this drug [217]. The re-analysis accounted for a mid-study protocol amendment that altered the dosing of many APOE-ε4 carrying participants who were originally on lower doses due to concerns regarding amyloid related imaging abnormalities (ARIA). Once these additional participants on the high dose of aducanumab were included, the larger EMERGE trial met its primary endpoint of slowed decline on the Clinical Dementia Rating-Sum of Boxes (CDR-SB). However, data from this trial has not yet been peer-reviewed and the re-analysis of the smaller aducanumab phase III trial ENGAGE did not replicate the positive findings of EMERGE, resulting in continued uncertainty as to the future of aducanumab [218].  Several BACE-inhibitors, drugs which block the cleavage of APP by β-secretase into Aβ precursors, have failed in clinical trials. Atabecestat was terminated in phase II/III due to liver toxicity [219], umibecestat was terminated in phase II/III due to a worsening in cognitive function in treated subjects [220] and verubecestat [221], LY3202626 (NCT02791191), and Atabecestat (NCT02406027) were all terminated in phase II or II/III due to futility. Elenbecestat is the last remaining BACE-inhibitor in development with a phase III clinical trial currently recruiting (NCT02956486).  Other researchers have turned their attention away from amyloid-based therapies to instead attempt to target tau or inflammation. A phase II trial of BIIB092 for early AD  has completed recruitment and is now ongoing (NCT03352557) while a phase II trial of ABBV-8E12 for early AD is currently enrolling by invitation (NCT03712787). In terms of neuroinflammation, a recent trial of naproxen, a non-steroidal anti-inflammatory drug, failed to slow cognitive decline and had dangerous intestinal and cardiovascular side effects, cooling enthusiasm for further research in this area [222].    1.11 The cerebrovascular system Despite comprising only 2% of total body mass, the brain consumes approximately 20% of total cardiac output [223]. The high metabolic activity of the neural network and lack of glucose stores in the central nervous system (CNS) means that the brain must be highly vascularized to enable oxygen and glucose influx, maintain ion balance, and remove neurotoxic waste products [224].   23 The BBB has a specialized anatomy compared to peripheral vessels to regulate the transport of waste products and nutrients into and out of the brain. Tight junctions between endothelial cells (ECs) create a selective barrier between the blood and the brain parenchyma through which transport is regulated by specialized proteins on ECs. ECs of the cerebrovasculature are surrounded by mural cells, either pericytes in capillaries or vascular smooth muscle cells in arteries and arterioles, and astrocytic end-feet [225–227].  1.11.1 The relationship between the cerebrovasculature and AD   The importance of the vasculature to dementia is underscored by the fact that most dementia cases exhibit vascular pathologies [228]. Early, anecdotal observations by Hassler et al. found arteriole and precapillary deformities to be very common in the brains of people with dementia, less common in aged people without dementia, and not present in people younger than 68 years old [229]. Subsequent systematic and quantitative analysis of post-mortem AD and healthy brains built upon these initial observations and found reduced vascular density [230,231], increased vessel tortuosity [231], and more vessel remnants lacking a lumen and EC coverage [146,232,233] in the brains of people with AD. More recently, the US-based National Alzheimer’s Coordinating Center studied nearly 6,000 brain autopsies with various neurodegenerative diseases and found that AD subjects had a greater burden of cerebrovascular pathologies including macro- and micro-infarcts, atherosclerosis, arteriosclerosis, and CAA than people with other types of neurodegenerative diseases [228]. Similarly, the Religious Orders Study and Rush Memory and Aging Project studied over 1,100 post-mortem brains and found that the severity of atherosclerosis or arteriosclerosis and presence of macro- and micro-infarcts was correlated with AD risk [234].   Technological advancements in imaging techniques and biomarker measurements have provided further evidence of cerebrovascular dysfunction in AD in living people. An elegant analysis of plasma and CSF biomarkers along with 7,700 brain images obtained through the Alzheimer’s Disease Neuroimaging Initiative (ADNI) tracked the progression of AD and suggested that cerebrovascular dysfunction, specifically reduced cerebral blood flow measured with arterial spin labelling magnetic resonance imaging (ASL-MRI) and reduced glucose metabolism measured with FDG-PET, is an early temporal event in the development of AD [235]. Many other groups   24 have also observed cerebrovascular dysfunction in living AD subjects with imaging techniques. Transcranial doppler techniques have shown that reduced cerebral blood flow predicts dementia risk and cognitive decline [236] and microbleeds observed on MRI have been associated with AD risk [237,238]. Pulse wave velocity tests have shown that greater arterial stiffness is associated with greater Aβ burden observed with PET imaging, lower brain volume in certain brains regions observed by MRI, and higher white matter hyperintensity burden [239]. Furthermore, dynamic contrast-enhanced MRI has shown that BBB breakdown occurs in the hippocampus in an age-dependent manner, worsens with MCI [240], and occurs even in the early stages of cognitive impairment independent of Aβ and/or tau biomarker changes [6]. Taken together, vascular pathology, including micro- and macro-infarcts, small vessel disease, white matter abnormalities, arteriosclerosis and CAA, can be the major pathology in 10-20% of dementia cases and are associated with slower perceptual speed and poorer episodic memory [5,241–244]. MRI sequences that can evaluate disrupted cerebral blood flow and cerebral small vessel disease were recently proposed as potential vascular biomarkers to be added into the 2018 ATN Research Framework for preclinical AD in recognition of their importance in identifying vascular dysfunction in the early stages of AD pathogenesis [4].  1.11.2 Cerebrovascular clearance of Aβ from the brain The cerebrovasculature plays a pivotal role in removing Aβ from the brain. This can occur through active transport across ECs of the BBB or via vascular drainage in mid- and large-sized arteries [245]. Direct transcytosis through the BBB involves the LRP1, LRP2, VLDLR, p-glycoprotein and phosphatidylinositol binding clathrin assembly protein (PICALM) [246–249] (Figure 1.1). Lipoproteins affect direct BBB clearance of Aβ by acting as receptor chaperones whereby apoE2 and apoE3 direct Aβ clearance through the faster LRP1 route whereas apoE4 prefers the slower VLDLR and apoJ assists in transport via LRP2. Aβ is also drained through the paravascular and intramural space of the brain artery wall toward the cervical lymph nodes [7]. The space between the pia mater and glia limitans forms the paravascular space where Aβ transport is facilitated by astroglial aquaporin 4 and diffusion, whereas for intramural drainage, soluble Aβ enters the basement membrane of the capillaries and is drained along the basement membrane surrounding   25 smooth muscle cells [7,250,251]. Modification in vascular compliance has been proposed as a cause of impairment leading to vascular Aβ accumulation known as CAA [252].    Figure 1.1 Aβ clearance via the cerebrovasculature. In addition to clearance of Aβ by intracellular and extracellular degradation, Aβ can be cleared via the cerebrovasculature. Direct Aβ transport across the BBB at the capillaries occurs via receptors such as LRP1, LRP2, and VLDLR. LRP1 receptor-mediated endocytosis of Aβ is assisted by PICALM. Aβ also drains to the lymph system via perivascular drainage pathways whereby Aβ enters the capillaries then drains through the paravascular or Virchow-Robin space with the assistance of AQP4 or drains through the intramural pathway between layers of smooth muscle cells of arteries. Aβ: amyloid beta, AQP4: aquaporin 4, LRP1: low-density lipoprotein receptor-related protein 1, VLDLR: very low-density lipoprotein receptor, PICALM: phosphatidylinositol binding clathrin assembly protein, P-gyp: P-glycoprotein, apo: apolipoprotein.  1.11.3 Cerebral amyloid angiopathy The deposition of Aβ in vessels, known as CAA, has many similarities to Aβ deposition in neuritic plaques including their similar genetic and sporadic origins. Mutations in APP, in particular in or near the Aβ coding regions, increased APP gene dosage, and mutations in PSEN1 and PSEN2 are all related to familial CAA while APOE is the greatest risk factor for sporadic CAA [253]. Unlike parenchymal Aβ deposits, which are predominantly composed of Aβ42, Αβ40 dominates in vascular amyloid deposits [253,254].    26 Sporadic CAA is found in two forms. In Type I CAA both arterioles and capillaries are affected while in Type II CAA capillaries are unaffected [255] although autopsy studies suggest that capillary CAA presence and severity is a better predictor of neuritic plaque and other AD pathologies and dementia [256,257]. Autopsy studies find a high prevalence of CAA both in normal aging and in dementia. For example, in the Religious Orders study, 85% of their aged cohort (average age 87 years old, N=404) had CAA pathology and CAA was found in 94% of cases with dementia [258].   CAA is diagnosed based on the Boston criteria. The original Boston criteria relied on post-mortem or biopsy diagnoses however the updated criteria now includes the diagnosis of probable or possible CAA based on MRI or CT findings. Specifically, the presence of multiple hemorrhages (in lobar, cortical, or subcortical regions or a single lobar, cortical, or subcortical hemorrhage with cortical superficial siderosis) found using neuroimaging in people at least 55 years of age with no other cause of hemorrhage constitutes probable CAA. Possible CAA uses the same criteria but with the presence of a single hemorrhage [259]. Although the presence of hemorrhages is used to assist in the diagnosis of CAA, ischemic lesions are also observed [260] and include white matter hyperintensities, structural dysfunction observed using diffusion tensor imaging MRI, and cerebral microinfarcts [253].   1.11.4 Tau and the cerebrovasculature In addition to Aβ, it is also important to consider the interaction of the cerebrovascular system with tau. Researchers have long observed NFT and p-tau in the perivascular regions of immunostained post-mortem AD brains [261–263]. This perivascular tau pathology correlates with the severity of vascular Aβ pathology [264] and vessels afflicted with CAA are more likely to show perivascular tau pathology than non-CAA vessels [265]. More recent work systematically quantified perivascular tau pathology and BBB impairment in post-mortem brains at each Braak stage of tau deposition and revealed that deposition of perivascular tau and vascular degeneration, specifically loss of α-smooth muscle actin (α-SMA) expression, occurs in early Braak stages and can be independent of CAA deposition [266]. While tau is typically an intracellular protein and the perivascular tau observed in the above studies is assumed to be intracellular, Castillo-Carranza et   27 al. recently demonstrated that extracellular tau oligomers can also deposit in perivascular spaces. They used a tau oligomer specific antibody for immunostaining on post-mortem brains and observed increased perivascular deposition of extracellular tau oligomers in AD brains and in Tg2576 AD model mice compared to non-AD humans and mice [267].   Studies using mice transgenic for human tau or injected with adeno-associated virus (AAV) tau constructs have shown that perivascular tau and cerebrovascular dysfunction can also occur in the absence of Aβ pathology. For example, increased BBB leakiness [268–270], abnormal blood vessel morphology [271], and perivascular astrocyte swelling [270] are observed in mice expressing human mutant tau. Also emerging from studies in mice is the role of the cerebrovasculature in clearing tau from the brain. Tau-specific BBB transporters have not yet been discovered and most tau clearance is assumed to occur via intracellular degradation [7,272]. However, it has been suggested that the cerebrovasculature may be involved in clearing soluble extracellular tau as peripheral administration of tau oligomer antibodies in transgenic human mutant tau mice reduced brain concentrations and increased serum concentrations of oligomeric tau [273]. Furthermore, extracellular human tau injected directly into the cerebral cortex of mice appears to be cleared via glymphatic pathways along perivascular spaces [274].   1.11.5 Cerebrovascular inflammation Inflammation is another component of cerebrovascular dysfunction known to occur in AD. Microvessels isolated from the brains of people with AD have elevated concentrations of a number of pro-inflammatory markers compared to age-matched control brain microvessels including hypoxia-inducible factor α (HIF-1α), IL-8 [275], monocyte chemoattractant protein 1 (MCP-1), IL-1β [276], NOS activity [277], TGFβ [278] and phosphorylated p38 MAPK [279]. More recent unbiased gene expression profiling supports these earlier observations with gene ontology analysis suggesting that 2,865 genes are differentially regulated in microvessels from AD brains, most of which overlap in immune system, defense, chemotaxis, and inflammatory response pathways [280].    28 Studies both in humans and in mice suggest that abnormal thrombosis and fibrinolysis contribute to cerebrovascular inflammation in AD brains. While the thrombosis and fibrinolysis are part of a process critical in preventing hemorrhage, the healthy balance of clot formation and degradation may be disrupted in disease resulting in prolonged fibrin deposition in tissues, which is problematic in the brain due to the inflammatory properties of fibrin [281]. Fibrin positive blood vessels [282] and parenchymal fibrin deposits [283,284] are observed more commonly in AD mouse and human brains than controls. Parenchymal fibrin deposits in the AD mouse brains are found in areas with BBB breakdown, increase with age, and correlate with Aβ deposition [283]. Reducing fibrinogen levels with genetic or pharmacological methods in AD mice results in improvements to AD pathologies including reduced BBB damage, neuroinflammation [283], CAA, and cognitive impairment [282]. Similarly, cerebrovascular thrombin levels are elevated in  human and mouse AD brains [275,285] and treatment of AD mice with a thrombin inhibitor reduces HIF-1α, IL-6, MCP-1, and MMP2 expression on cerebral vessels [285].   The cerebrovascular system also provides a route for immune cell entrance into the brain. While peripheral immune cell infiltration is a major component of some neurodegenerative diseases, such as MS and stroke [286], the role of this process is less clear in AD. Although evidence for infiltrating immune cells in human AD brains is limited [287], extravasation of neutrophils [288], monocytes [289], and T cells [290] into the parenchyma of AD mouse brains have been demonstrated. Neutrophils can cause cerebrovascular dysfunction in AD mice as they adhere to capillaries resulting in stall blood flow [291] and as they extravasate to areas with Aβ then release pro-inflammatory signals [288]. Antibodies against neutrophils, neutrophil depletion, or blocking of neutrophil receptors can reverse these effects and results in reduced cognitive decline [288,291]. Live two-photon microscopy reveals patrolling Ly6Clo monocytes similarly are attracted to Aβ as they are observed to bind and crawl into the walls of Aβ-positive veins, take up Αβ, and return to the blood stream. Blocking monocyte extravasation in AD mice by Ly6Clo monocyte ablation worsens plaque load [289] while deficiency in the CC chemokine receptor 2 (CCR2) required for monocyte recruitment both increases both plaque load and worsens memory deficits [292], suggesting that monocyte infiltration may play a protective role against Aβ pathologies.   29 Perivascular macrophages likely play a similar role as depletion of these cells in AD mice leads to increased CAA while perivascular macrophage stimulation reduces CAA [293].   In order for leukocytes to enter the brain, endothelial cells of inflamed tissues express adhesion proteins on their cell surface that bind patrolling immune cells [294]. Interestingly, although it is unclear whether leukocyte extravasation occurs in AD, plasma concentrations of several of these proteins are increased in AD subjects [295–297]. Vascular cell adhesion molecule 1 (VCAM-1) plasma concentrations in AD subjects increases depending on AD severity, correlates with CDR-SB cognitive scores, and is associated with white matter changes [297]. VCAM-1 has also been tied to the detrimental outcomes on the brain in aging. In experiments where aged plasma was administered to young mice, the young mice suffered increased microglial reactivity and cognitive deficits similar to those of old mice. These detrimental effects were suppressed with an anti-VCAM-1 antibody, suggesting that leukocyte extravasation is harmful to the aging brain [298].  Changes to leukocyte infiltration in AD may also be related to the microbiome. A recent study in AD mice found that microbiome altering drug sodium oligomannate reduced neuroinflammation and cognitive deficits [299] and  in an unpublished phase III clinical trial it was reported to improve cognition in AD patients [299]. Microbiome alterations have been reported in people with AD [300,301] but conclusions of the effectiveness of sodium oligomannate and its mechanisms of action are not possible until more data from the previous phase III trial is made available and until after the completion of the global phase III trial that begins in 2020. There is also skepticism among AD researchers of the connection between microbiome changes and inflammation in the treated mice as some reported microbiome alterations are in fact related to increased inflammation [302,303].  1.11.6 Vascular co-morbidities in AD The important role of the vasculature in AD is further put into perspective when one considers that CVDs such as stroke and CHD, are associated with increased AD risk [243,304,305]. Associations between CVD and AD might be due to direct causal effects, as cerebral perfusion is often reduced in CVD, or due to shared genetic and environmental risk factors. Indeed, numerous genetic studies   30 have shown that individuals who carry at least one copy of the APOE-ε4 allele have an increased risk for both AD and CVD compared to those without the APOE-ε4 allele [306]. AD and CVD share many cardiometabolic and lifestyle risk factors, especially in middle-aged and elderly people, including age, sex, smoking, blood pressure, physical activity, blood lipids, and T2DM [182,191,307]. Several of these factors have been compiled into composite measures of general cardiovascular risk known as the Cardiovascular Risk Factors Aging and Dementia (CAIDE) risk score [308,309], which has recently been shown to correlate with executive function, visual perception and construction, APOE, white matter hyperintensities and CSF AD biomarkers including Aβ and tau in healthy adults [308]. CAIDE dementia risk scores above 11 points were also recently associated with a 2.10 increase in the odds ratio for cerebral infarcts up to 10 years later [309]. In the population-based Rotterdam Study, a cerebral small vessel disease sum score based on MRI was recently found to be associated with higher risk of stroke, dementia and death [310]. The Framingham cardiovascular risk profile score also predicts conversion from amnestic MCI to AD within 24 months [311].   1.12 In vitro models of the BBB  Given the importance of the cerebrovascular system to the pathogenesis of AD, developing human-based vascular models that retain both anatomical and physiological similarities to humans are highly desirable. A wealth of studies on the BBB have been performed using traditional two dimensional (2D) cell culture methods from primary human or animal brain ECs, immortalized human or animal brain ECs, and brain ECs produced from human or mouse pluripotent stem cells sometimes in co-culture with other BBB cell types [312–320], however it is increasingly recognized that cells behave very differently in three dimensional (3D) compared to 2D environments [321]. Breakthroughs in stem cell biology and tissue engineering have resulted in several innovative in vitro models of the human BBB with increased relevance for human physiology.  The validity of each model is assessed by examining the hallmark properties of the BBB including the anatomical structure, microenvironment, barrier function, cell function and composition [322].      31 Traditional trans-well systems offer highly reproducible models for permeability assays [323,324].  A recent example using a BBB trans-well system demonstrated that apoA-I and apoJ increase Aβ40 efflux through the endothelium [325].  While this system displays high trans-endothelial electrical resistance and functional efflux, it lacks the complex cell-cell and cell-matrix interactions of the cerebral vasculature.   Advanced microfluidic devices have advantages over trans-well methods as they allow for cell-cell interactions and can be perfused. The first 3D neurovascular microfluidic model brought together primary murine neurons and glia cells with human cerebral ECs to study BBB transport [326].  To address concerns about the notable differences in drug transporters between murine cell types to human physiology, Kamm and colleagues have now populated microfluidic devices with human iPSC-derived ECs, primary human pericytes and astrocytes [327].  Maoz and colleagues have developed a distributed neurovascular unit organ-on-chip microfluidic system by linking a BBB chip to a brain chip composed of neurons and astrocytes [328].  This novel system successfully identified previously unknown metabolic coupling between the BBB and neurons but is unable to model the anatomical connections between cells of the neural vascular unit. Vatine et al. also used iPSC to create a microfluidic BBB with neural spheroids and brain EC separated by a porous barrier allowing for cell to cell interactions [329]. EC is this model stained positive for BBB markers and spontaneous neuronal activity was observed by calcium imaging. Interestingly, whole blood perfused through the lumen of this system was contained by the capillary wall and did not induce neuronal toxicity, demonstrating the formation of a physiological BBB.   Cerebral organoid models provide further advancements to physiological relevance in that they are capable of forming distinct cortical layers [330] and can be partially vascularized using a variety of strategies. Ham et al. aimed to vascularize their embryonic stem cell (ESC) cerebral organoids using treatments of recombinant vascular endothelial growth factor (VEGF) [331]. VEGF-treated organoids had increased ANGPT and PECAM1 mRNA expression levels and  increased cluster of differentiation 31 (CD31) staining with minimal changes to stem cell and neuroectoderm markers. At 32 days of age, CD31 and claudin-5 positive tubes were visible in the VEGF-treated organoids yet absent in untreated organoids. The addition of wnt7a, a component   32 of the wnt/βcatenin pathway crucial to BBB development [332], resulted in the presence of α-SMA-positive cells surrounding CD31-positive cells in 4-month old brain organoids.  Song et al. employed an entirely different method in organoid vascularization, opting instead to create a hybrid iPSC-derived neuro-vascular organoid from the addition of a vascular spheroid to a neuronal spheroid [333]. These hybrid organoids were positive for CD31, zona occludens 1 (ZO-1), and VE-cadherin markers, which were further increased by the addition of mesenchymal stem cells (MSC) at the time of spheroid combination. A third strategy employed by Cakir et al involved engineering embryonic stem cells to express human ETS variant 2 (ETV2), a transcription factor involved in vascular development [334]. Their vascularized organoids with 5-20% ETV2 expressing cells featured lumens positive for more CD31, claudin-5, ZO-1, GFAP, and occludin than organoids without ETV2 expressing cells and showed significantly less necrosis and hypoxia. Although perfusion with dextran demonstrated the perfusibility of CD31 and claudin-5 positive lumens, only 8% of the lumens were perfused in total suggesting a lack of connectivity of the vascular network. A final strategy attempted by several groups was simply to co-culture various cell types of interest to form an organoid with a neuronal core and endothelial surface. Pham et al. coated 34-day old iPSC-derived cerebral organoids with iPSC-derived EC and observed vascularization of the outer layers of the organoid after 3-5 weeks in vitro or two weeks after implantation into a mouse [335]. Cho et al. instead used primary astrocytes, pericytes, and EC cultured simultaneously and observed self-organization of organoids with an astrocyte core and a surface of pericytes and EC after 48 h [336,337]. The surface of these organoids stained positive for claudin-5, occludin, ZO-1, and LRP-1 and expressed functional P-glycoprotein as demonstrated with successful rhodamine 123 transport into the spheroid interior. Since then, Nzou et al. co-cultured organoids with a vascularized surface containing a total of six cell types from primary or iPSC sources [338]. They began by culturing neurons, astrocytes, oligodendrocytes, and microglia into spheroids then added pericytes and EC after 48 h to produce vascularized organoids with surface expression of ZO-1, claudin-5, P-glycoprotein, GLUT-1, β-catenin, and VE-cadherin. While these recent advances are innovative and exciting, 3D organoids with a completely perfusible vascular network have yet to be developed.    33 1.13 High-density lipoproteins (HDL)  As thoroughly described above, promoting vascular health is a key strategy in preventing dementia and may be a novel strategy in treating AD. Studying the effects of known vasoprotective agents on the brain may therefore lead to innovative therapeutic strategies for AD. One such vasoprotective agent is HDL. Like other mature lipoproteins, HDL consists of a core of hydrophobic lipids surrounded by a phospholipid and free cholesterol monolayer studded by proteins (Figure 1.2) [339]. A key protein found in most HDL particles is apoA-I, which makes up 70% of its protein content [340]. The major lipid classes found on HDL includes cholesterol and other steroids, phospholipids, cholesteryl esters, sphingolipids, and triglycerides [341]. Overall, HDL particles consist of approximately 85-95 distinct proteins [342]  and hundreds of lipid subtypes [343], that together mediate diverse HDL functions including lipid metabolism, anti-oxidation, immune response, hemostasis, metal binding, and vitamin transport [343–345].  Figure 1.2 HDL composition and function HDL is the smallest and densest of the plasma lipoproteins and contains an estimated 85-95 distinct proteins, 200 lipid species and several other nonpolar cargo molecules. HDL components and subclass distribution can vary between individuals and is altered by diseased states. EC: endothelial cell; PL: phospholipid; S1P: sphingosine-1-phosphate; SM: sphingomyelin; NO: nitric oxide; p: phosphate group; eNOS: endothelial nitric oxide synthase; apo:   34 apolipoprotein; SAA: serum amyloid A; PON-1: paraoxonase 1; miRNA: micro RNA; LDL: low density lipoprotein; oxLDL: oxidized LDL [346]. (modified from Boyce et al. 2017 [347])  1.13.1 HDL protective associations with cardiovascular diseases The association between low HDL cholesterol (HDL-C) concentration and elevated CVD risk was first suggested in the 1960s in the Framingham Heart Study [348]. Since then, a multitude of clinical studies have supported this relationship [349–351]. Although Mendelian randomization studies have demonstrated that HDL-C concentration per se has no causal relationship with CVD [352–354], the question remains as to whether HDL-C, a static measure of HDL’s cholesterol content, adequately reflects the beneficial functions of HDL on vascular health.   In humans, genetic deficiency of APOA1 and ATP-binding cassette transporter 1 (ABCA1) lead to very low concentrations of plasma HDL-C that can be associated with increased risk of and accelerated onset of coronary artery disease (CAD). Similar outcomes are observed in some but not all cases of lecithin cholesterol acyltransferase (LCAT) deficiency [355]. However, in other cases the genetic alteration of HDL-C concentrations has more complex outcomes. For example, the apoA-I Milano variant was discovered in a family in Northern Italy and has a substitution resulting in the formation of homodimers of apoA-I and heterodimers with apoA-II. Carriers of apoA-I Milano have very low plasma HDL-C concentrations but similar levels of atherosclerosis and CAD as controls with normal HDL-C concentrations [356,357]. Conversely, carriers of mutations in SCARB1, the gene encoding scavenger receptor class B type I (SR-BI), have abnormally high HDL-C concentrations yet are at increased risk for CAD [358].  The predominant plasma lipoprotein in human plasma is LDL [359] while the major plasma lipoprotein in mice is HDL, making them generally resilient to CVD and advanced atherosclerosis compared to humans [360]. As a result, atherosclerosis studies in mice are overwhelmingly based on genetically modified models to allow the therapeutic effect of HDL on atherosclerosis to be studied [361].  For example, genetic deletion of either apoA-I, ABCA1, LCAT or SR-BI on the pro-atherosclerotic backgrounds apoE or LDLR knockout alters murine HDL-C concentrations, however, the effects on atherosclerosis vary with genetic background and animal diet [362]. Nonetheless, HDL-targeted therapies in murine and rabbit models of atherosclerosis appear to be   35 beneficial. For example, transgenic overexpression of apoA-I, gene transfer of human apoA-I, adenoviral transfer of apoA-I, and infusion with recombinant apoA-I or HDL can reduce or stabilize atherosclerotic plaques in apoE or LDLR knockout mice [363]. Similar improvements have been observed in rabbit models treated with HDL-based therapeutics. Aortic plaque area [364] and aortic inflammation [364,365] in rabbits with high-fat diet induced atherosclerosis are reduced when they are treated intravenously with purified plasma human apoA-I. Likewise, atherosclerotic progression in LDLR knockout or LDLR mutant rabbits was inhibited with intravenous apoA-I mimetic [366] or adenoviral apoA-I treatment [367]. While the protective effect of HDL against CAD has long been assumed to be related to HDL’s function in reverse cholesterol transport, other protective functions of HDL may also play a role.  1.13.2 HDL function in health and disease  Circulating HDL is best known for its pivotal role in reverse cholesterol transport, the process by which HDL removes excess cholesterol from cells, the lipoprotein remodels, and the excess cholesterol is delivered to the liver for excretion [368]. However, only 1/3 of the identified 95 proteins on HDL [110] have roles in lipid metabolism [343,369] while others function in protease inhibition, complement regulation, hemostasis and inflammation [370]. Known vasoprotective functions of HDL include promoting endothelial nitric oxide synthase (eNOS) activity, reducing inflammation, and suppressing vascular adhesion molecule expression [347,371–374]. Several of these functions have been observed to change during disease.  1.13.2.1 HDL and cholesterol efflux capacity The cholesterol efflux capacity (CEC) of HDL is modified in CVD [375–379], metabolic syndrome [380], during acute inflammation [381], and, interestingly, in AD [382,383] (Table 1.1). In T2DM, the CEC of HDL is reduced in some [384,385], but not all [380], studies. How distinct CEC and HDL-C are as biomarkers of disease has been debated, with some studies observing a reduction in CEC independent of changes in HDL-C [375,376,378,381,386,387], while others find that CEC and HDL-C changes correlate [377,379,380]. Possible mechanisms to explain reduced CEC in most of these studies is increased HDL-associated serum amyloid A (SAA) [381] and reduced HDL-associated paraoxonase 1 (PON-1) [379] in inflammatory states, which will be   36 discussed below. Importantly, despite its obvious implication for reverse cholesterol transport, CEC is not the only known function of HDL.  37 Table 1.1 HDL compositional and functional heterogeneity in disease. Disease Proteome Lipidome Size Function Cardiovascular disease ↓/--  PON-1  [379,388,389] ↑ apoA-I,  apoA-II,  apoE [390] ↑ SAA [390–392] ↑S1P [393], SM [394] ↑/↓ PL [392,394,395] ↑ TG [392,394] ↑/↓ large HDL [376,396–398] ↓ small HDL [388] ↓ cholesterol efflux [378,379], eNOS phosphorylation [376,391,399] ↑ inflammatory activity [400] Inflammation ↓ PON-1 [381,401,402] ↑ SAA, apoA-II, complement C 3[401,402] ↑/↓ apoA-I [401,402] ↓S1P, SM[402,403] ↑ TG, FFA [403]   ↓/-- cholesterol efflux [381,401,402] ↓ NO production [401,402] ↑ inflammatory activity, oxidative activity [404] Chronic kidney disease ↓/--  PON-1 [399,405] ↑ SAA, SDMA, apoC-II [399,406–408] ↓apoA-I, apoA-II [399,408] ↓ PL [408] ↑ TG [408] ↓ small HDL[409] ↓ NO production [406], cholesterol efflux [406,408], EC proliferation and migration [410]  ↑ inflammatory activity [406,410], oxidative activity [406,407] Cirrhosis ↑/↓ PON-1 [411] ↓ apoA-I, apoA-II, apoC-II, apoC-III [412] ↑ SAA, apoE [412] ↑ PL[411] ↑ large HDL[412] ↓ cholesterol efflux [412], PON-1 activity [411] Aging -- PON-1 [413,414] ↓ apoE [413] ↑ SAA, complement C3 [413] ↑ SM[415]  ↑ oxidative activity [413,414] ↑/-- cholesterol efflux [413,415] ↓ PON-1 activity [413,414] Arthritis ↓ PON-1 [416] ↑ SAA [416] ↑ SM, PL [417]  ↑ inflammatory activity [418] Type 2 diabetes mellitus ↓ apoA-II, PON-1, apoE [122] ↑ SAA [419]  ↑ TG [420] ↑/↓ S1P [421,422]  --/↓ cholesterol efflux [380,384,385] ↓ eNOS activation [423,424] ↑ inflammatory activity [423,425–427] ↓ anti-oxidative [422,424,426,428] Alzheimer’s disease    ↑ inflammatory activity [383] ↓ cholesterol efflux [382,383]  Age-related Macular Degeneration ↑ SAA [429]    ↑ anti-inflammatory activity [429] PON-1: paraoxonase 1; SAA: serum amyloid A; apo: apolipoprotein; S1P: sphingosine-1-phosphate; SM: sphingomyelin; PL: phospholipid; TG: triglyceride; eNOS: endothelial nitric oxide synthase; FFA: free fatty acid; NO: nitric oxide; SMDA: symmetrical dimethylarginine.  38  1.13.2.2 HDL and inflammation Several mechanisms by which HDL exerts anti-inflammatory effects on ECs have been described. These include pathways dependent on the HDL receptor SR-B1 [430–432] as well as through vascular sphingosine-1-phosphate (S1P) receptor 1 and 3 [433], which trigger a signaling cascade through the phosphoionositide 3-kinase (PI3K)/Akt pathway leading to phosphorylation of eNOS. The vasoprotective effects of nitric oxide (NO) produced by eNOS phosphorylation are well-established, and include vasodilation, reduced EC permeability, and inhibition of vascular adhesion molecule expression via cells downregulation of the pro-inflammatory nuclear factor kappa beta (NFκB) signaling pathway [434]. Although moderate amounts of NO produced by EC through HDL signaling are vasoprotective, high or sustained NO production from inducible NOS signalling or uncoupled eNOS signalling contribute to oxidative stress [435].  HDL-S1P action can also directly inhibit NFκB signaling to suppress adhesion molecule expression [433], reduce endothelial exocytosis [431], and maintain annexin-1 expression [430]. Additionally, HDL can indirectly increase eNOS activity via actions of the lipid transporter ATP-binding cassette G1 (ABCG1) to maintain proper membrane fluidity for eNOS function [436,437]. The HDL-associated protein PON-1 prevents lipid and LDL oxidation thereby protecting ECs from oxidative damage, pro-inflammatory signaling, and apoptosis [438,439].  Many disease states, particularly those with an inflammatory component, can affect HDL’s vasoprotective functions (Table 1.1). For example, HDL isolated from CVD patients exhibits reduced ability to phosphorylate eNOS [344,391,440] and has a distinct repertoire of immune cell trafficking proteins [391]. HDL isolated from children with chronic kidney disease exhibits reduced ability to protect against EC activation [406,410]. Acute inflammation, as in the case of periodontal therapy, can also alter the ability of HDL to induce eNOS phosphorylation [404]. In abdominal aortic aneurysm, a quantitative reduction in apoA-I-mediated vasoprotection may result from the decrease in circulating HDL3, a process which itself may partially be due to the sequestering of apoA-I at the site of inflammation in thrombotic aortic tissue [441,442]. Lack of anti-inflammatory function has also been consistently observed in HDL from people with T2DM. For example, HDL from people with T2DM is less effective in inhibiting tumour necrosis factor   39 alpha (TNFα)-induced EC activation [423,425,426] and cytokine release by macrophages [427]. These anti-inflammatory deficits may be due to reduced activation of eNOS signalling by T2DM HDL [423,424]. The anti-oxidative functions of HDL are similarly impaired in T2DM [422,424,426,428].  1.13.3 HDL composition in health and disease HDL can be classified by a variety of schemes including apolipoprotein content, size, surface charge, and density (Figure 1.2) [340]. The distribution of HDL-C among different sizes has been observed to vary with exercise, CVD status, and lipid-lowering medications, however, the direction of the association between HDL subclass and disease outcomes has been controversial. Many recent studies have found that large HDL subclasses appear to be beneficial for cardiovascular health with stronger associations with disease than total HDL-C [376,396,397,443,444]. By contrast, other studies have observed that CAD patients have lower concentrations of small HDL (HDL3) [388] and elevated concentrations of large HDL (HDL2) [398].  Most studies investigating the effect of statins on HDL subclass find an increase in large HDL-C levels [445–451], although the number of HDL particles of any size may not change [452]. Similarly, investigations on the effect of statins on HDL subclass is not as clearly defined as their well-established ability to elevate HDL-C concentrations [453]. Statins have been shown to exert no effect [454] or increase the concentration of HDL2 while decreasing HDL3 concentration [447,455] although contrasting reports indicate an increase in HDL3 levels [456].  Fibrates increase the concentration of plasma HDL3 and decrease the concentration of HDL2 [453] whereas  niacin has the opposite effect on HDL subclass promoting conversion to mature HDL2 particles [453].  Combination treatments of lipid lowering drugs result in no change in HDL subclass compared to monotherapy [455] or an additive effect and improved HDL functionality [457]. Variations in cohort, drug regimen and experimental techniques likely contribute to the varying observations on the effect of medication on HDL subclass distribution. More consistent is the change in HDL subclass distribution in people with T2DM where HDL is smaller overall [458], concentrations of HDL3 are higher [458,459], and concentrations of HDL2 are lower [420,458–460]. HDL size has also been shown to predict T2DM incidence in a cohort of women where small HDL had a positive and large HDL a negative association with incidence [461].   40  The process by which various HDL subclasses form has also come under debate. Traditionally, it was understood that HDL is first secreted from the liver as small, lipid-poor, discoidal HDL and evolves into larger, more spherical forms over time as lipids are added to the particle. However, in a study monitoring HDL secretion with endogenous ectopic labelling, HDL appeared to be secreted from the liver in all of its unique sizes and remained in those size classes for several days before excretion [462,463].  Another measure of interest is the heterogeneity of the HDL proteome. The HDL proteome varies considerably between individuals based on disease, diet, age, and inflammatory status (Table 1.1). For example, PON-1 content or activity on HDL is reduced in patients with CVD [379,389], liver cirrhosis [411,412], acute inflammation [381,402,404], chronic kidney disease [399,405], rheumatoid arthritis [416] and in the elderly [413,414], and is elevated with exercise [464] or a diet rich in olive oil [465]. Conversely, the SAA content on HDL has been found to increase in chronic kidney disease [406,408], aging [413], acute inflammation [402,404], rheumatoid arthritis [416], T2DM [419], and cirrhosis [412]. Importantly, changes to the HDL proteome are sometimes observed in inflammatory or disease states without a change in total plasma HDL-C [404,413], again highlighting the importance of looking beyond HDL-C when considering lipoprotein function in the etiology of disease. Other alterations to the HDL proteome that have been observed in vascular and inflammatory pathologies include reduced or elevated apoA-I, apoA-II, apoC-II and apoE, and elevated complement C3 and apoC-III [390,399,402,404,408,412,413,466]. For example, apoE, apoA-II, and PON-1 levels on HDL are all reduced in young males with T2DM compared to nondiabetic youths [122].   The HDL proteome, and therefore also HDL function, is also subject to change by hypolipidemic agents such as statins, niacin and fibrates. Green et al. reported that CAD-associated changes in HDL3 apolipoprotein profile, including increased apoE and decreased levels of apoF, and phospholipid transfer protein, are reversed by combination therapy of atorvastatin and niacin [467]. Niacin is also shown in a separate study to exhibit a synergistic enhancement of apoA-I in concert with atorvastatin [457]. Fibrates increase plasma apoA-I and, to a greater extend, apoA-II   41 concentrations [468] . Gordon et al. [452] reported that rosuvastatin treatment dramatically increases the levels of α-1-antitrypsin in large HDL which in turn enhances HDL’s anti-inflammatory properties. Additionally,  statins and fibrates can alter the activity of HDL-associated anti-oxidant proteins such as PON-1 to augment its vasoprotective function [469,470].   The HDL lipidome is another field of emerging interest. While most of the work thus far has investigated the lipidome of HDL from healthy subjects, it is becoming clear that changes to the proportions of HDL lipids in disease can have functional consequences. For example, the CEC, anti-oxidant, and anti-inflammatory activities of HDL are impaired with excess triglyceride, cholesterol ester, oxidized lipids, and sphingomyelin content [369]. A well-studied bioactive lipid on HDL is S1P, which is well known to be at least partially responsible for the anti-inflammatory actions of HDL. S1P on HDL can be reduced in CVD [393], T2DM [422] and acute inflammation [402] resulting in impaired signaling to eNOS. In another study in T2DM, S1P was instead found to be elevated on HDL possibly as a compensatory mechanism [421]. Many other changes to the HDL lipidome have been observed including changes to triglyceride, phospholipid, and sphingomyelin content (Table 1.1).  Among other cargo carried on HDL are small non-coding ribonucleic acids (RNAs) including transfer RNA (tRNA)-derived RNA fragments, RNase P-derived RNA fragments, and microRNAs (miRNA) [471]. MiRNAs in particular have emerged as an exciting topic in lipid research for their potential as biomarkers and in therapeutic approaches. HDL has been found to be regulated by and to carry a number of miRNA that vary between individuals according to a number of factors including diet [472], weight loss [473], and CAD [389]. As with changes to the HDL proteome, changes to the miRNA profile of HDL can be observed even when there is no change in total plasma HDL-C [389,473] or apoA-I concentration [472]. Interestingly, HDL-carried miR-223, which can be altered with diet or weight loss [472,473], has even been found to be transferred to ECs [474] and to alter gene expression of intercellular adhesion molecule 1 (ICAM-1) in those cells [475].     42 HDL also carries a number of other nonpolar molecules including fat-soluble vitamins, vitamin binding proteins, carotenoids, steroids and other hormones, potentially serving as a transporter for delivery to other tissues [471]. Polar metabolites have also been found on HDL, some of which correlate with insulin resistance [476]. An additional molecule of interest on HDL is symmetric dimethylarginine, a metabolite that is increased in children with chronic kidney disease and may be partially responsible for impaired vasoprotective actions of HDL in these patients [407].   A further layer of complexity to HDL heterogeneity includes modifications to its protein components including the addition of aldehydes such as acrolein [477], modifications of apoA-I by myeloperoxidase [478], or carbamylation of HDL-associated proteins [405]. Oxidation of HDL lipids are commonly observed including in people with T2DM [424,426,479]. HDL protein modifications are associated with CVD and compromise HDL functions including cholesterol efflux, anti-oxidant properties, and promotion of EC migration and proliferation [405,477,478].   1.14 The relationship between HDL and AD 1.14.1 Mixed genetic evidence on HDL and cardiovascular disease and AD risk. Mendelian randomization aims to determine the causality of a modifiable risk factor on disease risk by measuring how disease risk changes based on randomly distributed genetic variants that affect the risk factor [405]. Although it is well-accepted that high plasma HDL-C concentrations associate with reduced heart disease mortality [480], Mendelian randomization questions the causality of this relationship. Several groups observe that genetic variants associated with HDL-C do not alter CHD, myocardial infarction, or carotid atherosclerosis risk [353,354,481], although one study found that an allele score based on all known genetic variants associated with HDL-C  was significantly associated with CHD risk [353]. Two Mendelian randomization studies also suggest plasma HDL-C concentration is not causal for AD risk [208,209]. Importantly, these studies address only a causal link between disease risk and elevated HDL-C mediated by particular genes; they do not take into account the complex changes to HDL function and composition that can occur in disease and that can be superior predictors of disease risk [376,413,423,424,482–486]. Recently, two large GWAS for AD found lipoprotein metabolism and HDL particle gene sets to be significantly associated with AD risk. Genes in these sets encode HDL biogenesis   43 proteins and HDL protein components such as APOE, ABCA1, APOC1, APOM, APOA2, PON1, CLU, LCAT, CETP and APOA1 [63,64].   1.14.2 Epidemiological evidence for a protective effect of HDL on AD  Several cross-sectional and prospective cohort studies suggest that AD risk is attenuated by the concentration of HDL-C or apoA-I and that these measures also have established associations with lower CVD risk [487]. For example, cross-sectional studies in AD patients and healthy controls have found that serum apoA-I and HDL-C concentrations are significantly lower in AD patients [488,489], serum apoA-I concentrations are inversely correlated with MMSE scores [488,489], and a serum apoA-I cut-off level of 1.5 g/L can distinguish between healthy people and those with AD with 71% sensitivity and 69% specificity [488]. Montañola and colleagues reported that circulating apoA-I concentrations positively correlate with plasma Aβ40 among CAA patients, suggesting a potential role of HDL in removing Aβ from the brain [490]. This role in Aβ clearance is further supported by an inverse correlation of HDL-C concentration and brain amyloid burden as measured by PIB-PET imaging in elderly people with mild dementia [491]. Interestingly, higher HDL-C may also promote cognitive function in people without dementia as positive associations have been found between HDL-C concentration and working memory [492,493], MMSE scores [492], and verbal learning scores [493] in cognitively normal subjects. However, other cross-sectional studies have found no association between HDL-C concentration and cognitive function, for example, in a cohort of 1,100 elderly subjects in the Framingham study [200] and in a small cohort of nonagenarians in Spain [494].   Large prospective cohort studies aiming to determine if circulating HDL-C or apoA-I concentration can predict future AD or cognitive function have also been mixed. The Honolulu-Aging study followed 929 Japanese-American men and found that those in the highest quartile of plasma apoA-I at baseline had the lowest risk of dementia 16 years later [495]. Another prospective cohort study followed 1,130 elderly people in the Manhattan area for a median of 4 years and found that those with the highest baseline HDL-C concentration had a reduced risk of AD [210]. Furthermore, a recent report on the Baltimore Longitudinal Study of Aging found that those with higher baseline HDL-C concentrations were less likely to be cognitively impaired 20 years later   44 and were protected from reductions in entorhinal cortex and parahippocampal gyrus volume [496]. On the other hand, two reports on the Adult Changes in Thought study and two studies in cognitively normal elderly women found that baseline HDL-C concentrations were not related to the risk of AD or cognitive impairment upon follow-up [497–500].   Some have suggested that differences in age at baseline measurement and length of follow-up may explain these inconsistencies in the literature [495,498]. Indeed, the studies discussed above with follow-up time greater than 10 years found significant associations between HDL-C concentration and AD risk [495,496] while some others with less than 10 years of follow-up did not [498,500]. Furthermore, those measuring baseline HDL-C concentration when subjects were on average 50 to 70 years old all found significant associations with AD risk [489,493,495] whereas those with baseline measures when subjects were on average 70 years old or older did not [499,500]. It is therefore possible that HDL has its greatest influence on AD risk at mid-life.  1.14.3 Potential protective mechanisms of action of HDL against AD The mechanism(s) by which HDL may influence AD risk remain unknown and is a major challenge in the field. Many HDL associated proteins such as apoA-I, clusterin (apoJ), apoE, apoC-III, apoD and apoA-IV are present within the brain parenchyma, CSF and cerebrovascular intima of leptomeningeal arteries [112,501–503]. With the exception of apoE, the CSF concentrations of these proteins correlate moderately with their respective concentrations in plasma, suggesting transport or diffusion from the periphery to the brain.  Although Fung and colleagues reported that HDL can be transported through human brain microvascular ECs in vitro via SR-BI [504], and it is known that CSF lipoproteins have a similar density to plasma HDL [505], there is currently no evidence that HDL can enter the brain as an intact lipoprotein particle in vivo. Furthermore, neither cholesterol nor apoE cross the BBB [113]. These observations suggest that HDL might indirectly influence brain health as a circulating factor that primarily acts from the cerebrovascular lumen and intima. This raises the possibility that blood may be a suitable and relatively non-invasive matrix to develop assays of HDL function(s) relevant to AD as better predictors of AD disease risk or progression than apoA-I or HDL-C concentration per se.    45 1.14.4 Vasoprotective functions of HDL in AD animal models   1.14.4.1 Mouse models of AD AD, as defined by the presence of Aβ plaques and neurofibrillary tangles in the brain resulting in cognitive and behavioural impairment, is a disease specific to humans [506]. Although mice naturally produce Aβ and tau, they do not accumulate plaques or tangles in their brains nor do they experience the dementia associated with AD [507]. Therefore, a number of genetically engineered mouse models have been developed in order to allow for AD research in an in vivo system. Mouse models of AD typically overexpress transgenes of mutated human amyloid processing proteins, specifically, APP, the precursor to Aβ, and PSEN1, a subunit of the γ-secretase complex that cleaves APP into Aβ [507–511]. Differences in the transgenes expressed, the specific mutations in the genes, the promoter used to drive expression of the transgene, the number of copies of the transgene expressed, and the wild type background the mice are bred onto all result in variation in the severity and onset of amyloid pathologies. Common AD mouse models include APP/PS1, which have the Swedish APP mutation (huAPP695) and a deletion in exon 9 of PS1 under the mouse prion promoter (PrP), Tg2576, which have the Swedish APP mutation under the hamster PrP, and PDAPP, which have the Indiana APP mutation (huAPP770) under the platelet derived growth factor beta (PDGFRB) promotor [511,512].   That all AD research in mice has relied on the expression of a small number of mutations is a major limitation of our current understanding of the disease, especially as the mutations used are rare and found only in people with the familial, early-onset form of AD and not the much more common late-onset form of AD [507,509,510,512]. AD in humans is a heterogeneous disease involving a complex interaction of genetics and lifestyle over a human life-span, factors which are not feasible to model in mice. Furthermore, the introduction of transgenes into mice has the potential to disrupt off-target genes unless it is introduced by gene targeting [508,512]. Such disruptions have been recently shown in transgenic rTg4510 tau mice [513] and compromise the validity of the observations made in this mouse model to human disease. Saito et al. aimed to avoid these issues with their  APP knock-in mouse model that expresses the human APP gene at wildtype levels but maintains overexpression of Aβ in order to eliminate artifacts resulting from disruption of off-target genes and overproduction of APP and APP fragments [514]. An additional limitation   46 to the validity of AD mice is the lack of tau pathology arising in amyloid-based models [510] despite observations in humans suggesting that tau pathology is a better predictor of cognitive decline than amyloid pathology [52,515]. Mouse models with tau mutations alone or in combination with amyloid mutations have been created that result in the formation of tau tangles, however, these tau mutations are not found in humans with AD [510,511].    Nonetheless, mice are a key model organism for research and many discoveries on brain function and AD have origins in mouse studies.  Recently, attempts have been made to increase the validity of AD mouse models by creating new mouse models combining various human mutations and gene isoforms [510]. Alternatively, efforts have been made to increase the variability of the genetic background of the mice by expressing the transgenes in wild mice rather than highly inbred mice as has been done in the past [512]. The development of non-human primate models injected with human AD brain homogenates [510] and transgenic minipig models [516] have been ongoing and some interest in studying spontaneous cognitive impairment in age canines exists [517]. Compared to mice, these animals have a more similar neuroanatomy to humans but cost, time, and ethical concerns remain important issues limiting their use [518].   1.14.4.2 Genetic alteration of HDL in AD animal models Studies in AD transgenic mice have been used to explore the impact of altering plasma HDL concentration on AD-relevant pathological and behavioral outcomes. Genetic ablation of apoA-I robustly reduces plasma HDL-C concentration, worsens memory deficits and increases CAA in 12-month old APP/PS1 mice [519], without altering parenchymal Aβ plaque load [519,520]. Transgenic overexpression of apoA-I from its native promoter in 12-month old APP/PS1 mice results in the opposite phenotype, including attenuation of memory deficits, selective reduction of CAA, and reduced neuroinflammation [521]. On the other hand, a recent study observed improvements in total and vascular amyloid burden in apoA-I knockout mice crossed to the Tg2576 mice compared to Tg2576 mice with normal concentrations of plasma apoA-I [522]. Differences between the Aβ transgenic mouse lines used and the methodology for the analysis of amyloid burden may explain these contradictory results, as will be discussed more thoroughly in section 2.5.   47  1.14.4.3 HDL-based therapeutics on AD-relevant outcomes in animal models  Although there are no HDL-based therapeutic strategies that have specifically been tested for AD in clinical trials, several preclinical studies have been performed in APP/PS1 or other AD model mice. For example, our group showed that raising plasma apoA-I concentration through intravenous administration of reconstituted pre-β HDL composed of human plasma-derived apoA-I and phospholipids acutely reduced soluble brain Aβ concentrations in APP/PS1 mice [523]. Another reconstituted HDL formulation based on apoE3 also improved AD pathology in SAMP8 mice by reducing Aβ deposition, microgliosis, and memory deficits when injected in the tail vein [524]. APP23 mice treated intravenously with recombinant apoA-I Milano similarly showed reduced microgliosis and Aβ deposition with a specific reduction in vascular Aβ [525]. Recently, APP23 mice treated intravenously with free apoJ or apoJ lipidated to form a reconstituted (rHDL) were shown to have reduced insoluble Aβ and CAA burden [526]. Finally, the orally administered apoA-I mimetic D-4F improves memory, Aβ deposition, microgliosis, astrogliosis, and other markers of inflammation in APPswe/PS1ΔE9 mice [527]. The effects of HDL on cerebrovascular health have also been studied in mice outside of the context of AD. For example, administration of D-4F after middle cerebral artery occlusion in mice reduced neuroinflammation and white matter damage while improving neurological functional outcomes [528]. Similarly, D-4F treated atherosclerotic mice lacking LDLR and fed a western diet show improved cognition and reduced brain arteriole inflammation as measured by arteriole association with macrophages, microglia, or inflammatory proteins [529].   1.15 Liver X receptor and its therapeutic potential for AD Liver X receptors (LXR) are nuclear receptors and master regulators of cholesterol metabolism and inflammation [530,531]. LXRs are found in two isoforms, LXRα is highly expressed in adipose tissue, adrenal tissue, intestine, kidney, liver, lung and spleen while LXRβ is more ubiquitously expressed [532]. LXRs form heterodimers with retinoid X receptors (RXR) and when activated by oxysterols, their endogenous agonist, or synthetic agonists, they transcribe the genes in the bound LXR response element [533,534]. Oxysterols are cholesterol metabolites therefore their activation of LXR signalling acts as a feedback loop on cholesterol metabolism [533].    48  LXR signalling pathways were first suggested to be protective against AD by two groups in 2003. These groups found that Aβ production or secretion was reduced when rat primary neuronal and non-neuronal cells [535] or mouse neuroblastoma cells [536] with the Swedish APP mutation were treated with natural or synthetic LXR agonists. Soon after, these results were translated into AD mouse models where treatment with the synthetic LXR agonists GW3965 or TO901317 reduced Aβ concentrations [537–539]  and prevented memory deficits [537,538]. Since then many studies have treated various AD mouse models at various ages with LXR agonists and observed a number of beneficial effects (summarized in Table 1.2 and Table 1.3). While LXR agonist treatment consistently improves memory in AD model mice across studies [537,538,540–546], the effect on Aβ and amyloid load has been more controversial. Some studies found reductions to soluble Aβ [537–539,541,546] while others found no difference [542,543,545,547] and others found an increase in certain brain regions [540]. Similarly, LXR agonist treatment has been found to reduce amyloid plaque load in some studies [537,540–542,546] while no difference was found in others [543–545,547,548]. Interestingly, Donkin et al. reported that the beneficial effects of LXR agonist treatment on memory and amyloid load is dependent on the presence of ABCA1, a well-established LXR target gene [540]. ABCA1 is a cell lipid transporter functioning to maintain lipid homeostasis within cells by effluxing cholesterol and phospholipids across the cell membrane to apolipoprotein acceptors such as apoA-I, to form HDL, and apoE [549]. LXR agonists have also been shown consistently to reduce neuroinflammation in AD model mice [537,542,543,546,550].    49 Table 1.2 Summary of studies treating AD model mice with GW3965 investigating amyloid, neuroinflammation, and memory deficits. Reference Dose (mg/kg/day) Route Age at treatment onset (mo) Duration AD model Sex Aβ and amyloid Neuroinflammation Memory Jiang (2008) [537] 33 (120*) Chow (oral gavage*) 12 4mo (6d*) Tg2576 -- ¯sol Αβ (hp) ¯plaques (hp) ¯CD45+ number, area ¯IL-6 mRNA ­contextual fear memory Donkin (2010) [540] 33 Chow 10 2mo APP/PS1 F ­sol Aβ (hp, cx) trend ¯ plaques (hp) --  ­novel object recognition, spatial memory Wesson (2011) [541] 33 Oral gavage 14-16 14d Tg2576 Both ¯sol and insol Aβ ¯plaques in ob, pcx  ­odour habituation behaviour Skerrett (2015) [542]  50 Oral gavage 6 9d APP/PS1 M No change in Aβ ¯ plaque (hp) ¯Iba1 mRNA, area ¯CD45, IL-6, IL-1β mRNA ­contextual fear memory Sandoval-Hernández (2015) [543]  33 Oral gavage 12 12w 3xTg-AD Both No change in Aβ, plaques ¯GFAP area ­spatial memory Sandoval-Hernández (2016a) [544]  50 Oral gavage 12 6d 3xTg-AD F No change in plaques -- ­spatial memory Sandoval-Hernández (2016b) [550] 50 Oral gavage 24 6d 3xTg-AD -- ¯perivascular Aβ ¯GFAP area -- M: male, F: female, mo: month, w: week, d: day, hp: hippocampus, cx: cortex, pcx: piriform cortex, GFAP: glial fibrillary acidic protein, CD45: cluster of differentiation 45, IL-6: interleukin 6, IL-1β: interleukin 1 beta, Iba1: ionized calcium binding adaptor molecule 1. * Sub-study with oral gavage. -- data not reported.  50 Table 1.3 Summary of studies treating AD model mice with TO901317 investigating amyloid, neuroinflammation, and memory deficits. Reference Dose (mg/kg/day) Route Age at treatment onset (mo) Duration AD model Sex Aβ and amyloid Neuroinflammation Memory Koldamova (2005) [539] 50 Gastric gavage 2 6d APP23 -- ¯soluble Aβ -- -- Riddel (2007) [538] 50 Oral gavage 4 6-7d Tg2576 M ¯Aβ42 (hp), no change Aβ (cx) -- ­contextual fear memory Lefterov (2007) [551] 50 Gastric gavage  6 25d APP23 -- ¯ insoluble Aβ ¯IL-1β, IL-6, TNFα mRNA -- Vanmierlo (2011) [545] 30 Chow 21 6-9w APPSLxPS1mut M No change in Aβ, plaque load -- ­novel object recognition, spatial memory Jansen (2012) [548] 30 Chow 12 3mo AβPP-PS1 M No change in plaque -- -- Cui (2012) [546] 30 Oral gavage 6 30d APP/PS1 M ¯Aβ42, plaque load  ¯GFAP, CD11b area  ¯COX-2, iNOS, p65 levels ­spatial memory Stachel (2016) [547] 10 Subcutaneous 12 3w Tg2576 -- No change in soluble Aβ   No change in locomotor activity M: male, mo: month, w: week, d: day, ISF: interstitial fluid, hp: hippocampus, cx: cortex, GFAP: glial fibrillary acidic protein, CD11b: cluster of differentiation 11b, COX-2: cyclooxygenase 2, iNOS: inducible nitric oxide synthase. -- data not reported.  51 Unfortunately, LXR agonists are unlikely to be a viable therapeutic option for treating human AD patients due to off-target effects resulting in hepatoxicity [552,553]. Clinical trials in human subjects have occurred testing bexarotene, an RXR agonist, for AD [554,555]. Some preliminary studies with bexarotene showed that treatment reduced amyloid burden and improved cognition in mouse models [556–560], although other rodent studies found no benefits [561–563]. Human clinical studies began with a promising case study showing improved memory and reduced CSF tau concentration in a single patient [554], however, a phase I clinical trial found no effect on Aβ metabolism and poor CNS penetrance [555].  1.16 Summary, research hypothesis, and specific objectives 1.16.1 Summary AD is a devastating, neurodegenerative disease characterized neuropathologically by amyloid plaques and NFTs and clinically by a decline in memory and executive function. Vascular health is also tied to brain health and AD. Vascular risk factors make up many of the modifiable risk factors that together may be responsible for 35% of dementia cases. The cerebrovasculature system is crucial for proper brain functioning as it is responsible for providing oxygen and nutrients to energy-hungry neurons and clearing brain waste, including a portion of the Aβ produced in the brain. Dysfunction in this system is well documented in AD and may even precede amyloid pathologies and clinical symptoms. Given the frequent failure of amyloid and tau-based therapeutics in clinical trials for AD, a therapeutic strategy instead targeting the cerebrovasculature is attractive. HDL is well documented to exert beneficial functions on peripheral arteries in healthy people, for example by acting to promote endothelial migration, inducing endothelial nitric oxide production, preventing oxidation, and preventing inflammation. Previous work in mouse models and human epidemiological research suggests that HDL may also protect against AD.   1.16.2 Research hypothesis HDL protects against AD by promoting cerebrovascular health acting from the lumen of brain blood vessels.    52 1.16.3 Specific objectives 1. Investigate amyloid pathology, neuroinflammation, cerebrovascular inflammation, and cognitive deficits in an AD mouse model (APP/PS1) deficient for apoA-I. 2. Investigate amyloid pathology, neuroinflammation, cerebrovascular inflammation, and cognitive deficits in an AD mouse model (APP/PS1) treated with a non-brain penetrant LXR agonist to selectively act on peripheral lipid and lipoprotein metabolism. 3. Investigate the vasoprotective functions of human HDL against Aβ on brain-derived ECs and in 3D bioengineered arteries. 4. Develop assays of AD-relevant HDL vasoprotective functions to determine whether the protective functions of HDL against AD are lost in disease.     53 Chapter 2: Amyloid pathology, neuroinflammation, cerebrovascular pathology, and cognition in apoA-I deficient, APP/PS1 mice   2.1  Summary Alzheimer’s disease (AD) is defined by amyloid beta (Aβ) plaques and neurofibrillary tangles and characterized by neurodegeneration and memory loss. Most AD patients also have Aβ deposition in cerebral vessels known as cerebral amyloid angiopathy (CAA), microhemorrhages, and vascular co-morbidities, suggesting cerebrovascular dysfunction contributes to AD etiology. Promoting cerebrovascular resilience may therefore be a promising therapeutic or preventative strategy for AD. Plasma high-density lipoproteins (HDL) have several vasoprotective functions, are associated with reduced AD risk in some epidemiological studies, and improve pathology and memory in AD model mice. Here we investigate the interaction of HDL, amyloid, and inflammation on the cerebrovasculature in AD model mice by breeding APP/PS1 mice with apolipoprotein (apo)A-I-deficient mice and aging up to 12 months. We found that APP/PS1 mice without apoA-I had worse amyloid pathology and neuroinflammation both globally and specifically on cerebral vessels. Specifically, apoA-I-deficient mice had increased parenchymal amyloid in the cortex and increased total brain levels of intercellular adhesion molecule 1 (ICAM-1) and of the reactive astrocyte marker glial fibrillary acidic protein (GFAP).  Additionally, apoA-I-deficient mice had significantly elevated cortical vascular amyloid as well as cortical and hippocampal ICAM-1 and GFAP specifically associated with endothelial cells (ECs). Loss of apoA-I increased the reactivity of astrocytes to parenchymal and vascular amyloid beta. We also confirmed previous observations of memory impairments in APP/PS1 mice lacking apoA-I with cued and contextual fear conditioning tests. In summary, apoA-I-containing HDL can reduce amyloid pathology, cerebrovascular inflammation, and astrocyte reactivity to parenchymal and vascular amyloid in Aβ expressing mice.      54 2.2 Introduction Cerebrovascular dysfunction is a common [5,225,234] and early event [6,235] in AD pathogenesis, therefore, methods to protect cerebrovascular health may present valuable therapeutic or preventative opportunities for AD. HDL is one such therapeutic target of interest as extensive research in the cardiovascular field has demonstrated that plasma HDL in healthy people has numerous vasoprotective functions. Furthermore, higher levels of plasma HDL-C, especially when measured at middle age, have been shown to be associated with reduced AD risk and memory impairments in later life [210,488–491,493,495,496] and GWAS studies find associations between HDL-related genes and AD risk [63,64].   Many researchers have turned to AD mouse models to show how plasma HDL may protect against AD pathologies by treating mice intravenously with HDL-based therapeutics. For example, intravenous administration of reconstituted apoA-HDL acutely reduces soluble brain Aβ concentrations [523], reconstituted apoJ-HDL reduces Aβ deposition and CAA burden [526], reconstituted apoE3-HDL reduces Aβ deposition, CAA, neuroinflammation, and memory deficits [524], and recombinant apoA-I Milano reduces parenchymal and vascular Aβ deposition, microgliosis, and memory deficits [525]. Similarly, oral administration of the apoA-I mimetic peptide D-4F reduces Aβ deposition and neuroinflammation [527].   Other researchers instead genetically altered the expression of endogenous murine apoA-I, the main protein component of HDL, to study the effect of plasma HDL in AD mice. Although apoA-I deficiency does not affect total amyloid burden in PDAPP [520] or APP/PS1 mice [156,519], a specific worsening of CAA has been observed [519]. Conversely, overexpression of apoA-I has been shown to reduce CAA but have no effect on total amyloid burden [521]. APP/PS1 mice overexpressing apoA-I also have also been shown to have reduced neuroinflammation and improved memory [521] while apoA-I deficient APP/PS1 mice have worsened memory deficits [519].   Although there is a wealth of evidence from the studies above suggesting that HDL can protect Aβ expressing mice from amyloid pathology and neuroinflammation, a recent study has found the   55 opposite. Contu et al. reported that apoA-I-deficient Tg2576 mice have less Aβ plaque load and CAA at 12 months of age compared to Tg2576 mice with normal apoA-I expression levels [522]. Therefore, the effect of apoA-I expression in Aβ expressing mice is not entirely clear. Furthermore, the specific effects of apoA-I on the cerebrovasculature in Αβ expressing mice have not been studied outside of CAA despite a wealth of research demonstrating vasoprotective functions of HDL in peripheral arteries. In the present study we aimed to fill these gaps in the research. We used APP/PS1 mice deficient or hemizygous for apoA-I to extend the previous in vivo studies, confirm a clear role for apoA-I on cortical CAA and on cortical and hippocampal inflammation, and report a novel interaction among apoA-I, astrogliosis and vascular and parenchymal amyloid.   2.3 Methods 2.3.1 Animals All procedures involving animals were approved by the Canadian Council of Animal Care and the University of British Columbia Committee on Animal Care. APP/PS1 mice (Jackson laboratories, B6.Cg-Tg(APPswe,PSEN1dE9)85Dbo/Mmjax, MMRRC stock no: 34832-JAX ) on a C57Bl/6 background co-express two transgenes from the murine prion promoter: a chimeric mouse/human amyloid precursor protein (APP) cDNA containing the Swedish (K670M/N671L) mutations, and the human presenilin-1 (PS1) gene deleted for exon 9 [564]. APP/PS1 mice were first bred with apoA-I deficient mice (Jackson Laboratories, B6.129P2-Apoa1tm1Unc/J, Stock no: 002055, also on a C57Bl/6 background) to produce an F1 generation hemizygous for both apoa1 and the APP/PS1 transgenes. These animals where then backcrossed to apoA-I knockouts to produce F2 male and female mice of four genotypes: APP/PS1 mice hemizygous (HEM) for apoa1 (APP/PS1 apoA-IHEM), APP/PS1 mice with complete apoa1 knockout (KO) (APP/PS1 apoA-IKO), nontransgenic littermates (wildtype, WT) hemizygous for apoa1 (WT apoA-IHEM), and nontransgenic littermates with apoa1 knockout (WT apoA-IKO).   Although APP/PS1 mice on a pure C57Bl/6 background have an increased seizure risk, we selected this strain to maintain genetic homogeneity throughout this breeding strategy. Overall, 80% of the animals survived until the end of the experiment, with specific survival rates of 66% for APP/PS1 mice, 63% for APP/PS1 apoA-IHEM mice, and 74% for APP/PS1 apoA-IKO mice (Figure 2.1).   56 While this breeding strategy was successful in generating N=6-8 mice per group, insufficient animals survived to yield sex-specific cohorts. We therefore used mixed sexes, which is an acknowledged limitation of this study, yet tracking of individual animals throughout the study showed no clear sex bias throughout the results and justifies the use of pooled sexes.   Figure 2.1 APP/PS1 survival rates.  Survival of (a) female and (b) male mice of each genotype was monitored through aging to the experiment endpoint. Only animals with known genotype upon death or experiment endpoint are included. Exact p-values for pair-wise comparisons by log-rank test and p-values after Bonferroni correction for multiple comparisons are shown below the graphs. apoA-I: apolipoprotein A-I, HEM: hemizygous apoA-I genotype, KO: knockout apoA-I genotype, WT: wildtype littermates for APP/PS1, APP/PS1: APP/PS1 transgenic.  2.3.2 Tissue collection Mice were fasted for 4 h prior to anesthetization by intraperitoneal injection of 20 mg/kg xylazine (Bayer) and 150 mg/kg ketamine (Bimeda-MTC). Blood was collected by cardiac puncture in ethylenediaminetetraacetic acid (EDTA)-containing syringes, centrifuged at 21,000 g for 10 min at 4°C, and the resulting plasma was stored at -80°C until use. Mice were then perfused for 6 min with ice-cold phosphate buffered saline (PBS) containing 2500 U/L heparin at 6 mL/min. Brains were excised and bisected in the sagittal plane. A 2x5 mm piece of parietal cerebral cortex was removed and stored separately for messenger ribonucleic acid (mRNA) analysis. The piece of brain for mRNA analysis and the remaining half-brain were snap-frozen on dry ice and stored at -80°C until use. The remaining half-brain was fixed in 4% paraformaldehyde (PFA) for 2 days at 4°C followed by storage in PBS containing 30% sucrose and 0.1% sodium azide at 4°C.    57 2.3.3 Plasma lipid measurements Plasma HDL-cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and total cholesterol concentrations were measured using Wako kits (Category no: 997-01301 for HDL-C, 993-00404 and 999-00504 for LDL-C, and 999-02601 for total cholesterol) as per the manufacturer’s directions adapted for 96-well microplates. Plasma specimens were randomized and the researcher performing the measurements was blinded to specimen genotype. Plasma lipids were analyzed in N=6-7 mice per genotype.  2.3.4 Histology Immunofluorescence triple co-staining was performed to visualize ECs (cluster of differentiation 31, CD31), Aβ plaques (1,4-Bis(3-carboxy-4-hydroxyphenylethenyl)benzene, X-34), and activated astrocytes (GFAP) as follows. Cryoprotected half brains were sectioned at 40 μm on a Leica CM3050 Research Cryostat and stored in PBS containing 0.1% sodium azide until staining. Three sections from each half-brain were selected from the anterior to posterior hippocampus at 200 μm intervals and mounted on Superfrost Plus slides. Antigen retrieval was performed with citrate buffer (10 mM citric acid, 0.05% Tween 20, pH 6.0) at 95°C for 5 min followed by washing with PBS, tissue permeabilization with 0.25% Triton X-100 in PBS for 30 min and blocking with 5% donkey serum and 1% bovine serum albumin (BSA) and 0.3% Triton X-100 in PBS for 60 min. Sections were incubated in primary antibodies for CD31 (Abcam, ab28364, 1:200 dilution), GFAP Alexa Fluor 488 (eBioscience, 53-9892-80 1:400 dilution), or ICAM-1 (R&D, AF796, 1:50 dilution) in 5% donkey serum, 1% BSA and 0.3% Triton X-100 in PBS overnight at 4°C. After washing with PBS, sections were incubated with secondary antibodies for 60 min at room temperature. Secondary antibodies were as follows: goat anti-rabbit Alexa Fluor 594 antibody (Life Tech, A11012, 1:600 dilution) to detect CD31 on its own and when co-stained with GFAP, goat anti-rabbit Alexa Fluor 647 antibody (Life Tech, A27040, 1:400 dilution) to detect CD31 co-stained with ICAM-1, and donkey anti-goat Alexa Fluor 594 antibody (Life Tech, A-11058, 1:400 dilution) to detect ICAM-1. Sections were then washed with PBS. Amyloid staining was subsequently performed with X-34 (Sigma-Aldrich, SML 1954, 2 μM in PBS with 0.5% Tween-20) for 20-30 min followed by washing with 40% ethanol in PBS then PBS alone. Coverslips were mounted onto slides in prolong antifade (Invitrogen, P36970) and slides were stored at 4 °C until   58 imaging with an Axio Scan.Z1 (Zeiss). Brain sections were randomized and the researcher performing the staining was blinded to mouse genotype. The cellular localization of ICAM-1 expression was visualized using an LSM 880 confocal laser scanning microscope (Zeiss).  Sections incubated with secondary antibody only were used as negative controls to rule out unspecific antibody staining. Sections from wildtype mice stained with X-34 were used as negative controls to rule out unspecific X-34 staining.  2.3.5 Image analysis Image analysis was performed with ImageJ (NIH) by a researcher blinded to mouse genotype as illustrated in Figure 2.9. Exported images for each channel were converted to 8-bit black and white images. The threshold for positive and negative pixels was manually determined for X-34 and GFAP images using histogram-based segmentation then applied to all of the images of that channel. Due to regional variation in CD31 staining, auto-local thresholding using the Bernsen method was performed on CD31 images. Cortical and hippocampal regions were manually selected and saved as “regions of interest” (ROIs) for each section.  Total CD31, X-34, and GFAP positive area was measured in the threshold segmented cortical and hippocampal ROIs and normalized to the total ROI area. Vascular GFAP was quantified as the GFAP positive area associated with CD31 in each region whereby a mask of the segmented CD31 image for a section was created then applied to the segmented GFAP image of the same section. The GFAP positive area within the CD31 mask was measured then normalized to the total CD31 positive area within the ROI. Plaque-associated GFAP was similarly quantified as the GFAP positive area associated with X-34 in each region whereby a mask of the segmented X-34 image for a section was created then applied to the segmented GFAP image of the same section. The GFAP positive area within the X-34 mask was measured then normalized to the total X-34 positive area within the ROI.  Total and vascular ICAM-1 quantification was performed using the same techniques as described above to measure total and vascular GFAP. Parenchymal ICAM-1 positive area was calculated as the difference between total and vascular ICAM-1 positive area in the cortical and hippocampal regions.    59 CAA was quantified using methods similar to that published in Nature Protocols by Wilcock et al. in 2006 [565], which has been used by many others [521,566,567] including in the study by Lefterov et al. using apoA-I deficient APP/PS1 mice [519]. A mask containing areas of vascular amyloid was created by manually discriminating vascular amyloid from parenchymal plaques in segmented X-34 images based on morphology, as illustrated in Figure 2.10. Identification of vascular versus parenchymal amyloid was performed on inverted images to ease visualization. These vascular amyloid masks were then applied back onto segmented X-34 images and the percent positive area was measured and normalized to total area of the cortical ROI to give percent cortical CAA area. The X-34 positive area within the vascular amyloid mask was then used to create a new CAA mask. This CAA mask was then applied to a mask of vascular GFAP created as above. The vascular GFAP positive area within the CAA mask was measured and normalized to CAA area to give the percent area of CAA associated with vascular GFAP.     The above analysis steps, excluding the drawing of cortical and hippocampal ROIs and vascular amyloid areas, were repeated for each section using an automated ImageJ macro. Figure 2.9 shows schematics illustrating the image analysis process. Analysis of vessel diameter and tortuosity was performed with Vesselucida 360 software (MBF Bioscience) using CD31 as the vessel marker. For amyloid staining, N=4-5 animals per genotype were used, for CD31, GFAP, and ICAM-1 staining, N=5-6 animals per genotype were used.  2.3.6 Protein extraction Protein from half-brains was serially extracted to produce soluble and insoluble protein fractions. Half-brains were first homogenized with a Tissuemite homogenizer for 20 s in carbonate buffer (10 mM carbonate, 50 mM sodium chloride, cOmplete protease inhibitor tablet, pH 11.5) then sonicated. Homogenates were cleared by centrifugation at 12,500 g for 45 min at 4°C. The resulting supernatant was neutralized with 1.5 volumes Tris (1 M Tris, pH 6.8) and labelled as the soluble fraction. The pellet was resuspended in 1 mL of guanidine hydrochloride (5 M guanidine hydrochloride in 1 M Tris, pH 8.0) by pipetting followed by rotation overnight at room temperature and labeled as the insoluble fraction. Soluble and insoluble extracts were stored at -80°C until use. Lysate protein concentrations were determined using the DC Protein assay kit (BioRad).    60  2.3.7 Enzyme linked immunosorbent assay (ELISA) Human Aβ40 (KHB-3482, ThermoFisher, 1:20, 1:2500), human Aβ42 (KHB-3442, ThermoFisher, 1:40, 1:2500), murine apoE (3752-1HP-2, ThermoFisher, 1:20, 1:200), and murine apoJ (ab199079, Abcam, 1:200, 1:200) concentrations were measured in carbonate soluble and guanidine hydrochloride insoluble half-brain lysates using commercial sandwich ELISA kits according to manufacturer’s instructions. Murine GFAP (NS830, Millipore, 1:400), intercellular adhesion molecule 1 (ICAM-1) (ab100688, Abcam, 1:2), vascular cell adhesion molecule (VCAM-1) (ab100750, Abcam, 1:2000), platelet derived growth factor receptor beta (PDGFRβ) (MBS919047, MyBioSource, 1:10), interleukin 1 beta (IL-1β) (K15245D, MesoScale Discovery, 1:2), and tau (K15121D, MesoScale Discovery, 1:50) concentrations were measured in carbonate soluble half-brain lysates using commercial sandwich or MesoScale Discovery ELISA kits according to the manufacturer’s instructions. Data points were interpolated from a standard curve using four parameter nonlinear regression curve fitting and normalized to total soluble protein concentration. Brain lysates were randomized and the researcher performing the ELISAs was blinded to mouse genotype. For apoE ELISA, N=6 mice per genotype were used. For Aβ ELISA, N=5-7 mice per genotype were used. For IL-1β, ICAM-1, VCAM-1, and PDGFRβ ELISA, N=5-19 mice per genotype were used. For GFAP ELISA, N=6-14 mice per genotype were used.  2.3.8 Isolation of RNA and real-time quantitative polymerase chain reaction (RT-qPCR) RNA was extracted from parietal cerebral cortices using Trizol (Invitrogen) and treated with DNaseI (Life Technologies) prior to complementary deoxyribonucleic acid (cDNA) synthesis. cDNA was generated using the Taqman Reverse Transcription Kit (ThermoFisher, N8080234). RT-qPCR was performed using the Light Cycler 96 Real-Time PCR System (Roche) and FastStart Essential DNA Green Master Mix (Roche). Expression of murine Il1b (Fwd: TTGACGGACCCCAAAAGA; Rev: CAGCTTCTCCACAGAGCCACA) was normalized to murine β-actin (Fwd: ACGGCCAGGTCATCACTATTG; Rev: CAAGAAGGAAGGCTGGAAAAG). RNA homogenates were randomized and the researcher performing the RT-qPCR was blinded to mouse genotype. For Apoe and Il1b mRNA analyses, N=7-12 mice per genotype were used.   61  2.3.9 Contextual and cued fear conditioning    2.3.9.1 Training During training, mice were placed in an illuminated compartment of a shuttle box (Med Associates Inc, St. Albans, VT) and allowed to habituate to the internal testing environment for 120 s. After the habituation period, an 80-dB auditory cue was played for 30 s. During the last 2 s of the auditory cue, a mild foot shock (0.3 mA) was administered. Shocked mice were left undisturbed in the testing chamber for an inter-trial interval of 60 s, after which mice were presented with a second identical tone, which was called the shock trial. After the second tone, shocked mice were again left undisturbed in the testing chamber for 60 s.  2.3.9.2 Context testing Mice were tested for their ability to remember the context in which they received the foot shock 24 h after training. Mice were placed in the same illuminated compartment and observed for the presence/absence of a freezing response over 5 min. During this time mice were not exposed to the tone or shock and the percent freezing time was measured using the ANY-Maze system (Stoelting Co., Wood Dale, IL). Contextual testing included N=7-23 mice per genotype.  2.3.9.3 Cued testing Mice were tested for their ability to remember the 80-dB auditory cue presented during the training session 1 h following context testing. Testing was conducted in an environment different from the training and context testing session (darkened compartment, false floor placed over the steel rods of the shuttle box, different cleaning solution). Mice were placed in the darkened compartment and allowed to habituate for 120 s, followed by the 80-dB auditory cue for 120 s. After the tone, mice were left undisturbed in the testing chamber for 60 s and percent freezing time was measured using the ANY-Maze system. Cued testing included N=7-23 mice per genotype.  2.3.10 Statistical analysis All statistical analyses, with the exception of fear memory analysis, were performed with GraphPad Prism 7, p-values <0.05 were considered statistically significant. For the analysis of all   62 non-amyloid data, statistical comparisons were made by two-way analysis of variation (ANOVA) considering APP/PS1 genotype as one factor and apoA-I genotype as a second factor (omnibus analysis) followed by Sidak’s multiple comparisons test if significant factor or interaction effects were observed. Analysis of Aβ concentration, amyloid plaque area, and CAA were performed with unpaired t-test or Mann Whitney test for parametric and non-parametric data respectively. Analysis of fear memory was performed with IBM SPSS Statistics Build 1.0.0.1347 using three-way ANOVA to analyze freezing time as the dependent variable and time, APP/PS1 genotype, and apoA-I genotype as fixed variables. Pairwise comparisons of estimated marginal means and standard error were performed and adjusted by Sidak’s multiple comparisons test, adjusted p-values <0.05 were considered statistically significant.  2.4 Results 2.4.1 Loss of apoA-I significantly reduces plasma total cholesterol and HDL cholesterol concentrations By omnibus two-way ANOVA, apoA-I deficiency significantly reduced plasma total cholesterol (p=0.013), HDL-C (p<0.0001), and LDL-C concentrations (p=0.048) (Figure 2.2a-c) relative to apoA-IHEM mice, as previously reported [115,568]. HDL-C was the cholesterol pool most affected by the complete loss of apoA-I with reductions from 48.3 mg/dL to 14.6 mg/dL (p=0.011) in wildtype (WT) apoA-IKO vs. WT apoA-IHEM mice and from 50.1 mg/dL to 8.3 mg/dL (p=0.003) in APP/PS1 apoA-IKO vs. APP/PS1 apoA-IHEM mice. Plasma cholesterol pools were not affected by APP/PS1 genotype. ApoA-I genotype did not affect soluble (Figure 2.2d) or insoluble (Figure 2.2e) brain apoE protein concentration or cortical Apoe mRNA expression (Figure 2.2f), although insoluble brain apoE protein concentrations were significantly higher in APP/PS1 mice compared to WT controls overall (p<0.0001 by omnibus two-way ANOVA) (Figure 2.2e). Soluble (Figure 2.2g) and insoluble (Figure 2.2h) apoJ protein concentrations in the brain were increased in APP/PS1 mice compared to WT controls overall (p=0.018 and p<0.0001 respectively by omnibus two-way ANOVA). Insoluble apoJ protein was also increased overall in apoA-IKO compared to apoA-IHEM (p=0.015 by omnibus two-way ANOVA).    63  Figure 2.2 Plasma lipid and brain apolipoprotein concentrations in apoA-I deficient APP/PS1 mice.  Plasma levels of (a) total cholesterol, (b) HDL-C, and (c) LDL-C were measured with commercially available kits. ApoE protein in (d) soluble and (e) insoluble half brain homogenates was measured by ELISA. (f) Apoe mRNA expression was measured in the parietal cerebral cortex by RT-qPCR. ApoJ protein in (g) soluble and (h) insoluble half brain homogenates was measured by ELISA. Points represent individual mice and bars represent mean values. Circles represent female mice and squares represent male mice. Omnibus analyses of apoA-I genotype effects (across APP/PS1 genotype) and APP/PS1 genotype effects (across apoA-I genotype) by two-way ANOVA are displayed as exact p-values below graphs. Sidak’s multiple comparisons test results are displayed within graphs as *p<0.05, **p<0.01, and ***p<0.0001. For plasma lipid and brain protein analysis N=6-7 mice per genotype, for mRNA analysis N=7-17 mice per genotype were used. apo: apolipoprotein, TC: total cholesterol, HDL-C: high-density lipoprotein cholesterol, LDL-C: low-density lipoprotein cholesterol, HEM: hemizygous apoA-I genotype, KO: knockout apoA-I genotype, WT: wildtype littermate for APP/PS1, APP/PS1: APP/PS1 transgenic.  2.4.2 ApoA-I deficiency increases cortical parenchymal and vascular amyloid burden of APP/PS1 mice Histological examination of X-34-stained sections showed that apoA-I deficiency led to significantly greater amyloid burden in the cortex, an increase from a mean of 0.28% to 0.91%   64 (Figure 2.3a-b) (p=0.001 by unpaired t-test), but not in the hippocampus (Figure 2.3c-d) of APP/PS1 mice. Consistent with a previous study [519], we also observed significantly increased cortical CAA from a mean of 0.01% to 0.05% cortical area in APP/PS1 apoA-IKO compared to APP/PS1 apoA-IHEM mice (p=0.016 by Mann-Whitney test) (Figure 2.3e-f). CAA was not observed in the hippocampus of any animal. These observations suggest that apoA-I protects against Aβ deposition in the walls of cortical arteries, which may in turn reduce parenchymal amyloid deposition in the surrounding cortex.   Figure 2.3 Amyloid plaques and vascular Aβ deposition in apoA-I deficient APP/PS1 mice.     65 Total (a) cortical and (c) hippocampal area occupied by amyloid deposits and (e) CAA area were evaluated by staining fixed cryosections with X-34. (b,d,f) Representative images are below graphs. (g) Soluble Aβ40, (h) soluble Aβ42, (i) insoluble Aβ40, and (j) insoluble Aβ42 protein concentrations were measured in half-brain homogenates by ELISA, values were normalized to total protein concentration in the homogenates. (k) Total tau and (l) phosphorylated tau proteins were measured by ELISA and used to calculate the (m) phosphorylated tau/total tau ratio, values were normalized to total protein concentration in the homogenates. All analytes were normalized to homogenate total protein concentration. Points represent individual mice and bars represent mean values. Circles represent female mice and squares represent male mice. Arrowheads in (f) indicate areas of CAA. Results of unpaired t-test (a) and Mann-Whitney test (e) are displayed within graphs as *p<0.05 and **p<0.01. Omnibus analyses of apoA-I genotype effects (across APP/PS1 genotype) and interaction effects by two-way ANOVA are displayed as exact p-values below graphs (k,l). Sidak’s multiple comparisons test results (l) are displayed within the graph as *p<0.05. For amyloid staining N=4-5 mice per genotype were used, for Aβ ELISA N=5-7 mice per genotype were used, for tau and p-tau ELISA N=6-19 mice per genotype were used. apoA-I: apolipoprotein A-I, HEM:  hemizygous apoA-I genotype, KO: knockout apoA-I genotype, WT: wildtype littermate for APP/PS1, APP/PS1: APP/PS1 transgenic, CAA: cerebral amyloid angiopathy, Aβ: amyloid beta, p-tau: phosphorylated tau, X-34: 1,4-Bis(3-carboxy-4-hydroxyphenylethenyl)benzene, amyloid stain.  Biochemical analysis of total soluble and insoluble Aβ40 and Aβ42 pools in half-brain homogenates revealed no differences by apoA-I genotype, although soluble Aβ40, insoluble Aβ40, and insoluble Aβ42 were all very slightly and non-significantly increased in APP/PS1 apoA-IKO vs. APP/PS1 apoA-IHEM mice (Figure 2.3g-j). The lack of a significant effect of apoA-I genotype on Aβ concentration, despite the significant increase in cortical amyloid deposition, may be due to a region-specific effect of apoA-I on amyloid that is not detectable in crude half-brain homogenates. Indeed, immunofluorescence analysis (described above) confirms that apoA-I genotype influences amyloid plaque burden in the cortex but not in the hippocampus.   Although no significant differences in total tau or tau phosphorylated at Thr231 (p-tau) were observed by APP/PS1 genotype, total tau concentration was significantly higher (Figure 2.3k) and p-tau (Figure 2.3l) concentration tended to be higher in apoA-IKO mice compared to apoA-IHEM mice (p=0.041 and p=0.054, respectively, by omnibus two-way ANOVA). ApoA-I genotype significantly interacted with APP/PS1 genotype with respect to p-tau (p=0.023). Post-hoc Sidak’s multiple comparison tests revealed a nearly significant increase in total tau in APP/PS1 apoA-IKO vs. APP/PS1 apoA-IHEM mice (p=0.063) and a significant increase in p-tau from 1.09 ng/mg in APP/PS1 apoA-IHEM to 1.40 ng/mg in APP/PS1 apoA-IKO (p=0.026). Concentration of p-tau showed a trend towards a reduction in APP/PS1 apoA-IHEM vs. WT apoA-IHEM mice (p=0.099 by Sidak’s multiple comparisons test). The total tau:p-tau ratio was unaffected by loss of apoA-I (Figure 2.3m).   66  2.4.3 ApoA-I deficiency increases pro-inflammatory protein and mRNA levels in the cortices of APP/PS1 mice IL-1β protein concentrations in half-brain homogenates were elevated in APP/PS1 compared to WT mice from a mean of 0.36 pg/mg and 0.42 pg/mg in WT apoA-IHEM and apoA-IKO, respectively, to 0.79 pg/mg and 0.82 pg/mg in APP/PS1 apoA-IHEM and apoA-IKO, respectively, (p<0.0001 by omnibus two-way ANOVA). IL-1β protein concentrations were not significantly affected by apoA-I genotype (Figure 2.4a). However, cortical Il1b mRNA expression levels were significantly elevated overall in APP/PS1 compared to WT mice (p<0.0001 omnibus by two-way ANOVA) and in apoA-IKO vs. apoA-IHEM mice (p=0.0021 by omnibus two-way ANOVA) (Figure 2.4b), where the mean expression ratio was 4.85e-11 in WT apoA-IHEM, 1.08e-10 in WT apoA-IKO, 1.36e-10 in APP/PS1 apoA-IHEM, and 2.17e-10 in APP/PS1 apoA-IKO. The discrepancy between IL-1β protein and mRNA observations may be due to both the lower limit of sensitivity for the protein assay and regional selectivity. Brain protein and mRNA expression of tumour necrosis factor alpha (TNFα) and interleukin 6 (IL-6) were also measured but were all below limits of detection (data not shown).   We next investigated markers of inflammation expressed specifically by cells of the blood-brain barrier. PDGFRβ  is a pericyte marker that is elevated in some cases of neuroinflammation [569]. We observed a significant interaction effect of APP/PS1 and apoA-I genotypes (p=0.026 by omnibus two-way ANOVA) on PDGFRβ protein concentrations in the brain, where PDGFRβ protein tended to be elevated only in APP/PS1 apoA-IKO mice compared to APP/PS1 apoA-IHEM mice (p=0.065 by Sidak’s multiple comparisons test) (Figure 2.4c). Mean PDGFRβ concentrations were 5.03 ng/mg in WT apoA-IHEM, 4.56 ng/mg in WT apoA-IKO, 3.76 ng/mg in APP/PS1 apoA-IHEM, and 5.61 ng/mg in APP/PS1 apoA-IKO. To evaluate EC specific inflammation, we measured protein concentrations of VCAM-1 in the brain. VCAM-1 is expressed on the cell surface of ECs and assists in the adhesion of circulating immune cells to the vessel wall [373,570]. We found that VCAM-1 tended to be elevated in apoA-IKO mice compared to apoA-IHEM mice overall (p=0.076 by omnibus two-way ANOVA) independent of APP/PS1 genotype (Figure 2.4d). ICAM-1 is another EC adhesion molecule that functions similarly to VCAM-1 and   67 can also be expressed in other cells such as astrocytes and microglia [570–574]. We observed an overall effect of apoA-I genotype on brain ICAM-1 concentrations with increases in apoA-IKO mice (p=0.013 by omnibus two-way ANOVA), a specific increase in APP/PS1 apoA-IKO mice compared to APP/PS1 apoA-IHEM mice from a mean of 0.72 ng/mg to 1.25 ng/mg (p=0.023 by Sidak’s multiple comparisons test), and a trend towards a significant interaction effect between apoA-I and APP/PS1 genotypes (p=0.079 by omnibus two-way ANOVA) (Figure 2.4e). Finally, we observed a significant interaction effect between APP/PS1 and apoA-I genotypes on total GFAP protein concentrations in half-brain homogenates (p=0.007 by omnibus two-way ANOVA). GFAP was elevated both in APP/PS1 mice compared to WT controls overall (p<0.0001 by omnibus two-way ANOVA) and in APP/PS1 apoA-IKO compared to APP/PS1 apoA-IHEM mice (p=0.030 by post-hoc Sidak’s multiple comparisons test) (Figure 2.4f).  Mean GFAP protein concentrations were 20.41 μg/mg in WT apoA-IHEM, 18.55 μg/mg in WT apoA-IKO, 25.75 μg/mg in APP/PS1 apoA-IHEM, and 30.59 μg/mg in APP/PS1 apoA-IKO. Together, these data suggest that loss of apoA-I can exacerbate neuroinflammation in APP/PS1 mice, particularly in cell types associated with the cerebrovasculature.  Figure 2.4 Pro-inflammatory protein and mRNA markers in apoA-I deficient APP/PS1 mice.  (a) IL-1β, (c) PDGFRβ, (d) VCAM-1, (e) ICAM-1, and (f) GFAP protein concentrations were measured by ELISA in soluble half-brain homogenates, values were normalized to total protein concentration in the homogenates. (b) Il1b mRNA expression was measured the cortex by RT-qPCR and normalized to β-actin expression. Points represent individual mice and bars represent mean values. Circles represent female mice and squares represent male mice.   68 Omnibus analyses of apoA-I genotype effects (across APP/PS1 genotype), APP/PS1 genotype effects (across apoA-I genotype), and interaction effects by two-way ANOVA are displayed as exact p-values below graphs. Sidak’s multiple comparisons test results are displayed within graphs as *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. For ELISA N=5-19 mice per genotype were used, for mRNA N=7-21 mice per genotype were used. apoA-I: apolipoprotein A-I, HEM:  hemizygous apoA-I genotype, KO: knockout apoA-I genotype, WT: wildtype littermate for APP/PS1, APP/PS1: APP/PS1 transgenic, IL-1β: interleukin 1 beta, VCAM-1: vascular cell adhesion molecule 1, PDGFRβ: platelet derived growth factor receptor beta, ICAM-1: intercellular adhesion molecule 1, GFAP: glial fibrillary acidic protein.  2.4.4 ApoA-I deficiency increases total ICAM-1 protein concentration in the brain, total ICAM-1 positive area in the hippocampus, and vascular ICAM-1 positive area in the cortex and hippocampus of APP/PS1 mice Confocal imaging confirmed co-localization of ICAM-1 with CD31 (Figure 2.5a). Co-localizations of parenchymal ICAM-1 with GFAP was minimal, suggesting expression of ICAM-1 in the parenchyma is predominantly by other cell types (Figure 2.5b). The effect of apoA-I genotype on total brain ICAM-1-positive area was found to be region specific using immunofluorescence. Although we observed no significant differences in total cortical ICAM-1 staining by apoA-I genotype (Figure 2.5c,e), total ICAM-1-positive area was increased in APP/PS1 apoA-IKO compared to APP/PS1 apoA-IHEM in the hippocampus, from 2.34% of hippocampal area to 4.67% (p=0.001 by Sidak’s multiple comparisons test) (Figure 2.5g,i). Interestingly, vascular specific ICAM-1, measured as the ICAM-1 positive area associated with CD31, was significantly elevated in both cortical (Figure 2.5d,f) and hippocampal regions (Figure 2.5h,j) in APP/PS1 apoA-IKO compared to APP/PS1 apoA-IHEM. Cortical vascular ICAM-1 positive areas were 7.87% and 12.26% of CD31 positive area in APP/PS1 apoA-IHEM and APP/PS1 apoA-IKO, respectively (p=0.001 by Sidak’s multiple comparisons test) (Figure 2.5d,f) and hippocampal levels were 15.02% and 24.19% (p=0.001 by Sidak’s multiple comparisons test) (Figure 2.5h,j).     69  Figure 2.5 Total and endothelial associated ICAM-1 in apoA-I deficient APP/PS1 mice. Confocal microscopy was used to identify (a) co-localization of ICAM-1 with the EC marker CD31 and (b) evaluate the co-localization of some ICAM-1 in the brain parenchyma with GFAP-positive astrocytes. Total ICAM-1 staining area was visualized by immunofluorescence in (c,e) cortical and (g,i) hippocampal regions and positive staining area was normalized to total region area. Vascular-specific and parenchymal ICAM-1 expression was visualized using immunofluorescence and shows association of ICAM-1 with CD31 in (d,f) cortical and (h,j) hippocampal regions, where positive co-stained area was normalized to total CD31 positive area. Representative images for immunofluorescent data are below the graphs, green arrowheads indicate examples of vascular ICAM-1. Points represent individual mice and bars represent mean values. Circles represent female mice and squares represent male mice. Omnibus analyses of apoA-I genotype effects (across APP/PS1 genotype), APP/PS1 genotype effects (across apoA-I genotype), and interaction effects by two-way ANOVA are displayed as exact p-values below graphs.  Sidak’s multiple comparisons test results are displayed within graphs as *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001.   70 For ELISA N=6-19 mice per genotype were used. apoA-I: apolipoprotein A-I, HEM:  hemizygous apoA-I genotype, KO: knockout apoA-I genotype, WT: wildtype littermate for APP/PS1, APP/PS1: APP/PS1 transgenic, ICAM-1: intercellular adhesion molecule 1, GFAP: glial fibrillary acidic protein, CD31: cluster of differentiation 31.   2.4.5 ApoA-I deficiency exacerbates Aβ-mediated increases in total and cerebrovascular GFAP levels in both cortex and hippocampus Immunofluorescent staining for GFAP confirmed the biochemical total protein data, with an additive effect of Aβ overexpression and apoA-I loss on GFAP staining area in the cortex (p=0.001 for APP/PS1 vs. WT omnibus two-way ANOVA analysis; p=0.001 for APP/PS1 apoA-IKO vs. WT apoA-IKO, and p=0.040 for APP/PS1 apoA-IKO vs. APP/PS1 apoA-IHEM by post-hoc Sidak’s multiple comparisons test) (Figure 2.6a,c) and a synergistic effect in the hippocampus (p=0.008 interaction effect by omnibus two-way ANOVA; p=0.004 for APP/PS1 apoA-IKO vs. WT apoA-IKO, and p=0.004 for APP/PS1 apoA-IKO vs. APP/PS1 apoA-IHEM by Sidak’s multiple comparisons test) (Figure 2.6e,g). Cortical GFAP positive area increased from 0.07% and 0.06% in WT apoA-IHEM and apoA-IKO, respectively, to 0.70% in APP/PS1 apoA-IHEM and 1.66% in APP/PS1 apoA-IKO. In the hippocampus, GFAP positive area was 2.27%, 1.77%, and 1.65% in WT apoA-IHEM, WT apoA-IKO, and APP/PS1 apoA-IHEM, respectively, and 4.28% in APP/PS1 apoA-IKO.   71  Figure 2.6 Reactivity of astrocytes associated with the cerebrovasculature in apoA-I deficient APP/PS1 mice.  Total GFAP staining area was visualized by immunofluorescence in (a,c) cortical and (e,g) hippocampal regions and positive staining area was normalized to total region area. Vascular-specific GFAP expression was visualized using immunofluorescence and observing the association of GFAP with CD31 in (b,d) cortical and (f,h) hippocampal regions, positive co-stained area was normalized to total CD31 positive area. (Representative images for immunofluorescent data are below graphs. Points represent individual mice and bars represent mean values. Circles represent female mice and squares represent male mice. Yellow arrowheads indicate examples of areas of CD31-associated GFAP in the cortex (d) and hippocampus (h) that are quantified in (b) and (f), respectively. Omnibus analyses of apoA-I genotype effects (across APP/PS1 genotype),  APP/PS1 genotype effects (across apoA-I genotype), and interaction effects by two-way ANOVA are displayed as exact p-values below graphs. Sidak’s multiple comparisons test results are displayed within graphs as *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. For GFAP total and vascular staining N=5-6 mice per genotype were used. apoA-I: apolipoprotein A-I, HEM: hemizygous apoA-I genotype, KO: knockout apoA-I genotype, WT: wildtype littermate for APP/PS1, APP/PS1: APP/PS1 transgenic, GFAP: glial fibrillary acidic protein, CD31: cluster of differentiation 31.  We next used immunofluorescence to examine the association of CD31 and GFAP and observed a robust and statistically significant increase in the cerebrovascular area associated with GFAP in both the cortex and hippocampus of APP/PS1 apoA-IKO mice compared to APP/PS1 apoA-IHEM   72 mice (p=0.008 in cortex, p=0.007 in hippocampus by Sidak’s multiple comparisons test) (Figure 2.6b,d,f,h). In addition, significant elevations were observed overall in APP/PS1 compared to WT mice (p<0.0001 in the cortex by Sidak’s multiple comparisons test) and in apoA-IKO compared to apoA-IHEM mice (p=0.025 in cortex, p=0.049 in hippocampus by Sidak’s multiple comparisons test). Mean cortical vascular GFAP levels were elevated from 0.21% and 0.30% in WT apoA-IHEM and apoA-IKO, respectively, to 1.80% in APP/PS1 apoA-IHEM mice and 3.86% in APP/PS1 apoA-IKO. In the hippocampus, mean vascular GFAP levels increased from 6.86%, 6.20%, and 5.20% in WT apoA-IHEM, WT apoA-IKO, and APP/PS1 apoA-IHEM, respectively, to 10.86% in APP/PS1 apoA-IKO. Strikingly, the elevation in vascular GFAP in apoA-IKO mice was statistically more robust than the changes in total GFAP positive area in both the cortex and the hippocampus. The total vascular area as defined by CD31 staining, vessel thickness, and vessel tortuosity were all unchanged by apoA-I genotype in the cortex and the hippocampus (Figure 2.11a-h).   A closer examination of APP/PS1 mice revealed significantly greater levels of cortical plaque-associated GFAP in APP/PS1 apoA-IKO compared to APP/PS1 apoA-IHEM, which increased from a mean of 0.02% to 0.51% of plaque area (p=0.030 by Mann-Whitney test) (Figure 2.7a-b), suggesting that apoA-I deficiency promotes astrocyte reactivity to Aβ plaques. Parenchymal plaque-associated GFAP was not significantly affected by apoA-I genotype in hippocampus (Figure 2.7c-d). We also investigated whether astrocyte reactivity was affected by CAA, and observed a significant elevation of GFAP area on CAA positive vessels from 0.17% of CAA area on average in APP/PS1 apoA-IHEM mice compared to 2.67% in APP/PS1 apoA-IKO controls (p=0.023 by unpaired t-test) (Figure 2.7e,g). Notably, this effect was larger in magnitude than the increases in either CAA levels or vascular GFAP expression alone in APP/PS1 apoA-IKO mice compared to APP/PS1 apoA-IHEM mice. Finally, we observed a trend toward increased CAA in the areas of vascular GFAP expression where CAA made up 0.04% of vascular GFAP area on average in APP/PS1 apoA-IHEM mice compared to 0.54% in APP/PS1 apoA-IKO controls (p=0.116 by unpaired t test) (Figure 2.7f-g). Taken together, our data support an interaction among Aβ, astrogliosis, and apoA-I.    73  Figure 2.7 Astrocyte reactivity to Aβ plaques and vascular Aβ deposits in apoA-I deficient APP/PS1 mice.  GFAP associated with Aβ plaques in the (a) cortex and (c) hippocampus was visualized with immunofluorescence and normalized to total area of the region. The association of vascular-specific GFAP (GFAP colocalized with CD31) with CAA was visualized using immunofluorescence and normalized to (e) total cortical CAA area and (f) total vascular GFAP area in the cortex. Representative images are (b,d) beside or (g) below graphs. Points represent individual mice and bars represent mean values. Circles represent female mice and squares represent male mice. Yellow closed arrowheads ( ) indicate examples of areas of plaque-associated or CAA-associated GFAP, white open arrowheads ( ) indicate examples areas of plaque or CAA not associated with GFAP. Results of Mann-Whitney test (a) and unpaired t-test (e) are displayed within graphs as *p<0.05. For plaque and CAA area associated with GFAP N=5-6 mice per genotype were used, for CAA area associated with vascular GFAP N=4-5 mice per genotype were used. apoA-I: apolipoprotein A-I, HEM: hemizygous apoA-I genotype, KO: knockout apoA-I genotype, APP/PS1: APP/PS1 transgenic, GFAP: glial fibrillary acidic protein, CD31: cluster of differentiation 31, X-34: 1,4-Bis(3-carboxy-4-hydroxyphenylethenyl)benzene , amyloid stain, CAA: cerebral amyloid angiopathy.  2.4.6 ApoA-I deficiency worsens fear memory deficits in APP/PS1 mice Others have observed that genetically altering apoA-I expression in APP/PS1 mice affects spatial learning and memory, with increased apoA-I improving performance and loss of apoA-I worsening performance in the Morris Water Maze [519,521]. We performed cued and contextual fear conditioning tests to evaluate fear memory and overall observed memory deficits in APP/PS1 mice compared to WT and a worsening of those deficits in APP/PS1 apoA-IKO mice (Figure 2.8a-b). For cued fear conditioning, a significant interaction between APP/PS1 and apoA-I genotype effects was observed (p=0.001 by three-way ANOVA) whereby APP/PS1 apoA-IHEM mice showed 1.8-fold more freezing behaviour than APP/PS1 apoA-IKO (p=0.026 by Sidak’s multiple comparisons test) yet for WT mice, apoA-IKO mice showed 1.4-fold more freezing behaviour than apoA-IHEM mice (p=0.009 by Sidak’s multiple comparisons test). Furthermore, WT apoA-IKO mice also   74 showed 2-fold more freezing behaviour than APP/PS1 apoA-IKO mice (p=0.002 by Sidak’s multiple comparisons test). Notably, non-significant differences in freezing behaviour were present between groups even before the presence of the auditory cue, suggesting that some of the differences in freezing in this test are not related to cued fear memory alone.   Figure 2.8 Cued and contextual fear memory in apoA-I deficient APP/PS1 mice.  (a) Cued and (b) contextual fear memory was measured using a foot shock as the stimulus, the foot shock box as the context, and a tone as the cue. During cued fear conditioning testing, the cue was played during min 3 and 4. Points and error bars represent estimated marginal means and ±standard error used to calculate pairwise comparisons for statistics. Results from three-way ANOVA are displayed as exact p-values below graphs. N=7-24 mice per genotype were used. apoA-I: apolipoprotein A-I, HEM: hemizygous apoA-I genotype, KO: knockout apoA-I genotype, WT: wildtype littermates for APP/PS1, APP/PS1: APP/PS1 transgenic.  For the contextual fear conditioning test, WT mice froze 1.5-fold longer than APP/PS1 mice overall (p=0.019 by three-way ANOVA) and apoA-IHEM mice froze 1.6-fold longer than apoA-IKO mice overall (p=0.004 by three-way ANOVA). More specifically, WT apoA-IKO mice tended to perform better than APP/PS1 apoA-IKO (1.8-fold, p=0.065 by Sidak’s multiple comparisons test), WT apoA-IHEM performed 1.4-fold better than WT apoA-IKO (p=0.020 by Sidak’s multiple comparisons test), and APP/PS1 apoA-IHEM tended to perform better than APP/PS1 apoA-IKO (2-fold, p=0.057 by Sidak’s multiple comparisons test). Taken together, this data suggests that APP/PS1 mice have memory impairments and apoA-I-deficiency exacerbates these impairments. Significant trial effects were also observed in cued and contextual fear conditioning tests (p=0.001 and p=0.019, respectively, by three-way ANOVA) with the time spent freezing tending to increase over the testing period (Cued: p=0.0255 for min 1 vs. min 3, p=0.008 for min 1 vs. min 4, p=0.002   75 for min 1 vs. min 5 by Sidak’s multiple comparisons test. Context: p=0.055 for min 1 vs. min 3, p=0.019 for min 1 vs. min 5 by Sidak’s multiple comparisons test).  2.5 Discussion A better understanding of the vascular contributions to AD may reveal attractive therapeutic targets that act systemically and therefore may not necessarily need to cross the blood-brain barrier to have beneficial effects on the brain. Circulating factors such as HDL, which has multiple vasoprotective properties, could promote cerebrovascular health by acting on the luminal side of cerebral vessels. We therefore aimed to confirm and extend previous findings by us and others on the role of apoA-I on AD pathology in mice. In agreement with Lefterov et al. [519], here we observed that complete loss of apoA-I worsens cognitive deficits and increases CAA in APP/PS1 mice. We also show for the first time that eliminating apoA-I in APP/PS1 mice increases total cortical Aβ deposition and astrogliosis in both cortex and hippocampus. Additionally, we observed that loss of apoA-I increased several markers neuroinflammation and vascular inflammation within the brain. Specifically, brain levels of Il1b mRNA, ICAM-1 protein, PDGFRβ protein, and GFAP protein were elevated in the absence of apoA-I. Furthermore, apoA-IKO mice showed elevated levels of ICAM-1 on ECs and increased GFAP-positive astrocytes in direct contact with the cerebrovasculature in the cortex and hippocampus. While Lefterov et al. did not observe changes in amyloid pathology or neuroinflammation in apoA-IKO mice, other groups have increased apoA-I expression in APP/PS1 mice using transgenic overexpression of apoA-I or by injecting apoA-I mimetics and observed reduced cerebral amyloid and astrogliosis [521,523–525,527]. Finally, we found that astrocyte reactivity to both parenchymal and vascular amyloid was increased in apoA-I deficient mice. Importantly, the previous study by Lefterov et al. compared APP/PS1 apoA-IKO mice to APP/PS1 apoA-IWT mice whereas the current work compares APP/PS1 apoA-IKO mice to APP/PS1 apoA-IHEM mice. Although others have shown that there is a greater difference in plasma apoA-I concentrations between apoA-IKO and apoA-IHEM mice compared to apoA-IHEM and apoA-IWT mice [115], the differences in genotypes used in each of these studies may contribute to some of the differences in the findings. Nonetheless, our observations support the hypothesis that plasma HDL acts to protect cerebral vessels not only from Aβ deposition but also from Aβ-induced astrocyte reactivity and endothelial activation.     76 Most studies altering plasma HDL-C concentration in Aβ expressing mice either find no effect on parenchymal amyloid deposition or a protective effect of raised HDL as well as a protective effect of HDL on CAA [519,521,523–525,527]. Contrary to the current report and others, Contu et al. reported a reduction in both amyloid plaques and CAA in apoA-IKO mice [522]. Several methodological differences between previous studies, the current study, and the study by Contu et al. may explain these inconsistent observations. A notable difference between the studies discussed are the lines of Aβ expressing mice. The present study as well as those by Lefterov et al. and Lewis et al. utilized APP/PS1 mice at 12 months of age [519,521]. APP/PS1 mice overexpress the Swedish amyloid precursor protein (APP) (APPSwe) mutation, have a deletion in the gene for presenilin 1 (PS1), and have extensive amyloid and moderate CAA pathology at 12 months of age [575,576]. However, the 12-month old Tg2576 mice used by Contu et al. only overexpress APPSwe and exhibit relatively mild amyloid pathology [575,576]. As suggested by the study authors, the observed reduction in CAA and amyloid deposition in apoA-IKO mice by Contu et al. may be only a temporary effect and evaluating amyloid pathology in older mice may produce results more in line with the other studies. Additionally, the histological methods used to detect amyloid pathology vary between studies. For example, the current study and those by Lefterov et al. and Lewis et al used amyloid fibril stains to detect amyloid plaques and CAA while Contu et al. used antibodies for Aβ40 and Aβ42 [519,521]. Therefore, the Aβ detected by Contu et al. may include soluble or oligomeric forms that would not be detected in the current study and others. Interestingly, Contu et al. observed elevated brain apoJ levels in WT apoA-IKO mice compared to WT apoA-IWT mice. They suggest elevated brain apoJ may explain their observed reduction in CAA as loss of apoJ in Aβ expressing mice shifts amyloid deposition from parenchymal plaques to the vasculature [248]. However, brain apoJ concentrations in the present study were elevated in the insoluble fraction of brain homogenates of apoA-IKO mice compared to apoA-IHEM mice while CAA levels worsened. Therefore, changes to brain apoJ levels in APP/PS1 apoA-IKO mice are unlikely to explain changes to CAA burden.  The mechanisms by which circulating HDL can reduce amyloid pathology and neuroinflammation are not completely understood but may include effects of apoA-I within the brain and effects of HDL/apoA-I from the periphery. ApoA-I has been shown to enter the brain parenchyma of mice   77 via the blood-CSF barrier at the choroid plexus [501]. ApoA-I that has crossed into the brain may thus have direct anti-inflammatory effects on astrocytes consistent with apoA-I’s anti-inflammatory effects on other cell types [434,577] or may reduce Ab-induced inflammation by inhibiting Aβ fibrillization as has been observed in vitro [578]. Alternatively, HDL may exert its effects on the brain indirectly through actions on the cerebrovasculature, where it would act from the lumen of cerebral vessels on cells of the neurovascular unit without crossing into the brain parenchyma. The latter explanation appears more plausible for several reasons. First, apoA-I concentrations observed within the brain are low compared to the concentrations circulating in the blood, ELISA and immunoblot levels of apoA-I in brain interstitial fluid are <0.005% of plasma levels [501]. Second, HDL has multiple vasoprotective properties including anti-inflammatory activity, modulation of vessel tone, and promotion of EC repair [579]. Finally, previous work genetically manipulating apoA-I expression levels in APP/PS1 mice found changes specifically in vascular amyloid deposition without any changes to parenchymal plaques suggesting that the cerebrovasculature is the most sensitive target of HDL in the brain [519,521]. As described above, we observed several vascular specific changes in APP/PS1 mice based on apoA-I genotype. Briefly, we found increased protein concentrations of PDGFRβ and ICAM-1 as well as ICAM-1 positive area on ECs and GFAP-positive astrocytes associated with the cerebrovasculature.   Our observations of increased plaque-associated GFAP in apoA-IKO APP/PS1 mice suggests an interaction among Aβ, astrogliosis, and apoA-I. CAA-associated GFAP, but not vascular GFAP-associated CAA, was significantly elevated in apoA-IKO mice. Therefore, we speculate that the loss of apoA-I increases the reactivity of astrocytes to amyloid, as opposed to loss of apoA-I increasing astrogliosis first and then leading to increased amyloid deposition as a result. In other words, the anti-inflammatory effects of HDL in the brain are working specifically against Aβ. In support of this hypothesis, loss of apoA-I did not significantly affect any neuroinflammatory marker in WT mice, which do not accumulate Aβ. Others have similarly observed that the elimination of apoA-I is not sufficient to observe pathological vascular changes. For example, apoA-I deficiency alone does not cause atherosclerosis in mice even at 15 months of age on an atherosclerotic diet, whereas apoA-I deficiency can worsen the atherosclerotic pathology of mice lacking LDL receptor (LDLR) [580]. Interestingly, loss of apoA-I expression may indeed be   78 sufficient to result in hypothalamic astrogliosis [581], however, we did not observe any significant apoA-I genotype effects on GFAP expression in the hypothalamus in the current study (Figure 2.12a-b).  If the primary mechanism of action of HDL on AD-relevant pathologies is through HDL’s peripheral activities, increasing circulating HDL may be an attractive therapeutic approach that may complement anti-amyloid approaches. Although HDL-based therapeutics have not yet been studied in humans for their potential benefits against cognitive decline, several studies have investigated whether treating vascular risk factors can affect dementia and AD risk. Potential HDL-based therapeutic approached for AD will be thoroughly discussed in section 6.3.2.  Our study has several limitations. First, our breeding strategy was designed to maximize the production of mice with a total absence of apoA-I rather than produce control animals with wildtype levels of apoA-I. Nevertheless, our observation of a robust reduction in plasma HDL-C concentration between apoA-IHEM and apoA-IKO mice is consistent with previous literature showing that the difference in plasma HDL-C concentration is indeed larger in magnitude between apoA-IHEM and apoA-IKO mice compared to apoA-IHEM and apoA-IWT mice. Therefore, we posit that the apoA-IHEM and apoA-IKO cohorts in this study are sufficient to test the hypothesis that reducing plasma HDL-C concentration worsens cerebrovascular phenotypes in APP/PS1 mice. Further limitations include the use of mixed sexes and analysis at a single age. It is well known that in humans, females have a higher risk of AD than males and their disease progresses more quickly upon diagnosis [582]. Although we were unable to make statistical comparisons based on sex due to limitations in breeding and survival of animals of each genotype and sex, male and female mice are distinguished visually in all of our presented data and no clear sex bias is evident even though the sex of the mice was not balanced between groups. Aging is another major risk factor for AD that we did not extensively investigate as our study examines mice only at 12 months of age when amyloid pathology is well developed. A technical limitation in this study is our limited investigation of the interaction of vascular amyloid and astrogliosis by the method of immunofluorescence. While we were conservative in our investigation by only analyzing amyloid, CD31, and GFAP that were completely associated, it is possible that some association could arise   79 from the overlap of these markers in different planes due to the z-stack imaging technique used to visualize the brain sections. A second technical limitation is that biochemical analysis was only evaluated in crude half-brain homogenates, therefore our ability to detect region-specific changes was limited for biochemical assessments.   Despite these limitations, this study makes significant progress in developing a strong rationale to test HDL as a therapeutic agent in AD that specifically targets the vasculature. This study also replicates the work by both Lewis et al. and Lefterov et al. showing that genetic manipulation of apoA-I can affect CAA and memory in APP/PS1 mice that has not otherwise been confirmed since the original publications in 2010. Our study also expands upon the role of apoA-I in cerebrovascular health by showing that loss of apoA-I increases Aβ within the brain and exacerbates the potentially pathological reactivity of astrocytes to vascular and parenchymal Aβ.  Some groups have already begun to evaluate the potential benefits of apoA-I mimetics or reconstituted HDL in AD mice and have observed benefits with respect to amyloid pathology, CAA, and whole brain neuroinflammation [523,525,527,583], although not all studies found significant improvements or reported data for all of these pathologies. Our current study demonstrates that the investigation of vascular-specific pathologies, including CAA,  cerebrovascular astrogliosis, and endothelial ICAM-1 expression, may be critical in future studies using HDL-based therapeutics to target AD pathology in general and vascular specific amyloid pathologies including CAA.     80 2.6 Supplemental figures  Figure 2.9 Analysis strategy for histology.  Images with bold, coloured outlines indicate steps where percent positive area was measured. The type of measurement and figure presenting data from each measured image is displayed in matching coloured text beside the colour-outlined images. Coloured arrows between images indicate steps where the pixels positive in both images were combined into a new image. A detailed description of each step of the analysis is found in section 2.3.5.     81   Figure 2.10 Morphological discrimination of vascular and parenchymal amyloid.  Morphological discrimination of vascular amyloid from parenchymal amyloid was necessary to analyze CAA area. Examples of inverted 8-bit X-34 images are shown with (right) and without (left) vascular (teal boxes) and parenchymal (green circles) amyloid indicated. A mask of areas containing vascular amyloid was created for each X-34 image and analyzed as described in methods section 2.3.5 and Figure 2.9. X-34: 1,4-Bis(3-carboxy-4-hydroxyphenylethenyl)benzene (amyloid stain), CAA: cerebral amyloid angiopathy.    82  Figure 2.11 Total vascular area in apoA-I deficient APP/PS1 mice.  Vascular area as defined by immunofluorescent staining of CD31 was visualized in the (a) cortex and (c) hippocampus. (b,d) Representative images are below graphs. Vessel thickness (e-f) and tortuosity (g-h) were measure using Vesselucida 360 software analysis of CD31 staining in the cortex and hippocampus. Points represent individual mice and bars represent mean values. Circles represent female mice and squares represent male mice. N=5-6 mice per genotype were used. apoA-I: apolipoprotein A-I, HEM:  hemizygous apoA-I genotype, KO: knockout apoA-I genotype, WT: wildtype littermates for APP/PS1, APP/PS1: APP/PS1 transgenic , CD31: cluster of differentiation 31.    83  Figure 2.12 GFAP staining area in the hypothalamus in apoA-I deficient APP/PS1 mice.  (a) GFAP staining area was visualized by immunofluorescence in the hypothalamus and positive staining area was normalized to total region area. (b) Representative images for immunofluorescent data are beside graphs. Points represent individual mice and bars represent mean values. Circles represent female mice and squares represent male mice. Scale bars represent 400 μm. N=5-6 mice per genotype were used. apoA-I: apolipoprotein A-I, HEM: hemizygous apoA-I genotype, KO: knockout apoA-I genotype, WT: wildtype littermates for APP/PS1, APP/PS1: APP/PS1 transgenic, GFAP: glial fibrillary acidic protein.    84 Chapter 3: Amyloid pathology, neuroinflammation, cerebrovascular pathology, and cognition in APP/PS1 mice treated with a non-brain penetrant LXR agonist  3.1 Summary The liver X receptor (LXR) is a master regulator for transcription of lipid and inflammatory pathways. Treatment of Alzheimer’s disease (AD) mouse models with LXR agonists can reduce brain Aβ concentrations, amyloid pathology, neuroinflammation, and memory deficits. It is known that the ATP-binding cassette transporter A1 (ABCA1) is important for these beneficial effects but the relative importance of ABCA1 within the brain and outside of the brain to these benefits is unknown. Brain ABCA1 is known to lipidate apolipoprotein (apo)E and promote amyloid beta (Aβ) degradation while peripheral ABCA1 lipidates peripheral apolipoproteins, including apoA-I as part of the rate-limiting step of high-density lipoprotein (HDL) biogenesis. We compared the effects of 7 days of treatment with a traditional, brain penetrant LXR agonist to a novel, non-brain penetrant LXR agonist on AD pathology in 12-month old female APP/PS1 mice. We found that neither LXR agonist significantly altered plasma lipid concentrations, brain amyloid beta (Aβ) concentrations, or brain amyloid burden. However, we observed that both LXR agonists, and in particular the non-brain penetrant agonist VTP-38443, significantly reduced neuroinflammation and VTP significantly reduced cerebrovascular inflammation and rescued cued fear memory deficits in APP/PS1 mice. These data suggest that therapeutics do not need to be designed to cross the blood-brain barrier (BBB) in order to have beneficial effects within the brain.  3.2 Introduction The human brain contains 20% of the body’s cholesterol, a higher proportion than any other organ [584]. Proper cholesterol metabolism within the brain is vital to form the cell membranes of neurons and glial cells and the myelin sheaths around axons that allow for efficient propagation of electrical signals in the brain [584]. ApoE is the main transporter of cholesterol in the brain [133] and is encoded by the  APOE gene,  the greatest genetic risk factor for AD [105]. Cholesterol in the blood can also affect brain health. Epidemiological studies in humans have found that high concentrations of total plasma cholesterol and low-density lipoprotein cholesterol (LDL-C) as well   85 as low concentrations of HDL cholesterol (HDL-C) can increase the risk of AD and cognitive impairment, at least in some studies [210,488,489,495,497,585]. Drugs that can modify cholesterol metabolism are therefore of great interest for AD researchers and LXR signalling pathways may provide targets for such drugs.  LXR is a ubiquitously expressed nuclear transcription factor that modulates a number of genes including many related to lipid metabolism and inflammation [586]. LXR agonists were first used in Aβ expressing transgenic mice in 2005 where reductions in soluble Aβ40 and Aβ42 concentrations and increased ABCA1 expression levels in the brains of treated mice were observed [539]. Following this report, LXR agonists were shown to also reduce Aβ plaque deposition [537,540–542], neuroinflammation [537,541–543,550,551], and memory deficits [537,540–542,544] in various AD mouse models (summarized in Table 1.2 and Table 1.3). The importance of ABCA1 in mediating the amyloid and memory improvements by LXR agonist treatment was confirmed by Donkin et al. They administered GW3965 to APP/PS1 mice with and without ABCA1 and observed that ABCA1 was required for the LXR agonist to increase brain and cerebrospinal fluid (CSF) apoE levels, improve amyloid burden in the hippocampus and whole brain, and improve novel object recognition memory [540]. This work together with that showing increased apoE lipidation by LXR agonists [537] lead to the assumption that LXR agonists improve pathology in AD mice by increasing apoE lipidation by ABCA1, which then promotes Aβ degradation. However, ABCA1 is also expressed in peripheral tissues [549] and LXR agonist treatment can increase plasma HDL-C concentration through increased lipidation of apoA-I by peripheral ABCA1 [587], the rate-limiting step in HDL biogenesis [588–591]. Interestingly, evidence from some human epidemiological studies [210,488–491,493,495,496], GWAS studies [63,64], and preclinical studies in AD mouse models [519,521,524–527], including the study in Chapter 2 [592], suggests that circulating HDL may protect against AD.   These observations raise the question of whether the beneficial effects of LXR agonists in AD mice are due to effects of LXR activation inside or outside of the brain. This is a crucial question as it will not only inform on whether targets related to peripheral LXR signaling may be of therapeutic benefit for AD but also whether drugs that do not cross the BBB are able to reduce   86 AD-relevant pathologies within the brain. To answer this question, we administered a non-brain penetrant LXR agonist, known as VTP-38443 (VTP), to 12-month old female APP/PS1 mice and compared its effects to GW3965 (GW), a brain penetrant LXR agonist [593].   3.3 Methods 3.3.1 Animals All procedures involving animals were approved by the Canadian Council of Animal Care and the University of British Columbia Committee on Animal Care. Female APPswe/PENS1dE9 (APP/PS1) mice (line85) were maintained on a mixed C3H/H3J × C57Bl/6 background until 12 months of age at which time they were randomly assigned to control, GW3965 (GW), or VTP-38443 (VTP) diets for 7 days before sacrifice and tissue collection. GW, a brain-penetrant LXR agonist, and VTP, a non-brain penetrant LXR agonist, were dissolved in ethanol and compounded in chow diet at 120 mg/kg and 15 mg/kg, respectively (Envigo, 2920X Teklad global soy protein-free extruded rodent diet). Dosing for VTP was chosen to optimize target engagement and minimize drug penetrance into the brain and dosing of GW was defined as previously reported [540]. Importantly, animals for the pilot study were housed at a different animal facility than animals in the larger study and differences in baseline body weights were observed (Figure 3.10).  3.3.2 Tissue collection Procedures were performed as described in section 2.3.2.  3.3.3 Plasma lipid measurements Procedures were performed as described in section 2.3.3.  3.3.4 Histology Procedures were performed as described in section 2.3.4. In addition, ionized calcium binding adaptor molecule 1 (Iba1) immunohistochemistry was performed on fixed sections. Sections were washed in tris-buffered saline (TBS), treated with 0.3% H2O2, blocked with 3% normal goat serum (NGS) in TBS with 0.25% Triton X-100 (TBS-X) for 30 min, incubated with Iba1 antibody (Wako Chemicals, #019-19741, 1:1000) in TBS-X with 1% NGS overnight, washed, incubated with   87 biotinylated goat-anti-rabbit secondary antibody (Jackson Laboratories, 111-035-003, 1:1000) in TBS-X for 1 h, then developed with horseradish peroxidase (Vectastain Elite ABC kit, PK-6120, Vector laboratories) and 3,3’Diaminobenzidine (DAB) substrate. Stained sections were then mounted and dehydrated with alcohol at 50, 70, 90, then 95% concentrations, cleared with xylene before imaging with an Axio Scan.Z1 (Zeiss).   3.3.5 Image analysis Procedures were performed as described in section 2.3.5. Analysis of Iba1 positive microglia were similarly performed using ImageJ to count positively stained particles in the cortex and hippocampus and measuring regional area for normalization. Positively stained debris greater than 160 pixels was filtered out of the analysis.  3.3.6 Protein extraction Procedures were performed as described in section 2.3.6.  3.3.7 Enzyme-linked immunosorbent assay (ELISA) Procedures were performed as described in section 2.3.7.   3.3.8 Ribonucleic acid (RNA) isolation and real-time quantitative polymerase chain reaction (RT-qPCR) Procedures were performed as described in section 2.3.8.  3.3.9 Contextual and cued fear conditioning Procedures were performed as described in section 2.3.9.  3.3.10 Measurement of plasma and brain drug concentration GW and VTP concentrations were measured in plasma and half-brains using semi-quantitative, label-free mass spectrometry. Briefly, frozen plasma and half-brains were thawed on ice, lysed, fractionated, digested, separated by gas chromatography, then analyzed by mass spectrometry    88 using spectral counts and a standard curve of GW or VTP of known concentrations for quantification.  3.3.11 Statistical analysis  All statistical analyses, with the exception of fear memory, were performed with GraphPad Prism 7, p-values <0.05 were considered statistically significant. Unless stated otherwise, statistical comparisons were made by two-way analysis of variation (ANOVA) considering APP/PS1 genotype as one factor and drug as a second factor followed by Sidak’s or Dunnett’s multiple comparisons test respectively if significant factor or interaction effect was observed. To analyze AD relevant pathologies, namely brain Aβ concentrations, amyloid plaque area, and cerebral amyloid angiopathy (CAA), APP/PS1 were analyzed separately of wild type by one-way ANOVA with Dunnett’s multiple comparisons test if a significant factor effect was observed. Analysis of fear memory was performed with IBM SPSS Statistics Build 1.0.0.1347 using three-way ANOVA to analyze freezing time as the dependent variable and time, APP/PS1 genotype, and apoA-I genotype as fixed variables. Pairwise comparisons of estimated marginal means and standard error were performed and adjusted by Sidak’s multiple comparisons test, adjusted p-values <0.05 were considered statistically significant.  3.4 Results 3.4.1 VTP does not enter the brain in aged WT or APP/PS1 mice  The non-brain penetrant LXR agonist VTP was originally tested by the drug developer for brain penetrance in young C57Bl/6 mice, however, aged APP/PS1 mice are expected to exhibit BBB dysfunction with a potential increase in BBB permeability [594]. We first investigated brain penetrance of VTP 12-month old female APP/PS1 mice treated for 7 days with the drugs at 15 mg/kg in chow. The brain concentration of VTP was below the lower limit of quantification the assay and estimated to be 0.66 nM in WT mice and 0.76 nM in APP/PS1 mice (Table 3.1). Plasma concentrations of VTP were on average 42 nM and 53 nM in WT and APP/PS1 mice, respectively. The brain:plasma ratio of VTP was on average 0.017 and 0.013 in WT and APP/PS1 mice, respectively. The brain penetrance of GW3965 administered in chow (120 mg/kg) for 7 days was confirmed in a separate study in 9-month old mice (Table 3.3).     89  Table 3.1 VTP brain penetrance in 12-month old APP/PS1 mice and wildtype littermates. Genotype WT (n=4) APP/PS1 (n=5) Brain (nM) (mean ± SD) 0.66 ± 0.43 0.76 ± 0.59 Plasma (nM) (mean ± SD) 42 ± 20 53 ± 26 Brain:Plasma 0.017 ± 0.012 0.013 ± 0.0079 WT: wildtype. Lower limit of quantification = 2 nM.  3.4.2 VTP induces Abca1 messenger RNA (mRNA) expression in the small intestine but not in the brain We next investigated the drugs ability to induce expression of Abca1, an LXR target gene, in cortical and intestinal tissue. Both drugs significantly increased the expression of Abca1 mRNA in the small intestine, by 4.5-fold for GW and 5.8-fold for VTP overall (p=0.0165 and p=0.0043 for overall vehicle vs GW and VTP respectively by Dunnett’s multiple comparisons test) (Figure 3.1a). In the cortex, VTP did not affect Abca1 mRNA expression levels (p=0.7727 for overall vehicle vs VTP by Dunnett’s multiple comparisons test) while GW caused a non-significant 1.8-fold increase compared to vehicle control treated mice (p=0.0929 for overall vehicle vs GW by Dunnett’s multiple comparisons test) (Figure 3.1b). The utility of the LXR agonist VTP to study the peripheral-specific effects of LXR signaling in AD mice was therefore confirmed.   Figure 3.1 Intestinal and brain Abca1 mRNA expression in VTP and GW treated APP/PS1 mice in pilot study. Abca1 mRNA expression levels were measured by RT-qPCR in the (a) small intestine and (b) cortex of 12-month old female APP/PS1 mice or WT treated with GW (120 mg/kg in chow) or VTP (15 mg/kg in chow) for 7 days. Points represent individual mice and bars represent mean. Results from omnibus two-way ANOVA of drug effect across genotype and Dunnett’s multiple comparisons test for drug effects across genotypes (in brackets) are displayed as exact p-values below graphs. Results from Dunnett’s multiple comparisons test within each genotype are displayed within graphs as *p<0.05 and **p<0.01. WT: wildtype, GW: GW3965 treated group, VTP: VTP treated group, Abca1: ATP-binding cassette transporter A1 gene, RT-qPCR: real-time quantitative polymerase chain reaction.     90 3.4.3 Soluble Aβ42 concentrations tend to be reduced in APP/PS1 mice treated with the non-brain penetrant LXR agonist VTP in pilot study Having confirmed that VTP is not brain penetrant in aged APP/PS1 mice and it elevates peripheral Abca1 mRNA without significantly affecting brain Abca1 mRNA expression levels, we next aimed utilize this pilot cohort of 12-month old female APP/PS1 mice to determine whether VTP shows any indication of improving AD pathologies. Although changes to brain Aβ concentrations are not consistently observed across LXR agonist studies (Table 1.2, Table 1.3), we chose to begin with evaluation of Aβ changes for our pilot study as it is a relatively high-throughput analysis. To our surprise, we observed a significant reduction in soluble and insoluble Aβ42 brain concentrations with VTP treatment in this pilot study of n=4 mice per group and therefore proceeded to a larger cohort to further probe differences to other AD pathologies. Specifically, the non-brain penetrant LXR agonist VTP significantly reduced soluble brain Αβ42 from by 1.8-fold (p=0.0322 for APP/PS1 vehicle vs VTP by Dunnett’s multiple comparisons test) (Figure 3.2b) and insoluble brain Aβ42 by 1.4-fold (p=0.0252 for APP/PS1 vehicle vs VTP by Dunnett’s multiple comparisons test) (Figure 3.2d). However, GW only resulted in a non-significant 1.5-fold reduction in soluble brain Aβ42 concentration (p=0.1032 for APP/PS1 vehicle vs GW by Dunnett’s multiple comparisons test) and did not significantly reduce insoluble Aβ42 concentration (p=0.5908 for APP/PS1 vehicle vs GW by Dunnett’s multiple comparisons test). Brain Aβ40 concentration was not significantly altered with drug treatment although there was a non-significant 1.8-fold reduction in soluble brain Aβ40 in VTP treated mice compared to vehicle control (p=0.0534 for APP/PS1 vehicle vs VTP by Dunnett’s multiple comparisons test) (Figure 3.2a, c).    Figure 3.2 Brain Aβ peptide concentrations in VTP and GW treated APP/PS1 mice in pilot study. (a) Soluble Aβ40, (b) soluble Aβ42, (c) insoluble Aβ40, and (d) insoluble Aβ42 concentrations were measured in half-brain homogenates by ELISA. Points represent individual mice and bars represent mean. Results from one-way ANOVA are displayed as exact p-values below graphs. Results from Dunnett’s multiple comparisons test are   91 displayed within graphs as *p<0.05. WT: wildtype, GW: GW3965 treated group, VTP: VTP treated group, Aβ: amyloid beta.  As VTP was confirmed in this pilot study to be non-brain penetrant, significantly upregulated peripheral Abca1 mRNA expression levels, and significantly reduced soluble and insoluble Aβ42 concentrations, a study with a larger number of mice was initiated to further probe the changes to amyloid, inflammation, and memory with classical and non-brain penetrant LXR agonists. Although we had originally intended to use our pilot study to perform power calculations, a significant difference was already found in soluble and insoluble brain Aβ42 concentrations with the pilot sample size thereby eliminating the utility of a power analysis for this outcome. Therefore, we instead aimed to develop a cohort with n=6-10 mice per group, as has been used in similar studies, [538,540,542,543], to investigate other neuroinflammatory and cognitive changes.   3.4.4 VTP induces Abca1 messenger RNA (mRNA) expression in the small intestine but not in the brain A larger cohort of 12-month old female APP/PS1 mice was created and the lack of BBB penetrance of VTP was confirmed in a subset of animals (Table 3.2). The effect of the drugs on Abca1 mRNA induction in peripheral tissues and the brain (Figure 3.3a-b) was also confirmed.  Both drugs significantly increased the expression of Abca1 mRNA in the small intestine, overall by 4.5-fold for GW and 4.3-fold for VTP (p<0.0001 and p=0.0001 for overall vehicle vs. VTP and GW respectively by Dunnett’s multiple comparisons test), (Figure 3.3a) whereas only the brain penetrant agonist GW increased Abca1 mRNA expression in the cortex by 1.6-fold (p=0.0006 for overall vehicle vs. GW, p=0.2186 for overall vehicle vs. VTP by Dunnett’s multiple comparisons test) (Figure 3.3b). APP/PS1 genotype had no effect on Abca1 mRNA expression levels in the intestine or cortex. Despite the successful target engagement, neither agonist had any significant effect on plasma apoA-I, apoE total cholesterol, HDL-C, or LDL-C concentrations (Figure 3.3c-f, i). Brain apolipoproteins were also generally unaffected by either agonist. There was no drug effect on insoluble apoA-I or soluble apoE concentrations in the brain although there was an overall drug effect on insoluble apoE in the brain (p=0.0207 for overall drug effect by two-way ANOVA) driven by a 1.2-fold increase in the GW treated mice (p=0.0275 for overall vehicle vs. GW by Dunnett’s multiple comparisons test) Figure 3.3h) and in soluble apoA-I (p=0.0113 for   92 overall drug effect by two-way ANOVA) driven by a 1.6-fold reduction in the VTP treated mice (p=0.0115 for overall vehicle vs. VTP by Dunnett’s multiple comparisons test, p=0.0286 APP/PS1 vehicle vs. VTP)  (Figure 3.3j). APP/PS1 genotype did affect both apoE and apoA-I in the brain. Soluble apoE, insoluble apoE and soluble apoA-I protein concentrations were significantly elevated on average by 1.2-, 2.6-, and 1.5-fold (p=0.0026, p<0.0001, and p=0.0047 for genotype effect by two-way ANOVA omnibus analysis) (Figure 3.3g-h,j) while insoluble apoA-I showed a nearly significant 1.2-fold reduction in (p=0.0534 for genotype effect by two-way ANOVA omnibus analysis) (Figure 3.3k) in the brains of transgenic APP/PS1 mice compared to non-transgenic controls.  Table 3.2 VTP brain penetrance in 12-month old APP/PS1 mice and wildtype littermates. Genotype WT (n=2) APP/PS1 (n=2) Brain (nM) (mean ± SD) <LLOQ <LLOQ Plasma (nM) (mean ± SD) 35 ± 21 63 ± 23 WT: wildtype, N/A:  Lower limit of quantification (LLOQ) = 2 nM.   93  Figure 3.3 Lipoprotein metabolism in VTP and GW treated APP/PS1 mice.  (a) Intestinal and (b) cortical Abca1 mRNA expression levels were measured by RT-qPCR. Plasma apoA-I was measured by ELISA. Plasma (c) total cholesterol, (d) HDL-C, and (e) LDL-C were measured with commercially available kits. Plasma (f) apoE and (i) apoA-I were measured by ELISA. Soluble (g) apoE and (j) apoA-I were measured in carbonate soluble fractions of half-brain homogenates and insoluble (h) apoE and (k) apoA-I were measured in guanidine soluble fractions all by ELISA. Points represent individual mice and bars represent mean. Results from two-way ANOVA omnibus analyses of APP/PS1 genotype effects (across drug treatment groups) and drug effects (across APP/PS1 genotype) are displayed as exact p-values below graphs. Results from Dunnett’s multiple comparisons test for specific drug effects across genotypes are displayed as exact p-values below graphs in brackets. Results from Dunnett’s multiple comparisons test within each genotype are displayed within graphs as *p<0.05 ****p<0.0001. WT: wildtype, GW: GW3965 treated group, VTP: VTP treated group, Abca1: ATP-binding cassette transporter A1 gene, TC: total cholesterol, HDL-C: high-density lipoprotein cholesterol, LDL-C: low-density   94 lipoprotein cholesterol, apoE: apolipoprotein E, apoA-I: apolipoprotein A-I, real-time quantitative polymerase chain reaction.  3.4.5 VTP does not significantly alter Aβ or tau pathologies Cortical amyloid plaque load (Figure 3.4a-b), hippocampal amyloid plaque load (Figure 3.4c-d), and CAA burden (Figure 3.4e-f) remained similar between vehicle control and LXR agonist treated mice. We then measured Aβ concentrations using ELISA and similarly found no changes in either soluble or insoluble pools after VTP treatment (Figure 3.4g-j). We finally investigated whether LXR agonists alters total tau protein concentration and tau phosphorylation at Thr231 and found no differences (Figure 3.4k-m).  Figure 3.4 Amyloid and tau pathology in VTP and GW treated APP/PS1 mice.  (a-b) Cortical and (c-d) hippocampal amyloid plaques and (e-f) cortical CAA were visualized with X-34 staining. (g) Soluble Aβ40, (h) soluble Aβ42, (i) insoluble Aβ40, and (j) insoluble Aβ42 concentrations were measured in half-brain homogenates by ELISA. (k) Total tau, (l) phosphorylated tau, and (m) the ratio of phosphorylated tau or total tau were measured in soluble half-brain homogenates by ELISA. Points represent individual mice and bars represent mean. Scale bars represent 200 μm. WT: wildtype, GW: GW3965 treated group, VTP: VTP treated group, Aβ:   95 amyloid beta, CAA: cerebral amyloid angiopathy, p-tau: phosphorylated tau, Thr231: phosphorylation site at threonine 231, X-34: 1,4-Bis(3-carboxy-4-hydroxyphenylethenyl)benzene, amyloid stain.  3.4.6 VTP treatment reduces some markers of neuroinflammation Despite the lack of effect of either LXR agonists on plasma lipids or AD pathology, drug treated APP/PS1 mice exhibited reduced neuroinflammation. We first investigated the effect of LXR agonists on pro-inflammatory cytokines expression by RT-qPCR and ELISA. Both GW and VTP subtly reduced cortical Il1b mRNA expression and half-brain soluble interleukin 1 beta (IL-1β) protein concentration in APP/PS1 mice. The expression levels of cortical Il1b mRNA showed a nearly significant 2.9-fold increase in vehicle control APP/PS1 mice compared to WT (p=0.0558 for vehicle WT vs. APP/PS1 by Sidak’s multiple comparisons test). Both GW and VTP significantly reduced Il1b mRNA expression levels in APP/PS1 by 3.5- and 2.6-fold for GW and VTP treated APP/PS1 mice respectively (p=0.0114 and p=0.0282 for APP/PS1 vehicle vs. GW and VTP, respectively, by Dunnett’s multiple comparisons test) resulting in expression levels close to those observed in the WT animals (Figure 3.5a). IL-1β protein concentrations were also significantly elevated by 14.3-fold in vehicle control APP/PS1 mice compared to WT mice (p=0.0028 for vehicle WT vs. APP/PS1 by Sidak’s multiple comparisons test). However, only VTP treatment had a significant effect on IL-1β protein concentrations in APP/PS1 mice where concentrations were reduced to by 3-fold (p=0.0256 for APP/PS1 vehicle vs. VTP by Dunnett’s multiple comparisons test) while GW treatment only resulted in a non-significant 1.8-fold reduction (p=0.1798 for APP/PS1 vehicle vs. GW by Dunnett’s multiple comparisons test) (Figure 3.5b).    96  Figure 3.5 Biochemical markers of neuroinflammation in VTP and GW treated APP/PS1 mice. (a) Cortical Il1b mRNA expression levels were measured with RT-qPCR. Protein concentrations of (b) IL-1β, (c) PDGFRβ, (d) ICAM-1, and (e) VCAM-1 in soluble half-brain homogenates were measured by ELISA.  Points represent individual mice and bars represent mean. Results from two-way ANOVA omnibus analyses of APP/PS1 genotype effects (across drug treatment groups) and drug effects (across APP/PS1 genotype) are displayed as exact p-values below graphs. Results from Dunnett’s or Sidak’s multiple comparisons test, for drug and genotype effects respectively, are displayed within graphs as *p<0.05 and **p<0.01. WT: wildtype, GW: GW3965 treated group, VTP: VTP treated group, IL-1β: interleukin 1 beta, PDGFRβ: platelet-derived growth factor beta, ICAM-1: intercellular adhesion molecule 1, VCAM-1: vascular cell adhesion molecule 1.  We then investigated the effect of LXR agonists on platelet-derived growth factor receptor beta (PDGFRβ) protein concentration, a pericyte marker that is elevated by inflammatory stimuli and in AD [569]. PDGFRβ protein concentration was increased from 2-fold in vehicle treated APP/PS1 mice compared to WT (p=0.0239 for vehicle WT vs. APP/PS1 by Sidak’s multiple comparisons test) and nearly restored to WT levels in VTP treated APP/PS1 mice (1.8-fold, p=0.0457 for APP/PS1 vehicle vs. VTP by Dunnett’s multiple comparisons test) (Figure 3.5c). A non-significant 1.5-fold reduction in PDGFRβ protein concentration was observed in  GW treated APP/PS1 mice compared to vehicle control APP/PS1 (p=0.15 for APP/PS1 vehicle vs. GW by Dunnett’s multiple comparisons test).   Because endothelial activation is reported in AD brain [578], we then analysed expression of the endothelium adhesion proteins ICAM1 and VCAM1 by ELISA.  Neither LXR agonist   97 significantly affected brain protein concentrations of intercellular adhesion molecule 1 (ICAM-1) (Figure 3.5d) or vascular cell adhesion molecule 1 (VCAM-1) (Figure 3.5e).  3.4.7 VTP does not alter the number of cortical or hippocampal Iba1+ microglia Microglia are the principal producer of Il-1β in the brain [595]. We therefore used immunohistochemistry staining against the microglial marker ionized calcium-binding adaptor molecule 1 (Iba1) to investigate if the LXR agonists reduce the number of microglia in the brain. No differences in microglia number were observed in cortical or hippocampal regions based on drug treatment or APP/PS1 genotype (Figure 3.6).   Figure 3.6 Iba+ microglia numbers in VTP and GW treated APP/PS1 mice.  Iba+ microglia numbers in (a) cortical and (c) hippocampal regions visualized with immunohistochemistry. Points represent individual mice and bars represent mean. Representative images for (b) cortex and (d) hippocampus displayed below graphs, scale bars represent 200 μm. WT: wildtype, GW: GW3965 treated group, VTP: VTP treated group, Iba1: ionized calcium binding adaptor molecule 1.   3.4.8 VTP reduces total and vascular ICAM-1 expression in the hippocampus Although total Iba1+ area in APP/PS1 was unchanged by VTP-treatment, Iba+ staining does not discriminate between resident microglia and infiltrating myeloid cells [596]. A critical step in immune cell infiltration into surrounding tissues is the expression of adhesion molecule proteins by activated endothelial cells [570]. As VTP most likely exerts its effects on neuroinflammation by acting from the lumen of cerebral vessels, we therefore next investigated whether VTP-treated mice showed changes to endothelial ICAM-1 expression in the brain. ICAM-1 and CD31 immunofluorescent staining was performed and the CD31-associated (vascular) and unassociated (parenchymal) ICAM-1-positive area in the cortex and hippocampus was quantified. A significant   98 interaction between APP/PS1 genotype and drug treatment was observed for both total and vascular ICAM-1 in the hippocampus (p=0.0428 and p=0.0421 interaction effect by two-way ANOVA omnibus analysis) (Figure 3.7e-h). Specifically, VTP-treatment significantly reduced total and vascular ICAM-1-positive staining area in the hippocampus of APP/PS1 mice to 0.5-fold and 0.6-fold, respectively, the levels in APP/PS1 mice on the vehicle control diet (p=0.0062, p=0.0184 APP/PS1 vehicle vs. VTP for total and vascular ICAM-1, respectively, by Dunnett’s multiple comparisons test). Total and vascular ICAM-1 were also reduced in the cortices of VTP-treated APP/PS1 mice to 0.5-fold and 0.7-fold the levels in vehicle control treated mice, respectively, although the effect was not statistically significant (p=0.0984 for vascular ICAM-1 and p=0.0818 for parenchymal ICAM-1 in APP/PS1 vehicle vs. VTP by Dunnett’s multiple comparisons test) (Figure 3.7a-d). We also observed strongly elevated total and vascular ICAM-1 levels in cortical and hippocampal regions in APP/PS1 mice compared to WT mice, as expected (p<0.0001 genotype effect by two-way ANOVA omnibus analysis) (Figure 3.7a-h). APP/PS1 mice treated with GW were not significantly different from vehicle treated APP/PS1 mice for any of the ICAM-1 related measures (p=0.9903, p=0.8754, p=0.1036, and p=0.2408 for APP/PS1 vehicle vs. GW by Dunnett’s multiple comparisons test for cortical total, cortical vascular, hippocampal total, and hippocampal vascular ICAM-1, respectively).    99    100 Figure 3.7 Endothelial and parenchymal ICAM-1-positive staining area in VTP and GW treated APP/PS1 mice. (a,c,e,g) Total ICAM-1 staining area was visualized by immunofluorescence in the (a,c) cortex and (e,g) hippocampus. (b,d,f,h) Vascular ICAM-1 was quantified as ICAM-1 staining associated with the CD31-positive endothelial cells in the (b,d) cortex and (f,h) hippocampus normalized to total CD31-positive area in each region. Points represent individual mice and bars represent mean values. Representative images from the (c-d) cortex and (g-h) hippocampus are displayed below graphs, green arrows indicate examples of vascular ICAM-1. Results from two-way ANOVA omnibus analyses of APP/PS1 genotype effects (across drug treatment groups) and interaction effects are displayed as exact p-values below graphs. Sidak’s and Dunnett’s multiple comparisons test results are displayed within graphs as *p<0.05, **p<0.01, and ****p<0.0001. WT: wildtype, GW: GW3965 treated group, VTP: VTP treated group,  ICAM-1: intercellular adhesion molecule 1, CD31: cluster of differentiation 31. Green arrowheads indicate examples of vascular ICAM-1.  3.4.9 VTP does not alter total levels of GFAP+ astrocytes in the brain but may subtly reduce cerebrovascular astrogliosis We next evaluated the contribution of astrogliosis to cerebrovascular inflammation in LXR-agonist treated APP/PS1 mice. Previous studies have reported that plasma lipoproteins reduce brain astrogliosis in APP/PS1 [521,525,527] and we recently showed that lack of apoA-I worsens vascular astrogliosis[592]. Therefore, we investigated whether VTP decreased astrogliosis in total brain lysate using GFAP ELISA and around the cerebrovasculature using immunofluorescence staining against GFAP and the endothelium marker CD31. We observed that GFAP protein concentration in total brain lysate was elevated by 2.5-fold in APP/S1 overall compared to WT controls (p<0.0001 APP/PS1 genotype effect by two-way ANOVA) but neither VTP nor GW altered GFAP concentration (p=0.6700 and 0.3665 for APP/PS1 vehicle vs. GW and VTP, respectively, by Dunnett’s multiple comparisons test) (Figure 3.8a). We confirmed using immunostaining that cortical and hippocampal GFAP levels were enhanced 49- and 2.6-fold overall, respectively, in APP/PS1 compared to WT (p<0.0001 for cortical staining, p=0.0004 for hippocampal staining APP/PS1 genotype effect by two-way ANOVA). VTP and GW did not alter GFAP-positive staining area in both the cortex and hippocampus (Figure 3.8b-c). Finally, using co-immunostaining for GFAP and CD31, we found that cerebrovascular associated GFAP level was increased by 8.7fold overall in the cortex of APP/PS1 mice compared to WT (p<0.0001 APP/PS1 genotype effect by two-way ANOVA) but not in the hippocampus (p=0.2174 APP/PS1 genotype effect by two-way ANOVA) (Figure 3.8d-e). Vascular GFAP levels in the cortex of VTP treated APP/PS1 mice were non-significantly reduced by 1.6-fold reduced compared to APP/PS1 vehicle control mice,  (p=0.1068 for APP/PS1 vehicle vs. VTP by Dunnett’s multiple   101 comparisons test) (Figure 3.8d). The total vascular area defined by CD31 staining was unchanged by either drug in the cortex and hippocampus (Figure 3.11a-b).    102  Figure 3.8 Total and vessel associated GFAP levels in VTP and GW treated APP/PS1 mice.  (a) Total GFAP protein concentrations were measured in soluble half-brain homogenates by ELISA. GFAP was visualized in (b-c) the cortex and (d-e) hippocampus by immunofluorescence. Vessel-associated GFAP was visualized in the (f-g) cortex and (h-i) hippocampus by immunofluorescent co-staining of GFAP with CD31. Points represent   103 individual mice and bars represent mean. Scale bars represent 200 μm in (c) and (e), 50 μm in (g), and 100 μm in (i). Results from two-way ANOVA omnibus analyses of APP/PS1 genotype effects (across drug treatment groups) are displayed as exact p-values below graphs. Results from Sidak’s multiple comparisons test are displayed within graphs as *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001. WT: wildtype, GW: GW3965 treated group, VTP: VTP treated group, GFAP: glial fibrillary acidic protein, CD31: cluster of differentiation 31.  3.4.10 VTP rescues deficits in cued fear memory in APP/PS1 mice LXR agonists have previously been shown to improve memory deficits in AD mouse models in as little as six days when treated by oral gavage [537,541,542,544], although treatment in chow is typically for at least one month [537,540,545,548]. We evaluated cued and spatial context memory in WT and APP/PS1 mice on each diet using fear conditioning tests. On day one, mice were trained to associate a cue (auditory tone) or a context (testing box environment) with a foot shock. The next day, the fear memory to the same cue or context were evaluated by measuring the time spent frozen upon exposure. In the cued fear memory test, significant genotype, drug, and trial effects were observed (p=0.023, p=0.012, p=<0.0001, respectively, by three-way ANOVA) (Figure 3.9a). Overall, WT mice spent 1.3-fold as much time freezing as APP/PS1 mice over the entire 5 min testing period, suggesting a memory deficit conferred by the transgenes (1.4-fold, p=0.049 for vehicle, 1.6-fold, p=0.003 for GW by Sidak’s multiple comparisons test). Over the 5 min testing period and across both genotypes, GW-treated mice froze for 1.4-fold as long as VTP treated mice (p=0.012 by Sidak’s multiple comparisons test) although a significant interaction between drug and genotype effects was observed (p=0.014 by three-way ANOVA) whereby GW only improved memory in WT mice (1.9-fold, p=0.001 by Sidak’s multiple comparisons test). Comparing the drug effects during only min 3 of testing, the first min of cue exposure, VTP treated APP/PS1 mice performed significantly better than vehicle control APP/PS1 mice (1.9-fold, p=0.030 by Sidak’s multiple comparisons test) while VTP had no effect on WT mice (p>0.999 by Sidak’s multiple comparisons test ). GW-treated mice also demonstrated increased freezing during the third min of testing, a 1.7-fold increase over vehicle animals of both genotypes although the effect was only significant in WT mice (p=0.023, p=0.164 for WT vehicle vs. GW and APP/PS1 vehicle vs GW, respectively, by Sidak’s multiple comparisons test). Overall, these data suggest that 7 days of treatment with VTP can rescue the fear memory deficits in 12-month old female APP/PS1 mice and GW can improve the fear memory of WT and APP/PS1 mice, although the effect is only significant in the WT.  In terms of the significant trial effect observed, the time spent freezing   104 increased across all groups up to min 3 then decreased for the last 2 min of testing (p=0.048 min 1 vs. 2, p<0.001 min 1 vs. 3, p<0.001 min 1 vs. 4, p<0.001 min 1 vs. 5, p<0.001 min 2 vs. 3, p<0.001 min 2 vs. 4, p<0.001 min 3 vs. 5 by Sidak’s multiple comparisons test) suggesting an initial fear response to the cue exposure then acclimation.   Figure 3.9 Cued and contextual fear conditioning in VTP and GW treated APP/PS1 mice.  (a) Cued and (b) contextual fear memory was measured using a foot shock as the stimulus, the foot shock box as the context, and a tone as the cue. During cued fear conditioning testing, the cue was played during min 3 and 4. Points and error bars represent estimated marginal means and ±standard error used to calculate pairwise comparisons for statistics. Results from three-way ANOVA are displayed as exact p-values below graphs. At min indicated: *p<0.05 vs. APP/PS1 vehicle, #p<0.05 vs. WT vehicle, †p<0.05 vs. WT VTP by Sidak’s multiple comparisons test. WT: wildtype, GW: GW3965 treated group, VTP: VTP treated group.  Significant drug, genotype, and time effects were also observed in the contextual fear conditioning test (p=0.015, p=0.010, p=0.046, respectively, by three-way ANOVA) (Figure 3.9b). Overall, time spent freezing by WT mice was 1.3-fold longer than APP/PS1 mice. However, a significant difference between WT and APP/PS1 mice was only observed in the GW-treated group (1.7-fold, p=0.001 by Sidak’s multiple comparisons test) suggesting a lack of sensitivity of this test to detect cognitive deficits in this cohort. Across genotypes, GW-treated mice spent 1.4-fold as much time freezing as vehicle control mice and 1.3-fold as much time as VTP-treated mice (p=0.022 and p=0.048, respectively, by Sidak’s multiple comparisons test). However, a significant drug and genotype interaction effect was also observed (p=0.035 by three-way ANOVA) whereby GW treatment only improved the performance of WT mice (1.6-fold, p=0.006 by Sidak’s multiple   105 comparisons test). Minute-by-minute analysis revealed a significant increase in time spent freezing by GW-treated WT mice compared to vehicle control during minute 2 of testing (2.2-fold, p=0.026 by Sidak’s multiple comparisons test). Post-hoc testing did not reveal any differences between specific minutes of testing although an overall reduction in time spent freezing over the testing period was observed. Together, these fear conditioning tests suggest that GW may improve the memories of WT mice while VTP can rescue some memory function in APP/PS1 mice.  3.5 Discussion It is well established that synthetic LXR agonists can have beneficial effects on amyloid pathology, neuroinflammation, and memory deficits in AD model mice [537,541–544,550] and these benefits are at least in part dependent on the lipid transporter ABCA1 [540]. However, the relative importance of ABCA1 upregulation in the brain compared to upregulation in peripheral tissues has not yet been evaluated. To address this question, we used a pharmacological approach with a non-brain penetrant LXR agonist that selectively upregulates peripheral ABCA1 in 12-month old female APP/PS1 mice and observed improved neuroinflammation and cued fear memory.  Despite the benefits of LXR agonist treatment for AD model mice in this study and others, the utility of LXR agonists as therapeutics for human AD patients is limited due to the detrimental off-target effects, namely hepatic steatosis and hypertriglyceridemia [597]. Although no LXR agonists have been tested in clinical trials, a retinoid X receptor (RXR) agonist known as bexarotene made it as far as a phase II clinical trial. RXR forms a heterodimer with LXR and together they are nuclear transcription factors for the LXR response element [586]. The failure of bexarotene in clinical trials to reduce brain amyloid, except in APOE-ε4 noncarriers, and the detrimental rise in serum triglycerides associated with treatment has further diminished the enthusiasm of researchers for LXR or RXR based AD therapeutics [598]. Instead of suggesting that non-brain penetrant LXR agonists are a valuable therapeutic target, this study proposes more generally that therapeutics that do not cross the BBB can still have beneficial effects on AD-relevant outcomes in the brain. This is of interest as the need to cross the BBB has proven to be an obstacle for drug development [599]. The failure rate of AD drugs in development was 99.6% between 2002 and 2012 [600], suggesting   106 that alternative strategies for drug development are crucial, including pursuing drugs that do not need to cross the BBB.    Another important point of consideration is our use of a pharmacological approach to study the importance of peripheral LXR agonist signaling rather than a genetic approach. We preferred a pharmacological approach for this study as ABCA1flox-/- mice, including those on an AD background, are difficult to breed due to perinatal lethality of ABCA1flox-/- pups [601–603] and poor fertility of ABCA1flox-/- females due at least in part to placental malformation [601]. Furthermore, a pharmacological approach provides proof-of-concept for the translation into a pharmaceutical approach in human AD subjects.  After confirming the lack of brain penetrance of VTP in APP/PS1 mice, we proceeded to perform a pilot study to evaluate the ability of VTP to induce peripheral and brain Abca1 expression and probe whether it exerts any benefits against AD pathologies in 12-month old female APP/PS1 mice. We observed a significant reduction to soluble and insoluble brain Aβ42 with VTP treatment even with the relatively small sample size of the pilot cohort and decided to proceed with the same treatment regimen in a larger cohort aiming for n=6-10 mice per group to more closely reflect the sample size used by similar studies [538,540,542,543]. However, the reduction in soluble brain Aβ concentrations was not replicated in this larger study of 12-month old female APP/PS1 mice. One possible reason for this discrepancy is that the animals were bred and housed at different animal facilities and maintained on different diets until the experimental diet was administered due to centralization of animal care facilities at UBC during the course of these studies. It has recently become clear that the microbiome of research animals can significantly alter their response to experimental procedures and that different housing environments and diets can result in changes to the microbiome [604]. While specific measurements of the gut bacteria of the mice at each facility was not possible, we did observe a significant difference in body weight between the 12-month old female mice housed at each facility (Figure 3.10). Specifically, APP/PS1 mice for the larger study were significantly heavier at baseline compared to APP/PS1 mice in the pilot study (p=0.012 by Sidak’s multiple comparisons test). Furthermore, APP/PS1 mice for the larger study were significantly heavier than WT mice (p=0.012 by Sidak’s multiple comparisons test), whereas   107 there was no significant difference in body weight by genotype in the pilot study (p=0.24 by Sidak’s multiple comparisons test). These body weight differences suggest that mice at each facility may have other metabolic differences that could contribute to differences in AD pathology and response to LXR agonists, in particular as LXR agonists act on cholesterol metabolism and the mice were treated with agonists compounded in chow.   That GW failed to improve amyloid pathology in this study was consistent with other previous studies (Table 1.2). While some have observed that GW treatment can reduce Aβ concentrations or amyloid burden in AD model mice, others have not observed such improvements. For example, Jiang et al. reported reduced Aβ40 and Aβ42 as well as reduced plaque number and area in Tg2576 mice treated with GW at 120 mg/kg in chow [537]. However, Donkin et al. similarly treated APP/PS1 mice with GW and observed an increase in soluble Aβ40 and Aβ42 concentrations in the cortex and only a trend towards reduced plaque area [540]. Furthermore, Sandoval-Hernandez et al. saw no differences in Aβ concentration or plaque burden in 3xTg-AD mice treated with GW by oral gavage at an equivalent dosage [543,544]. A key difference between the present study and previous work is the specific AD mouse model used. It is known that the genetic background of AD model mice and the specific transgenes present can alter the severity and timing of amyloid pathology. While Tg2576, APP/PS1, and 3xTg-AD mice all carry the Swedish APP mutation, the onset of amyloid plaques in these mice are 7 to 10-months old, 6 to 7-months old, and 6-months old, respectively [576]. Additionally, differences in the sex of animals, route of administration, drug dosage, treatment duration, age at treatment, protein and staining analysis techniques used, and brain regions analyzed may all contribute to inconsistencies in the observed effects of GW treatment on brain Aβ concentrations and amyloid pathology.  Although no benefits to amyloid or Aβ concentrations with LXR agonist treatment were observed in the 12-month old female APP/PS1 mice in this study, we did find reduced neuroinflammation in these mice. Specifically, we found reduced brain IL-1β protein and mRNA, brain PDGFRβ protein,  hippocampal ICAM-1-positive area and also a trend towards a reduced vessel-associated astrogliosis in VTP-treated mice. Reduced neuroinflammation by LXR agonist treatment has been consistent across the literature when reported (Table 1.2), in line with our observations. For   108 example, others have observed reduced microglial number [537,550], IL-6 expression [537,542,551], IL-1β expression [542,551], and reduced astrogliosis [543,550]. Contrary to others, we did not observe any changes to GFAP expression or microglia counts, possibly due to differences in the AD mouse models used, age at treatment initiation, or route of drug administration, as discussed above.   Although it is known that LXR agonists such as GW can reduce neuroinflammation in AD model mice, it has been assumed that this occurs through direct inhibition of inflammatory gene transcription in glial cells [537,605]. However, VTP is non-brain penetrant and therefore cannot directly alter the transcriptional profiles of microglia or astrocytes. Instead, the effects of VTP on neuroinflammation must occur via peripheral mechanisms, for example, by limiting the infiltration of circulating immune cells into the brain. IL-1β in the brain is produced by many different cell types including microglia, astrocytes, endothelial cells, and infiltrating immune cells. IL-1β is also an inducer of peripheral immune cell recruitment into the brain [606]. Taken together, this suggests that Aβ in the brains of APP/PS1 mice increases production of IL-1β by glial cells and recruitment of peripher