Characterisation of Lipids in the Human Hippocampus

Libraries and Learning Services

University of Auckland Research Repository, ResearchSpace Copyright Statement The digital copy of this thesis is protected by the Copyright Act 1994 (New Zealand). This thesis may be consulted by you, provided you comply with the provisions of the Act and the following conditions of use:

• • •

Any use you make of these documents or images must be for research or private study purposes only, and you may not make them available to any other person. Authors control the copyright of their thesis. You will recognize the author's right to be identified as the author of this thesis, and due acknowledgement will be made to the author where appropriate. You will obtain the author's permission before publishing any material from their thesis.

General copyright and disclaimer In addition to the above conditions, authors give their consent for the digital copy of their work to be used subject to the conditions specified on the Library Thesis Consent Form and Deposit Licence.

Distribution of Lipids in the Human Brain and their Differential Expression in Alzheimer's Disease: A Matrix-Assisted Laser Desorption/IonisationImaging Mass Spectrometry (MALDI-IMS) Study

B. Lakshini Hishanthi Sharitha Mendis

A thesis submitted in fulfilment of the requirements for the degree of Doctor of Philosophy in Anatomy, The University of Auckland, 2016

How much better to get wisdom than gold, to get insight rather than silver! -- Proverbs 16:16

“Anyone could be seduced by research when the results poured in. The trick was to love it when the results weren't forthcoming, and the reasons why were elusive.” -- Lisa Genova, Still Alice

Abstract Alzheimer’s disease (AD), the leading cause of dementia, is pathologically characterised by β-amyloid plaques and tau tangles. However, there is also evidence of lipid-dyshomeostasis-mediated AD pathology. Given the structural diversity of lipids, mass spectrometry (MS) is a useful tool for studying lipid changes in AD. The use of matrix-assisted laser desorption/ionization (MALDI)-imaging mass spectrometry (IMS) circumvents the limitation of traditional MS, allowing users to visualise the distribution of lipids. Thus, I optimised MALDI-IMS to image the distribution of lipids in the postmortem human middle temporal gyrus (MTG) and hippocampus, and analyse its differential expression in AD. In order to study a large number of cases, compared to previously published MALDI-IMS papers, I developed an analysis workflow to efficiently evaluate large, heterogeneous datasets and accurately detect lipids that were differentially expressed in AD. I hypothesised that the MTG would show similar lipid differences to those previously reported in other cortical regions. Further, given that each hippocampal anatomical sub-field has its own function, I postulated that there would be lipid differences unique to each sub-field. Both positively- and negatively-charged lipid ion species were abundantly detected in the control and AD cohorts. Grey matter and white matter had unique lipid profiles. However, there were variations in the distribution of lipids even within the same region, especially in the grey matter in the MTG and the CA1 region in the hippocampus. In AD, while the distribution patterns of lipids were comparable to the control cohort, some lipids were expressed at different levels. For example, the expression of some phosphatidylethanolamine (PE) lipids was decreased in the MTG. The majority of lipids that were differentially expressed in the hippocampus were found in the CA1 region. Further, there were differences in eight lipids that were specific to the dentate gyrus (DG) region. High-resolution MALDI IMS revealed that these lipids showed a heterogeneous distribution amongst the three DG layers. Finally, I quantified PE lipids with MALDI-IMS, using a lipid-spiked tissue homogenate approach. This is the first time that this approach has been successfully used to quantify lipids the human MTG and the DG. The concentration of PE did not change in the DG in AD; however, the concentrations of four PE species, namely PE 38:4, PE 39:5, and PE 40:6, were reduced in the grey matter in the MTG. Thus, the MALDI-IMS technique, the analysis workflow, and the lipid quantification approach, provided a novel method to investigate specific lipid differences in the postmortem human brain in AD. This work extends the understanding of the lipid composition of the human brain and how it differs in AD. Future work will focus on elucidating if these lipid differences are a driver, or consequence, of AD pathogenesis.

ii | P a g e

Acknowledgements The past four years would not have been possible without my faith and trust in God and the incredible people who I am blessed to have in my life. I am truly indebted to all of you. First and foremost, I’d like to thank my primary supervisor, Associate Professor Maurice Curtis, for giving me the opportunity to work on this project. Maurice, I appreciate your open-mindedness in allowing me to pursue a completely new field, your patience with my bouts of ‘perfectionism’, and your constant belief in my abilities, which has helped me soar to new heights. I am a better researcher because of you. Thank you for encouraging my extracurricular pursuits, all the sound career advice, and for peppering my postgrad years with your ‘tearable puns’. Secondly, I would like to thank my co-supervisor Dr Angus Grey, for taking the time to teach me the intricacies of MALDI and continuously being available to bounce ideas and troubleshoot. Gus, you have been a constant source of encouragement throughout my project, especially during times of high-panic! I’m so grateful that I got to work with you and hope I can contribute to expanding the MALDI Empire in NZ. To all the members of the Curtis group, past and present: Bonnie, Brigid, Daniel, Hector, Helen, Mandana, Molly, Natacha, Sheryl, Steph, Tom, Vicky, and Victor, it has been a privilege to work and play alongside you. A special thank you to Brigid, for helping polish bits of this thesis, to Nat, for being an incredible role-model, to Victor, for the quality banter, brilliant input, and chocolate mousse week, to Vicky, for years of encouragement, and last but not least, to Bonnie, who was there from our first lab induction and shared the ups and downs of the PhD journey. I’d like to thank Distinguished Professor Richard Faull for inspiring me to join the Centre for Brain Research (CBR) family and Associate Professor Henry Waldvogel for continually expanding my neuroanatomical knowledge. A special thanks to Marika Eszes, the technical manager of the Neurological Foundation Human Brain Bank, who allowed me access to this precious resource and enabled me to have well-matched cases for this study. I’d also like to thank Claire Lil and Kristina Hubbard for all their technical support throughout the years. To all the members of the Faull group, especially, Christine, Sam, Jane, Jennifer, and Brittney for all the chats, the constant encouragement, and making sure my chores were completed while I was busy writing this thesis. A special shout-out to my desk-neighbours, Nasim (Salim) and Malvindar (Mervinder) who have provided laughs, high drama, and lots of colour over the past few years.

iii | P a g e

Finally, an acknowledgment to the wider CBR family, including the graduate students, and technical and administrative staff, who have all helped make this place a second home. I would also like to acknowledge Susana Quilodran-Valdebenito for her wise counsel during my final year. Thank you for not belittling what I thought were obstacles in my path and for helping me overcome them. I was able to grow as a person because of you. A special thanks to my friends who have continuously supported me and helped me achieve better work/life balance. Narisara and Hozefa Dhruv, I appreciate all the times you welcomed me into your home whenever I needed a respite from PhD life. Arwa Dhruv, you are a ray of sunshine and I cannot wait to do fun science experiments with you! Dr Ankita Umapathy, thank you for your academic advice and thesis editing. Our chats, just like our mutual love of karaoke and travel, are priceless. To my gym-buddy, Dilini Perera, your classic one-liners have me in stiches. Thank you for motivating me on my laziest days! Justin Wong and Jennifer Chin, I’m grateful to have shared undergrad lab experiences with you. Thanks for sticking by me through the past few years. Finally, to Drs Pulasthi Mithraratne, Shyam Sankaran, and Leo Lam, and soon-to-be Dr Sanjaya Gamage… here’s to all the pub quizzes, arty excursions, and copious amounts of banter. I could not ask for a better squad! A HUGE THANK YOU to my parents, Hemal and Lorraine Mendis, and my sister, Priyesha Mendis. Words cannot express my gratitude for your ongoing love, support, and encouragement, even during the days when I am at my worst. Ammi and Thaththi, you inspired me to dream big and taught me to be resilient. I would be nothing without you. I appreciate all the sacrifices you have made, which have opened so many doors for me. Nangi, thank you for always being there for me and for putting up with me. You continue to teach me so much (Hamilton!) and inspire me to be better. I always have the best time hanging out with you and the laughs we share always brighten my day… “Ahhhhh GENE!” Last, but not least, I’d like to acknowledge the New Zealand Federation of Graduate Women and the University of Auckland for their generous scholarships that allowed me the financial freedom to pursue my goals.

iv | P a g e

Table of contents

Abstract ....................................................................................................................................................ii Acknowledgements ................................................................................................................................. iii Table of contents ......................................................................................................................................v List of figures ............................................................................................................................................ x List of tables........................................................................................................................................... xiii List of abbreviations .............................................................................................................................. xiv Chapter One: General introduction and literature review ............................................................... 1 Section One: Literature review ................................................................................................................ 2 1.1.1 Alzheimer’s disease ...................................................................................................................................... 2 1.1.1.1 Prevalence and incidence ..................................................................................................................... 2 1.1.1.2 Clinical symptoms ................................................................................................................................ 2 1.1.1.3 Pathology of Alzheimer’s disease: Macroscopic brain changes ........................................................... 3 1.1.1.4 Pathology of Alzheimer’s disease: Microscopic changes ..................................................................... 3 1.1.1.5 Apolipoprotein ε4 ................................................................................................................................. 5 1.1.2 Lipids in the human brain ............................................................................................................................. 6 1.1.2.1 Glycerophospholipids: Structure and biosynthesis ............................................................................... 7 1.1.2.2. Glycerophospholipids: function .......................................................................................................... 10 1.1.2.3. Sphingolipids: Structure and biosynthesis .......................................................................................... 11 1.1.2.4. Sphingolipids: Function ...................................................................................................................... 14 1.1.2.5. Lipidomic changes in Alzheimer’s disease .......................................................................................... 15 1.1.3. The human hippocampus ........................................................................................................................... 21 1.1.3.1. Cornu Ammonis (CA) anatomy ........................................................................................................... 21 1.1.3.2. Dentate gyrus (DG) anatomy ............................................................................................................. 24 1.1.3.3. Functional circuitry of the hippocampal formation in memory ......................................................... 24 1.1.3.4 The human hippocampus in Alzheimer’s disease ............................................................................... 25 1.1.4. Matrix-assisted laser desorption/ionisation–imaging mass spectrometry ................................................ 27 1.1.4.1. Matrix-assisted laser desorption/ionisation (MALDI)–time of flight (TOF) mass spectrometry (MS) 27 1.1.4.2. Imaging mass spectrometry ............................................................................................................... 30 1.1.4.3. Sample preparation............................................................................................................................ 30 1.1.4.4. Matrix application .............................................................................................................................. 32 1.1.4.5. Data acquisition ................................................................................................................................. 33 1.1.4.6. Data analysis ...................................................................................................................................... 34

Section Two: Project overview and thesis aims ..................................................................................... 37 1.2.1 Rationale..................................................................................................................................................... 37 1.2.2 Aims ........................................................................................................................................................... 37 1.2.3 Originality of study ..................................................................................................................................... 38

Chapter Two: Materials and methods .......................................................................................... 39 Section One: General methods .............................................................................................................. 40 2.1.3. Acquisition and processing of postmortem human brain tissue ................................................................ 40 2.1.3.1. Human Brain Bank ............................................................................................................................. 40 2.1.1.2 Processing of human brain tissue ...................................................................................................... 40 2.1.1.3 Cases used .......................................................................................................................................... 40 2.1.2. Matrix-assisted laser desorption/ionisation–imaging mass spectrometry (MALDI–IMS) .......................... 43 2.1.2.1. Tissue preparation ............................................................................................................................. 43 2.1.2.2. Choice of matrix ................................................................................................................................. 44

v|Page

Table of contents 2.1.2.3. Matrix application ............................................................................................................................. 44 2.1.2.4. Instrument settings ............................................................................................................................ 45 2.1.2.5. MALDI imaging .................................................................................................................................. 47 2.1.3. Histological staining and imaging ............................................................................................................... 47 2.1.3.1. Haematoxylin & eosin/luxol fast blue staining .................................................................................. 47 2.1.3.2. Imaging .............................................................................................................................................. 48 2.1.4. Lipid identification using tandem mass spectrometry (MS/MS) ................................................................ 48 2.1.4.1. On-tissue matrix-assisted laser desorption/ionisation (MALDI) – tandem mass spectrometry (MS/MS) ............................................................................................................................................. 48 2.1.4.2. Liquid-chromatography (LC) – tandem mass spectrometry (MS/MS) ............................................... 49

Chapter Three: Developing MALDI-IMS to study lipids in the human brain.................................... 51 Section One: Introduction ...................................................................................................................... 52 Section Two: Methods ............................................................................................................................ 53 3.2.1 Histological staining ................................................................................................................................... 53 3.2.1.1 Oil Red O staining .............................................................................................................................. 53 3.2.1.2 Periodic-Acidic Schiff staining ............................................................................................................ 53 3.2.1.3 Imaging .............................................................................................................................................. 53 3.2.2 Mass spectrometry .................................................................................................................................... 53 3.2.2.1. Sample preparation ........................................................................................................................... 53 3.2.2.2. Liquid-chromatography (LC)-mass spectrometry (MS) ...................................................................... 54 3.2.3 Matrix-assisted laser desorption/ionisation–imaging mass spectrometry (MALDI–IMS) .......................... 54 3.2.3.1 Matrix optimisation ........................................................................................................................... 54 3.2.3.2 Sample for imaging mass spectrometry ............................................................................................ 54 3.2.3.3 Homogenised tissue standards preparation ...................................................................................... 54 3.2.3.4 Spectral realignment.......................................................................................................................... 55 3.2.3.5 Spectral normalisation ....................................................................................................................... 55 3.2.3.6 Spectral denoising and hotspot removal ........................................................................................... 56 3.2.3.7 Receiver-operating characteristic (ROC) analysis .............................................................................. 56 3.2.3.8 Six-set Venn diagram ......................................................................................................................... 56

Section Three: Results ............................................................................................................................ 57 3.3.1. Lipid analysis .............................................................................................................................................. 57 3.3.1.1. Histological stains for lipid analysis ................................................................................................... 57 3.3.1.2. Analysis of lipid classes using liquid chromatography-mass spectrometry (LC-MS) .......................... 58 3.3.2. Lipid analysis using matrix-assisted laser desorption/ ionisation (MALDI) ................................................ 60 3.3.2.1. Matrix selection ................................................................................................................................. 60 3.3.2.2 Spectral alignment ............................................................................................................................. 62 3.3.2.3 Lipid spectra detected in the middle temporal gyrus (control cases) ................................................ 63 3.3.3. Lipid distributions in the middle temporal gyrus using MALDI-IMS ........................................................... 65 3.3.3.1 Image processing: Normalisation and denoising ............................................................................... 65 3.3.3.2 Distributions of lipids in the middle temporal gyrus (MTG) ............................................................... 66 3.3.3.3 Lipid expression in the middle temporal gyrus (MTG) in Alzheimer’s disease (AD) ........................... 69 3.3.4. Developing the analysis workflow.............................................................................................................. 71 3.3.4.1 Technical variation ............................................................................................................................. 71 3.3.4.2 Pair-wise receiver operating characteristic (ROC) comparison .......................................................... 72 3.3.4.3 Determining m/z values of interest ................................................................................................... 73 3.3.4.4 Summary of analysis workflow .......................................................................................................... 75 3.3.5. Differentially expressed lipid species in Alzheimer’s disease..................................................................... 77 3.3.5.1 Negative ion mode ............................................................................................................................. 77 3.3.5.2 Positive ion mode ............................................................................................................................... 81

vi | P a g e

Table of contents Section Four: Discussion ........................................................................................................................ 85 3.4.1. General discussion ...................................................................................................................................... 85 3.4.1.1. Methods to analyse lipids .................................................................................................................. 85 3.4.1.2. Optimising matrix-assisted laser desorption/ionisation (MALDI)–imaging mass spectrometry (IMS)85 3.4.1.3. Distribution of lipids in the human middle temporal gyrus (MTG)..................................................... 87 3.4.1.4. Differential lipid expression in the middle temporal gyrus (MTG) in Alzheimer’s disease (AD) ......... 89 3.4.2. Summary of findings ................................................................................................................................... 91

Chapter Four: Lipid changes in the postmortem human hippocampus in Alzheimer’s disease......... 93 Section One: Introduction ...................................................................................................................... 94 Section Two: Methods ........................................................................................................................... 95 4.2.1. Matrix-assisted laser desorption/ionisation-imaging mass spectrometry (MALDI-IMS) ........................... 95 4.2.2. Data analysis ............................................................................................................................................... 95 4.2.3. Lipid assignments ....................................................................................................................................... 96

Section Three: Results ............................................................................................................................ 97 4.3.1. Lipids in the postmortem human hippocampus ........................................................................................ 97 4.3.1.1. Liquid-chromatography data ............................................................................................................. 97 4.3.1.2. Matrix-assisted laser desorption/ionisation (MALDI)–imaging mass spectrometry (IMS) data ........ 99 4.3.1.3. Lipid profiles of different hippocampal sub-fields ............................................................................ 105 4.3.2. Differentially expressed negatively charged lipids in Alzheimer’s disease............................................... 108 4.3.2.1. Whole hippocampus ........................................................................................................................ 108 4.3.2.2. White matter.................................................................................................................................... 108 4.3.2.3 Grey matter ...................................................................................................................................... 108 4.3.2.4. Cornu Ammonis regions ................................................................................................................... 108 4.3.2.5. Dentate gyrus ................................................................................................................................... 110 4.3.2.6. Summary .......................................................................................................................................... 110 4.3.3. Differentially expressed positively charged lipids in Alzheimer’s disease ................................................ 113 4.3.3.1. Whole hippocampus ........................................................................................................................ 113 4.3.3.2. White matter.................................................................................................................................... 113 4.3.3.3. Grey matter ...................................................................................................................................... 113 4.3.3.4. Cornu Ammonis regions ................................................................................................................... 113 4.3.3.5. Dentate gyrus ................................................................................................................................... 115 4.3.3.6. Summary .......................................................................................................................................... 115

Section Four: Discussion ...................................................................................................................... 118 4.4.1. General discussion .................................................................................................................................... 118 4.4.2. Summary of dindings ................................................................................................................................ 123

Chapter Five: Quantifying lipids in the middle temporal gyrus using MALDI-IMS ......................... 125 Section One: Introduction .................................................................................................................... 126 Section Two: Methods ......................................................................................................................... 127 5.2.1 Generation of lipid-spiked tissue homogenate standards ....................................................................... 127 5.2.1.1 Lipid standards ................................................................................................................................. 127 5.2.1.2 Tissue sample used for brain homogenates ..................................................................................... 127 5.2.1.3 Incorporation of lipid standard in brain homogenates .................................................................... 127 5.2.1.4 Data acquisition ............................................................................................................................... 129 5.2.1.5 Data analysis .................................................................................................................................... 129 5.2.1.6 Standard curves ............................................................................................................................... 129 5.2.1.7 Histological staining ......................................................................................................................... 129 5.2.2 Quantifying phosphatidylethanolamine in the middle temporal gyrus ................................................... 129 5.2.2.1 Sample preparation and data acquisition ........................................................................................ 129 5.2.2.2 Standard curves ............................................................................................................................... 130 5.2.2.3 Determining m/z values to analyse .................................................................................................. 130

vii | P a g e

Table of contents 5.2.2.4 Calculation of the amount of lipid in the MTG ................................................................................. 130 5.2.2.5 Statistical analysis............................................................................................................................ 130

Section Three: Results .......................................................................................................................... 131 5.3.1. Quality assessment of lipid-spiked tissue standards ................................................................................ 131 5.3.1.1. Analysis of lipid standards using matrix-assisted laser desorption/ionisation (MALDI) .................. 131 5.3.1.2. Morphology of tissue homogenates ................................................................................................ 131 5.3.1.3. Analysis of lipid standards using imaging mass spectrometry (IMS) ............................................... 131 5.3.1.4. Standard curves ............................................................................................................................... 135 5.3.2. Lipid quantification in the middle temporal gyrus ................................................................................... 137 5.3.2.1. Phosphatidylinositol quantification ................................................................................................. 137 5.3.2.2 Phosphatidylethanolamine quantification....................................................................................... 140 5.3.2.3 Phosphatidylethanolamine differences in Alzheimer’s disease ....................................................... 141

Section Four: Discussion ....................................................................................................................... 146 5.4.1. General discussion ................................................................................................................................... 146 5.4.1.1. Use of lipid-spiked tissue homogenates for quantification .............................................................. 146 5.4.1.2. Phosphatidylethanolamine changes in the grey matter in the middle temporal gyrus in Alzheimer’s disease .......................................................................................................................... 147 5.4.1.3. Considerations for future use ........................................................................................................... 148 5.4.2. Summary of findings ................................................................................................................................ 149

Chapter Six: Lipid distribution and quantification in the dentate gyrus ........................................151 Section One: Introduction .................................................................................................................... 152 Section Two: Methods .......................................................................................................................... 153 6.2.1 High-resolution imaging of the dentate gyrus ......................................................................................... 153 6.2.1.1 Cases used and tissue preparation .................................................................................................. 153 6.2.1.2 Data acquisition ............................................................................................................................... 153 6.2.1.3 Data processing ............................................................................................................................... 153 6.2.1.4 Histological staining ........................................................................................................................ 155 6.2.2 Phosphatidylethanolamine quantification in the dentate gyrus ............................................................. 155 6.2.2.1 Standard curves ............................................................................................................................... 155 6.2.2.2 Calculation of phosphatidylethanolamine concentration in the dentate gyrus ............................... 155 6.2.2.3 Statistical analysis............................................................................................................................ 155

Section Three: Results .......................................................................................................................... 156 6.3.1 High-resolution imaging of lipid distribution in the dentate gyrus .......................................................... 156 6.3.1.1 Mismatch between the defined area for data acquisition and the actual measured area ............. 156 6.3.1.2 Lipid distribution in the dentate gyrus in control sections ............................................................... 156 6.3.1.3 Distribution of lipids that are differentially expressed in Alzheimer’s disease ................................. 158 6.3.2 Phosphatidylethanolamine quantification in the dentate gyrus ............................................................. 161 6.3.2.1 Standard curves ............................................................................................................................... 161 6.3.2.2 Concentration of phosphatidylethanolamine in the dentate gyrus ................................................. 165

Section Four: Discussion ....................................................................................................................... 167 6.4.1. General discussion ................................................................................................................................... 167 6.4.1.1 High-resolution matrix-assisted laser desorption (MALDI)-imaging mass spectrometry (IMS) ....... 167 6.4.1.2 Lipid distribution in postmortem human dentate gyrus .................................................................. 167 6.4.1.3 Quantification of the phosphatidylethanolamine concentration in the dentate gyrus ................... 169 6.4.2. Summary of findings ................................................................................................................................ 169

viii | P a g e

Table of contents Chapter Seven: General discussion and conclusion ..................................................................... 171 Section One: General discussion ......................................................................................................... 172 7.1.1 Introduction .............................................................................................................................................. 172 7.1.2 Using MALDI-IMS to study the postmortem human brain ....................................................................... 173 7.1.2.1 MALDI-IMS as a qualitative tool ...................................................................................................... 173 7.1.2.2 MALDI-IMS as a semi-quantitative tool ........................................................................................... 173 7.1.2.3 MALDI-IMS as a quantitative tool .................................................................................................... 174 7.1.3 Lipid distribution in the postmortem human brain .................................................................................. 174 7.1.4 Differential lipid expression in Alzheimer’s disease ................................................................................. 175

Section Two: Concluding remarks ........................................................................................................ 177 Appendix: Materials and solutions ...................................................................................................... 178 Appendix: Spectral calibration method ............................................................................................... 182 Appendix: Supplementary data (MS/MS)..................................................................................... 183 List of references.................................................................................................................................. 188

ix | P a g e

List of figures Chapter 1: General introduction and literature review Figure 1.1: Glycerophospholipid structure ........................................................................................... 8 Figure 1.2: The biosynthesis of glycerophospholipids in the brain ...................................................... 9 Figure 1.3: Sphingolipid structure ...................................................................................................... 12 Figure 1.4: The biosynthesis of sphingolipids in the brain ................................................................. 13 Figure 1.5: Location and structure of the human hippocampus ........................................................ 23 Figure 1.6: Schematic diagram outlining the functional circuitry of the hippocampal formation in memory in the human brain ............................................................................................. 26 Figure 1.7: Principle of matrix-assisted laser desorption/ionisation (MALDI)-time of flight (TOF) mass spectrometry .................................................................................................................... 29 Figure 1.8: Schematic outline of the matrix-assisted laser desorption/ionisation (MALDI)-imaging mass spectrometry (IMS) workflow.................................................................................. 31 Chapter 2: Materials and methods Figure 2.1: Mass spectra from three hippocampus sections from Alzheimer’s disease (AD; case AZ71) acquired in reflector negative ion mode ................................................................ 46 Chapter 3: Developing MALDI-IMS to study lipids in the human brain Figure 3.1: Oil Red O and Periodic Acid-Schiff (PAS) histological staining.......................................... 57 Figure 3.2: Abundance (as area under the curve measurement) of different lipid classes detected in the middle temporal gyrus (MTG) using liquid chromatography-mass spectrometry (LC-MS) ....................................................................................................... 59 Figure 3.3: Optimisation of matrix choice and tissue washing ........................................................... 61 Figure 3.4: Spectral realignment ........................................................................................................ 62 Figure 3.5: Overview of lipid spectra acquired using matrix-assisted laser desorption/ionisation (MALDI) ............................................................................................................................. 64 Figure 3.6: Normalisation and denoising of imaging mass spectrometry (IMS) data ........................ 67 Figure 3.7: Distribution of selected lipids in the postmortem human middle temporal gyrus (MTG) in a representative control section (H190)............................................................ 68 Figure 3.8: Overview lipid spectra and distribution of selected lipids in Alzheimer’s disease (AD) ... 70 Figure 3.9: Data from technical replicates in reflector negative ion mode ........................................ 71 Figure 3.10: Receiver operating characteristic (ROC) curve for m/z 717.2 in grey matter .................. 73 Figure 3.11: Six-set Venn diagram to determine m/z value differences common to all datasets ....... 74 Figure 3.12: Data analysis workflow ..................................................................................................... 76 Figure 3.13: Relative mean intensity changes of selected m/z values, detected in negative ion mode, in Alzheimer’s disease (AD; as a % change from control) ............................... 78 Figure 3.14: Relative mean intensity change of selected m/z values, detected in negative ion mode, in Alzheimer’s disease (AD; as a % change from control), grouped by sex ..... 79 Figure 3.15: Spatial distribution of selected lipids, which were detected in reflector negative ion mode and differentially expressed in the middle temporal gyrus (MTG) in Alzheimer’s disease (AD) ...................................................................................................................... 80 Figure 3.16: Relative mean intensity changes of selected m/z values, detected in positive ion mode, in Alzheimer’s disease (AD; as a % change from control), grouped by sex ..... 82 x|Page

List of figures Figure 3.17 Spatial distribution of selected lipids, which were detected in reflector positive ion mode and differentially expressed in the middle temporal gyrus (MTG) in Alzheimer’s disease (AD)................................................................................................... 83 Chapter 4: Lipid changes in the postmortem human hippocampus in Alzheimer’s disease Figure 4.1: Abundance (as area under the curve measurement) of different lipid classes detected in the hippocampus in the liquid chromatography-mass spectrometry (LC-MS) trial ..... 98 Figure 4.2: Spatial distribution of selected lipid detected in negative ion mode in the control human hippocampus ...................................................................................................... 100 Figure 4.3: Spatial distribution of selected lipids detected in negative ion mode in the Alzheimer’s disease (AD) human hippocampus ................................................................................. 101 Figure 4.4: Spatial distribution of selected lipids detected in positive ion mode in the control human hippocampus ...................................................................................................... 103 Figure 4.5: Spatial distribution of selected lipids detected in positive ion mode in Alzheimer’s disease (AD) .................................................................................................................... 104 Figure 4.6: Segmentation data .......................................................................................................... 105 Figure 4.7: Overview of lipid spectra detected in the control hippocampus using matrix-assisted laser desorption/ionisation (MALDI)............................................................................... 107 Figure 4.8: Mean intensity difference of natively charged lipids in Alzheimer’s disease (AD; as a % change from control) ................................................................................... 109 Figure 4.9: Spatial distribution of selected lipids detected in reflector negative ion mode in the human hippocampus ...................................................................................................... 112 Figure 4.10: Mean intensity difference of positively charged lipids in Alzheimer’s disease (AD; as a % change from control) ................................................................................... 114 Figure 4.11: Spatial distribution of selected lipids detected in positive ion mode in the human hippocampus................................................................................................................... 117 Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS Figure 5.1: MALDI analysis of lipid standards ................................................................................... 132 Figure 5.2: Morphology of lipid-spiked tissue homogenate ............................................................. 133 Figure 5.3: MALDI-imaging mass spectrometry (IMS) of tissue standards spiked with different concentrations of lipids................................................................................................... 134 Figure 5.4: Linear regression analysis of lipid standards .................................................................. 136 Figure 5.5: Linear regression analysis of the phosphatidylinositol (PI) lipid standard for each dataset .................................................................................................................... 138 Figure 5.6: Residual plots of each linear regression analysis of the phosphatidylinositol (PI) lipid standard curves shown in Figure 5.5 .............................................................................. 139 Figure 5.7: Unprocessed MALDI data from tissue homogenate sections spiked with 0 µg/g (black) and 10,000 µg/g (red) PE 16:0/18:1 lipid standard ........................................................ 141 Figure 5.8: Linear regression analysis of the phosphatidylethanolamine (PE) lipid standard for each dataset .............................................................................................................. 142 Figure 5.9: Residual plots of each linear regression analysis of the phosphatidylethanolamine (PE) lipid standard curves shown in Figure 5.8 ............................................................... 143 Figure 5.10: Concentration and distribution of selected PE species in the grey matter of the middle temporal gyrus (MTG) in control and Alzheimer’s disease (AD) cases ............... 145

xi | P a g e

List of figures Chapter 6: Lipid distribution and quantification in the dentate gyrus Figure 6.1: Mismatch between user-defined area and actual area used for data acquisition ........ 157 Figure 6.2: Distribution of selected lipids in the dentate gyrus (DG) ............................................... 159 Figure 6.3: Distribution of selected lipids in the dentate gyrus in Alzheimer’s disease (AD) ........... 160 Figure 6.4: Linear regression analysis of the phosphatidylethanolamine (PE) lipid standard for control cases ................................................................................................................... 162 Figure 6.5: Linear regression analysis of the phosphatidylethanolamine (PE) lipid standard for Alzheimer’s disease (AD) cases ....................................................................................... 163 Figure 6.6: Residual plots of standard curves shown in Figure 6.4 .................................................. 164 Figure 6.7: Concentration of selected phosphatidylethanolamine (PE) species in the dentate gyrus (DG) in control and Alzheimer’s disease (AD) cases ....................................................... 165

xii | P a g e

List of tables Chapter 1: General introduction and literature review Table 1.1: Glycerophospholipid changes in the human brain in Alzheimer’s disease ......................... 17 Table 1.2: Sphingolipid changes in the human brain in Alzheimer’s disease....................................... 19 Chapter 2: Materials and methods Table 2.1: Summary of control and Alzheimer’s disease (AD) cases used in this thesis ...................... 42 Table 2.2: Summary of pathology of Alzheimer’s disease (AD) cases used in this thesis .................... 43 Table 2.3: Software programs used to acquire and analyse the data presented in this thesis ........... 50 Chapter 3: Developing MALDI-IMS to study lipids in the human brain Table 3.1: Matrix solutions used for optimisation ............................................................................... 54 Table 3.2: Lipid assignments for abundant m/z values ........................................................................ 65 Table 3.3: Summary of the relative change of selected lipid species in the middle temporal gyrus (MTG) in Alzheimer’s disease (AD), detected in negative and positive ion modes............. 84 Chapter 4: Lipid Changes in the postmortem human hippocampus in Alzheimer’s disease Table 4.1: Summary of differential negatively charged lipid expression ........................................... 111 Table 4.2: Summary of differential positively charged lipid expression ............................................ 116 Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS Table 5.1: Volumes of lipid standard stock solutions added to each 250 mg tissue homogenate to create lipid-spiked tissue standards .............................................................................. 128 Table 5.2: Phosphatidylethanolamine (PE) quantification in grey matter in the middle temporal gyrus (MTG) in control and Alzheimer’s disease (AD) ....................................... 144 Chapter 6: Lipid distribution and quantification in the dentate gyrus Table 6.1: Cases used for Chapter Six ................................................................................................ 154 Table 6.2: Lipids that were differentially expressed in the dentate gyrus (DG) alone....................... 158 Table 6.3: Phosphatidylethanolamine (PE) quantification in the dentate gyrus in control and Alzheimer’s disease (AD) ................................................................................................... 166

xiii | P a g e

List of abbreviations AA Aβ ACN AD APOE APP ARP AUC BACE CA CDP Cer Cer1P CERAD CerG1 CerS CHCA CID DAG DAN DG dH2O DHA DHB DPX EOAD ER GalCer GCL GlcCer GM GPI H&E HP IMS ITO LA ICP-MS LCM LC-MS LC-MS/MS LFB

Arachidonic acid Beta amyloid Acetonitrile Alzheimer’s disease Apolipoprotein ε4 Amyloid precursor protein Age-related plaque scores Area under curve β-site APP cleaving enzyme Cornu Ammonis (region) Cytidine diphosphate Ceramide Ceramide-1-phosphate Consortium to Establish a Registry for Alzheimer’s Disease Neutral glycosphingolipid 1 Ceramide synthase α-cyano-4-hydroxycinnamic acid Collision Induced Dissociation Diacylglycerol 1,5-diaminonaphthalene Dentate gyrus Deionised water Docosahexaenoic acid 1,5-dihydroxybenzoic acid Dibutyl phthalate (mounting medium) Early-onset Alzheimer’s disease Endoplasmic reticulum Galactosylceramide Granule cell layer Glucosylceramide Ganglioside Glycosylphosphatidylinositol Haematoxylin and eosin stain Hippocampus Imaging mass spectrometry Indium-tin oxide Laser ablated inductively-coupled plasma mass spectrometry Laser capture microdissection Liquid chromatography-mass spectrometry Liquid chromatography-tandem mass spectrometry Luxol fast blue stain

xiv | P a g e

List of abbreviations LIFT LOAD m/z MACS MALDI MALDI-TOF ML MRI MS MS/MS MTG NFTs NMR OCT PA PAS PBS PC PCIS PE PG PHFs PI PL pPE PS PUFA ROC ROI RT sAPPα SF SM SMS STG TIC TMA

1

This is not an abbreviation1 Late-onset Alzheimer’s disease mass-to-charge ratio Magnetic-activated cell sorting Matrix-assisted laser desorption/ionisation Matrix-assisted laser desorption/ionisation-time of flight Molecular layer (in dentate gyrus) Magnetic resonance imaging Mass spectrometry Tandem mass spectrometry Middle temporal gyrus Neurofibrillary tangles Nuclear magnetic resonance spectroscopy Optimum cutting temperature compound Phosphatidic acid Periodic-acid Schiff stain Phosphate-buffered saline Phosphatidylcholine PreCursor Ion Selector Phosphatidylethanolamine Phosphatidylglycerol Paired helical filaments Phosphatidylinositol Polymorphic layer (in dentate gyrus) Ethanolamine plasmalogen Phosphatidylserine Polyunsaturated fatty acid Receiver-operating characteristic analysis Region of interest Room temperature Soluble APPα Sulfatide Sphingomyelin Sphingomyelin synthase Superior temporal gyrus Total ion current Tissue microarray

LIFT™ is a proprietary trademarked term of Bruker Daltonics that is used with their TOF-TOF mass spectrometer and is not an abbreviation.

xv | P a g e

Chapter One

General introduction and literature review

Section One: Literature review........................................................................................................2 1.1.1 Alzheimer’s disease........................................................................................................................ 2 1.1.1.1 1.1.1.2 1.1.1.3 1.1.1.4 1.1.1.5

Prevalence and incidence..................................................................................................................... 2 Clinical symptoms ................................................................................................................................ 2 Pathology of Alzheimer’s disease: Macroscopic brain changes ........................................................... 3 Pathology of Alzheimer’s disease: Microscopic changes ..................................................................... 3 Apolipoprotein ε4 ................................................................................................................................ 5

1.1.2 Lipids in the human brain .............................................................................................................. 6 1.1.2.1 1.1.2.2. 1.1.2.3. 1.1.2.4. 1.1.2.5.

Glycerophospholipids: Structure and biosynthesis............................................................................... 7 Glycerophospholipids: function .......................................................................................................... 10 Sphingolipids: Structure and biosynthesis ......................................................................................... 11 Sphingolipids: Function ...................................................................................................................... 14 Lipidomic changes in Alzheimer’s disease.......................................................................................... 15

1.1.3. The human hippocampus ............................................................................................................ 21 1.1.3.1. 1.1.3.2. 1.1.3.3. 1.1.3.4

Cornu Ammonis (CA) anatomy........................................................................................................... 21 Dentate gyrus (DG) anatomy ............................................................................................................. 24 Functional circuitry of the hippocampal formation in memory ......................................................... 24 The human hippocampus in Alzheimer’s disease............................................................................... 25

1.1.4. Matrix-assisted laser desorption/ionisation–imaging mass spectrometry ................................. 27 1.1.4.1. 1.1.4.2. 1.1.4.3. 1.1.4.4. 1.1.4.5. 1.1.4.6.

Matrix-assisted laser desorption/ionisation (MALDI)–time of flight (TOF) mass spectrometry (MS) 27 Imaging mass spectrometry .............................................................................................................. 30 Sample preparation ........................................................................................................................... 30 Matrix application ............................................................................................................................. 32 Data acquisition ................................................................................................................................. 33 Data Analysis ..................................................................................................................................... 34

Section Two: Project overview and thesis aims..............................................................................37 1.2.1 Rationale ...................................................................................................................................... 37 1.2.2 Aims ............................................................................................................................................. 37 1.2.3 Originality of study ....................................................................................................................... 38

Chapter 1: General introduction and literature review

Section One

Literature review

1.1.1 Alzheimer’s disease Although the history of senile dementia dates back to Grecian and Roman times (Berchtold and Cotman, 1998), Alzheimer’s disease (AD) was first described by Alois Alzheimer in 1906 in his seminal paper noting his observations on 51-year-old Auguste Deter (Alzheimer, 1906). AD is the leading cause of dementia, accounting for an estimated 60-80% of cases (Alzheimer's Association, 2015). It can be classified as late-onset AD (LOAD), which typically affects people over the age of 65, and earlyonset AD (EOAD), which usually affects people between 50 and 60 years, but may also be observed in someone as young as 15 years (Duyckaerts and Dickson, 2011). EOAD is the familial form of AD and is caused by rare and highly penetrant genes that are transmitted in an autosomal dominant manner. In contrast, common polymorphisms with a relatively low penetrance, but high prevalence, leads to LOAD (Bertram and Tanzi, 2011). Given the complex and multifactorial aetiology of this disease, there is currently no cure for AD. However, recent research has focused on understanding disease pathogenesis. Additionally, there has also been a focus on determining early biomarkers of AD, in order to test putative therapeutic interventions.

1.1.1.1 Prevalence and incidence In 2015, there were 46.8 million people worldwide who were living with AD or another form of dementia, of which two-thirds were women (Alzheimer's Association, 2015, Prince et al., 2015). The number of people with AD is projected to double every 20 years, to 74.4 million in 2030, and 131.5 million in 2050 (Prince et al., 2015). As age is the biggest risk-factor for AD, the growing incidence of AD can be linked to the growing aging population. Reports of the AD prevalence in New Zealand, from 2008, reflect the trends that have been reported worldwide, with an estimated 28,000 New Zealanders affected with AD, and this number is predicted to reach 70,000 by 2031 (Tobias et al., 2008). In addition to the estimated economic cost of US$818 billion in 2015, the disease also has a huge impact on the quality of life of, both, individuals living with dementia, and their families and carers (Prince et al., 2015).

1.1.1.2 Clinical symptoms The criteria for a clinical diagnosis was first established in 1983, but was more recently revised by the National Institute of Aging and Alzheimer’s Association (NIA-AA) in 2011 (Knopman, 2011, McKhann et al., 2011). The updated criteria take into account the genetic information, imaging, and spinal fluid biomarkers, of AD (Knopman, 2011). Individuals with dementia caused by AD can be classified as: probable AD dementia or possible AD dementia. A third classification, probable or possible AD

2|Page

Chapter 1: General introduction and literature review dementia, is currently used for research purposes. The specific criteria for each classification have been outlined in depth by McKhann et al. (2011). Generally, AD can be clinically characterised by: progressive memory loss that disrupts daily life, personality and mood changes, and problems with language, judgement and abstract thinking (Costa et al., 1997). Symptoms are usually present for one to three years before being brought to medical attention, and often it is the deficit in recent memory that is the hallmark of cognitive dissonance. Patients under the age of 65 years usually often present with language deficits at the time of diagnosis and usually have a faster rate of disease progression (Knopman, 2011).

1.1.1.3 Pathology of Alzheimer’s disease: Macroscopic brain changes Despite a clinical profile of dementia by AD, given the other concomitant factors that may lead to dementia, a definitive diagnosis of AD can only be made microscopically after autopsy. Macroscopically, there is visible brain shrinkage, which is particularly evident in EOAD. A decrease in brain weight of 41%, 30%, and 14%, has been previously reported in the temporal lobe, parietal lobe and frontal lobe, respectively (Najlerahim and Bowen, 1988). The significant loss in the temporal lobe can be linked to the atrophy of medial temporal lobe structures such as the hippocampus, amygdala, fusiform gyrus and temporal pole (Mirra and Markesbery, 1996, Duyckaerts and Dickson, 2011). The reduced cortex volume is linked to the decrease in the length and thickness of the cortical ribbon. Finally, there is an enlargement of the ventricular system, particularly of the third and fourth ventricle, which is evident in coronal sections (Mirra and Markesbery, 1996, Fischl et al., 2002).

1.1.1.4 Pathology of Alzheimer’s disease: Microscopic changes Microscopically, AD is characterised by: the accumulation of beta amyloid (Aβ) plaques extracellularly, and the formation of neurofibrillary tangles (NFTs), which are composed of hyperphosphorylated tau protein (Alzheimer, 1906, Mirra and Markesbery, 1996, Duyckaerts and Dickson, 2011). Amyloid precursor protein (APP) is a transmembrane protein that plays a role in cell adhesion and neurotrophy (Thinakaran and Koo, 2008). In normal cell physiology, APP is first cleaved, by αsecretase, an extracellular protease, in the transmembrane region known as the ‘Aβ domain’. This cleavage liberates the large, soluble APPα (sAPPα) ectodomain, from the smaller, carboxy-terminal fragment (C83), which consists of 83 amino acids and can be subsequently cleaved by γ-secretase. The second excision liberates a small peptide, P3, from an intracellular peptide known as the APP intracellular domain (AICD; LaFerla et al., 2007, Thinakaran and Koo, 2008, Harrington, 2012). In the AD brain, however, APP is first processed by β-site APP cleaving enzyme (BACE), rather than αsecretase, separating sAPPα from the last 99 amino acids within the membrane, known as C99. The 3|Page

Chapter 1: General introduction and literature review first amino acid of C99 is the first amino acid of Aβ. C99 is subsequently cleaved 38-43 amino acids from the amino terminus by the γ-secretase complex, which consists of presenilin 1 or 2, nicastrin, anterior pharynx defective, and presenilin enhancer 2, separating the Aβ peptide from the AICD. This cleavage predominantly produces Aβ peptides with 40 amino acids (Aβ1-40) and its more amyloidogenic counter-part with 42 amino acids (Aβ1-42) at a ratio of 10:1. Mutations in APP, which is known to be associated with EOAD, are known to increase the relative production of Aβ1-42 (LaFerla et al., 2007, Thinakaran and Koo, 2008, Harrington, 2012). The aggregation of extracellular Aβ1-42 creates the Aβ plaques, which can be diffuse, primitive, neuritic or burnt-out. Of these, the neuritic plaque, which is approximately 30-50 µm in diameter, consists of a dense central Aβ core, surrounded by processes, which are known as neurites. The neuritic plaque is considered the ‘classic plaque’ and is the most important for disease scoring (Mirra and Markesbery, 1996). The processes that surround the neuritic plaques consist of dense bodies of lysosomes, lipofuscin, degenerating mitochondria and paired helical filaments (Duyckaerts and Dickson, 2011). In contrast, diffuse deposits can be up to several hundred microns in diameter, generally lack dystrophic neurites, and often have a low Aβ content (Duyckaerts and Dickson, 2011). The distinct pattern of Aβ deposition has been previously described in five stages by Thal et al. (2002). Briefly, Aβ is first deposited in the neocortex (Stage 1), followed by the hippocampus and entorhinal cortex (Stage 2), striatum and diencephalic nuclei (Stage 3), various brainstem nuclei (Stage 4), and finally the cerebellum and additional brainstem nuclei (Stage 5; Thal et al., 2002). The Thal et al. (2002) staging criteria determine the ‘A’ of the ABC scoring system used to categorise AD. The ABC scoring system has been previously outlined in detail by Montine et al. (2012). Aβ deposits are normally found in grey matter, with some evidence of streaks of diffuse deposits in white matter (Duyckaerts and Dickson, 2011). The amyloid cascade hypothesis, which was proposed in the early 1990s, states that the Aβ plaques are the linear driving force in both familial and sporadic AD (Selkoe, 1991, Hardy and Higgins, 1992). The microscopic Aβ plaque evidence is supported by the fact that mutations in the APP, PSEN1, and PSEN2, genes are known to confer EOAD (Dickson and Weller, 2011). However, more recently, bioactive Aβ oligomers, rather than the Aβ plaques, have been posited to be neurotoxic (Walsh et al., 2002, Mucke and Selkoe, 2012). Further, since the Aβ plaque load of the human brain does not directly correlate with AD or dementia, Herrup (2015) has argued that the amyloid cascade hypothesis alone is not sufficient to explain AD pathogenesis. Rather, Aβ plaques must be studied alongside NFTs to explain this complex, multifactorial disease. Tau proteins are a family of microtubule-associated phosphoproteins. Predominantly expressed in neuronal axons, tau proteins promote the polymerisation of tubulin monomers into microtubules 4|Page

Chapter 1: General introduction and literature review that make up the neuronal cytoskeletal network (Duyckaerts and Dickson, 2011, Harrington, 2012). The six isoforms of tau, ranging from 352 to 441 amino acids, which are expressed in the human central nervous system are derived from alternative mRNA splicing of the Microtubule-Associated Protein Tau gene (MAPT; Harrington, 2012). In AD, inflammation and stress drive the mislocalisation and hyperphosphorylation of tau (Krstic and Knuesel, 2013). Hyperphosphorylated tau accumulates to form intraneuronal inclusions known as NFTs, which often precedes Aβ pathology by a couple of decades (Duyckaerts, 2011, Krstic and Knuesel, 2013). The presence of NFTs impairs normal axonal transport, which leads to the slow but progressive retrograde degeneration of neurons (Wang et al., 2013). This is the basis of the neuronal cytoskeletal degeneration hypothesis of AD that was proposed by Kosik et al. (1986). The majority of NFTs are composed of bundles of paired helical filaments (PHFs) that are approximately 22 nm in diameter; others may also contain straight filaments that are 15 nm in diameter (Harrington, 2012). The formation of NFTs shows a region-specific hierarchical order, comparable to the deposition of Aβ plaques, which Braak and Braak (1991) described in six stages. Briefly, NFTs form in the transentorhinal and entorhinal areas (Stage I and II), followed by the hippocampus (Stage III and IV), and finally the neorcortex (Stages V and VI; Duyckaerts and Dickson, 2011). The Braak and Braak (1991) stage determines “B” of the of the ABC scoring system used to categorise AD (Montine et al., 2012). In some severe cases up to 30% of the total volume of some brain regions can be occupied by NFTs, exceeding the maximal burden of Aβ (Duyckaerts and Dickson, 2011). Further, tau pathology correlates better with cognitive dysfunction and can cause frontotemporal dementia syndromes, even in the absence of Aβ pathology (Gómez‐Isla et al., 1997, Harrington, 2012). Given the complexity of AD, it is likely that there is an amalgamation of disease mechanisms. In fact, in addition to the pathological hallmarks of AD, i.e. Aβ plaques and tau neurofibrillary tangles, Alzheimer also noted a higher incidence of adipose saccules in many glial cells (Alzheimer, 1906). Moreover, he described a direct relationship between the disturbances in the fibril structure of ganglion cells and the lipid accumulation in the same (Alzheimer, 1911), thus suggesting aberrant lipid metabolism in AD (Foley, 2010, Di Paolo and Kim, 2011). Lipid dyshomeostasis in AD will be discussed in detail in Section 1.1.2.5.

1.1.1.5 Apolipoprotein ε4 While age remains the biggest risk-factor for developing AD, there are also several genetic riskfactors linked to the disease. However, of all the polymorphisms linked to increasing the chance of developing AD, the presence of the ε4 allele of the apolipoprotein gene (Apoε4) poses the largest risk

5|Page

Chapter 1: General introduction and literature review (Corder et al., 1993, Strittmatter et al., 1993). Human ApoE, a 299 amino acid (34 kDa) protein, is one of the primary lipid transport proteins in the central nervous system. It mediates the transport of cholesterol, phospholipids, and sulfatides (SF) through binding to the low density lipoprotein receptors (Han, 2005, Kanekiyo et al., 2014). Of the three ApoE alleles, Apoε3 is the most common isoform with a cysteine and arginine at positions 112 and 158, respectively. Apoε4, which has an arginine in both positions, is present in 52% of all LOAD cases with a family history and 40% of all sporadic cases (Anderson et al., 2007). There is evidence that Apoε4 may drive AD pathogenesis via a number of methods. The first theory, which has been outlined in detail by Kanekiyo et al. (2014), proposes that Apoε4 binds easily to Aβ, enhancing its aggregation and reducing its clearance (Mattson, 2004). The second stipulates that Apoε4 can destabilise microtubule assembly (Strittmatter et al., 1994) and induce tau phosphorylation, driving the formation of NFTs, at least in animal models (Kobayashi et al., 2003, Harris et al., 2004, Huang, 2010). Apoε4 also inhibits neurite branching and extension, causing synaptic deficits (Kim et al., 2014), and may play a role in impairing mitochondrial function (Gibson et al., 2000, Huang, 2010). However, given that the phospholipid changes in the human AD brain cannot be attributed to Apoε4 alone (Mulder et al., 1998, Pettegrew et al., 2001), it is important to understand what these changes are and how they occur.

1.1.2 Lipids in the human brain The human brain is the most lipid-rich organ in the human body, with lipids accounting for up to half the dry weight of brain tissue (Agranoff et al., 1999). Lipids are naturally occurring organic compounds, which are hydrophobic and only soluble in nonpolar organic solvents, like chloroform and acetone. Currently, the mammalian lipidome consists of an estimated 10,000–100,000 individual lipid species, making it more diverse than the proteome or genome (Sparvero et al., 2012). Each lipid species belongs to one of eight different categories: 1) sterol lipids, 2) glycerophospholipids, 3) sphingolipids, 4) fatty acids, 5) glycerolipids, 6) prenol lipids, 7) saccharolipids, and 8) polyketides. The majority of lipids in the human brain belong to one of the first three categories listed above: 1) cholesterol,

2) glycerophospholipids

(phosphatidic

acid,

phosphatidylcholine,

phosphatidylethanolamine, phosphatidylinositols, and phosphatidylserine), and 3) sphingolipids (i.e. sphingosine, ceramide, sphingomyelin, and glycosphingolipids; Jackson et al., 2005, Woods and Jackson, 2006, Wood, 2012). The role that cholesterol plays in normal functioning and pathology of the central nervous system has been previously outlined by Orth and Bellosta (2012), while Puglielli et al. (2003b) present its particular role in AD, which has been debated by Wood et al. (2014). The focus of this thesis, however, will be on glycerophospholipids and sphingolipids alone.

6|Page

Chapter 1: General introduction and literature review

1.1.2.1 Glycerophospholipids: Structure and biosynthesis Glycerophospholipids consist of a glycerol backbone, with C1 and C2 hydroxyls esterified to (usually non-identical) fatty acids, which constitute the non-polar moeity of the molecule. In addition, the C3 hydroxyl is esterified to a phosphate, which in turn may be esterified to an alcohol of one of the following polar headgroups: serine, choline, ethanolamine, glycerol, or inositol (IUPAC, 1997, Frisardi et al., 2011). This constitutes the polar end of the lipid, which is charged due to the ionisation of the phosphate group and the presence of the nitrogenous base. An overview of the glycerophospholipid structures is illustrated in Figure 1.1. The majority of all neural membrane glycerophospholipids are made of: phosphatidylcholine (PC), phosphatidylethanolamine (PE), phosphatidylserine (PS), and phosphatidylinositol (PI; Farooqui et al., 2000). Phosphatidic acid (PA) is the main precursor of all neural membrane glycerophospholipids. It can be synthesised in two ways, i.e the glycerol-3-phosphate pathway, or the dihydroxyacetone phosphate pathway, which have been illustrated in Figure 1.2. The former is present in bacteria and all eukaryotes, while the latter is restricted to yeast and mammalian cells (Athenstaedt and Daum, 1999). As illustrated in Figure 1.2, the conversion of PA to diacylglycerol (DAG) is required for the synthesis of PC, PE, and PS (Athenstaedt and Daum, 1999, Farooqui et al., 2000). PC and PE are synthesised by the cytidine diphosphate (CDP)-choline, and the CDP-ethanolamine pathways, respectively. Both pathways involve three enzymatic reactions. The first reaction is a phosphorylation step by choline or ethanolamine kinase, which produces phosphocholine or phosphoethanolamine, respectively. The second

reaction,

which

is

the

rate

limiting-step,

then

converts

phosphocholine

or

phosphoethanolamine, to CDP-phosphocholine or CDP-phosphoethanolamine, respectively, via the appropriate cytidylyltransferase. The final step is catalysed by a phosphotransferase, which transfers phosphocholine or phosphoethanolamine, to 1,2-diacylglycerol, from CDP-choline or CDPethanolamine, producing PC and PE (Farooqui et al., 2000). PC can also be synthesised from the methylation of PE, via PE N-methyltransferase, which occurs at a higher rate in glial cells rather than neurons, at least in the rat brain (Tsvetnitsky et al., 1995). Finally, PS can be produced from both PC and PE, via a reversible, energy-dependent, base-exchange reaction involving PS synthase (Farooqui et al., 2000). Plasmalogens constitute a subclass of glycerophospholipids, which have a vinyl ether fatty alcohol substituent at the sn-1 position of the glycerol backbone (Naudí et al., 2015). PI is primarily synthesised in the endoplasmic reticulum (ER), before being delivered to other membranes by vesicular transport or cytosolic PI transfer proteins (Di Paolo and De Camilli, 2006). The de novo synthesis of PI relies on the conversion of PA to cytidine diphosphate CDP-DAG, via

7|Page

Chapter 1: General introduction and literature review

Figure 1.1: Glycerophospholipid structure. This is an overview highlighting the structural similarity between the different glycerophospholipid species. All glycerophospholipids consist of a central glycerol moiety, in which the C1 and C2 hydroxyls are esterified to fatty acids (black; stearic acid in this example), and the C3 hydroxyl is esterified to a phosphate group, making up the glycerophosphate group (red). The fatty acid tail makes up the hydrophobic end of the molecule. This is the basic structure of phosphatidic acid (PA). The phosphate group can be further esterified to a polar headgroup

(shown in blue), which determines the type of glycerophospholipid species. This polar headgroup can be a choline, ethanolamine, glycerol, inositol, or serine group, which make up the phosphatidylcholine (PC), phosphatidylethanolamine (PE), phosphatidylglycerol (PG), phosphatidylinositol (PI), and phosphatidylserine (PS) species, respectively. Structures were drawn using ADC/ChemSketch Freeware Version 14.01 (Advanced Chemistry Development, Inc., Canada).

phosphatidate cytidylyltransferase. CDP-DAG and inositol then undergo a reaction catalysed by PI synthase, to produce PI (Athenstaedt and Daum, 1999, Farooqui et al., 2000). Free myoinositol can also undergo an exchange reaction in the presence of manganese producing PI. Finally, the hydroxyl group at position 3 of the inositol ring in PI can be phosphorylated by different kinases (as illustrated in Figure 1.2) to produce seven phosphoinositides, of which PI-3-phosphate, PI (3,4)-bisphosphate, PI (4,5)-bisphosphate, and PI (3,4,5)-triphosphate are abundant in the human brain (Pacheco and Jope, 1996, Di Paolo and De Camilli, 2006).

8|Page

Chapter 1: General introduction and literature review

Figure 1.2: The biosynthesis of glycerophospholipids in the brain. This figure illustrates the relationship between different lipid classes in the synthesis of others. The enzymes required for each reaction, numbered 1-24 (in blue) are: 1) acyl transferase, 2) acyl transferase, 3) dihydroxyacetone phosphate acyltransferase, 4) acyl-dihydroxyacetone phosphate reductase, 5) phosphatidate cytidylyltransferase, 6) PI synthase, 7) PI-3-kinase, 8) PI-3-P 4-kinase, 9) PI-3,4-P 5-kinase, 10) Type I PI 5-kinase, 11) Type II PI 4-kinase, 12) PI 3-kinase, 13) PA phosphatase, 14) diacylglycerol kinase, 15) ethanolamine kinase, 16) CTP, phosphoethanolamine cytidylyltransferase, 17) CDP-ethanolamine, DAG, phosphoethanolamine transferase, 18) choline kinase, 19) phosphocholine

cytidylyltransferase, 20) CDP-choline, DAG phosphocholine transferase, 21) PE Nmethyltransferase, 22) and 23) PS synthase, 24) PSdecarboxylase. The co-factors required for each reaction are shown in green. Adapted from Athenstaedt and Daum (1999). ADP, adenosine diphosphate; ATP, adenosine triphosphate; CDP, cytidine diphosphate; CMP, cytidine monophosphate; CTP, cytidine triphosphate; CO2, carbon dioxide; CoA, Co-enzyme A; DAG, diacylglycerol; P, phosphate; PA, phosphatidic acid; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol; PS, phosphatidylserine

9|Page

Chapter 1: General Introduction and Literature Review

1.1.2.2. Glycerophospholipids: function The structure of a glycerophospholipid, which consists of a polar head-group and non-polar tail, allows for the marked amphilicity of these molecules. The non-polar ends tend to aggregate in aqueous environments, forming the characteristic structure of the lipid bilayer, which is a highly dynamic structure that plays a central role in an array of cellular processes (Söderberg et al., 1991, Schiller et al., 2004, Mencarelli and Martinez–Martinez, 2013). In addition to these hydrophobic bonds, hydrogen bonds, and coulombic and van der Waal forces also hold the lipid bilayer together (Farooqui et al., 2000). PA makes up about 2% of the total glycerophospholipids found in the brain, while PC and PE make up about 32.8% and 35.6%, respectively. Choline plasmalogen accounts for about 2% of the total PC, while ethanolamine plasmalogen (pPE) constitutes about 50-60% of total PE found in the human brain. The PS concentration is about 16.6%, while PIs account for about 2.6% of total glycerophospholipids (O'Brien and Sampson, 1965a, O'Brien and Sampson, 1965b, Naudí et al., 2015). The asymmetric distribution of glycerophospholipids along the plane of the plasma membrane, i.e. PE and PS concentrated in the inner leaflet, and PC concentrated on the outer leaflet, contributes to neural membrane stability (Piomelli et al., 2007, Naudí et al., 2015). Further, the polyunsaturated fatty acids (PUFAs) in the non-polar tail determine the physical properties of the lipid bilayer, such as: phase transition temperature, bilayer thickness, acyl chain packing free volume, and lateral domains (Farooqui et al., 2000). Thus, as glycerophospholipids determine the fluidity and permeability of neural membranes, they play a role in controlling the function of integral membrane proteins, receptors and ion-channels, which span the lipid bilayer (Farooqui et al., 2000). Not surprisingly, given these varied roles, the lipid composition of different cell types and regions in the human brain differs. For instance, myelin has a much higher total glycerophospholipid content (7881%) than white matter (49-66%) or grey matter (36-40%; O'Brien and Sampson, 1965b). However, of the specific glycerophospholipid classes, grey matter contains a higher abundance of PE and PC species compared to myelin (Farooqui et al., 2000). In addition to its fundamental structural role, some PEs and PIs also serve as ‘docking stations’ for membrane proteins, such as acetylcholinesterase and neural cell adhesion molecules (Hooper, 1997). The seven phosphoinositides derived from PI also have distinctive roles in the cell nucleus and cytosol, which have been previously reviewed by Hammond et al. (2004) and Di Paolo and De Camilli (2006). Briefly, these functions include (but are not limited to) vesicle docking and secretion (Martin, 2015), endocytosis (Posor et al., 2015), regulation of ion channel activity (Hille et al., 2015), and actin filament assembly (Saarikangas et al., 2010).

10 | P a g e

Chapter 1: General introduction and literature review

1.1.2.3. Sphingolipids: Structure and biosynthesis Named after the sphinx by JLW Thudichum in 1884 for their enigmatic nature, sphingolipids, like glycerophospholipids, are amphipathic molecules with both hydrophobic and hydrophilic moieties. The hydrophobic tail in sphingolipids consists of a sphingoid long chain base, which is connected to a fatty acid via an amide bond to carbon 2. There are at least five different sphingoid bases in mammalian cells, with more than 20 species of fatty acids that vary in chain length, and the degree of saturation and hydroxylation. As illustrated in Figure 1.3, the polar headgroup can consist simply of two hydroxyl (OH) groups or phosphate or carbohydrate residues in more complex sphingolipids. Depending on its headgroup, sphingolipids can be classified as: 1) sphingosine, 2) ceramides (Cer), 3) sphingomyelin (SM), and 4) glycosphingolipids. Glycosphingolipids are complex structures that include galactosylceramides (GalCer), SF, glucosylceramides (GlcCer), lactosylceramides, and gangliosides (GM). The large diversity in the hydrophobic moiety and the polar headgroup, together with the different combinations of the two, give rise to a large range of different sphingolipid species (Buccoliero and Futerman, 2003, Futerman and Hannun, 2004). Cer, the main precursor of all other sphingolipids, can itself be generated via de novo synthesis, via the degradation of complex sphingolipids, or re-acylation of sphingosine (Ben-David and Futerman, 2010). The de novo synthesis of ceramide occurs in the ER (Futerman and Hannun, 2004). As illustrated in Figure 1.4, serine and activated fatty acids, usually palmitoyl-CoA or stearoyl-CoA first undergo a condensation reaction producing 3-ketosphinganine, which is rapidly reduced to sphinganine in a reaction that consumes nicotinamide adenine dinucleotide phosphate (NADPH). The saturated long chain base sphinganine is then N-acylated to form dihydroceramide (DHCer) in a reaction that is catalysed by a family of six ceramide synthases (CerS). CerS1 is abundant in the neocortex and hippocampus, at least in the mouse brain, and predominantly mediates the synthesis of C18-ceramide (Becker et al., 2008). DHCer is then desaturated by DHCer-desaturase, in a NADPHdependent reaction, producing Cer (van Echten-Deckert and Walter, 2012). Cer can then undergo other reactions, and be bound to different functional groups, to produce other sphingolipids. The phosphorylation of Cer via Cer kinase leads to the production of Cer-1-phosphate (Giussani et al., 2014). SM is produced from Cer in a reaction catalysed by sphingomyelin synthase (SMS), which transfers a phosphocholine from PC to Cer (Futerman and Hannun, 2004). There are at least two isoforms of SMS in mammalian cells. The first is SMS1 that is localised primarily on the luminal side of the cis/medial Golgi apparatus, which is the site of 90% of de novo SM synthesis. The second is SMS2, which is localised at the plasma membrane (Huitema et al., 2004).

11 | P a g e

Chapter 1: General introduction and literature review

Figure 1.3: Sphingolipid structure. This overview illustrates the structural relationship between different sphingolipid species. All sphingolipids consist of a sphingosine backbone (red), which is a long acyl chain with an amino group and two hydroxyl groups at one end (blue). Ceramide is formed when a second fatty acid is covalently linked to the amino group. The polar head group attached to the terminal hydroxyl, which is illustrated in blue within the dashed line box, determines the sphingolipid group. Attachment of a phosphocholine produces a

12 | P a g e

sphingomyelin, while attachment of a galactose, or glucose, produces galactosylceramide, or glucosylceramide, respectively. Sulfatides contain a sulphated galactocerebroside headgroup. Gangliosides consist of a ceramide attached to an oligosaccharide (blue) and sialic acid (in green). Structures were drawn using ADC/ChemSketch Freeware Version 14.01 (Advanced Chemistry Development, Inc., Canada).

Chapter 1: General introduction and literature review

Figure 1.4: The biosynthesis of sphingolipids in the brain. This figure illustrates the relationship between different sphingolipid species. The enzymes required for each reaction, numbered 1-21 (in blue) are: 1) serine-palmitoyl transferase, 2) ketosphinganine reductase, 3) ceramide synthase, 4) dihydroceramide desaturase, 5) ceramide kinase, 6) ceramide-1phosphate phosphatase, 7) sphingomyelin synthase, 8) sphingomyelinase, 9) ceramidase, 10) ceramide synthase, 11) sphingosine kinase, 12) sphingosine-1phosphate phosphatase, 13) sphingosine-1-phosphate lyase, 14) ceramide galactosyl transferase, 15) galactosylceramidase,16) cerebrosides sulfotransferase, 17) arylsulfatase A, 18) glucosylceramide synthase, 19) glucosylceramidase, 20) globotriaosylceramide synthase, 21) globotetraosylceramide synthase,

22) GM3 synthase, 23) Neuraminidase 3, 24) GD3 synthase, 25) GT3 synthase, 26) GalNAc-T, β-1,4-Nacetylgalactosaminyltransferase-1, 27) Gat-IIgalactosyltransferase, 28) ST-IV,ST6-Nacetylgalactosaminide-α-2,6-sialyltransferase, 29) Neuraminidase 3. Co-factors required for reactions are shown in green. Figure adapted from Giussani et al. (2014). Cer1P, ceramide-1-phosphate; CoA, Co-enzyme A; DAG, diacylglycerol; Gb, globotriaosylceramide; GD, disialoganglioside; GalCer, galactosylceramide; GlcCer, glycosylceramide; GM, ganglioside; GT, triganglioside; PC, phosphatidylcholine; SF, sulfatides; SM, sphingomyelin; UDP-Gal, uridine-diphosphategalactose; UDP-Glu, uridine-diphosphate-glucose

13 | P a g e

Chapter 1: General introduction and literature review Although all SM share a common headgroup, they are produced from a variety of Cer species, and can thus have varying acyl chains attached to their C2 amino group (Gault et al., 2010). The conjugation of a primary alcohol residue of Cer with one or more saccharides through a β-glycosidic bond produces the glycosphingolipids. GalCer is produced from Cer and uridine-diphosphate(UDP)galactose by a reaction catalysed by ceramide galactosyltransferase, which is primarily localised in the ER of Schwann cells and oligodendrocytes in the brain (Gault et al., 2010). GalCer can undergo a further reaction catalysed by the GalCer sulfotransferase enzyme to produce SF, which are synthesised almost exclusively by oligodendrocytes (Gault et al., 2010, van Echten-Deckert and Walter, 2012). GlcCer, in contrast, is produced from Cer and UDP-glucose in a reaction catalysed by GlcCer synthase, which is located in the cytosolic leaflet of the Golgi apparatus. GlcCer is then transported directly to the plasma membrane, or undergoes subsequent glycosylations, producing lactosylceramide, which in turn can generate more complex sphingolipids, namely the gangliosides, as outlined in Figure 1.4 (Gault et al., 2010, Giussani et al., 2014). The main gangliosides found in the brain belong to the ‘a ganglio series, which is characterised by the tetrasaccharide β-Gal-1,3-β-GalNAc-1,4-β-Gal-1,4-β-Glc to which residues of sialic acid (derivatives of neuraminic acid) are α-ketosidically linked (Naudí et al., 2015).

1.1.2.4. Sphingolipids: Function Like glycerophospholipids, sphingolipids also play a key structural role in neural membranes. SM, which constitutes approximately 2-15% of the total lipids detected in mammalian tissue, is a major lipid found in myelin (O'Brien and Sampson, 1965a, Svennerholm and Vanier, 1973). Like PC, it is particularly concentrated in the outer leaflet of the membrane (Farooqui et al., 2000, Ramstedt and Slotte, 2002). SF, which are synthesised almost exclusively by oligodendrocytes, are another lipid constituent of the myelin sheath that surround the axons (Vos et al., 1994). Unlike most glycerophospholipids, however, sphingolipids tend to have longer, more saturated hydrocarbon chains. Thus, sphingolipids, like SM, readily associate with the rigid sterol ring of cholesterol, forming dynamic nanoscale micro-domains, known as lipid rafts, within the lipid membrane (Ariga et al., 2008, Lingwood and Simons, 2010). Hippocampal neurons in particular have a high SM content, increasing their ability to produce these lipid rafts, particularly in the axonal membrane (Brown and London, 1998, Simons and Toomre, 2000). Since lipid rafts are enriched with palmitolylated

transmembrane

proteins,

glycosylphosphatidylinositol

(GPI)-linked

proteins,

cholesterol-linked proteins like Hedgehog, and doubly acylated proteins like the Src-family kinases or the α-subunit of heterotrimeric G proteins, they favour specific protein-protein interactions, facilitating the activation of signalling cascades (Hakomori, 2000, Simons and Toomre, 2000). Cer 14 | P a g e

Chapter 1: General introduction and literature review serve to cluster receptors within these lipid rafts, enabling the critical density required for signalling to be achieved (van Blitterswijk et al., 2003, Mencarelli and Martinez–Martinez, 2013). Further, Cer can also bend the bilayer, which facilitates the formation of pores, and drives vesicle formation and budding (van Blitterswijk et al., 2003, Mencarelli and Martinez–Martinez, 2013) Unlike SF and SM, which are predominantly located in white matter, adult cerebral and cerebellar grey matter contain about three times more GMs than white matter (Svennerholm et al., 1989). GMs, particularly GM1, like SM, readily associates with cholesterol, forming lipid rafts (Ariga et al., 2008, Lingwood and Simons, 2010). In addition to its role in signal transduction via lipid rafts, GMs also play a role in cell-cell recognition and adhesion (Hakomori et al., 1998, Robert et al., 2011). Further, Svennerholm et al. (1989) reported a large accumulation of GMs in development, particularly during the period of dendritic arborisation and synaptogenesis, with a change in GM composition (i.e. an increase in GM1) in adulthood. Subsequent work has highlighted the role GM1 plays in neuronal differentiation, neuritogenesis, and calcium homeostasis (Hakomori and Igarashi, 1995, Ledeen et al., 1998, Ariga et al., 2008). Finally, there is evidence, mainly from in vitro and animal studies, that Cer, Cer-1-phosphate, sphingosine, and sphingosine-1-phosphate, may also play a role as secondary messengers in driving differentiation, growth, cell migration, and apoptosis (Spiegel and Milstien, 1995, Naudí et al., 2015).

1.1.2.5. Lipidomic changes in Alzheimer’s disease Now, more than a century after Alzheimer first described lipid aberrations in the AD brain, the advent of new tools and technologies has contributed to the growing number of publications detailing lipid dyshomeostasis in the human brain. Table 1.1 and Table 1.2 outline changes in specific glycerophospholipid and sphingolipid groups, respectively. In AD, PA is decreased in the superior temporal gyrus (STG), but remains unchanged in the frontal and parietal cortices (Nitsch et al., 1992, Pettegrew et al., 2001). PC is depleted in the hippocampus, frontal and parietal cortices, and in cerebrospinal fluid, but remains unchanged in the interior parietal lobe (IPL), subiculum, and hippocampal gyrus (Nitsch et al., 1992, Wells et al., 1995, Prasad et al., 1998, Pettegrew et al., 2001, Mulder et al., 2003). The levels of PE and pPE were decreased in all the regions that have been analysed to date in AD (Nitsch et al., 1992, Wells et al., 1995, Prasad et al., 1998, Han et al., 2001, Pettegrew et al., 2001). PI was decreased in the STG and hippocampal formation, particularly in synaptosomes (Wells et al., 1995, Prasad et al., 1998, Pettegrew et al., 2001). PS was decreased in the IPL, where it was externalised to the outer leaflet, particularly in synaptosomes, but was increased in the hippocampus and unchanged in the frontal and parietal cortices (Nitsch et al., 1992, Wells et al., 1995, Lange et al., 2008). Although the study design cannot

15 | P a g e

Chapter 1: General introduction and literature review be overlooked when considering the discrepancy seen in the PA, PC, and PS changes in AD, it highlights the need to study lipid composition variations in AD in a region-dependent manner. The decrease in PE, pPE, and PI is accompanied by a marked increase in the relative amount of saturated fatty acids (14:0, 16:0 and 18:0), and a decrease in PUFAs (20:4, 22:4, 22:6; Söderberg et al., 1991, Han et al., 2001). Since PUFAs are easily oxidisable, neural and glia cell membranes, which have a high concentration of PUFA-rich PEs and PIs, are primary targets for oxidation and the subsequent lipid peroxidation in AD (Naudí et al., 2015). The frontal cortex and hippocampus in particular, which are highly atrophied in AD, have a high peroxidisability index (Naudí et al., 2012). The elevation of lipid peroxidation products, such as 4-hydroxynonenal and acrolein, can inhibit proteins that actively maintain phospholipid asymmetry, which may explain the externalisation of PS in AD (Lange et al., 2008). This disruption in membrane homeostasis can lead to neuronal dysfunction and cell death, which can be detected even during early mild cognitive impairment (Lange et al., 2008). Further, even small alterations in pPE content is known to affect the critical temperature, disrupting membrane stability, which could lead to loss of synapses seen early in AD (Ginsberg et al., 1998, Han et al., 2001). In addition to the effect on membrane structure, the role glycerophospholipids play as second messengers are also affected by their change in AD. These effects have been previously outlined in detail by Frisardi et al. (2011). Briefly, the substantial decrease in PUFAs, namely arachidonic acid (AA) and docosahexaenoic acid (DHA), can impede production of active metabolites such as prostaglandins and leukotrienes, which normally play an anti-inflammatory role (Söderberg et al., 1991, Frisardi et al., 2011, Bazinet and Layé, 2014). Further, DHA is known to inhibit β-secretase, thus preventing the amyloidogenic processing of APP. However, a decrease in the DHA pool may exacerbate this pathway, driving Aβ aggregation (Bazinet and Layé, 2014). Finally, the loss of PI may also account for the impaired phosphoinositide second messenger system seen in AD (Jope et al., 1994, Martın et al., 2010). While this affects signalling pathways related to synaptic survival, plasticity, and long-term potentiation (Osborne et al., 2001), Fowler (1997) also proposes that this will decrease Protein kinase C activity, driving the production of Aβ. However, this remains to be confirmed in vivo in humans.

16 | P a g e

Chapter 1: General Introduction and Literature Review Table 1.1: Glycerophospholipid changes in the human brain in Alzheimer’s disease. An overview of findings from previous publications. 31

Analytical Method

Sample Size

Brain Regions

Change

References

Silica gel chromatography

10 Control; 10 AD

Frontal and parietal cortices

Unchanged

Nitsch et al. (1992)

16 Control ; 40 AD

STG

Decreased

Pettegrew et al. (2001)

Frontal and parietal cortices

Decreased

HP

Decreased in SPM fraction

IPL, Subiculum, HPG

Unchanged

Frontal cortex

Decreased

Temporal cortex

Specific decrease in PC (18:0/22:6)

Frontal and parietal cortices

Decreased

HP

Decreased

Phosphatidylserine (PS)

Phosphatidylinositol (PI)

Ethanolamine plasmalogen (pPE)

Phosphatidylethanolamine (PE)

Phosphatidylcholine (PC)

Lipid Class Phosphatidic Acid (PA)

P NMR, Phosphorus-31 nuclear magnetic resonance; ESI-MS, electrospray ionisation-mass

spectrometry; GM, grey matter; HP, hippocampus; HPG, hippocampal gyrus; HPLC, High Performance Liquid Chromatography; IPL, inferior parietal lobe; SPM, synaptosome plasma membrane; STG, superior temporal gyrus

31

P NMR analysis

Silica gel chromatography HPLC Gas chromatography HPLC LC-MS (ntot=18) MALDI-IMS (ntot=2) Silica gel chromatography HPLC Gas chromatography HPLC 31

P NMR analysis HPLC ESI-MS

10 Control; 10 AD 3 Control; 4 AD 9 Control; 9 AD 13 Control; 15 AD 10 Control; 10 AD 10 Control; 10 AD 3 Control; 4 AD 9 Control; 9 AD 13 Control; 15 AD 16 Control ; 40 AD 13 Control; 15 AD 6 Control; 6 AD

IPL, Subiculum, HPG Frontal cortex, HP STG Frontal cortex, HP Frontal, parietal and temporal cortices

Decreased Decreased Decreased Decreased

Nitsch et al. (1992) Wells et al. (1995) Prasad et al. (1998) Guan et al. (1999) Yuki et al. (2014) Nitsch et al. (1992) Wells et al. (1995) Prasad et al. (1998) (Guan et al., 1999) Pettegrew et al. (2001) Guan et al. (1999)

Decreased

Han et al. (2001)

16 Control ; 40 AD

STG

Decreased

Pettegrew et al. (2001)

HPLC

3 Control; 4 AD

HP

Decreased in SPM fraction

Wells et al. (1995)

Gas chromatography

9 Control; 9 AD

Subiculum, HPG

Decreased

Prasad et al. (1998)

16 Control; 40 AD

STG

Decreased

Pettegrew et al. (2001)

Frontal and parietal cortices

Unchanged

HP

Increased

IPL

Decreased

IPL

PS exposure on outer leaflet

31

P NMR analysis

31

P NMR analysis

Silica gel chromatography HPLC 31

P NMR analysis

Fluorescence assay; Western Blotting

10 Control; 10 AD 3 Control; 4 AD 7 Control; 37 AD 5 Control; 5 AD

Nitsch et al. (1992) Wells et al. (1995) Pettegrew et al. (2001) Lange et al. (2008)

17 | P a g e

Chapter 1: General introduction and literature review Table 1.2 outlines the changes in sphingolipids in the human brain in AD. Generally, there was an elevation in ceramide levels in AD in all the regions studied to date, except in the white matter of the medial frontal gyrus (Han et al., 2002, Cutler et al., 2004, Bandaru et al., 2009, He et al., 2010). However, little is known about the distribution of ceramide in the human hippocampus, and if the level of ceramide changes in this region in AD. The increase in ceramide could be linked to the accelerated degradation of other sphingolipids species (Goñi and Alonso, 2002, van Echten-Deckert and Walter, 2012). Although there is also a reported increase in SM (Wells et al., 1995, Pettegrew et al., 2001, Bandaru et al., 2009), Nitsch et al. (1992) found unchanged levels of SM in the frontoparietal cortex, and He et al. (2010) reported decreased levels of SM in the fronto-temporal cortex. Given that SM is found abundantly in white matter, this discrepancy might reflect the differential composition of the samples analysed in each study. These findings remain to be confirmed in the hippocampus. Although early studies report increased SF in AD (Majocha et al., 1989), there is now evidence of significant SF depletion (up to 58% in white matter and up to 93% in grey matter), even at the early stages of disease (Han et al., 2002, Cheng et al., 2013). This decrease can be linked to greater SF degradation and an increase in its transport from oligodendrocytes (Wood, 2012). Neither the composition ratio of SF species, nor their patterns of distribution, however, is significantly different in AD, at least in the human cerebral cortex (Yuki et al., 2011). Finally, elevated levels of GM1, GM2 and GM3, have been reported in hippocampal grey matter, the frontal cortex, and the entorhinal cortex, respectively (Valdes-Gonzalez et al., 2011, Chan et al., 2012). Svennerholm and Gottfries (1994), however, found a decrease in hippocampal gangliosides in AD, which might reflect the decrease found in b-series GM alone (Valdes-Gonzalez et al., 2011, Chan et al., 2012, Pernber et al., 2012). Very recently, Hirano-Sakamaki et al. (2015) reported a specific decrease in the GM1 d20:1/C18:0 in relation to GM1 d18:0/C18:0 in AD, in the molecular layer of the dentate gyrus.However, Pernber et al. (2012) reported a decrease in GM1 in the frontal cortex using immunohistochemistry. Further,

Svennerholm and Gottfries (1994), also found a decrease in

gangliosides in AD in the human hippocampus, however, this might reflect the decrease found in bseries GM alone (Valdes-Gonzalez et al., 2011, Chan et al., 2012). The change in sphingolipids seen in AD affects the structural role it plays in the brain. For instance, the decrease in SF reflects the hypomyelination seen in mild cognitive impairment and AD patients (Wood, 2012). Further, given that Cer, SMs, and GMs, are core constituents of lipid rafts, the

18 | P a g e

Chapter 1: General introduction and literature review Table 1.2: Sphingolipid changes in the human brain in Alzheimer’s disease. An overview of findings from previous publications. 31

P NMR, Phosphorus-31 nuclear magnetic resonance; CDR, clinical dementia rating; CN, caudate nucleus; ESI, electrospray ionisation; HP, hippocampus; HPG, hippocampal gyrus; HPLC, high performance liquid Lipid Class

Analytical Method

Sphingomyelin (SM)

Ceramide (Cer)

ESI-MS ESI-MS ESI-MS/MS HPLC Silica gel chromatography HPLC 31

P NMR analysis ESI-MS/MS HPLC

Sulfatide (SF)

HP-TLC

Brain Regions

Change

5 Control; 3 AD (CDR = 0.5) 7 Control; 7 AD 30 Control; 30 AD 6 Control; 9 AD 10 Control; 10 AD 3 Control; 4 AD 16 Control ; 40 AD 30 Control; 30 AD 6 Control; 9 AD 7 Control; 5 AD

Temporal and cerebellar WM

Increased

MFG

Increased C24:0 species

MFG grey matter MFG WM Frontotemporal grey matter Frontal/parietal grey matter Parietal/temporal /frontal cortices Cerebellum and IPL

Increased C24:0 Decreased

MFG grey matter

Increased

Frontotemporal grey matter

Decreased

Temporal cortex

Increased

5 Control; 3 AD (CDR = 0.5)

MALDI-IMS

3 Control; 3 AD

HP-TLC

Ganglioside (GM)

Sample Size

ESI-MS

MALDI-MS

chromatography; HP-TLC, high performance thin layer chromatography; IHC, immunohistochemistry; IPL, inferior parietal lobe; IMS, imaging mass spectrometry; MALDI, Matrix-assisted laser desorption/ionisation mass spectrometry; MFG, medial frontal gyrus; MS, mass spectrometry; SFG, superior frontal gyrus; WM, white matter

8 control; 6 AD (pre-clinical) Varied sample sizes; control vs EOAD vs LOAD

Increased Unchanged Increased Increased

References Han et al. (2002) Cutler et al. (2004) Bandaru et al. (2009) He et al. (2010) Nitsch et al. (1992) Wells et al. (1995) Pettegrew et al. (2001) Bandaru et al. (2009) He et al. (2010) Majocha et al. (1989)

Frontal/temporal/ parietal grey matter Frontal, temporal and parietal cortices (unclear)

Decreased

Han et al. (2002)

Hydroxylated : non-hydroxylated ratio unchanged

Yuki et al. (2011)

SFG

Decreased

Cheng et al. (2013)

Varied frontotemporal regions, CN and HP

Decreased (regional variation between EOAD and LOAD)

Svennerholm and Gottfries (1994)

TLC-Blot MALDI-MS

4 Control; 4 AD

HP grey matter

Increased GM1 Decreased b-series GM

ValdesGonzalez et al. (2011)

HPLC

8 Control; 10 AD

Entorhinal cortex

Increased GM3 Decreased b-series GM

Chan et al. (2012)

IHC

5 Control; 5 AD

Frontal grey matter

Decreased GM1 Increased GM2

Pernber et al. (2012)

MALDI-IMS

2 Control; 2 AD

Molecular layer of dentate gyrus

Decreased GM1 d20:1/C18:0 to GM1 d18:0/C18:0 ratio

HiranoSakamaki et al. (2015)

19 | P a g e

Chapter 1: General introduction and literature review change in the composition of these sphingolipids also impacts the function of these microdomains (Fabelo et al., 2014, Díaz et al., 2015). Although pools of APP and presenilin are present in cell membranes, both in raft and non-raft regions, evidence that the amyloidogenic processing of APP primarily occurs in lipid rafts is now well-established and has been previously reviewed extensively (Cordy et al., 2006, Rushworth and Hooper, 2010, Di Paolo and Kim, 2011). This occurs because lipid rafts also contain a significant pool of BACE1 and γ-secretase, which undergo S-palmitoylation to target its transport to these microdomains (Benjannet et al., 2001, Vetrivel and Thinakaran, 2010). Cer helps stabilise BACE1 within lipid rafts, driving the production of Aβ1-40 and Aβ1-42 (Puglielli et al., 2003a). Aβ1-42 can subsequently bind to GMs, such as GM1, which alters the conformation of Aβ1-42 from random coils to more ordered structures with increased β-sheet content, leading to its aggregation (Yanagisawa et al., 1995, Kakio et al., 2002, Yanagisawa, 2007). Thus, early changes in sphingolipids may play a role in the production and formation of Aβ plaques, driving AD pathogenesis. Aβ aggregation in lipid rafts also drives the accumulation of phosphorylated tau in these microdomains, at least in a mouse model of AD (Kawarabayashi et al., 2004). Further, the presence of NFTs in the brain of patients with lipid storage disorders, such as Niemann-Pick disease, also suggests a link between lipid dyshomeostasis and tau pathology (Nixon, 2004). This link, however, in contrast to that between lipid dyshomeostasis and Aβ pathology, is more tenuous, and remains to be confirmed in the AD brain. Nonetheless, there is now evidence that most, if not all, classes of lipids are implicated in AD. The studies reviewed here, however, apart from those conducted by Yuki et al. (2011), Pernber et al. (2012) and Hirano-Sakamaki et al. (2015), are not conducive to visualising the anatomical distribution of these lipids in the brain regions that were analysed. Further, only five studies of the 19 that were reviewed, report changes in the hippocampus, a primary site of atrophy in AD (Wells et al., 1995, Guan et al., 1999, Valdes-Gonzalez et al., 2011, Chan et al., 2012, Hirano-Sakamaki et al., 2015). Thus, little is known about the changes in the level of Cer, SM, and SF that occur in this region in AD. Further, as outlined in the next section, the hippocampus consists of sub-fields, each with their own function. Therefore, there is merit in visualising the distribution of lipids in the hippocampus, to study the composition of each sub-field, and understand how changes in the composition of lipids might be implicated in AD.

20 | P a g e

Chapter 1: General introduction and literature review

1.1.3. The human hippocampus The hippocampus, which is part of the limbic lobe, is a bilaminar structure consisting of the Cornu Ammonis (CA) and the dentate gyrus (DG). While the hippocampal formation also includes the subiculum, presubiculum, parasubiculum, and entorhinal cortex (Anderson et al., 2007), these regions are not a central focus of this thesis. Much of the detailed CA and DG neuroanatomy, which is described in Section 1.1.3.1 and 1.1.3.2, respectively, is based on historical work utilising histological methods (Cajal, 1911, de Nó, 1928, de Nó, 1934, Duvernoy et al., 2013). More recently, techniques such as magnetic resonance imaging (MRI) and nuclear magnetic resonance spectroscopy (NMR) have helped extend the study of the hippocampus and how it changes in AD (Manganas et al., 2007, Yushkevich et al., 2009, Wisse et al., 2012, Winterburn et al., 2013). The central role that the hippocampal formation plays has been well-documented, ever since the famous report on patient HM who developed severe anterograde and partial retrograde amnesia, following bilateral removal of his hippocampi (Scoville and Milner, 1957). The hippocampus is also thought to play a role in spatial navigation; a theory first proposed by O'Keefe and Dostrovsky (1971). The hippocampal pathways that play a role in memory function will be outlined in Section 1.1.3.3.

1.1.3.1. Cornu Ammonis (CA) anatomy The CA, first used by Rafael Lorente de Nó and named using the Latin term for Ammon’s horn, is now known as the hippocampus proper (Dudek et al., 2016). Though the CA essentially has only one cellular layer, the plexiform layers above and below this layer have been given different names (Insausti and Amaral, 2004). Thus, the CA consists of six layers, from the deepest level (the ventricular cavity) to the surface (towards the vestigial hippocampal sulcus): 1) the alveus, 2) stratum oriens, 3) stratum pyramidale, 4) stratum radiatum, 5) stratum lacunosum, and 6) stratum moleculare, or molecular layer (ML; Duvernoy et al., 2013). The alveus is located at the intraventricular surface, and contains efferent axons from hippocampal and subicular neurons that enter the fimbria. The axons of these efferent neurons cross the stratum oriens, which also consists of scattered basket cells (Duvernoy et al., 2013). The limits of the stratum oriens in humans, however, is poorly defined as it blends with the underlying stratum pyramidale (Stephan and Manolescu, 1979). The stratum pyramidale is mainly composed of pyramidal neurons. However, there are also basket-type interneurons and stellate neurons scattered throughout this area (Olbrich and Braak, 1985, Duvernoy et al., 2013). The stratum radiatum contains axons from associational connections or Schaffer collaterals (Insausti and Amaral, 2004). The stratum lacunosum contains numerous axonal fasciculi, which mainly consists of performant fibres and Schaffer collaterals. Finally, the ML adjoins the vestigial hippocampal sulcus, and contains interneurons and the original arborisations of the apical dendrites of pyramidal neurons (Duvernoy et al., 2013). 21 | P a g e

Chapter 1: General introduction and literature review When studied coronally, the main element of the CA, the stratum pyramidale shows a heterogeneous composition due to the different aspects of its pyramidal neurons. Based on these differences, the stratum pyramidale can be further divided into four sub-fields: CA1, CA2, CA3, and

CA4 or the hilus, as illustrated in Figure 1.5 (Duvernoy et al., 2013, Winterburn et al., 2013). The human CA1 is a large area, which continues from the subiculum. It primarily contains small and scattered pyramidal somata, which are typically triangular in shape (Dam, 1979). The base of a CA1 pyramidal neuron soma faces the alveus, while the apex extends toward the vestigial hippocampal sulcus. Axons from these neurons mainly project to the septal nucleus; however, some of these projections can act as association fibres for other pyramidal neurons and often have Schaffer collaterals, which curve back into the stratum radiatum (Schaffer, 1892). The CA1 pyramidal neurons also contain apical dendrites that typically traverse the entire thickness of the CA, and basal dendrites that can arborise in the stratum oriens (Duvernoy et al., 2013). In contrast to the CA1 region, the CA2 area contains larger pyramidal cells, which are more ovoid and densely packed (Anderson et al., 2007, Duvernoy et al., 2013). Pyramidal neurons in the CA2 primarily project to the basal dendrites of CA1 pyramidal neurons, constituting almost 20% of the input to the stratum oriens (Dudek et al., 2016). The CA3 contains pyramidal cell bodies similar to the CA2 region, albeit at a lower density. The dendrites of these cells contain specialised thorny excrescences associated with input from mossy fibres from the DG (Duvernoy et al., 2013, Dudek et al., 2016). The primary output target of pyramidal neurons in the CA3 is to the apical dendrites of CA1 pyramidal neurons in an area known as the stratum radiatum (Dudek et al., 2016). Finally, the CA4 region contains pyramidal somata that are ovoid, larger, few in numbers and scattered amongst large, mossy myelinated fibres. While functionally similar to the CA3 region (Insausti and Amaral, 2004), the CA4 region can be distinguished from the CA3 region as it is situated within the concavity of the DG (Duvernoy et al., 2013). There have been several MRI-based volumetric studies on hippocampal volume (Yushkevich et al., 2009, Wisse et al., 2012, Winterburn et al., 2013). Wisse et al. (2012) studied 14 subjects and reported an average volume of 3.1 ± 0.84 cm3 for the whole hippocampus, 1.71 ± 0.84 mm3 for the CA1 region, 0.077 ± 0.027 cm3 for the CA2 region, 0.71 ± 0.07 cm3 CA3 region, and 0.93 ± 0.22 cm3 for the CA4/DG region.

22 | P a g e

Chapter 1: General introduction and literature review

Figure 1.5: Location and structure of the human hippocampus. (A) The location of the human hippocampus on a coronal whole brain structure. (B) Stylised diagram of an enlarged view of hippocampal structures. The hippocampus has two main components: the Cornu Ammonis (CA) and the dentate gyrus (DG). CA1-4: fields of the Cornu Ammonis: 1) alveus, 2) stratum oriens, 3) stratum pyramidale, 4) stratum radiatum, 5) stratum

lacunosum, 6) stratum moleculare or molecular layer (ML). DG: 7) ML, 8) stratum granulosum or granule cell layer (GCL), 9) polymorphic layer (PL). (C) A coronal human hippocampus section from a control case stained with haematoxylin, eosin and luxol fast blue, with an indication of white matter (WM), CA, and DG regions. Inset: An enlarged view of the DG, indicating the PL, GCL and ML (C, inset).

23 | P a g e

Chapter 1: General introduction and literature review

1.1.3.2. Dentate gyrus (DG) anatomy The hippocampal sulcus, which becomes vestigial after development, separates the DG from the CA1-3 regions, while the concavity of the DG envelopes the CA4 area (Duvernoy et al., 2013). The DG, with its characteristic U-shape, is a trilaminar structure (Insausti and Amaral, 2004, Anderson et al., 2007). It consists of: 1) the strata molecular, which is also known as the ML, 2) the strata granulosum, or the granule cell layer (GCL) and 3) the polymorphic layer (PL; Duvernoy et al., 2013). This basic formation of the DG is conserved across many species, including rat, monkey, and human (Anderson et al., 2007). The ML is a relatively cell-free layer that mainly consists of fibres that originate in the entorhinal cortex, along with some interneurons (Amaral et al., 2007). The GCL is the principal cell layer, which consists of small, round granule cells that are densely packed (Amaral et al., 2007, Duvernoy et al., 2013). These granule cells send distinct axonal projections, known as mossy fibres, to the CA3 and CA4 regions, and may also extend basal dendritic trees to the final layer of the DG, i.e. the PL, which is also known as the plexiform layer (Insausti and Amaral, 2004, Duvernoy et al., 2013). While the PL is traversed by granular cell projections, it also contains a number of other cells, of which the most prominent is the mossy cell (Amaral et al., 2007, Duvernoy et al., 2013). Although the volume of the DG and the hippocampus is approximately 100 times larger in humans than rats, there are only about 15 times more granule cells in humans in comparison to rats (Amaral et al., 2007). Nonetheless, the human GCL is estimated to have approximately 15 x 106 granule cells (Insausti and Amaral, 2004). Granule cells continue to be generated, even in adulthood, from a population of continuously dividing progenitor cells that are found in the subgranular zone, a one-cell thick area at the GCL-polymorphic layer border (Eriksson et al., 1998).

1.1.3.3. Functional circuitry of the hippocampal formation in memory The hippocampus plays a central role in converging information from different cortical regions to rapidly form context-dependent memories (Rolls, 1996). The pathways that are central to this role are categorised as the polysynaptic pathway and the direct pathway, based on the organisation of the intrahippocampal fibres (Duvernoy et al., 2013). Figure 1.6 provides an overview of both pathways. The entorhinal cortex is the principal input to the hippocampus in both pathways (Insausti et al., 1987, Witter and Groenewegen, 1992). In the polysynaptic pathway, the entorhinal cortex first receives input from the posterior parietal association cortex and its neighbouring temporal and occipital cortices, via the parahippocampal gyrus. Projections from layer II of the entorhinal cortex, which make up the perforant path, then travel caudally, perforating the subiculum and terminating in the outer two-thirds of the GCL (Insausti and Amaral, 2004, Duvernoy et al., 2013). Since the perforant path is made up of glutamatergic fibres, it provides a powerful, excitatory action on the DG. Then, the distinctive axons

24 | P a g e

Chapter 1: General introduction and literature review of the granule cells, i.e. the mossy fibres, in turn provide glutaminergic input to the proximal dendrites of the pyramidal cells in CA3 and CA4. The axons of CA3 and CA4 then enter the alveus and fimbria (Witter, 2007). However, Schaffer collaterals of these CA3 and CA4 pyramidal neurons also reach other levels of the same areas and the apical dendrites of CA1 in the strata radiatum and lacunosum (Insausti and Amaral, 2004). CA1 axons that enter the alveus also produce collaterals, which reach the subiculum. The efferent axons from CA1, which travel via the alveus and then fimbria of the fornix, however, are considered the main output of the hippocampus. These efferent axons from the fornix then reach the anterior thalamic nucleus, either directly or via the mammillary bodies, before reaching the posterior cingulate cortex, retrosplenial cortex, and anterior cingulate cortex (O'Keefe and Nadel, 1978, Devinsky and Luciano, 1993). The polysynaptic pathway plays a key role in episodic memory, the memory of facts in relation to one another, and in memorising spatial perception (Duvernoy et al., 2013). While the direct pathway has received less attention, in comparison to the polysnaptic pathway, it nonetheless plays a crucial role in semantic memory in humans (Insausti and Amaral, 2004, Duvernoy et al., 2013). The fibres in this pathway originate from neurons in layer III of the entorhinal cortex, which itself receives input from the inferior temporal association cortex, via the perirhinal cortex, in a topographically organised manner (Insausti et al., 1987, Witter and Groenewegen, 1992). The afferent fibres from the layer III neurons project almost exclusively to pyramidal neurons in CA1. The axons from the CA1 neurons then project onto the subiculum, and returning to the deep layers of the entorhinal area, before projecting to the inferior temporal association cortex, the temporal pole, and the prefrontal cortex (MacLean, 1992, Duvernoy et al., 2013).

1.1.3.4 The human hippocampus in Alzheimer’s disease In AD, the accumulation of Aβ and hyperphosphorylated tau leads to the disruption of axonal transport, inducing widespread metabolic decline. This leads to neuronal loss, and thus, the profound atrophy of the temporoparietal association cortices and the medial temporal lobe. The entorhinal cortex and hippocampus, in particular, undergo atrophy, to the greatest extent, and at the earliest stages of disease (Thompson et al., 2003). Within the hippocampus, the CA1 shows the most extensive damage, with a pyramidal cell loss of about 58-68% (West, 1993, Fukutani et al., 1995, West et al., 2000, Rössler et al., 2002). West et al. (2000) also report a moderate (29%) but significant loss of cells in CA4. Although there is no loss of granule cells in the DG in AD (West et al., 2000), a loss of dendritic processes (Flood et al., 1987), and a decline in the density of synaptic contacts in the ML (Scheff and Price, 1998), has been previously described. Further, there have been a number of publications on changes in proteins (Jansen et al., 1990, Dewar et al., 1991, Schonberger et al., 2001)

25 | P a g e

Chapter 1: General introduction and literature review

Figure 1.6: Schematic diagram outlining the functional circuitry of the hippocampal formation in memory in the human brain. Blue: Polysynaptic pathway. The parietal lobe, neocortex, and prefrontal cortex, project to the entorhinal cortex, via the parahippocampal gyrus. Glutamatergic fibres from layer II of the entorhinal cortex then travel via the perforant path, which perforates the subiculum, and terminates in the outer two-thirds of the granule cell layer (GCL). The GCL provides glutaminergic input to axons of CA3 and CA4, which reach apical dendrites in CA1, which in turn project to the subiculum. The main output, however, is to the fornix, via the alveus and fimbria. The axons from the fornix then reach the anterior thalamic nucleus, either directly or via the

26 | P a g e

mammillary bodies, before reaching the cingulate cortex. Red: Direct Pathway. The entorhinal cortex receives input from the inferior temporal association cortex via the perirhinal cortex. Fibres from layer III then project almost exclusively to CA1 neurons, before reaching the subiculum, which project back to the cortex, via the entorhinal cortex. Solid arrows represent forward connections, while dashed arrows represent back projections (particularly to the cortex from the subiculum, via the entorhinal cortex). Figure adapted from Rolls (1996). CA, Cornu Ammonis

Chapter 1: General introduction and literature review and metals (Becker et al., 2005), in the human hippocampus, in AD. There has, however, only been one study on lipidomic changes in particular sub-fields of the hippocampus in AD (Hirano-Sakamaki et al., 2015). In this study Hirano-Sakamaki et al. (2015) only report the change of two GM1 species in relation to each other in the DG alone. In contrast, I optimised the central technique used in their investigation, i.e. matrix-assisted laser desorption/ionisation (MALDI)-imaging mass spectrometry (IMS), and developed an analysis workflow that enabled me to detect the differential expression of multiple lipid classes in AD, in the cortex and across all the hippocampal sub-fields.

1.1.4. Matrix-assisted

laser

desorption

ionisation–imaging

mass

spectrometry 1.1.4.1. Matrix-assisted laser desorption/ionisation (MALDI)–time of flight (TOF) mass spectrometry (MS) The direct laser desorption of intact endogenous biomolecules, which are strongly absorbing, is usually limited to masses that are less than 1500 daltons (Da). However, when non-absorbing molecules are irradiated with a laser, this produces excessive fragmentation, due to the surplus of deposited energy (Karas et al., 1990). To address these challenges, Hillenkamp and colleagues introduced an absorbing matrix, which reduced the irradiation required for desorption by about a tenth (Karas et al., 1985), thus, developing the MALDI technique (Karas et al., 1985, Karas and Hillenkamp, 1988, Karas et al., 1990). For MALDI-MS, the analyte is mixed with an abundance of matrix, which is an organic compound of crystallized molecules. It acts like a buffer between the fragile analyte and powerful laser, and it helps ionise the biomolecules in the sample, which can then undergo mass measurement. According to Sparvero et al. (2012), a good matrix will typically have the following characteristics: 1. A strong optical absorption of the laser energy upon irradiation by a UV-laser (typically a 337 nm N2 laser2). 2. A fairly low molecular weight, which ensures easy evaporation, but one that is still large enough so that the matrix is stable under high vacuum. 3. Ability to promote the ionisation of analytes and isolation of ions, with minimal cluster formation. 4. Ability to form a homogeneous surface layer with analyte incorporation. 5. Typically contains a chromophore, which is a useful visual aid.

27 | P a g e

Chapter 1: General introduction and literature review In contrast to other ionisation methods, a pulsed ion generation rather than a continuous one is produced in MALDI. As illustrated in Figure 1.7, when the pulsed laser beam hits the sample, part of the matrix is vaporised, carrying intact ionised analyte molecules into the vapour phase, which then undergo mass measurement (Schiller et al., 2004). Usually, a TOF analyser, with an ion detector at the end of the flight tube, enables one to determine an ion’s mass-to-charge ratio (m/z). The length of the field-free time-of-flight pathway determines the mass resolution, with equally charged low mass ions being detected earlier than equally charged high mass ions (Caprioli et al., 1997, Schiller et al., 2004). Mass resolution can be improved with the use of longer flight tubes, and by the use of the reflector ion mode, which allows for a longer flight-path, as seen in Figure 1.7. The resolving power (Δm) of a typical linear TOF mass spectrometer is 5 000 (Watson and Sparkman, 2007), while the resolving power in the reflector mode in current instruments can be up to 40 000 (Bruker, 2016). Greater mass resolution allows the generation of isotopically resolved spectra (Schiller et al., 2004). Finally, it is important to note that the possible drift in the calibration of the different components could lead to a systematic error when measuring the analyte m/z (Rubakhin and Sweedler, 2010). Thus, it is crucial to calibrate recorded masses to known standards. Calibrants may be external, i.e. a mixture of standards such as red phosphorus that is analysed separate from the sample, or internal, i.e. endogenous analytes in the sample with a known m/z. The latter is preferable as it accounts for other external factors, such as sample thickness or analyte environment (Rubakhin and Sweedler, 2010). The limitation of MALDI, and other traditional MS techniques, however, is that they require homogenisation and purification during preparation, which leads to the loss of detail about specific anatomical distribution of biomolecules of interest (Seeley and Caprioli, 2008, Veloso et al., 2011a, Veloso et al., 2011b). Thus, these techniques need to be accompanied by histological and immunohistochemical techniques to visualise the anatomical distribution of endogenous biomolecules in a complex region of interest. Imaging mass spectrometry (IMS) can be used to overcome this limitation.

28 | P a g e

Chapter 1: General introduction and literature review

Figure 1.7: Principle of matrix-assisted laser desorption/ionisation (MALDI)-time of flight (TOF) mass spectrometry. (A) The matrix creates a mixture with analyte molecules of the fresh, frozen brain tissue. When placed in the instrument, brief UV laser pulses hit the mixture, desorbing and ionising molecules. (B) The ionised molecules (denoted by the + sign) are accelerated through the electrostatic field in the ion source and directed to the flight tube.

These molecules are then separated based on mass and charge, as they accelerate at different speeds. The reflector mode enables better spectral resolution as the ionised molecules travel in the flight tube for a longer duration of time. The intensity of each ionised molecule is measured as it is detected, resulting in a mass spectrum. m/z,

mass

to

charge

ratio;

UV, ultraviolet.

29 | P a g e

Chapter 1: General introduction and literature review

1.1.4.2. Imaging mass spectrometry The use of MALDI as an imaging mass spectrometry (IMS) method, which was introduced by Caprioli et al. (1997), allows thin tissue sections to be raster scanned, with mass spectra acquired in two dimensions. Thus, the anatomical distribution of any biomolecule of interest in the mass spectrum can be directly correlated with the underlying histology (Stoeckli et al., 2001, Seeley and Caprioli, 2008). The application of MALDI-IMS has been demonstrated for the detection of lipids (Veloso et al., 2011b), peptides (Hanrieder et al., 2012), proteins (Anderson et al., 2015), and administered pharmaceuticals (Takai et al., 2013). Further, its use to complement histopathological evaluation has been previously demonstrated in the field of oncology (Rauser et al., 2010, Calligaris et al., 2015), which has been extensively reviewed by Kriegsmann et al. (2015). Longuespée et al. (2016) has also reviewed the application of MALDI-IMS to other clinical fields such as neurology, ophthalmology, diabetology, and cardiology. Depending on the set spatial resolution, a typical data array can contain approximately 1,000–30,000 spots. Each spot contains the intensity of desorbed ions within the molecular m/z range of focus, which can extend from about 500 Da to over 80 kDa (Stoeckli et al., 2001). Thus, molecular weightspecific maps of the samples at any desired molecular weight value can be generated, with hundreds of discrete molecular weight image maps being produced from a single raster of a piece of tissue (Stoeckli et al., 2001). This is especially advantageous for researchers using postmortem human samples as it enables the economical use of precious tissue. The relative speed, convenience (as one can automate the analysis of many samples and sites), and high sensitivity, are some of the other benefits of using MALDI-IMS (Schiller et al., 2004). Additionally, the lack of a priori knowledge allows it to be used as a tool for novel biomarker detection (Debois et al., 2010, Grey et al., 2010). A typical experimental workflow for a MALDI-IMS experiment is illustrated in Figure 1.8. The four key principles of MALDI-IMS: sample preparation, matrix application, data acquisition, and data analysis (Zaima et al., 2010) have been outlined in detail in Sections 1.1.4.3–1.1.4.6.

1.1.4.3. Sample preparation The preparation of samples for MALDI-IMS must ensure tissue integrity, while avoiding the delocalisation and degradation of analytes (Walch et al., 2008). MALDI-IMS commonly utilises fresh, frozen tissue to bypass any ion suppression effects that the standard fixation process may generate (Walch et al., 2008). Nonetheless, the feasibility of its use for proteomic work on formalin-fixed

30 | P a g e

Chapter 1: General introduction and literature review

Figure 1.8: Schematic outline of the matrix-assisted laser desorption/ionisation (MALDI)-imaging mass spectrometry (IMS) workflow. (A) Fresh, frozen tissue is cut and mounted on to conductive target, i.e. a MALDI target plate, which is shown in this figure, or an indium-tin oxide (ITO)-coated slide. The matrix is applied using sublimation. (B) The region of interest is then raster-scanned with a user-defined spacing between spectra. (C) A MALDI mass spectrum is acquired at each x,y coordinate. (D) The acquired data can then be used to produce a spectral intensity

summary of a user-delineated region of interest. (E) The relative intensity of an observed ion can then be plotted as a function of the sampling location, allowing the user to plot distribution images. The images show the distribution of five different ions, which were all acquired in the same dataset. The heat intensity colour scale indicates the relative intensity of the chosen ion at each point in the region of interest. m/z, mass to charge ratio

31 | P a g e

Chapter 1: General introduction and literature review paraffin embedded tissue has been successfully demonstrated recently (Gorzolka and Walch, 2014, De Sio et al., 2015). The optimal thickness of tissue sections used for MALDI-IMS is between 1020 µm, which allows the easy manipulation of the sample without adversely affecting analysis (Caldwell and Caprioli, 2005). Once the tissue section is cut, it is thaw-mounted on either a conductive MALDI plate or indium tin oxide (ITO)-coated glass slides. The advantage of using ITOcoated slides is the ability to perform optical microscopy and MS on the same sample (Chaurand et al., 2004, Walch et al., 2008).Once the tissue has been mounted, it can be treated with solutions, such as ammonium formate or ammonium acetate, to enhance signal detection (Angel et al., 2012). Samples that are being prepared for protein analysis will also often be washed with a series of solvents, including alcohols and Carnoy’s fixative (1887), to remove salts and abundant lipids that can hinder protein detection (Agar et al., 2007, Watson and Sparkman, 2007). Samples can also be treated with trypsin to digest proteins with higher molecular weights, which may otherwise not be detected given their low abundance and poor ionisation (Groseclose et al., 2007, Stauber et al., 2010).

1.1.4.4. Matrix application Matrix selection plays an important role in obtaining high-quality mass spectra and MALDI images (Schwartz et al., 2003). Given the large range of available matrices, the ideal matrix and optimal tissue coating conditions must be empirically determined for each individual experiment (McDonnell and Heeren, 2007). An early review by Zenobi and Knochenmuss (1998) provides an in-depth summary of general recommendations of matrices for each molecular class. More recently, Thomas et al. (2012) evaluated the use of a novel matrix, 1,5-diaminonaphthalene (DAN), for imaging lipids in particular, with significant results in both positive and negative polarities. The use of DAN to analyse lipids has also been successfully replicated by other recent studies (Anderson et al., 2014, Korte and Lee, 2014, Weishaupt et al., 2015). To obtain high-quality MALDI-IMS, the entire surface of the tissue region of interest must be homogeneously coated with a thin layer of matrix (Schwartz et al., 2003, Römpp and Spengler, 2013). Early matrix application was first done using manual methods such as dipping tissue sections in matrix solutions (Caprioli et al., 1997), and spraying using an airbrush or thin layer chromatography sprayer (Todd et al., 2001). However, given the poor reproducibility of these procedures from sample-to-sample, there was a shift to using automated devices that either spot (e.g. Labcyte Portrait, Shimadzu ChiP) or spray (e.g. HTX TM-Sprayer, Bruker Daltonics ImagePrep) the matrix onto the sample (Walch et al., 2008). The use of a modified inkjet printer for matrix application has also been previously reported (Baluya et al., 2007). Nonetheless, all these matrix deposition techniques, both manual and automated, require a solution of matrix that is dissolved in a mixture of water and

32 | P a g e

Chapter 1: General introduction and literature review organic solvents, such as methanol, ethanol, and acetonitrile (Schwartz et al., 2003). Given the low molecular weight of lipids, and their high solubility in organic solvent commonly used in these protocols, matrix deposition by these techniques could lead to the inadvertent delocalisation of analytes (Hankin et al., 2007, Watson and Sparkman, 2007). The method of vapour-phase deposition of matrix via sublimation, which was introduced by Kim et al. (1998) and first successfully applied to a MALDI-IMS experiment by Hankin et al. (2007), overcomes this challenge. Additionally, in comparison to dried-droplet methods, sublimation produces a more homogeneous coating with considerably smaller crystals, which increases the quality of the data acquired (Schwartz et al., 2003, Jaskolla et al., 2009). The successful use of sublimation as a matrix deposition method, in experiments analysing lipids, has been previously demonstrated (Wang et al., 2008, Murphy et al., 2011, Thomas et al., 2012, Anderson et al., 2014).

1.1.4.5. Data acquisition The first commercial instruments with MALDI-IMS capability were introduced in 2004. There have been many advances since then, with companies such as Bruker, ABSciex, Shimadzu, Thermo Fisher Scientific, and Waters, producing many of the instruments that are currently used (MS-Imaging Society, 2015). To acquire a dataset, the MALDI laser is set to raster scan a user-determined region of interest, at a set spatial resolution, which can range from approximately 10-200 µm, in the x and y planes. A mass spectrum is collected at each sampling location. When analysing lipids, the majority of phospholipid ions are detected within a m/z range of 700–950, while fatty acyls are detected lower at m/z 250– 350 (Schiller et al., 2004). Gangliosides, which have a larger mass, are detected from m/z 1100–2000 (Sugiyama et al., 1997). The quality of the acquired spectrum should be optimised for the mass range being analysed in each experiment. This can be done by adjusting the following instrument parameters: laser power, laser beam diameter, number of laser shots, pulse delay, and grid voltage (Schwartz et al., 2003, Watson and Sparkman, 2007, Rubakhin and Sweedler, 2010). The first two parameters, laser power and laser beam diameter, determine laser fluence (Chaurand et al., 2007). The laser fluence, which is measured in J/cm2, is the amount of energy delivered per unit area of the sample surface (Dreisewerd et al., 1995, Guenther et al., 2010). The threshold fluence is one that is required to achieve analyte ion signals with sufficient signal-to-noise ratio (s/n). Increasing the fluence above this threshold, by increasing the laser power, proportionally increases the intensity of analyte ion signals (Spengler et al., 1988). However, care must be taken, as the ion signal intensity can become saturated if the laser fluence is too high. This saturation is thought to be due to secondary processes such as ion fragmentation (Guenther et al., 2010).

33 | P a g e

Chapter 1: General introduction and literature review Laser fluence is also affected by the diameter of the laser beam. A larger diameter increases the intensity of analyte ion signals as more of the sample is ionised. However, the size of the ablated area at each point determines the effective spatial resolution. Thus, for high-resolution imaging, it is better to use a narrow beam. However, focussing the laser too tightly will far exceed the fluence required for ablation and ionisation (Chaurand et al., 2007). Additionally, a beam that is too narrow can limit the number of molecules being ionised at a time. Thus, a laser beam diameter less than 10 µm is rarely used for lipid IMS studies (Rubakhin and Sweedler, 2010). The number of ions that are desorbed and ionised when forming a detectable mass signal is known as the limit of detection (LOD) of the instrument (Koestler et al., 2008). The area that is ablated is exponentially related to the number of laser shots per sample site (Chaurand et al., 2007). Thus, averaging the signal from multiple laser shots allows more ions to reach the LOD of the instrument, improving the intensity of ion signals (Koestler et al., 2008). Typically, the spectrum for each sampling spot is acquired by averaging 50–100 laser shots (Chaurand et al., 2007). The fourth parameter that is used to improve the quality of the spectrum is the delayed extraction, which is the time between the laser pulse ionising the sample and the application of the full accelerating voltage to nascent ions (Schwartz et al., 2003). The delayed extraction is measured in nanoseconds (ns; Watson and Sparkman, 2007). Increasing the delayed extraction allows ions with lesser initial kinetic energy to traverse a greater distance in the electric field, thereby picking up as much kinetic energy as other ions of the same mass. Since all ions of the same mass have the same kinetic energy, they arrive at the detector at a similar time, producing a sharp peak, i.e. greater spectral resolution. Typically, the delayed extraction is set between 120–400 ns (Watson and Sparkman, 2007). The final parameter, the grid voltage, can be adjusted in conjunction with the delayed extraction to optimise the quality of the spectrum (Schwartz et al., 2003). It is typically set to about 72-78%, at least to detect lipids in the 500-2000 Da mass range, in reflector mode, using a Voyager Biospectrometry Workstation (Applied Biosystems, 2000).

1.1.4.6. Data analysis The analysis of the MALDI-IMS dataset, following data acquisition, can be separated into two distinct steps, which are pre-processing and data processing/statistical analysis (Norris et al., 2007). A typical mass spectrum consists of three fundamental components: 1) the intensity peaks, 2) a baseline, and 3) noise (Trede et al., 2011). The baseline and noise can be attributed to random analytical and technical variation, which is present even when care is taken during sample preparation and measurement. Analytical and technical variation arise from a number of factors including metastable matrix clusters, fluctuations in detector gain, ion source contamination, or

34 | P a g e

Chapter 1: General introduction and literature review micro-environment chemical inhomogeneities including salt and pH gradients and phospholipid structures (Deininger et al., 2011, Sun and Markey, 2011, Trede et al., 2011, Fonville et al., 2012). The baseline and noise can lead to spectral intensity variations, even within the same tissue sample (Williams et al., 2005). The aim of pre-processing MALDI-MS data is to distinguish the intensity peaks from the baseline and noise, thereby conditioning subsequent statistical analyses to examine differences in the biomolecule expression between samples alone (Norris et al., 2007). MALDI-MS data is usually always pre-processed using baseline correction and normalisation. The use of denoising, peak detection, and alignment, processes are optional (Seeley and Caprioli, 2008, Alexandrov, 2012). The baseline generally appears as an exponential that decays with the m/z value (Williams et al., 2005). The simplest and typically used baseline correction method consists of heuristically finding the baseline slope and offset and subtracting them from the signal (Yang et al., 2009, Sun and Markey, 2011). There are a number of different baseline correction mathematical algorithms, some of which have been discussed in detail by Yang et al. (2009) and Trede et al. (2011). The most frequently applied baseline correction functions are ones that are integrated into commonly used software, such as DataExplorer, FlexAnalysis, and SciLS (Norris et al., 2007). These mathematical models generally provide a good correction of the baseline for a m/z window of about 10 kDa (Norris et al., 2007). Normalisation aims to minimise spectral intensity variations, transforming measured intensities to comparable scale, without altering biological information (Norris et al., 2007, Deininger et al., 2011, Fonville et al., 2012). Normalisation algorithms can either be individual or collective (Norris et al., 2007). Arguably, the individual approach, where each spectrum is normalised to a reference that is independent of the collective dataset, such as an internal, labelled calibration molecule, would be ideal (Norris et al., 2007). However, to date, the majority of MALDI IMS studies have used the collective approach, where all profile spectra in a dataset are first processed into a singular representation against which each spectrum is then normalised (Veloso et al., 2011a, Veloso et al., 2011b, Anderson et al., 2015, Hong et al., 2015). Methods based on this approach, including total ion current (TIC), median, noise level, and variance, each have their own strengths and limitations as demonstrated by several recent studies (Norris et al., 2007, Deininger et al., 2011, Sun and Markey, 2011, Fonville et al., 2012). Nevertheless, the TIC normalisation method remains the most popular (Trede et al., 2011). In the TIC approach each spectrum intensity is divided by the sum of all its intensities (Trede et al., 2011). It is based on the assumption that there are comparable numbers of signals present in each spectrum, and may not be suitable for widely heterogeneous tissue types (Deininger et al., 2011). Additionally, TIC may not be appropriate for examining a small subset of

35 | P a g e

Chapter 1: General introduction and literature review peaks in a large dataset as those changes would be lost (Sun and Markey, 2011). Nonetheless, Deininger et al. (2011) recommend visually comparing exemplary m/z images, following normalisation, with the histologically stained section, to determine if the correct normalisation method has been applied. High-frequency chemical noise that is inherent to MALDI data can be further reduced by denoising the signal, which will significantly improve subsequent m/z visualisation (Trede et al., 2011). Since this noise is non-white and non-stationary, it is better to use techniques that involve wavelet transformation (WT) rather than Fourier transformation (Coombes et al., 2005, Alexandrov et al., 2009, Sun and Markey, 2011). WT describes the change in frequency distribution over time, thus, taking into account the non-stationary nature of high-frequency chemical noise (Sun and Markey, 2011). Further, edge-preserving imaging denoising, which adjusts the level of denoising to the local noise level and to the local scale of the features to be resolved, is also available and is a key feature of the SCiLs lab software (Alexandrov et al., 2010, Trede et al., 2011). Peak detection, which reduces the number of m/z values by neglecting those that correspond to baseline or noise, is another pre-processing method that is sometimes considered separately from those listed above (Trede et al., 2011). A number of peak detection methods have been implemented in commonly used MS software packages (Alexandrov, 2012) and have been reviewed extensively by Yang et al. (2009). Finally, while it is important to include a calibrant during data acquisition, large datasets acquired separately can be recalibrated using a different calibration profile before analysis (Rubakhin and Sweedler, 2010). This recalibration step removes experimental error, ensuring that all datasets being analysed are aligned with each other. The pre-processing methods and algorithms that were specifically applied to this work will be outlined in Chapter Three. Once data have been pre-processed, the intensity of any m/z signal observed in the mass spectrum can then be plotted as a function of sampling position, where each m/z signal represents a biomolecule of interest and each sampling position denotes one pixel in the resultant display. In its fullest extent, a single raster of a region of interest can generate hundreds of image maps, each at a discrete m/z value (Stoeckli et al., 2001). In addition to visualising the distribution of biomolecules of interest, MALDI datasets allow researchers to answer two analytical questions: 1) the classification of samples into two or more classes such as diseased/non-diseased, and 2) the identification of biomarkers that are characteristic to each class (Norris et al., 2007).

36 | P a g e

Chapter 1: General introduction and literature review

Section Two

Project overview and thesis aims

1.2.1 Rationale Given the structural complexity of lipids and the plethora of molecular species, the literature review demonstrates the power of chromatography and MS in studying lipidomic changes in AD tissue. These techniques, however, involve the extraction and purification of the sample, which can be timeconsuming, and eliminates the ability to visualise the anatomical distribution of lipids of interest (Woods and Jackson, 2006). To circumvent this limitation, I chose to use MALDI-IMS for this study, which has been previously used successfully to visualise the anatomical distribution of different lipids in the human brain (Veloso et al., 2011a, Veloso et al., 2011b, Yuki et al., 2011, Hirano-Sakamaki et al., 2015). I specifically wanted to address the lack of understanding of the lipid composition of hippocampal sub-fields, and its change in AD. However, since a typical MALDI-IMS dataset contains approximately 108 to 109 intensity values and is larger than 1 GB in size, the analysis and interpretation of this data can be mathematically, statistically, and computationally challenging (Trede et al., 2011). Therefore, before I could apply this method to study lipidomic changes in AD, I first needed to build an efficient data-analysis workflow. Finally, although MALDI-IMS is a powerful tool to generate qualitative data, it is still at best, a semiquantitative method, with the majority of studies reporting relative changes in biomolecules of interest. However, recently, Jadoul et al. (2015) quantified PCs in the rodent brain using a tissuemimetic model first introduced by Groseclose and Castellino (2013). Thus, I investigated the use of this approach to quantify lipids in the MTG and DG, and analyse its change in AD.

1.2.2 Aims The overall objective of this project is to understand lipid expression changes in the human brain in AD, particularly in the MTG and hippocampus which are adversely affected by the disease. The specific aims of this thesis are directed towards developing MALDI-IMS as a tool to study lipid changes in the human brain. The three specific aims of this thesis are to: 1. Develop MALDI-IMS to image lipids in the postmortem human brain, namely in the MTG, hippocampus and DG. 2. Develop a workflow to analyse large MALDI imaging datasets to find differences in lipid expression in AD. 3. Quantify lipids in spatially-defined regions in the postmortem human MTG and hippocampus in control and AD cases.

37 | P a g e

Chapter 1: General introduction and literature review

1.2.3 Originality of study Although MALDI-IMS has been used previously to image the distribution of lipids in the control human cortex and hippocampus (Veloso et al., 2011a, Veloso et al., 2011b), this is the first study to compare distribution maps of all lipid classes, at high spatial resolution, in both these regions in control and AD tissue. Further, in comparison to previous MALDI-IMS studies on the AD brain (Yuki et al., 2011, HiranoSakamaki et al., 2015), this study uses a larger sample size in the control and AD cohorts. The increased number of cases, led to the generation of large datasets, with greater biological variation within each cohort. Therefore, I first developed an analysis workflow that allowed the efficient evaluation of large datasets. Using this workflow I was able to detect differential lipid expression, in the spatially-delineated sub-fields of the MTG and hippocampus, in AD. My results from the hippocampus study have been peer-reviewed and published (Mendis et al., 2016). Finally, since the use of MALDI-IMS as a tool for quantification is rare, it has never been used to quantify lipids in the postmortem human brain. Thus, this is the first time lipid-spiked tissue homogenates, an approach first introduced by (Groseclose and Castellino, 2013), has been used to quantify lipid classes of interest in the cortex and hippocampus in control and AD cases.

38 | P a g e

Chapter Two

Materials and methods

Section One: General.....................................................................................................................40 2.1.1. Acquisition and processing of postmortem human brain tissue ................................................. 40 2.1.1.1. Human Brain Bank ............................................................................................................................. 40 2.1.1.2 Processing of human brain tissue ...................................................................................................... 40 2.1.1.3 Cases used.......................................................................................................................................... 40

2.1.2. Matrix-assisted laser desorption/ionisation (MALDI)–imaging mass spectrometry (IMS).......... 43 2.1.2.1. 2.1.2.2. 2.1.2.3. 2.1.2.4. 2.1.2.5.

Tissue preparation ............................................................................................................................. 43 Choice of matrix ................................................................................................................................. 44 Matrix application ............................................................................................................................. 44 Instrument settings ............................................................................................................................ 45 MALDI imaging .................................................................................................................................. 47

2.1.3. Histological staining and imaging ................................................................................................ 47 2.1.3.1. Haematoxylin & eosin/luxol fast blue staining .................................................................................. 47 2.1.3.2. Imaging .............................................................................................................................................. 48

2.1.4. Lipid identification using tandem mass spectrometry (MS/MS) ................................................. 48 2.1.4.1. On-tissue matrix-assisted laser desorption/ionisation (MALDI)–tandem mass spectrometry (MS/MS) ............................................................................................................................... 48 2.1.4.2.Liquid-chromatography (LC) – tandem mass spectrometry (MS/MS) .................................. 49

Chapter 2: Materials and methods

Section One

General methods

2.1.3. Acquisition and processing of postmortem human brain tissue

2.1.3.1. Human Brain Bank The postmortem human brain tissue used for this thesis was obtained from the Neurological Foundation of New Zealand Human Brain Bank (Dep artment of Anatomy with Medical Imaging and Centre for Brain Research, University of Auckland). The use of this tissue was approved by the University of Auckland Human Participants Ethics Committee Ref #011654. All tissue was obtained with full informed consent of the families.

2.1.1.2 Processing of human brain tissue Fresh-frozen postmortem human brain tissue used in this study was processed according to a detailed, previously published protocol (Waldvogel et al., 2006, Waldvogel et al., 2008). In brief, once the brain is removed from the cranium, it is separated into the two hemispheres. One hemisphere is fixed with formalin perfusion, while the other is dissected into blocks that are snap-frozen with powdered dry ice, double-wrapped in foil and stored at -80°C. Both the fixed and unfixed hemispheres are dissected in the same way, with each 1-2 cm thick blocks representing the different functional regions. For instance, the sensory and motor cortex of the pre- and post-central gyrus, respectively, are separated together from the underlying white matter and divided into four equally sized blocks. An overview of the location of the specific blocks used in this study has been previously detailed by Waldvogel et al (2006).

2.1.1.3 Cases used Table 2.1 details the full list of control and Alzheimer’s disease (AD) cases used in this thesis, and Table 2.2 outlines the neuropathological information about the AD cases. Cases used to investigate lipid changes in the cortex in AD (Chapter Three). The results in Chapters 3-6 were based on six age- and sex-matched control and AD cases as outlined in Table 2.1. Given the evidence of the effect of sexual dimorphism on hippocampal size (Pruessner et al., 2001), and the well-known inverse relationship between age and hippocampal volume (Schuff et al., 1999), this coupling was important for the pair-wise analysis. The six control cases had no history of neurological disease and no evidence of neuropathology, and had an average age of 74.2 ± 6.46 years. There were three females and three males and the average postmortem delay was 21.8 ± 6.11 hours. The six AD cases had an average age of 73 ± 7.13 years, and also included three females and three males. The average postmortem delay for these cases was 4.8 ± 1.44 hours.

40 | P a g e

Chapter 2: Materials and methods Tissue for mass spectrometry analysis was taken from the middle temporal gyrus (MTG), which corresponds to Brodmann area 21. When first processed, the MTG is divided into four equally sized blocks, labelled 0-3 in the caudal-rostral direction (Waldvogel et al., 2006). Where possible, sections were taken from the MTG1 block. However, given the high demand for postmortem tissue, where the MTG1 block was unavailable, tissue was taken from the caudal end of the MTG2 or MTG3 block, as indicated in Table 2.1. Thus, sections were within 1 cm of each other in the coronal plane. Sections from the same cases and tissue blocks were also used for quantification (Chapter Five). Cases used to investigate lipid changes in the hippocampus AD (Chapter Four). Three control and three AD cases were first used to acquire pilot matrix-assisted laser desorption/ionisation (MALDI) – imaging mass spectrometry (IMS) data of the hippocampus. The three AD cases had an average age of 88.3 ± 7.23 years, with an average postmortem delay of 12.67 ± 4.86 hours, and included one female and two males. The severity of the pathological grade of AD was determined by a neuropathologist using the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) neuropathology criteria and Braak and Braak staging criteria (Table 2.2). The three control cases had no history of neurological disease, and no evidence of neuropathology, and had an average age of 87 ± 8.19 years, an average postmortem delay of 15.7 ± 6.66 hours), and included one female and two males too. The cases used to acquire the datasets that were utilised to determine lipid differences in the hippocampus, were the same as those used to investigate lipid changes in the cortex (outlined in Table 2.1). When first processed, the hippocampus is separated from the basal ganglia and divided into equal 1-cm blocks in a rostro-caudal direction, labelled consecutively starting from 0 (Waldvogel et al., 2006). The area of the hippocampus examined for this thesis is synonymous with Brodmann area 34. The HP2 block was used, where possible. However, when this was not possible, tissue was taken either from either caudal end of the HP1 block or the rostral end of the HP3 block. Sections were within 1 cm of each other in the coronal plane. Sections from the same cases and blocks were used for the higher spatial resolution imaging of the dentate gyrus and the quantification of selected lipids in this region (Chapter Six). Karelson et al. (2001) previously demonstrated that there is no significant correlation between the level of lipid peroxidation, total antioxidant capacity, reduced glutathione, superoxide dismutase, and catalase, and postmortem delays of less than 24 hours, in four cerebrocortical regions in AD and age-matched control brains. Thus, the postmortem delay of all the cases used in this thesis was limited to being less than 24 hours. Despite the large difference between the control and AD cases used in this thesis, there was no correlation between the postmortem delay of cases and the recorded intensities of the m/z values of interest.

41 | P a g e

Chapter 2: Materials and methods Table 2.1: Summary of control and Alzheimer’s disease (AD) cases used in this thesis. This table summarises the age, sex, and postmortem delay of the control and AD cases used in this thesis. The blocks used from each case, and their use, are also

Control

AD

42 | P a g e

indicated. Age- and sex-matched pairs have been listed consecutively in each section. HP, hippocampus; MTG, middle temporal gyrus

Blocks used in thesis

Use

Chapter Four, Pilot data

Case

Age (years)

Sex

Postmortem delay (hours)

H085 H156 H181

94 89 78

M M F

8 19 20

HP HP3 HP2

H137

77

F

21

HP1 MTG2

H152

79

M

18

HP2 MTG2

H169

81

M

24

HP2 MTG2

H180

73

M

33

HP3 MTG1

H190

72

F

19

HP2 MTG1

H238

63

F

16

HP3 MTG1

AZ92 AZ86 AZ82

93 92 80

M M F

11.5 8.5 18

HP2 HP1 HP2

AZ32

75

F

3

HP2 MTG3

AZ80

77

M

4.5

HP2 MTG1

AZ45

82

M

4.5

HP2 MTG2

AZ90

73

M

4

HP2 MTG1

AZ72

70

F

7

HP2 MTG1

AZ71

61

F

6

HP2 MTG1

Lipid changes in AD and lipid quantification

Chapter Four, Pilot data

Lipid changes in AD and lipid quantification

Chapter 2: Materials and methods Table 2.2: Summary of pathology of Alzheimer’s disease (AD) cases used in this thesis. Table summarising Braak and Braak stage, atrophy, tangles, age-related plaque scores (ARP), and Consortium to

Establish a Registry for Alzheimer's Disease (CERAD) scores, for the Alzheimer’s disease cases used in this thesis.

Case

Braak

Atrophy

Tangles

Plaques

ARP

CERAD

AZ32

Unknown

2/3

1/3

2/3

B

Probable Alzheimer’s

AZ45

Unknown

1/3

2/3

2/3

B

Probable Alzheimer’s

AZ71

VI

2/3

3/3

3/3

C

Definitive Alzheimer’s

AZ72

V

0/3

1/3

3/3

C

Indicative of Alzheimer’s

AZ80

VI

3/3

3/3

3/3

C

Definitive Alzheimer’s

AZ82

IV

3/3

3/3

-

B

Probable Alzheimer’s

AZ86

III

0/3

1/3

-

A

Possible Alzheimer’s

AZ90

IV

3/3

3/3

3/3

C

Definitive Alzheimer’s

AZ92

VI

3/3

3/3

-

C

Definitive Alzheimer’s

2.1.2. Matrix-assisted laser desorption/ionisation (MALDI)–imaging mass spectrometry (IMS) 2.1.2.1. Tissue preparation Fresh-frozen postmortem human brain blocks from the hippocampus and the MTG were used for this study. Blocks from the control and AD cases outlined in Table 2.1 were warmed from -80°C to -20°C for an hour and attached to cold specimen chucks with a small amount of TissueTek® OCT mounting medium (Sakura Finetek, CA, USA), at the base of the tissue. Twelve-µm-thick coronal sections were cut at -20°C on the Leica Bright OTF5000 Cryostat (A-M Systems, WM, USA) or the Leica CM3050 Cryostat (Leica Microsystems, Wetzlar, Germany). Cryosections were then mounted on a conductive MALDI target, i.e. an ABI-PerSpective Voyager DE STR metal MALDI Plate (Applied Biosystems, CA, USA) at room temperature (RT), or an indium-tin oxide (ITO)-coated MALDI glass slide (Hudson Surface Technology, NJ, USA) that had been pre-cooled to -20°C. Since polymers such as OCT are known to cause mass interference when analysing small molecules using MALDI (Schwartz

43 | P a g e

Chapter 2: Materials and methods et al., 2003, Chaurand et al., 2006, Thomas and Chaurand, 2014), care was taken to prevent OCT contamination during cutting and mounting. The advantage of using ITO-coated slides over a MALDI target plate is their compatibility with post-MALDI analysis microscopy (post-hoc histochemistry). Once the tissue had been mounted onto the plate or ITO-coated slides, it was dried for an hour in a dry-seal, grease-free desiccator (Jencons, PA, USA), under vacuum. Of several wash conditions that were trialled, the 2 min wash with cold 50 mM ammonium formate solution, followed by a 30 s wash with deionised water, was selected for this study. As demonstrated in Section 3.3.2.1, this wash condition reduced sodium and potassium adducted species and increased spectra intensity as previously reported by Angel et al. (2012).

2.1.2.2. Choice of matrix Following optimisation (as outlined in Section 3.3.2.1), 1,5-diaminonaphthalene (DAN) matrix (SigmaAldrich, MO,USA) was chosen for this study. The use of this matrix has been previously demonstrated to have the highest efficiency in providing rich lipid signatures in both positive and negative ion modes (Thomas et al., 2012, Thomas and Chaurand, 2014).

2.1.2.3. Matrix application The production of high quality MALDI ion images depends greatly on the homogeneous deposition of matrix on the tissue section (Thomas and Chaurand, 2014). Since the vacuum sublimation technique, first detailed by Hankin et al (2007), has been successfully used to coat dry matrix onto sectioned tissue, all our matrices were applied using a sublimation apparatus (CG-3038 Sublimation Apparatus, Chemglass Life Sciences, NJ, USA). Briefly, target plates or ITO-coated glass slides were fixed onto the flat-bottomed condenser using thermally conductive tape (3M Copper Foil Shielding Tape, Element14, NZ). 300 mg of DAN was added to the bottom section of the apparatus, and the two pieces were then assembled with an O-ring seal and connected to a vacuum pump. The cold trap was placed in liquid nitrogen. The condenser was filled with ice after 5 min at reduced pressure, followed by another 10 min to allow cooling and vacuum establishment. The vacuum level was set to ~50 mTorr. The base of the sublimator was heated to 140°C for 5 min using a sandbath placed on a heating mantle. Temperature was monitored using a thermocouple and the time was measured from the onset of heat application. Deposition of the solid matrix could be visually observed on the sides of the glassware up to the level of the sample plate. Following this heating period, the sublimation apparatus was removed from the heat, and allowed to slowly cool for 10 min, whilst still under vacuum. The system was then brought to ambient pressure, and the condenser coolant was removed. When the apparatus had returned to RT, it was disassembled and the sample plate/slide was carefully collected from the condenser. The plate/slide was then dried in a vacuum desiccator, and a 70% ethanol solution was used to remove matrix from regions of the plate that had no tissue 44 | P a g e

Chapter 2: Materials and methods on it. The plate/slide was then stored in a vacuum desiccator until analysis. Since a global decrease in lipid composition over a few days, as a function of temperature, humidity, and time of storage, has been previously reported (Thomas and Chaurand, 2014), matrix application was performed on the day before or the same day as analysis, where possible.

2.1.2.4. Instrument settings An UltrafleXtreme MALDI TOF/TOF mass spectrometer (Bruker Daltonics Gmbh, Bremen, Germany), equipped with a 2 kHz Smartbeam IITM UV MALDI laser, was used to acquire the majority of data presented in this thesis. A Voyager DE Pro MALDI-TOF mass spectrometer (Applied Biosystems, Life Technologies, CA, USA), with a 337 nm nitrogen laser, however, was used to acquire the pilot data presented in Chapter Four. Hippocampal pilot data were acquired using the Voyager DE Pro MALDI-TOF mass spectrometer, operating in reflector mode, with a laser repetition rate of 20 Hz, at an accelerating voltage of +20 kV or -20 kV. Red phosphorous (1 mg/mL; Sigma-Aldrich Chemistry, USA) was used as an external calibrant before data collection (Sládková et al., 2009). The mass range was m/z 350 to 1600 and data was acquired using a raster step-size was set to 150 µm, with 15 laser shots acquired at each sampling site. The MALDI MS Imaging Tool software (Novartis, Basel, Switzerland) was used for image acquisition. The majority of data presented in this thesis were acquired the Bruker UltrafleXtreme MALDI TOF/TOF mass spectrometer (Bruker Daltonics Gmbh, Bremen, Germany), operating in reflector mode, with a laser repetition rate of 2 kHz. Slides were first inserted in the holder, and the mass spectrometer was externally calibrated using red phosphorus, prior to data collection (Sládková et al., 2009). Pre-loaded FlexControl 3.0 (Bruker Daltonics Gmbh, Bremen, Germany) methods were optimised to acquire spectra in both negative and positive ion modes for each case. Delayed extraction parameters were optimised for signal intensity and mass resolution, and set to 120 ns. Data used to determine lipid changes in the cortex and hippocampus in AD were acquired using a raster step-size of 50 µm and 100 µm and a laser beam size of 40 µm and 60–70 µm, respectively. A raster step-size of 20 µm was used for higher-resolution dentate gyrus imaging (Chapter Six) using a laser beam size of 10 µm. At each sampling point, a total of 75 and 100 laser shots per spectrum were accumulated for negative and positive ion modes, respectively. All data were collected in the mass range of m/z 400 to 2000. Imaging datasets for age- and sex-matched control and AD cases were acquired in the same run using FlexImaging (version 4.1) software (Bruker Daltonics GmbH, Bremen, Germany). The three-point teaching procedure on this software was also used to accurately match the microscopic image of the tissue section and the actual target position in the instrument. Since lipid spectra were found to be highly reproducible between sections from the same case 45 | P a g e

Chapter 2: Materials and methods (Figure 2.1), one dataset from each matched pair was used for subsequent data analysis. Following data acquisition, DAN matrix was removed by 70% ethanol, and tissue sections were stained with haematoxylin and eosin (H&E) and luxol fast blue (LFB) for histological analysis (see Section 2.1.3).

Figure 2.1: Mass spectra from three hippocampus sections from Alzheimer’s disease (AD; case AZ71) acquired in reflector negative ion mode. A spectrum ranging from m/z 500 to m/z 2000 was recorded for three separate sections of the same AD case (AZ71). This example shows a snapshot of the recorded mass spectra, with m/z values ranging from 885 to 912. While the absolute intensity (shown on the y-axis)

46 | P a g e

differed amongst the three sections, peaks at the same m/z values were present in all three. The most abundant peaks are labelled in the first panel. Smoothing and baseline subtraction was applied to the raw data using FlexAnalysis 3.4 software (Bruker Daltonics GmbH, Bremen, Germany). a.u., arbitrary unit; MS, mass spectrum; m/z, mass-tocharge ratio

Chapter 2: Materials and methods

2.1.2.5. MALDI imaging For hippocampus pilot data, after data acquisition, molecular images were reconstituted using BioMAP3x Functional Image Analysis (Novartis, Basel, Switzerland). Data were re-binned and normalised to total ion current (TIC). Lipid assignments were based on previous publications (Veloso et al., 2011b, Yuki et al., 2011). Signals at selected mass-to-charge ratio (m/z; ± 0.05 Da) peaks were plotted. For display purposes data were interpolated and pixel intensities were plotted on the same relative intensity scale. For the remaining datasets, raw spectra from all datasets were first aligned to a specific mass control list based on previous publications (Veloso et al., 2011b, Yuki et al., 2011) using FlexAnalysis 3.4 software (Bruker Daltonics GmbH, Bremen, Germany). Datasets were then imported into SCiLS lab 2015b software (SCiLS GmbH, Germany), with a TopHat baseline removal, and normalised to TIC, before being analysed. Distribution maps of these selected m/z values were generated using SCiLS lab 2015b, with automatic hotspot removal and edge-preserving weak image denoising.

2.1.3. Histological staining and imaging 2.1.3.1. Haematoxylin & eosin/luxol fast blue staining H&E, and LFB, staining was used to compare anatomical detail to MALDI data. For tissue mounted on a metal MALDI target plate from which pilot data was obtained, H&E and LFB staining was performed on a sister section. However, the ability to use ITO-coated slides to acquire subsequent data on the Bruker mass spectrometer allowed these tissue staining protocols to be performed on the same tissue sections that were used to acquire MALDI-IMS data. These procedures were performed after MALDI-IMS data acquisition since H&E dyes are known to suppress MALDI signals (Thomas and Chaurand, 2014). The deposited matrix was first washed off by immersion in 70% ethanol for 5 min at RT. Sections were then incubated in LFB solution overnight at RT. Excess LFB stain was then washed off the next day with deionised water and differentiated using a hydroquinone/sodium sulphite reducer solution. Sections were then stained with Gill II Haematoxylin solution (Surgipath, IL, USA) for 30 s at RT and decolourised using acid alcohol (7 quick dips in this solutions sufficed). Sections were then immersed in running dH2O before being submerged in 1% lithium carbonate for 30 s at RT. Excess lithium carbonate was washed off using running dH2O. Sections were then stained using Eosin solution (Surgipath, IL, USA) for no more than 10 s in total, before being differentiated using 95% ethanol (approximately 3 min was sufficient for this step). Sections were dehydrated using 100% ethanol (3 x 5 min) and cleared using two changes of xylene (2 x 10 min), before being

47 | P a g e

Chapter 2: Materials and methods coverslipped using dibutyl phthalate (DPX) and No.1 coverslips (Lomb Scientific, NSW, Australia). Slides were left to dry for approximately a week prior to imaging.

2.1.3.2. Imaging Overview images of whole sections were photographed using a digital camera (Nikon DXM1200F), mounted on a Nikon Eclipse E800 microscope (Nikon, NY, USA). A calibration slide was also photographed at the same magnification of each brain section to produce accurate scale bars. Higher-resolution images of sections were acquired on a Nikon Eclipse Ni microscope (Nikon, NY, USA), using NIS Elements v.4.30 software (Nikon, NY, USA), using the 10x and 20x lenses.

2.1.4. Lipid identification using tandem mass spectrometry (MS/MS) Tandem mass spectrometry (MS/MS), which was first described by McLafferty (1981), was used to assign putative lipid IDs (Chapter Four). MS/MS consists of two steps. Charged lipid species are separated according to their m/z ratios in the first step (MS1). In the second step (MS2), the user sets the m/z value of the ‘parent’ ion of interest, which is isolated in a collision cell after other ion species are filtered away.

2.1.4.1. On-tissue matrix-assisted laser desorption/ionisation (MALDI)–tandem mass spectrometry (MS/MS) Twelve-µm coronal sections from cases AZ71 and H238 were processed as outlined in Section 2.1.2.1, and used for on-tissue MS/MS. Prior to data collection, the mass spectrometer was externally calibrated using red phosphorus (Sládková et al., 2009). Thereafter, MS/MS was performed using the LIFT method in FlexControl 3.0 (Bruker Daltonics GmbH, Germany). LIFT is not an abbreviation; rather, it is the term Bruker Daltonics uses to describe the process of elevating the potential of a collision cell above that of the ion source (Watson and Sparkman, 2007). The parent mass of interest was set manually and the PreCursor Ion Selector (PCIS) Window Range was set to ±2 Da. Collision Induced Dissociation (CID), using argon gas, was used in MS/MS. The argon gas imparts internal energy to the precursor ion, causing it to fragment, producing a spectrum containing the fragment ions provides a ‘structural fingerprint’ of the parent ion (Rauniyar, 2010). It is important to note, however, that not all the fragment ions of an ion of interest were present at detectable levels. MS/MS data were interpreted using the LIPID MAPS database (Fahy et al., 2007). Lipid assignments, where possible, were then subsequently confirmed using LC-MS/MS (Section 2.1.4.2), or using previously published data on the mammalian lipidome (Ariga et al., 1982, Fujiwaki et al., 1999, Han et 48 | P a g e

Chapter 2: Materials and methods al., 2001, Han et al., 2002, Jackson et al., 2005, Hicks et al., 2006, Jackson et al., 2007, Fuchs et al., 2008, Chan et al., 2009, Dill et al., 2010, Veloso et al., 2011b, Whitehead et al., 2011, Yuki et al., 2011, Samhan-Arias et al., 2012, Nunez et al., 2016).

2.1.4.2. Liquid-chromatography (LC)–tandem mass spectrometry (MS/MS) Two 12-µm thick MTG sections from six control cases, which were used to detect lipid changes, were scraped off glass slides and added to 600 µL Eppendorf tubes. The same procedure was repeated for AD MTG sections, and control and AD hippocampus sections, separately. The tissue was then weighed, and the extraction solution, made of a 1:2:1 ratio of chloroform: methanol: water, was added at a ratio of 1:3 (tissue: solution). The tissue was left steeping in the lipid extraction solution overnight at 4°C. The tubes were then centrifuged and the tissue mass was removed. The lipid extraction was then separated into its hydrophobic and hydrophilic layers using a mixture of chloroform and water at a 1:1 ratio. The samples of lipid extraction were then submitted to a core facility for analysis. Briefly, the analysis settings they used are as follows. LC–MS/MS analysis was conducted using a LTQ-FT (Linear Ion TrapFourier Transform) on ion cyclotron resonance (ICR) mass spectrometer (Thermo Electron, Bremen, Germany) under electrospray conditions in negative and positive ion modes. Each full MS1 scan was acquired in the ICR cell over a range of m/z 500–2000, followed by a parallel MS2 scan of the top three ions in the MS1 scan. The MS3 acquisition of the top three ions in each MS2 ion scan was then performed in the ion trap. Lipid assignments were made using LipidSearch software (Thermo Fisher Scientific, MA, USA).

49 | P a g e

Chapter 2: Materials and Methods Table 2.3: Software programs used to acquire and analyse the data presented in this thesis. This table summarises the software programs that were used to acquire and analyse the data presented in this thesis, along with their manufacturers, the output and supported file types, and function. The list has

been organised according to the order in which they appear in Chapter Two. IMS, imaging mass spectrometry; MALDI, matrixassisted laser desorption/ionisation; TOF, time of flight

Software

Designed

Output/supported

program

by

file types

MALDI MS Imaging Tool

FlexControl 3.0

Markus Stoeckli (Novartis, Basel,

Automated imaging of samples on Output: Analyze 7.5

Switzerland) Bruker Daltonics GmbH, Bremen, Germany

Function

Applied Biosystems Voyager mass spectrometry instruments

Output: *.par (for standard MS spectra) *.lft (for MS/MS spectra)

Configuration and operation of the Bruker TOF flex-series mass spectrometers Acquisition of IMS data on the

FlexImaging 4.1

Bruker Daltonics GmbH, Bremen, Germany

Supported: *.mis (imaging sequence files)

Bruker TOF flex-series mass spectrometers in conjunction with FlexControl; Visualisation of biomarker distribution

Martin Rausch & BioMAP3x

Markus Stoeckli (Novartis, Basel, Switzerland)

FlexAnalysis 3.4

SCiLS Lab

NIS Elements 4.30

Bruker Daltonics GmbH, Bremen, Germany

Supported:

Visualisation and storage platform

Analyze 7.5, *.TIF,

that supports numerous imaging

*.DICOM, *.PNP

modalities including IMS

Supported: XMASS-file formats *.fid (raw data) *.1r (processed data)

SCiLS GmbH,

Supported: *.sl

Bremen,

(requires conversion of

Germany

*.mis FlexImaging files)

Nikon, USA

Output: *.TIFF

Scientific, MA, USA

50 | P a g e

using the Bruker TOF flex-series mass spectrometers Visualisation, statistical analysis, and interpretation of MALDI-IMS data acquired using Bruker mass spectrometers Image acquisition software linked to Nikon Eclipse microscope Processing LC-MS data generated by

Thermo Fisher LipidSearch

Post-processing of spectra acquired

Output: *.XML

Thermo Fisher Scientific Orbitrapbased mass spectrometers; Accurate lipid identification

Chapter Three

Developing MALDI-IMS to study lipids in the postmortem human brain

Section One: Introduction .................................................................................................................. 52 Section Two: Methods........................................................................................................................ 53 3.2.1 Histological staining ................................................................................................................................... 53 3.2.1.1 Oil Red O staining .............................................................................................................................. 53 3.2.1.2 Periodic-Acidic Schiff staining ............................................................................................................ 53 3.2.1.3 Imaging .............................................................................................................................................. 53 3.2.2 Mass spectrometry .................................................................................................................................... 53 3.2.2.1. Sample preparation ........................................................................................................................... 53 3.2.2.2. Liquid-chromatography (LC)-mass spectrometry (MS) ...................................................................... 54 3.2.3 Matrix-assisted laser desorption/ionisation (MALDI)–imaging mass spectrometry (IMS) ........................ 54 3.2.3.1 Matrix optimisation ........................................................................................................................... 54 3.2.3.2 Sample for imaging mass spectrometry ............................................................................................ 54 3.2.3.3 Homogenised tissue standards preparation ...................................................................................... 54 3.2.3.4 Spectral realignment.......................................................................................................................... 55 3.2.3.5 Spectral normalisation ....................................................................................................................... 55 3.2.3.6 Spectral denoising and hotspot removal ........................................................................................... 56 3.2.3.7 Receiver-operating characteristic (ROC) analysis .............................................................................. 56 3.2.3.8 Six-set Venn diagram ......................................................................................................................... 56

Section Three: Results........................................................................................................................ 57 3.3.1. Lipid analysis .............................................................................................................................................. 57 3.3.1.1. Histological stains for lipid analysis ................................................................................................... 57 3.3.1.2. Analysis of lipid classes using liquid chromatography-mass spectrometry (LC-MS) .......................... 58 3.3.2. Lipid analysis using matrix-assisted laser desorption/ionisation (MALDI) ................................................. 60 3.3.2.1. Matrix selection ................................................................................................................................. 60 3.3.2.2 Spectral alignment ............................................................................................................................. 62 3.3.2.3 Lipid spectra detected in the middle temporal gyrus (control cases) ................................................ 63 3.3.3. Lipid distributions in the middle temporal gyrus using MALDI-imaging mass spectrometry (IMS) ........... 65 3.3.3.1 Image processing: Normalisation and denoising ............................................................................... 65 3.3.3.2 Distributions of lipids in the middle temporal gyrus (MTG) ............................................................... 66 3.3.3.3 Lipid expression in the middle temporal gyrus (MTG) in Alzheimer’s disease (AD) ........................... 69 3.3.4. Developing the analysis workflow.............................................................................................................. 71 3.3.4.1 Technical variation ............................................................................................................................. 71 3.3.4.2 Pair-wise receiver operating characteristic (ROC) comparison .......................................................... 72 3.3.4.3 Determining m/z values of interest ................................................................................................... 73 3.3.4.4 Summary of analysis workflow .......................................................................................................... 75 3.3.5. Differentially expressed lipid species in Alzheimer’s Disease .................................................................... 77 3.3.5.1 Negative ion mode ............................................................................................................................. 77 3.3.5.2 Positive ion mode ............................................................................................................................... 81

Section Four: Discussion .................................................................................................................... 85 3.4.1. General discussion ..................................................................................................................................... 85 3.4.1.1. Methods to analyse lipids .................................................................................................................. 85 3.4.1.2. Optimising MALDI-IMS ...................................................................................................................... 85 3.4.1.3. Distribution of lipids in the human middle temporal gyrus (MTG) .................................................... 87 3.4.1.4. Differential lipid expression in the middle temporal gyrus (MTG) in Alzheimer’s disease (AD) ......... 89 3.4.2. Summary of findings .................................................................................................................................. 91

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Section One

Introduction

Histological staining can be used to image the distribution of lipids and has been previously used to study changes in lipid metabolism in Alzheimer’s disease (AD; Burger and Vogel, 1973, Bugiani et al., 1999, Mandas et al., 2012). However, histological staining does not allow different lipid classes, or their specific fatty acid chains, to be easily distinguished. In contrast, mass spectrometry (MS) overcomes this challenge, and matrix-assisted laser desorption/ionisation (MALDI)-imaging mass spectrometry (IMS) in particular, can provide in-depth information about the distribution of specific lipid species. While MALDI-IMS is now routinely used to image the distribution of lipids in the rodent brain (some recent examples: Wang et al., 2008, Chan et al., 2009, Cerruti et al., 2012, Hanrieder et al., 2012, Weishaupt et al., 2015, Hong et al., 2016), there have been far fewer publications that have used this technique to study the human brain (Veloso et al., 2011a, Veloso et al., 2011b, Yuki et al., 2011, Hirano-Sakamaki et al., 2015). The distribution of selected lipids has been previously mapped in the physiologically-normal frontal cortex and cerebral cortex in the human brain (Veloso et al., 2011a, Veloso et al., 2011b, Yuki et al., 2011). Although the lipid profile, and its changes in AD, in the postmortem human temporal lobe have been studied using extractive mass spectrometry methods (Han et al., 2001, Han et al., 2002), the distribution of lipids in the middle temporal gyrus (MTG), which is one of the first, and most extensively, affected neocortical areas in AD (Convit et al., 2000, Galton et al., 2001), has still not been mapped. Hence, one aim of the current chapter was to use MALDI-IMS to investigate the profile and distribution of lipids in the MTG. The second aim was to analyse the differential expression of lipids in AD. To do this, however, I had to first troubleshoot parameters and develop an analysis pipeline that enabled the efficient processing of the large datasets generated with MALDIIMS. The analysis was then applied to the MTG dataset as an example of how it works. Thus, this chapter has been organised in three parts: 1) the optimisation of sample preparation, and data acquisition and processing parameters, 2) the development of the analysis workflow, and 3) the application of this workflow to detect differential lipid expression in grey and white matter in the MTG in AD. The main statistical method used to detect differential lipid expression was the receiver operating characteristic (ROC) function — a univariate measure that has been previously used to determine the predictive power of putative biomarkers associated with cancer tissue (Cazares et al., 2009, Rauser et al., 2010) and to find discriminative m/z values that characterise selected pathophysiological regions (Klein et al., 2014).

52 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Section Two

3.2.1

Methods

Histological staining

3.2.1.1 Oil Red O staining Fresh, frozen sections (12 µm) were mounted on glass slides and dried for 10 min at room temperature (RT), before being rehydrated in 1 x phosphate-buffered saline (PBS) solution for 2 min. Excess 1 x PBS was blotted, without allowing sections to dry, and stained with Oil Red O working solution for approximately 10–12 min at RT. The sections were then rinsed in: deionised water (dH2O) for 30 sec, 60% isopropanol for 1 min, and then in dH2O again for 2 min. Sections were then counter-stained with Gill’s II haematoxylin for approximately 1–2 min, before being cover-slipped using aqueous mounting medium and sealed with clear nail polish. Stained sections can be stored for up to 12 weeks at 4oC.

3.2.1.2 Periodic-Acidic Schiff staining Fresh, frozen sections (12 µm; mounted on glass slides) were first fixed using 95% ethanol for 5 min, before being oxidised for 10 min with 1% periodic acid solution. Sections were then rinsed in running dH2O for 5 min, before being placed in Schiff reagent for 20 min at RT. Sections were then rinsed in running dH2O for 15 min (stain may become darker during this step), before being counter-stained with Gill’s II haematoxylin for approximately 1–2 min. Excess staining was rinsed off in dH2O, and sections were dehydrated using a series of alcohols (3 x 5 min with 100% alcohol) and xylenes (2 x 10 min xylene). Slides were then cover-slipped using dibutyl phthalate (DPX) and allowed to dry for up to a week before being imaged.

3.2.1.3 Imaging Images of stained sections were acquired on a Nikon Eclipse Ni microscope (Nikon, NY, USA), using NIS Elements v.4.30 software (Nikon, NY, USA).

3.2.2

Mass spectrometry

3.2.2.1. Sample preparation MTG sections from the cases outlined in Table 2.1 were used to produce the results in this chapter. Fresh-frozen coronal sections were cut at 12 µm on a Leica CM3050 Cryostat (Leica Microsystems, Wetzlar, Germany).

53 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.2.2.2. Liquid-chromatography (LC)-mass spectrometry (MS) The protocol outlined in Section 2.1.4.2 was followed. Data were analysed using LipidSearch software (Thermo Fisher Scientific, MA, USA).

3.2.3

Matrix-assisted

laser

desorption/ionisation–imaging

mass

spectrometry (MALDI–IMS) 3.2.3.1 Matrix optimisation The dried-droplet method was used to apply the matrix solutions outlined in Table 3.1 to either untreated postmortem human MTG sections, or those that had been treated with 50 mM ammonium formate for 2 min, followed by a deionised water immersion. The lipid detection for each matrix was assessed in both positive and negative ion modes. The signal enhancement following the ammonium formate wash was also evaluated. Table 3.1: Matrix solutions used for optimisation. Five matrix solutions were assessed for lipid detection in the postmortem human brain in positive and negative ion modes. All matrix solutions were made in acetonitrile (ACN) and trifluoroacetic acid (TFA) as outlined below.

ACN, acetonitrile; CHCA, α-cyano-4-hydroxycinnamic acid; DAN, 1,5-diaminonaphthalene, DHB, 2,5dihydroxybenzoic acid; SA, sinapinic acid; TFA, trifluoroacetic acid; THAP, 2,4,6Trihydroxyacetophenone

Matrix

Solvent

CHCA (5 mg/mL)

60% ACN/0.1% TFA

DAN (100 mg/mL)

50% ACN/0.1% TFA

DHB (60 mg/mL)

70% Acetone/0.1% TFA

SA (15 mg/ml)

50% ACN/0.1% TFA

THAP (20 mg/mL)

50% ACN/0.1% TFA

3.2.3.2 Sample for imaging mass spectrometry MTG sections were prepared for imaging mass spectrometry using the protocol described previously in Section 2.1.2.3.

3.2.3.3 Homogenised tissue standards preparation Homogenised tissue standards were prepared to measure the technical replicability between different acquisition runs. Approximately 250 mg of visual cortex tissue was finely diced with a 54 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain polytetrafluoroethylene-coated blade (GEM, USA), on a parafilm-coated surface to avoid possible contamination. The diced tissue was then transferred into a 2.0 mL tube (Axygen, CA, USA), 15 µL of a methanol/chloroform solution (50/50 v/v) was added, and the tube was centrifuged. Approximately three to four 2 mm zirconium oxide homogenisation beads (Next Advance, NY, USA) were then added to each tube. Tissue was homogenised for 2 x 5 min at speed 8, using a Bullet Blender homogeniser (Next Advance, NY, USA), with brief centrifuging between cycles to ensure complete homogenisation. Each tube was then snap-frozen using CO2 ‘snow’, and then briefly warmed in the hand to extract the homogenised tissue pellet and transferred into a 1 cm2 histology mould (Tissue-Tek CryoMold, Sakura Finetek, CA, USA). The zirconium oxide beads were removed from the homogenised tissue as it defrosted. The tissue homogenate was then snap-frozen in the moulds using CO2 snow. The homogenate standard was then removed from the mould and stored at -80°C until further use. When required to measure technical replicability, 12 µm sections were cut from the homogenate standard using a Leica CM3050 Cryostat (Leica Microsystems, Wetzlar, Germany).

3.2.3.4 Spectral realignment Since spectral drift can occur in a MALDI-Time of flight (TOF) system, the thousands of spectra in an imaging dataset must be aligned to ensure accurate peak detection and data interpretation. A subset of peaks that were present in all datasets, based on known m/z values of lipid species that have been previously detected in the human brain (Veloso et al., 2011b, Yuki et al., 2011), were selected as an internal calibrant list. I ensured that this list spanned the entire mass range of interest (i.e. m/z 400 to 2000) to prevent extrapolation outside the chosen alignment points. Given the large number of spectra in each MALDI-IMS dataset, I first needed to automate the alignment of each spectrum to the internal calibrant. Thus, a pre-installed algorithm method in FlexAnalysis 3.4 software (Bruker Daltonics Gmbh, Bremen, Germany) was updated to automatically apply the list of chosen peaks as an internal calibrant to the raw spectra. The updated script that was used for spectral realignment is outlined in the Appendix (Figure A1).

3.2.3.5 Spectral normalisation Two normalisation methods were tested using SCiLS lab 2015b software (SCiLS GmbH, Bremen, Germany): 1) the median normalisation, and 2) Total Ion Current (TIC) normalisation, to minimise spectral intensity variations that are inherent to MALDI. In the first approach, each spectrum is divided by the median intensity value of the spectrum, while in the TIC approach each spectrum in the dataset is divided by the sum of all spectral intensities (Deininger et al., 2011).

55 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.2.3.6 Spectral denoising and hotspot removal Weak edge-preserving denoising was applied using SCiLS lab 2015b software (SCiLS GmbH, Bremen, Germany), where the filter is applied in a 3 x 3 point spectral neighbourhood. The ‘hotspot removal’ function of the SCiLS lab 2015b software (SCiLS GmbH, Bremen, Germany) was used to adjust the detected maximum, which was useful when analysing less intense peaks.

3.2.3.7 Receiver-operating characteristic (ROC) analysis Using the ‘Find Discriminative m/z Values’ on the SCiLS lab 2015b software (SCiLS GmbH, Bremen, Germany), I analysed all data-points in individual spectra in each age- and sex-matched pair separately. The AD dataset was set as Class 1 to detect m/z values that were higher for AD, while the AD dataset was set as Class 2 to detect m/z values that were lower in AD. In the parameters tab, the ‘Random subset’ option was selected in the ‘Spectra to Work on’ panel, with the subset size reflecting the lowest number of spectra in each age- and sex-matched pair. The minimal interval width was set to 0.05 Da. The threshold was set to 0.7 when detecting discriminative m/z values.

3.2.3.8 Six-set Venn diagram ROC analysis data were exported into InteractiVenn, an interactive web visualization tool for analysing lists of elements using Venn diagrams, which supports up to six different sets (Heberle et al., 2015). The generated Venn diagram was exported as an image file (*.png format), while the text data were exported into an Excel spreadsheet for further analysis.

56 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Section Three

Results

3.3.1. Lipid analysis

3.3.1.1. Histological stains for lipid analysis Two histological methods that have been previously used for lipid analysis in the human brain, i.e. the Oil Red O stain and the Periodic-Acidic Schiff (PAS) stain, were trialled for use in this study. Figure 3.1 shows the results from this trial for representative control and AD tissue. Both control and AD tissue showed positive Oil Red O staining (Figure 3.1A and B), as indicated by the red-brown colour. Staining, however, was more intense in the AD tissue. The staining pattern indicates the accumulation of lipofuscin, which consists of lipid-containing residues of lysosomal digestion, seen in normal aging. Both control and AD tissue also showed positive PAS staining, as indicated by the magenta colour (Figure 3.1C and D). The pattern of staining, at least in AD tissue, was similar to that seen with Oil Red O, and is again indicative of lipofuscin accumulation. Although both these stains could be used effectively to investigate lipid accumulation, they did not allow me to elucidate the different lipid classes that accumulate in normal aging, or whether lipid accumulation changes in AD.

Figure 3.1: Oil Red O and Periodic Acid-Schiff (PAS) histological staining. Micrographs in the first row demonstrate positive Oil Red O staining in (A) control

and (B) Alzheimer’s disease (AD) tissue sections. Micrographs in the second row show positive PAS staining in (C) control and (D) AD tissue sections.

57 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.3.1.2. Analysis of lipid classes using liquid chromatography-mass spectrometry (LC-MS) Next the use of liquid chromatography-mass spectrometry (LC-MS) was trialled to detect lipids in the human postmortem MTG in control and AD. The lipid extraction from MTG sections of all six control cases were done together, as outlined in Section 2.1.4.2. Lipids from all six AD cases were also extracted together, but separately from the control cohort. The results from the trial are shown in Figure 3.2, which shows the relative abundance of fatty acids, diglycerides, triglycerides, glycerophospholipids, and sphingolipids, detected in the MTG, in the control and AD cohorts. Depending on their class, different lipids can be protonated or deprotonated more easily. Deprotonated lipids are detected in negative ion mode, while protonated lipids are detected in positive ion mode. Therefore, data from both ion modes were pooled together. Further, the area under the curve of the extracted ion chromatogram was used as a measure of abundance. Generally, both control and AD tissue showed a similar variety of all lipid classes, albeit with differences in relative abundance between control and AD tissue. Phosphatidylcholine (PC) was the most abundant glycerophospholipid class and was decreased in AD (Figure 3.2A). Phosphatidic acid (PA) was the least abundant glycerophospholipid, but showed a relative increase in AD. Similarly, phosphatidylglycerol (PG), phosphatidylinositol (PI), and phosphatidylserine (PS), were also increased in AD. As shown in Figure 3.2B, of the sphingolipid classes detected in the MTG, Ceramide-1phosphate (Cer1P) and CerG1, which is a simple glycosphingolipid, were the most abundant. However, these lipid classes were far less abundant than the glycerophospholipids. Ceramide (Cer), CerG1, and CerG2, were all increased in AD. Cer1P, in contrast, showed a reduction in AD. Finally, Figure 3.2C showed the abundance of fatty acids, diglycerides, and triglycerides, of which the latter was the most abundant. Generally, in AD, fatty acids were decreased, while diglycerides and triglycerides were increased. Given that a relative difference was detected in these lipid classes in AD, next, I optimised MALDIIMS to image the distribution of these lipid classes, and their changes, in the MTG region.

58 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Figure 3.2: Abundance (as area under the curve measurement) of different lipid classes detected in the middle temporal gyrus (MTG) using liquid chromatography-mass spectrometry (LC-MS). (A) Abundance of all glycerophospholipid classes detected in the MTG region. Phosphatidic acid (PA) and phosphatidylglycerol (PG) levels are enlarged and shown inset. (B) Abundance of all sphingolipid classes detected in the MTG region. (C) Abundance of fatty acids (FA), diglyceride (DG), and triglyceride (TG).

Data from the control cohort is shown in blue, while data from the Alzheimer’s disease (AD) cohort is shown in red. A.U., arbitrary unit; AUC, area under the curve; Cer, ceramide; Cer1P, ceramide 1 phosphate; CerG1 and CerG2, simple glycosphingolipids; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol; PS, phosphatidylserine; SM, sphingomyelin

59 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.3.2. Lipid

analysis

using

matrix-assisted

laser

desorption/

ionisation (MALDI) 3.3.2.1. Matrix selection Droplets of different matrices (as indicated in Table 3.1) were applied to postmortem human MTG cortex tissue sections to evaluate their use in detecting signals in the expected lipidomic range, i.e. m/z 500–2000. I tested both positive and negative ion modes to maximise the number of different lipids studied. Control and AD tissue was tested and showed similar results. However, for simplicity, only the results from control tissue have been illustrated in Figure 3.3. In negative ion mode a number of signals were detected using α-Cyano-4-hydroxycinnamic acid (CHCA) and 1,5-Diaminonapthalene (DAN). Although CHCA produced a few signals in the m/z 600– 700 range, DAN yielded abundant signals in the m/z 500–950 range. In positive ion mode, spectral signals in the expected lipidomic range were detected using 1,5-dihydroxybenzoic acid (DHB) and DAN. Abundant signals in the m/z 730–850 range, and in the m/z 450–950 range, were detected using DHB and DAN, respectively. I also tested if a 2 min 50 mM ammonium formate wash during sample preparation, first described by Angel et al. (2012), enhances sensitivity. In negative ion mode, the ammonium formate wash decreased the overall intensity of signals recorded with CHCA. However, the wash step enhanced signals in the m/z 880–910 range, which previously was not detected in the unwashed tissue. The wash step also enhanced the signals detected with DAN. In positive ion mode, the ammonium formate wash step increased the intensity of signals detected with both DHB and DAN. It was particularly effective at enhancing signals detected in the m/z 600–900 range with DAN, i.e. the range in which many phospholipids are detected. Given the ability to detect abundant spectral signals in the lipidomic range, in both ion modes, with DAN, I chose to use this matrix for subsequent analysis. Further, I chose to incorporate the ammonium formate as part of the sample preparation to enhance lipid detection in both ion modes. The use of DAN also offers the advantage of a single sublimation preparation step (outlined in Section 2.1.2.3) being sufficient to detect lipids in both modes from the same section.

60 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Figure 3.3: Optimisation of matrix choice and tissue washing. This figure shows mass spectra acquired with different matrices, with and without 50 mM ammonium formate, in negative and positive ion modes. The first row illustrates α-Cyano-4hydroxycinnamic acid (CHCA), 1,5-dihydroxybenzoic acid (DHB), and 1,5-Diaminonapthalene (DAN), matrix

crystals. The next four rows show example mass spectra acquired in negative and positive ion modes, using unwashed tissue, and tissue washed with 50 mM ammonium formate solution. The m/z values are on the x-axis, while raw intensity values are on the yaxis. a.u., arbitrary units; m/z, mass-to-charge ratio

61 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.3.2.2 Spectral alignment The mass spectrometer was externally calibrated using red phosphorous, prior to data collection, for each acquisition run. However, despite this careful calibration, when imported into SCiLS lab 2015b software (SCiLS GmbH, Bremen, Germany), there was considerable misalignment between the overview spectra of each dataset (Figure 3.4A), which can be attributed to instrument drift that occurs through the duration of each acquisition run, and to small irregularities in the flatness or thickness of tissue sections (Norris et al., 2007). The misalignment between spectra makes it impossible to ensure that the intensities of the same ions are being compared across all the datasets. To effectively realign raw spectra, I applied the recalibration outlined in Section 3.2.3.4. Following the application of this calibration, the datasets were once again imported into the SCiLS lab 2015b software (SCiLS GmbH, Bremen, Germany). Figure 3.4B shows the improved alignment of calibrated spectra. Given this improvement, the same calibration method was applied to all subsequent datasets acquired on an UltrafleXtreme MALDI TOF/TOF mass spectrometer (Bruker Daltonics Gmbh, Bremen, Germany).

Figure 3.4: Spectral realignment. These images show representative mass spectra ranging from m/z 885 to m/z 891. (A) Misaligned raw data. (B) Result following spectral realignment using an updated method on FlexAnalysis 3.4 software (Bruker Daltonics Gmbh, Bremen, Germany). The script used for realignment is

62 | P a g e

shown in Figure A1. Each colour represents a summary spectrum of the data acquired from different acquisition runs. m/z, mass-to-charge ratio

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.3.2.3 Lipid spectra detected in the middle temporal gyrus (control cases) Figure 3.5A shows the overview or summary lipid profiles, detected in positive ion mode (red) and negative ion mode (black), which were acquired by averaging the datasets from all six control postmortem MTG sections that were analysed. For each region, the intensity of each m/z value was plotted on a relative abundance scale, which ranged from zero to one. Labelled peaks indicate m/z values to which lipid assignments were made based on previous publications. These are listed in Table 3.2. The majority of positively-charged lipids that were abundant and could be identified were all PC species. The most abundant peak in positive ion mode was m/z 760 (PC 34:1+H+). In contrast to positive ion mode, more peaks were observed in negative ion mode. Of the seven abundant negatively-charged lipids that were identified, two were phosphatidylethanolamine (PE) species, one was a PI, three were sulfatides (SF), and two were GM1. The most abundant peak was m/z 885 (PI 38:4-H-). Next, I compared the summary lipid profile of grey matter and white matter. A co-registered image of the same section, stained with H&E and LFB, was used to define grey and white matter regions. Figure 3.5B shows the overview of the lipid spectrum detected in grey matter, in both ion modes. Again, the positive ion mode dataset is shown in red and the negative ion mode dataset is shown in black. Given that the whole MTG region analysed was comprised of a greater grey:white matter ratio, it is not surprising that the spectral profile in both ion modes for grey matter is similar to that seen in the whole MTG area. There were, however, slight differences in the relative intensity of lipids such as m/z 790 (PE 40:6-H-) and m/z 834 (PS 40:6-H-). Figure 3.5C illustrates the summary lipid profile of white matter. In positive ion mode (shown in red), like grey matter, the most abundant peak was m/z 760 (PC 34:1+H+). Both m/z 734 (PC 32:0+H+) and m/z 788 (PC 36:1+H+) were also detected, while m/z 810 (PC 38:4+H+) was detected at a lower intensity (not labelled). The negative ion mode spectrum, in contrast, shows several differences to those detected in grey matter. For instance, the most abundant lipid in white matter was m/z 888 (SF 24:1-H-). Further, m/z 744 (PE 36:1-H-), m/z 790 (PE 40:6-H-), m/z 834 (PS 40:6-H-), m/z 1544 (GM1 36:1-H-), and m/z 1572 (GM1 38:1-H-) were not abundantly detected in this region. In contrast, m/z 726 (pPE 36:2-H-) and 788 (PS 36:1-H-) were more abundant in this region. Given the differences seen in the lipid spectral profile acquired from grey and white matter, I then plotted the distributions of these ions from the imaging data.

63 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Figure 3.5: Overview of lipid spectra acquired using matrix-assisted laser desorption/ionisation (MALDI). Average overview spectra, between m/z 500 to 1800, acquired from all six control sections in the (A) whole middle temporal gyrus (MTG) area that was analysed, (B) grey matter alone, and (C) white matter alone. Spectra acquired using reflectron positive mode are

64 | P a g e

shown in red (above the x-axis), while spectra using reflectron negative mode are shown in black (below the x-axis). The relative intensity for each m/z value was calculated using the most abundant peak in each dataset. The most abundant peaks for each region are labelled. m/z, mass-to-charge ratio

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Mode

Positive Ion

Negative Ion Mode

Table 3.2: Lipid assignments for abundant m/z values. Lipid identifications for some of the most abundant peaks were made based on previous publications. The m/z values from both ion modes are included. The publications used to make the lipid assignment have been referenced within brackets and are listed below.

m/z

Lipid Assignments [Ref]

726

pPE 36:2-H- [1, 2]

744

PE 36:1-H- [1-3]

788

PS 36:1-H- [1]

790

PE 40:6-H- [1, 2]

834

PS 40:6-H- [1]

885

PI 38:4-H- [1, 4]

888

SF 24:1-H- [1, 4]

904

SF 24:1 (OH)-H- [1, 5]

932

SF 26:1 (OH)-H- [5]

1544

GM1 36:1-H- [6-8]

1572

GM1 38:1-H- [6-8]

734

PC 32:0+H+ [7, 9, 10, 12]

760

PC 34:1+H+ [7, 10-12]

788

PC 36:1+H+ [7, 9, 10, 12]

810

PC 38:4+H+ [9, 11, 12]

GM1, ganglioside 1; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol; pPE, phosphatidylethanolamine plasmalogen; PS, phosphatidylserine; SF, sulfatide

References: 1 Jackson et al. (2007) 2 Han et al. (2001) 3 Melo et al. (2012) 4 Cha and Yeung (2007) 5 Yuki et al. (2011) 6 Sugiyama et al. (1997) 7 Woods and Jackson (2006) 8 Hirano-Sakamaki et al. (2015) 9 Hicks et al. (2006) 10 Fuchs et al. (2008) 11 Schiller et al. (2001) 12 Jackson et al. (2005)

3.3.3. Lipid distributions in the middle temporal gyrus using MALDI– IMS 3.3.3.1 Image processing: Normalisation and denoising Before generating images of lipid distributions I first examined the most suitable normalisation and denoising parameters to apply to the datasets. Normalisation aims to minimise spectral intensity variations, transforming measured intensities to comparable scale, without altering biological information (Norris et al., 2007, Deininger et al., 2011, Fonville et al., 2012).

65 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain Figure 3.6A shows representative data from a control section (H190) stained with H&E and LFB, and the distribution of raw data for m/z 834 (PS 40:6-H-) is shown in Figure 3.6B. The raw data showed the predicted pattern of distribution (based on the overall spectra seen in Figure 3.5), with high abundance of the lipid in grey matter. However, it is unclear if the most intense points are linked to biological variation or if they can be attributed to experimental variation. Thus, I tested the application of two normalisation methods on the dataset. Figure 3.6C illustrates the result following the application of the median normalisation to all datasets (i.e. six control and six AD cases). The distribution of m/z 834 (PS 40:6-H-) did not reflect spectral findings when this normalisation was applied. Thus, the TIC normalisation method was applied. Figure 3.6D illustrates how this normalisation method removes experimental artefact, but allows one to visualise the expected distribution pattern. Subsequently, a weak edge-preserving denoising filter was applied to reduce high-frequency chemical noise. Figure 3.6E shows the smoothing effect this method has on the raw data, with no effect on the detected intensity. The ‘hotspot removal’ function in the SCiLS lab 2015b software (SCiLS GmbH, Bremen, Germany) was used to adjust the detected maximum, which is particularly useful when studying the distribution of less intense lipid peaks. As illustrated in Figure 3.6F, this feature allows one to visualise the area where the lipid of interest is most abundant. Thus, in all subsequent lipid distribution images, TIC, weak denoising, and hotspot removal, were all applied, except where stated otherwise.

3.3.3.2 Distributions of lipids in the middle temporal gyrus (MTG) I then plotted images of the distribution of the lipids listed in Table 3.2. As illustrated in Figure 3.7, the patterns of distribution of these lipids clearly showed the difference between grey and white matter, reflecting the findings from the overall lipid spectra. There were six negatively-charged lipids that were abundant in grey matter. These were: m/z 744 (PE 36:1-H-), m/z 790 (PE 40:6-H-), m/z 834 (PS 40:6-H-), m/z 885 (PI 38:4-H-), m/z 1544 (GM1 36:1-H-), and m/z 1572 (GM1 38:1-H-). While most abundant in grey matter, these lipids showed a heterogeneous pattern of distribution in this region itself. In contrast, m/z 726 (pPE 36:2-H-), m/z 788 (PS 36:1-H-), m/z 888 (SF 24:1-H-), m/z 904 (SF 24:1 (OH)-H-), and m/z 932 (SF 26:1 (OH)-H-), were all abundant in white matter. The hydroxylated SF species, however, were also expressed in some grey matter layers, confirming results from a previous study (Yuki et al., 2011).

66 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Figure 3.6: Normalisation and denoising of imaging mass spectrometry (IMS) data. (A) Micrograph illustrating middle temporal gyrus anatomy. This section was stained with haematoxylin and eosin (H&E; grey matter in purple), and luxol fast blue (LFB; white matter in blue), after IMS acquisition. (B) Raw data showing the distribution of m/z 834. These raw data were then transformed using: (C) median

normalisation, (D) total ion current (TIC) normalisation, (E) TIC normalisation and weak denoising, and (F) TIC normalisation, weak denoising, and automatic hotspot removal GM, grey matter; Min, minimum; Max, maximum; WM, white matter

67 | P a g e

Chapter 3: Developing MALDI-IMS to Study Lipids in the Human Brain

Figure 3.7: Distribution of selected lipids in the postmortem human middle temporal gyrus (MTG) in a representative control section (H190). (A) Section stained post-acquisition with haematoxylin and eosin (H&E) and luxol fast blue (LFB), illustrated grey matter (in purple) and white matter regions (in blue). (B)–(L) Distribution of negatively-charged lipids. (M)–(P) Distribution of positively charged lipids. All data were normalised to total ion current (TIC), and was processed using a weak denoising filter and

68 | P a g e

automatic hotspot removal. The intensity scale with the adjusted maximum, shown by the black arrow, is shown for each image. The space between consecutive spectra was set to 50 µm. Scale bar is 0.5 mm. PC, phosphatidylcholine; GM, ganglioside; m/z, massto-charge ratio; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol pPE, phosphatidylethanolamine plasmalogen; PS, phosphatidylserine; SF, sulfatide

Chapter 3: Developing MALDI-IMS to study lipids in the human brain All positively-charged lipids, with the exception of m/z 734 (PC 32:0+H+), were detected across the whole MTG region (i.e. in both grey and white matter). Again, however, each lipid showed a unique heterogeneous pattern of distribution. For instance, m/z 788 (PC 36:1+H+) was intensely expressed in white matter, while m/z 810 (PC 38:4+H+) was abundantly detected in grey matter, reflecting the findings from the overview spectra.

3.3.3.3 Lipid expression in the middle temporal gyrus (MTG) in Alzheimer’s disease (AD) Figure 3.8A and B show the overview lipid spectra detected in the MTG region, in grey matter and white matter, respectively, in AD (averaged across all six AD datasets). The intensity of each m/z value was plotted on a relative intensity scale, from zero to one. The same lipid peaks outlined in Table 3.2 were compared. The spectra detected in positive ion mode (red) in both grey matter and white matter, were similar, with m/z 760 (PC 34:1+H+) being the most abundant signal. When calculated relative to this peak, the rest of the lipid peaks showed intensity levels that were comparable to those detected in control cases (Figure 3.5B and C). In negative mode, the most abundant peak in both regions was m/z 888 (SF 24:1-H-). This is comparable to results seen in the white matter in control cases. It is, however, different from the results seen in the grey matter in control cases, where m/z 885 (PI 38:4-H-) was the most abundant lipid peak. Differences in the intensity of lipid peaks between control and AD in negative ion mode might reflect this discrepancy. However, generally, the most abundant lipid peaks detected in both regions were similar to those observed in control cases. I then plotted images of the distribution of these lipids in AD and compared them to control cases. Figure 3.8(D–H) shows the distribution of five selected m/z values in a representative AD section (AZ72). The automatic hotspot removal for each AD case was applied to the same level as its age- and sex-matched control case to ensure accurate comparison. As seen in Figure 3.8, the pattern of distribution of the same lipid in AD was generally the same as that seen in the control case (Figure 3.7). This was true of all the lipids listed in Table 3.2. There were, however, some differences in the intensity of each lipid. For instance, the intensity of m/z 790 (PE 40:6-H-) was much lower in AD than in control cases. Next, given the large datasets, I wanted to devise an analysis workflow that would allow the efficient detection of lipid species that were differentially expressed in AD.

69 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Figure 3.8: Overview lipid spectra and distribution of selected lipids in Alzheimer’s disease (AD). (A) Overview lipid spectrum of grey matter in AD detected in positive ion mode (red) and negative ion mode (black). (B) Overview lipid spectrum of white matter in AD detected in positive ion mode (red) and negative ion mode (black). (C) Representative AD section (AZ72) stained with haematoxylin and eosin (grey matter stained purple), and luxol fast blue (white matter stained blue). The dark line through the

70 | P a g e

section is a sectioning artefact. (D)–(F) Distribution of negatively charged lipids: m/z 790, m/z 885, and m/z 888. (G) and (H) Distribution of positivelycharged lipids: m/z 788 and m/z 810. GM, grey matter; PC, phosphatidylcholine; H&E, haematoxylin and eosin; LFB, luxol fast blue; m/z, mass-to-charge ratio; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol; SF, sulfatide; WM, white matter

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.3.4. Developing the analysis workflow 3.3.4.1 Technical variation In addition to the intra-sample variation that is normally characteristic of human cohorts, I also observed considerable technical variation that is inherent to the MALDI-MS technique. This run-torun variation is illustrated in Figure 3.9, which shows data from five replicates of the same homogenised tissue standard. For the selected m/z values shown, the variation between technical replicates was higher for those m/z values that were more abundant (i.e. had a higher intensity). For example, the mean intensity and standard deviation for m/z 862.6 was 2.650 ± 1.1, while for m/z 888.7 it was 79.39 ± 28.53. Given the large variation between acquisition runs, I analysed each age- and sex-matched pair separately when using the receiver operating characteristic (ROC) option in SCiLS 2015b lab software (SCiLS GmbH, Bremen, Germany).

Figure 3.9: Data from technical replicates in reflector negative ion mode. The x-axis shows selected m/z values, with the recorded intensity (a.u.) on the yaxis. The table below indicates the mean, standard deviation, and standard error of mean for each m/z

value. This figure was prepared using GraphPad Prism 6 software (CA, USA). a.u., arbitrary unit; m/z, mass-to-charge ratio; std, standard

71 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.3.4.2 Pair-wise receiver operating characteristic (ROC) comparison The discriminative capability of each m/z peak was evaluated using the ROC analysis. Figure 3.10 shows an example of a ROC curve, which was generated for m/z 717.2 in grey matter for an age- and sex-matched pair, i.e. AZ32 and H137. The inset in Figure 3.10 shows the intensities of m/z 717.2 that were recorded at each point, in the grey matter in both cases. For each intensity level, for example 1.29127, which is highlighted in red, all spectra with intensities that are higher than this level are categorised as Class 1, while those with lower intensities are assigned to Class 2. This classification is then compared with the user-designated classification that was done prior to conducting the analysis. In this example, I had designated AZ32 as Class 1 and H137 as Class 2. At the highlighted intensity level, i.e. 1.29217, the majority of spectra will be correctly classified as AD and control. Thus, the true positive rate at this level, which is also known as the sensitivity, is high (0.815). However, because some of the lower intensity AZ32 spectra will be incorrectly classified into the Control group (Class 2), and vice versa, the false positive rate at this intensity level, which is calculated by subtracting the specificity or true negative rate from 1, is 0.158. The ROC curve is the graphical output of the calculated true positive rate against the false positive rate, at each intensity level (SCiLS, 2015). If the expression of a chosen m/z peak is significantly different between the AD and control case, there is little overlap in the distribution of intensities for that m/z peak between the datasets. Thus, the sensitivity and specificity at each level is likely to be high, and so, the area under the generated ROC curve will be large and close to 1.0. The area under the curve (AUC) for the example illustrated in Figure 3.10 was 0.910. Therefore, m/z 717.2, which was more abundant in grey matter in AD, is an excellent discriminator between AZ32 and H137. Since an AUC of 0.7 signifies m/z values that are of fair discriminatory quality (Carter et al., 2016), this was set as the threshold for all the ROC analyses in this thesis. Each region in each age- and sexmatched pair was analysed separately to find discriminatory m/z peaks that were differentially expressed in AD. The detected m/z values were then compared between all six datasets, as outlined in the next section.

72 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Figure 3.10: Receiver operating characteristic (ROC) curve for m/z 717.2 in grey matter. ROC analysis output for m/z 717.2 in the grey matter of an Alzheimer’s disease case (AZ32) and control case (H137). The false positive rate, which was calculated by subtracting the specificity from one, is graphed on the x-axis, while the true positive rate (sensitivity) is graphed on the y-axis. Inset: The intensity box plot showing all the intensities generated for m/z 717.2 in

all spectra acquired from grey matter in AZ32 and H137. The ROC curve and intensity boxplots were generated using SCiLS lab 2015b software (SCiLS GmbH, Bremen, Germany). AUC, area under the curve; Da, daltons; m/z, mass-tocharge ratio

3.3.4.3 Determining m/z values of interest A list of m/z peaks detected using the ROC function was then exported into the six-set Venn feature of the InteractiVenn online tool (Heberle et al., 2015) to determine m/z peaks that were consistently detected in all six datasets. Each histologically-defined region was analysed separately.

73 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain Figure 3.11 shows an example of the six-set Venn diagram output that was used to analyse m/z values

that were increased in grey matter in the MTG region in AD. The number of m/z values that were detected using the ROC function (given in brackets) was different for each age- and sex-matched pair. The largest number of discriminatory m/z values (453) was detected between the AZ71/H238 pair, while the lowest (37) was detected between AZ72/H190. The quadrant circled in red indicates the seven m/z values that were higher in AD in all six datasets. Once the m/z values were selected, the changes in relative intensity between AD and control were calculated and graphed using GraphPad Prism 6 software (CA, USA). To show the differential expression of each m/z value, MALDI images for each m/z value were generated using SCiLS lab 2015b software (SCiLS GmbH, Bremen, Germany).

Figure 3.11: Six-set Venn diagram to determine m/z value differences common to all datasets. This example shows the output of the analysis conducted to detect m/z values that were expressed at a higher intensity in the grey matter of the middle temporal gyrus (MTG) in Alzheimer’s disease (AD). The discriminative m/z values detected using the receiver operating characteristic (ROC) function for each data

74 | P a g e

set were exported into the InteractiVenn online tool (Heberle et al., 2015). The quadrant circled in red indicates the m/z values that were increased in the MTG grey matter in AD in all six datasets. Each ageand sex-matched data pair is shown by a different colour and the numbers within brackets by each pair indicate the number of m/z values detected using the ROC function.

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.3.4.4 Summary of analysis workflow The final analysis workflow that has been developed thus far is illustrated in Figure 3.12. Briefly, all raw data are first aligned to an internal calibration list, which is based on previous publications, using FlexAnalysis 3.4 software. Then, the data are imported into SCiLS lab (SCiLS GmbH, Bremen, Germany), using a TopHat baseline removal algorithm. Next, data are normalised using the TIC, and the weak denoising algorithm is applied. The regions of interest (i.e. the grey and white matter) are then delineated based on a co-registered image of the same section that has been stained using H&E and LFB, or can be defined according to the results of the spatial segmentation function, which analyses clusters of similar spectra. Regions of each age- and sexmatched pair are then analysed using the ROC function, to find m/z values that are increased or decreased in AD. The data are then imported into a six-set Venn diagram to find m/z values that are consistently different in all six datasets. The relative intensity of these m/z values in AD are then calculated as a percentage change from control cases, in each region of interest. Finally, the distribution of each m/z value is generated using SCiLS lab software (SCiLS GmbH, Bremen, Germany). I then applied this analysis workflow to the MTG dataset to detect differentially expressed lipid species in this region in AD.

75 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Figure 3.12: Data analysis workflow. Once data were acquired, spectra were aligned to an m/z list based on previous publications (Veloso et al., 2011b, Yuki et al., 2011), using FlexAnalysis 3.4 software. The datasets were then imported into SCiLS lab 2015b software, and total ion current (TIC) normalisation and weak denoising was applied. Next, age- and sex-matched control and Alzheimer’s disease (AD) cases were statistically analysed using the receiver operating characteristic (ROC) function. The m/z values that

76 | P a g e

were consistently statistically different across all six datasets were chosen. Distribution maps of the selected m/z values were generated using SCiLS lab 2015b software (SCiLS GmbH, Bremen, Germany). Graphs showing relative intensity changes in Alzheimer’s disease, for each region of interest, were generated using GraphPad Prism 6 software (CA, USA). m/z, mass-to-charge ratio

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.3.5. Differentially expressed lipid species in Alzheimer’s disease 3.3.5.1 Negative ion mode Figure 3.13A shows m/z values that were detected as being differentially expressed across the whole MTG region that was analysed. Although these lipids could not be identified based on the MS/MS analysis, I was able to assign putative lipid assignments based on previous publications (Ariga et al., 1982, Han et al., 2001, Chan et al., 2009, Whitehead et al., 2011). I detected nine lipids that were differentially expressed, of which eight showed a decrease in AD. These were m/z 743.6, 751.6, 778.7 (pPE 40:4-H-), 779.7, 791.7, 795.7, 1545.0, and 1574.0 (GM1 38:1-H-). The only lipid that was increased was m/z 909.7. Figure 3.13B shows m/z values that were changed in the grey matter alone. These changes reflect those that were detected in the whole MTG region. The increase in m/z 717.2, however, was only detected when grey matter was analysed alone. Figure 3.13C shows m/z values that were changed in the white matter alone. Only m/z 795.7 (PE 40:4-H-) was consistently decreased across all six datasets in this region. Figure 3.14 shows relative changes of selected m/z values, grouped by sex. The arrows indicate a discrepancy seen in the relative changes between the male and female cohorts. When grouped by sex, the male cohort showed an increase in m/z 717.2 across the whole MTG region too (Figure 3.14A). Since all six datasets consistently showed the same changes in grey matter region for the chosen m/z values, when grouped by sex, both male and female cohorts showed the same relative change for these m/z values (Figure 3.14B). Interestingly, in the white matter region the male cohort alone showed a decrease in m/z 791.7, 1545.0, and 1574.0 (GM1 38:1-H-), and an increase in m/z 909.7 (Figure 3.14C). Figure 3.15 illustrates the distribution of these lipids in the MTG in representative control (H190) and AD (AZ72) cases. With the exception of m/z 717.2, all the lipids that were differentially expressed in AD, were abundantly expressed in grey matter. Again though, even within the grey matter, each lipid showed a unique heterogeneous distribution. The intensity of all lipids, except m/z 909.7, was lower in AD, reflecting the trends seen in Figure 3.13 and Figure 3.14.

77 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Figure 3.13: Relative mean intensity changes of selected m/z values, detected in negative ion mode, in Alzheimer’s disease (AD; as a % change from control). Graphs showing the mean intensity difference of selected m/z peaks in AD, relative to the intensity in control in: (A) the whole middle temporal gyrus (MTG) region, (B) grey matter region in the MTG, and (C) white matter region in the MTG. Black

78 | P a g e

arrows, specifying an increase or decrease, indicate the lipid species that were consistently change in all six cases. An upward arrow indicates an increased intensity in AD, while a downward arrow indicates a decrease. Individual cases are shown as indicated. m/z, mass-to-charge ratio

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Figure 3.14: Relative mean intensity change of selected m/z values, detected in negative ion mode, in Alzheimer’s disease (AD; as a % change from control), grouped by sex. Arrows indicate relative changes in just one sex group (and not the other), in the (A) whole middle temporal gyrus (MTG) region,

(B) grey matter, and (C) white matter. The direction of the arrow indicates an increase or decrease in the mean intensity (as a % change from control). m/z, mass-to-charge ratio

79 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Figure 3.15: Spatial distribution of selected lipids, which were detected in reflector negative ion mode and differentially expressed in the middle temporal gyrus (MTG) in Alzheimer’s disease (AD). (A) Representative MTG sections (AZ72 and H190) stained with haematoxylin and eosin (H&E) and luxol fast blue (LFB), indicating grey matter (purple) and white matter (blue). (B)–(K) Edge-preserving image denoising and automatic hotspot removal (see rainbow intensity colour-bar) was applied using SCiLS

80 | P a g e

Lab 2015b software (SCiLS GmbH, Bremen, Germany). The spatial distance between adjacent spectra is 50 μm. Each image shows the distribution of a specific m/z value. The distribution of each lipid is shown in representative control (black box) and AD (red box) sections. GM, grey matter; GM1, ganglioside; m/z, mass-tocharge ratio; PE, phosphatidylethanolamine; pPE, phosphatidylethanolamine plasmalogen; WM, white matter

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.3.5.2 Positive ion mode Figure 3.16 shows the differential expression of positively-charged lipids in the MTG in AD. The values have been grouped by sex. Although these lipids could not be identified based on the MS/MS analysis, I was able to make putative lipid assignments based on previous publications (Jackson et al., 2005, Hicks et al., 2006, Woods and Jackson, 2006, Jackson et al., 2007, Fuchs et al., 2008). There were nine positively-charged lipids that were differentially expressed across the whole MTG region, particularly in the grey matter, in the AD male cohort. These were m/z 763.7, m/z 786.6 (PC 36:2+H+), m/z 787.8, m/z 788.8 (PC 36:1+H+), m/z 789.7, m/z 810.7 (PC 38:4+H+), m/z 811.7, m/z 812.7 (PC 38:3+H+), and m/z 834.7 (PC 40:6+H+). All these lipids, with the exception of m/z 763.7, were also decreased in the white matter region in the male AD cohort (Figure 3.16C). All the lipids, which could be identified, based on previous publications were PC species. Since I did not correct for isotopes and given that some of the unidentified m/z values are 1 Da apart, the m/z values that could not be identified could be isotopes of the species to which lipid assignments have been made. Further, since the relative intensity change in the AZ32/H137 dataset (female) seemed to be consistently higher for all these m/z values, sex may play a role in these lipid differences in AD. However, this remains to be confirmed in future studies. Figure 3.17 shows the distribution of these positively-charged lipids in the MTG. Lipids m/z 763.7 and m/z 834.7 (PC 40:6+H+) were highly abundant in grey matter. Lipids m/z 786.6 (PC 36:2+H+), m/z 787.8, m/z 788.8 (PC 36:1+H+), and m/z 789.7 were more abundantly expressed in white matter. Apart from m/z 788.8, the signal for all these lipids was quite low in grey matter. Lipids m/z 810.7 (PC 38:4+H+), m/z 811.7, and m/z 812.7 (PC 38:3+H+), were expressed across the MTG section, but were more intensely expressed in grey matter. Although the distribution of these lipids did not change in AD, the decreased abundance of these lipids is evident. 3.3.5.3 Summary Table 3.3 summarises the differential expression of lipids in AD in the MTG. Overall, 19 lipids (within the m/z range from 500 to 2000) were differentially expressed in AD in the MTG. The only two negatively-charged lipids that could be identified were a pPE and GM1 species, while all of the positively-charged lipids were PC species. The pPE and GM1 species were highly expressed in grey matter, while the distribution of the PC species in the MTG varied. Finally, the positively-charged lipids only showed a consistent change in the male AD cohort.

81 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Figure 3.16: Relative mean intensity changes of selected m/z values, detected in positive ion mode, in Alzheimer’s disease (AD; as a % change from control), grouped by sex. Arrows indicate relative changes in the (A) whole middle temporal gyrus

82 | P a g e

(MTG) region, (B) grey matter, and (C) white matter. The direction of the arrow indicates an increase or decrease the mean intensity (as a % changes from control). m/z, mass-to-charge ratio

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Figure 3.17: Spatial distribution of selected lipids, which were detected in reflector positive ion mode and differentially expressed in the middle temporal gyrus (MTG) in Alzheimer’s disease (AD). (A) Representative MTG sections (AZ72 and H190) stained with haematoxylin and eosin (H&E) and luxol fast blue (LFB), indicating grey matter (purple) and white matter (blue). (B)–(J) Edge-preserving image denoising and automatic hotspot removal (see

rainbow intensity colour-bar) was applied using SCiLS Lab 2015b software (SCiLS GmbH, Bremen, Germany). The spatial distance between adjacent spectra is 50 μm. Each image shows the distribution of a specific m/z value. The distribution of each lipid is shown in representative control (black box) and AD (red box) sections. GM, grey matter; m/z, mass-to-charge PC, phosphatidylcholine; WM, white matter

ratio;

83 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain Table 3.3: Summary of the relative change of selected lipid species in the middle temporal gyrus (MTG) in Alzheimer’s disease (AD), detected in negative and positive ion modes. An indication of how the mean intensities of selected lipid species change in AD in the analysed MTG region, and in grey matter and white matter, is given. Arrows indicate either an increase (↑) or decrease (↓), with the mean percentage change from control given in

Positive Ion Mode

Negative Ion Mode

Observed m/z

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.

717.2 743.6 751.6 778.7 779.7 791.7 795.7 909.7 1545.0 1574.0 763.7 786.6 787.8 788.8 789.7 810.7 811.7 812.7 834.7

Lipid assignment

pPE 40:4-H- [1, 2]

GM1 38:1-H- [3-5] PC 36:2+H+ [6, 7] PC 36:1+H+ [6-10] PC 38:4+H+ [6, 7, 11] PC 38:3+H+ [6] PC 40:6+H+ [6, 7]

Han et al. (2001) Han (2005) Ariga et al. (1982) Chan et al. (2009) Whitehead et al. (2011) Hicks et al. (2006) Jackson et al. (2005) Jackson et al. (2007) Fuchs et al. (2008) Woods and Jackson (2006) Lehmann et al. (1997)

84 | P a g e

brackets. For differentially expressed lipids detected in positive ion mode, the mean percentage change in brackets indicates the change in the male cohort alone. Lipid assignments are based on previous publications [indicated in brackets and listed below]. GM, ganglioside; PC, phosphatidylcholine; m/z, mass-to-charge ratio; PE, phosphatidylethanolamine; pPE, phosphatidylethanolamine plasmalogen

Whole MTG ↓ (-57.04) ↓ (-42.49) ↓ (-44.55) ↓ (-47.14) ↓ (-56.54) ↓ (-63.43) ↑ (23.99) ↓ (-60.96) ↓ (-72.80) ↓ (-38.50) ↓ (-17.61) ↓ (-20.87) ↓ (-10.27) ↓ (-14.21) ↓ (-54.65) ↓ (-57.66) ↓ (-49.58) ↓ (-51.63)

Region Grey matter ↑ (69.35) ↓ (-52.86) ↓ (-39.34) ↓ (-39.99) ↓ (-45.33) ↓ (-50.87) ↓ (-60.52) ↑ (36.68) ↓ (-55.35) ↓ (-69.09) ↓ (-25.46) ↓ (-14.75) ↓ (-17.44) ↓ (-14.02) ↓ (-18.31) ↓ (-40.95) ↓ (-46.09) ↓ (-40.64) ↓ (-38.23)

White matter

↓ (-38.29)

↓ (-27.66) ↓ (-31.55) ↓ (-14.93) ↓ (-21.04) ↓ (-48.12) ↓ (-48.24) ↓ (-34.64) ↓ (-34.56)

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

Section Four

Discussion

3.4.1. General discussion

3.4.1.1. Methods to analyse lipids First introduced by French (1926), Oil Red O, a lysochrome (fat soluble dye), stains neutral lipids, while PAS is used extensively to investigate structures containing neutral hexose sugars and/or sialic acids (Kiernan, 1999, Bancroft and Gamble, 2008). These stains have been previously used to investigate changes in lipid metabolism in AD (Burger and Vogel, 1973, Bugiani et al., 1999, Mandas et al., 2012). In Figure 3.1, Oil Red O and PAS show lipofuscin staining, which was present in both neurologically normal, but aged, postmortem human brain sections, as well as in AD. The accumulation of lipofuscin, which has been implicated in normal post-mitotic cell aging, consists of polymeric lipid and phospholipids structures, and amino acids (Taubold et al., 1975, Double et al., 2008), which explains the positive Oil Red O and PAS staining. Although these stains enable us to visualise the accumulation of lipids in normal aging and AD, they do not allow us to differentiate the different lipid classes that are present in lipofuscin. In contrast, a trial I conducted using whole MTG sections that were analysed using LC-MS, provided detailed information about the abundance of different lipid classes and their relative changes in AD. However, given the extraction method required for this analytical approach, I could not analyse differences between grey and white matter post-hoc or visualise its anatomical distribution. Thus, I chose to optimise and develop the use of MALDI-IMS to conduct a spatially-resolved lipidomic analysis of the postmortem human brain in AD.

3.4.1.2. Optimising matrix-assisted laser desorption/ionisation (MALDI)–imaging mass spectrometry (IMS) In keeping with previous reports (Thomas et al., 2012) of the different matrices that were available for use, DAN provided high-quality spectra in both negative and positive ion modes, and ensured that a single preparatory step was sufficient to detect a multitude of different lipid species. Further, Angel et al. (2012), demonstrated the efficacy of using 50 mM ammonium formate (pH 6.4) in enhancing the acquired signal. My study replicated this finding, even with the use of DAN as a MALDI matrix, instead of DHB. Thus, this protocol, i.e. the sublimation of DAN following a 50 mM ammonium formate wash, was chosen for subsequent analyses. Laser power, laser beam diameter, number of laser shots, pulse delay, and grid voltage, were optimised to acquire high-quality spectra. The same parameters were used for each acquisition run.

85 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain However, even within the same acquisition run, there is a possible drift due to the calibration of different components, which is inherent to mass spectrometry and can lead to a systematic error (Rubakhin and Sweedler, 2010). Further, other sources of variation such as metastable matrix clusters, fluctuations in detector gain, ion source contamination, or micro-environment chemical inhomogeneities including salt and pH gradients, produce the baseline and noise that are characteristic of mass spectra (Deininger et al., 2011, Sun and Markey, 2011, Trede et al., 2011, Fonville et al., 2012). Despite the application of an external calibrant prior to each MALDI imaging run, mass drift still occurred, leading to misalignment between the datasets when imported to the SCiLS lab software (SCiLS GmbH, Bremen, Germany), as illustrated in Figure 3.4. Therefore, I chose to realign all the datasets to an internal calibrant, based on a list of known m/z values of lipid species previously detected in the human brain (Veloso et al., 2011b, Yuki et al., 2011). Other pre-processing approaches, as detailed below, were also used to distinguish the intensity peaks from the baseline and noise, thereby conditioning subsequent statistical analysis to examine differences in the biomolecule expression between samples alone (Norris et al., 2007). The baseline generally appears as an exponential that decays with the m/z value (Williams et al., 2005). A TopHat morphological algorithm was applied to correct the baseline when importing datasets into SCiLS lab software (SCiLS GmbH, Bremen, Germany). This algorithm subtracts the morphological opening of data (a mathematical term to define baseline signals) from the original spectrum, performing a contrast enhancement (Sauve and Speed, 2004, SCiLS, 2015). Normalisation, in contrast, aims to minimise spectral intensity variations, transforming measured intensities to comparable scale, without altering biological information (Norris et al., 2007, Deininger et al., 2011, Fonville et al., 2012). On the recommendation of Deininger et al. (2011), I visually compared exemplary m/z images, following two different approaches of normalisation, with the histologically stained section. Since the TIC normalisation provided m/z images that were more comparable, it was subsequently applied for all datasets. In the TIC normalisation approach each spectral intensity is divided by the sum of all its intensities (Trede et al., 2011). Finally, when imaging the distribution of selected m/z values, I chose to apply an edge-preserving imaging denoising filter, which removes the high-frequency chemical noise that is inherent to MALDI data (Trede et al., 2011). The edge-preserving imaging denoising filter can be adjusted to the level of denoising to the local noise level and to the local scale of the features to be resolved (Alexandrov et al., 2010, Trede et al., 2011). Thus, weak denoising, which applies the denoising filter in a 3x3 neighbourhood (SCiLS, 2015), was chosen for use in this thesis. 86 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.4.1.3. Distribution of lipids in the human middle temporal gyrus (MTG) Once the MALDI-IMS parameters and data pre-processing techniques were optimised, the lipid expression in control MTG sections was analysed. Grey matter and white matter showed unique lipid spectra profiles (Figure 3.5), which were reflected in the images showing the distribution of m/z values of known lipid species (Figure 3.7). PC constitutes about 32.8% of total glycerophospholipids found in the human brain, and of all the PC species expressed in the human brain, PC 16:0/18:1 (34:1) is most predominant (O'Brien and Sampson, 1965a, O'Brien and Sampson, 1965b, Naudí et al., 2015). The results from this thesis reflect this finding, since m/z 760 (PC 34:1+H+) was the predominant peak detected in positive ion mode, in both grey and white matter. This PC species showed a homogeneous distribution across the MTG. This pattern of distribution has also been previously reported in the human frontal cortex (Veloso et al., 2011a). The rest of the high-intensity peaks, to which lipid assignments could be made, were all PC species too, reflecting the high abundance of PC in the MTG, confirming results from the LC-MS trial (Figure 1.2), and previous publications on the rodent (Jackson et al., 2005, Hicks et al., 2006) and human brain (Isaac et al., 2003, Han, 2005). The advantage of using MALDI-IMS over more conventional mass spectrometry methods, however, is that the distribution of each PC species could be imaged. Lipid m/z 788 (PC 36:1+H+) was more abundant in white matter, while m/z 734 (PC 32:0+H+) and m/z 810 (PC 38:4+H+) were highly expressed in grey matter. These distribution patterns reflect those previously reported in the human frontal cortex (Veloso et al., 2011a). The predominant peak detected in grey matter in the MTG, in negative ion mode, was m/z 885 (PI 38:4-H-), in agreement with previous findings in the rat brain (Jackson et al., 2007, Cerruti et al., 2012). Although the distribution of PI in the human cortex has not previously been imaged using MALDI-IMS, this pattern reflects the pattern that was previously detected in the human putamen and hippocampus (Veloso et al., 2011b). PE constitutes approximately 35.6% of the total glycerophospholipids in the human brain, and 42.2% and 34.1% of those found in grey matter and white matter, respectively (Naudí et al., 2015). PPE makes up about 50-60% of the total PE content. (O'Brien and Sampson, 1965a, O'Brien and Sampson, 1965b, Naudí et al., 2015). Both PE species detected in this work, i.e. m/z 744 (PE 36:1-H-) and m/z 790 (PE 40:6-H-), were abundantly expressed in grey matter, while m/z 726 (pPE 36:2-H-) was highly expressed in white matter. These results reflect previous findings in the rat brain (Kino et al., 1982, Jackson et al., 2007). A similar pattern of distribution of PE and pPE was also seen in the human superior frontal cortex, superior temporal cortex, inferior parietal cortex, and the cerebellum, using electrospray ionisation-mass spectrometry requiring sample extraction (Han et al., 2001). An advantage of using the MALDI-IMS approach is that the heterogeneous expression of m/z 744 (PE 87 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain 36:1-H-) and m/z 790 (PE 40:6-H-), even within grey matter, i.e. that it is more abundant in higher cortical layers, is evident. Further, Veloso et al. (2011a) described the presence of positively-charged PE species in white matter, at least in the frontal cortex (Brodmann area 8). However, the m/z peaks for positively-charged PE lipids reported by Veloso et al. (2011a) could not be detected in this study. Thus, the different pattern of distribution of positively-charged lipids could not be confirmed using this data. This discrepancy could be attributed to an anatomical variation between the different cortical regions, or to their use of 2-mercaptobenzothiazole as a matrix, instead of DAN. However, this remains to be confirmed. Both GM1 species, m/z 1544 (GM1 36:1-H-) and m/z 1572 (GM1 38:1-H-), were also abundantly expressed in grey matter, confirming previous findings in the human brain, particularly in the frontal and temporal cortex in the human brain (Svennerholm, 1963, Kracun et al., 1989, Pernber et al., 2012). The ratio of m/z 1544 (GM1 36:1-H-) to m/z 1572 (GM1 38:1-H-), however, was different to that detected in the rat cerebellar cortex (Woods and Jackson, 2006), and the human hippocampus (Hirano-Sakamaki et al., 2015), where m/z 1544 (GM1 36:1-H-) was the more abundant GM1 species. Although this discrepancy might reflect case-specific differences, this remains to be confirmed in future work. The most abundant m/z peak in white matter (in negative ion mode) was m/z 888 (SF 24:1-H-). There were also two hydroxylated SF species, m/z 904 (SF 24:1 (OH)-H-) and m/z 932 (SF 26:1 (OH)-H-), which were abundant in white matter. However, the expression of these hydroxylated SF extended into the grey matter region, reflecting previous findings in the human cortex (region not well defined; Yuki et al., 2011). Although previous publications have demonstrated the high abundance of sphingomyelin (SM) in white matter and its change in AD (Nitsch et al., 1992, Wells et al., 1995, Pettegrew et al., 2001, Bandaru et al., 2009, He et al., 2010, Chan et al., 2012), SM was scarcely detected in the human postmortem middle temporal gyrus tissue used in this thesis. The distribution of some SM species in the postmortem human brain tissue has been imaged using MALDI-IMS (Veloso et al., 2011a, Veloso et al., 2011b). However, since a saturated solution of 2-mercaptobenzothiazole in methanol was used as the matrix for these studies, the low detection of SM in the current thesis could be attributed to the use of DAN as a matrix. The PS species, m/z 788 (PS 36:1-H-) and m/z 834 (PS 40:6-H-), showed a contrasting pattern of distribution, in that the former was abundant in white matter, while the latter was highly expressed in grey matter. This trend has also been previously seen in the rat cerebellum (Jackson et al., 2007),

88 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain and the distribution of m/z 788 matches that reported in the human caudate nucleus and hippocampus (Veloso et al., 2011b). With the exception of positively-charged PE species and SM, the lipid distribution results from the MTG replicate those from previous studies on the lipidome of the mammalian brain. This helped validate the optimisation protocol and the use of DAN as a MALDI-IMS matrix for imaging lipids in the human brain. This body of work also extends the understanding of the similarities and differences in the distribution of lipids in the MTG region, in comparison to other regions in the human brain. Finally, these results highlight the advantage of imaging lipid distributions in a region of interest, given that some lipid species that belong to the same class can show varied patterns of distribution.

3.4.1.4. Differential lipid expression in the middle temporal gyrus (MTG) in Alzheimer’s disease (AD) The overall lipid spectrum acquired in positive ion mode, from both grey matter and white matter in AD, was similar to that acquired in control cases (Figure 3.8). In negative ion mode, however, the most abundant lipid peak was m/z 888 (SF 24:1-H-), in both regions, which could have led to the change in relative intensities of other lipids peaks. Nonetheless, the overall lipid profile, i.e. the rest of the most abundant lipids in each region, was generally the same in AD, and the pattern of distribution for each remained the same too. Given the variation between datasets, the workflow developed in this chapter enabled us to efficiently analyse the large MALDI-IMS datasets and detect differential lipid expression in each anatomical region. The majority of lipids that were differentially expressed in AD, to which lipid assignments could be made, were PE, pPE, GM1, and PC species, which were all decreased. There were also two lipids, m/z 717.2 and m/z 909.7, which were elevated in AD. However, given the limitation in identifying these lipids, their role in AD cannot be speculated. The decreases in pPE and GM1 in the MTG in AD (Figure 3.14, Figure 3.15, and Table 3.3) reflect the changes seen in all regions analysed in the human brain in AD, including the frontal, parietal and temporal cortices (Nitsch et al., 1992, Wells et al., 1995, Prasad et al., 1998, Han et al., 2001, Pettegrew et al., 2001, Martın et al., 2010). Since the AD cohort used in this study represents endstage disease, the pPE that were differentially expressed, were highly abundant in the grey matter, an area that has previously shown a gradual increase in the pPE deficiency from 10% to 30% (Han et al., 2001). PE and pPE species that are linked to polyunsaturated fatty acids (PUFAs), namely arachidonic acid (AA; 20:4n-6) and docosahexaenoic acid (DHA; 22:6n-3), are highly oxidizable (Butterfield and Lauderback, 2002, Hartmann et al., 2007, González-Domínguez et al., 2014). Given the highly oxidative environment of the AD brain, these PUFAs could serve as substrates for lipid

89 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain peroxidation (Butterfield and Lauderback, 2002). The depletion in these lipids can lead to membrane destabilisation. Further, even small alterations in pPE content are known to affect the critical temperature, disrupting membrane stability, which could lead to loss of synapses seen early in AD (Ginsberg et al., 1998, Han et al., 2001). Lipid peroxidation further generates aldehydes, such as 4hydroxynonenal, which in turn can oxidize other proteins and inhibit glycolysis, driving AD pathogenesis (González-Domínguez et al., 2014). However, this pathological function remains to be confirmed in the MTG. The decrease in m/z 1574 (GM1 38:1-H-) in the MTG in AD, reflects the depletion of GM1 in grey matter of the human frontal and temporal cortices (Svennerholm and Gottfries, 1994, Pernber et al., 2012). GM1 is also expressed in large neurons that are widely spread in grey matter, where it aggregates in lipid rafts in the presence of apolipoprotein E4 and cholesterol (Ariga et al., 2008, Lingwood and Simons, 2010, Pernber et al., 2012). Here it can bind Aβ1-40 and Aβ1-42, driving its conformational change and elevating plaque formation (Yanagisawa et al., 1995, Kakio et al., 2002, Yanagisawa, 2007). The depletion of GM1 38:1-H- might reflect the neurodegeneration of these cells. However, given that I could only detect this GM species, further work, specifically investigating the expression of other accurately identified lipids in the same class is required. The decrease in PC in the MTG in AD (Figure 3.16, Figure 3.17, Table 3.3) reflects previous results based on the human frontal and parietal cortices (Nitsch et al., 1992, Wells et al., 1995, Guan et al., 1999). The five PC species detected in this study were expressed across the MTG, in both grey and white matter, albeit at varying intensities, reflecting the decrease seen in both regions. The fatty acid tails of two of the PC species were AA and DHA, reflecting the previously reported decrease of these PUFAs in the human prefrontal cortex (Igarashi et al., 2011). Give the high susceptibility of these PUFAs to oxidative stress, and their role in determining the physical properties of the lipid bilayer, their loss may lead to neurodegeneration (Farooqui et al., 2000). However, again, given that I am studying an end-stage AD cohort, it is difficult to determine if these PC changes are the agents or consequence of disease.

90 | P a g e

Chapter 3: Developing MALDI-IMS to study lipids in the human brain

3.4.2. Summary of findings The aim of this chapter was to study the lipidome of the postmortem human MTG, and its changes in AD, using MALDI-IMS. However, the sample preparation, data acquisition parameters, and preprocessing methods, had to be first optimised. In the normal MTG, grey and white matter had a lipid spectral profile that was unique to each region, especially in negative ion mode, which was reflected when the distribution of selected lipids were imaged. In AD, a similar lipid profile was detected in the grey matter and white matter in the MTG. However, given that the signal for m/z 888 (SF 24:1-H-) was the most abundant in both regions, there were differences in the relative signal intensity in these regions. Nonetheless, the pattern of distribution of these lipids did not change in the AD MTG. However, given the large variation in raw data, which is inherent to this technique, none of the abundant lipid peaks, to which lipid assignments could be made, were statistically differently expressed in AD. Thus, I developed a workflow to efficiently analyse true differences between the large control and AD datasets. Briefly, a statistical test, i.e. the ROC function, was used to detect m/z values that could be used to discriminate control and AD tissue, in each age- and sex-matched pair. The relative intensity changes in AD of the detected m/z values, which were consistently detected in each pair (n=6), were then graphed as a percentage change from control. Finally, the distribution of these lipids was imaged and reflected the results seen in the graph. The majority of lipids that were differentially expressed in AD were PC, pPE, and GM1 species, which were all reduced in the grey matter of the MTG.

91 | P a g e

Chapter Four

Lipid changes in the postmortem human hippocampus in Alzheimer’s disease

Section One: Introduction..............................................................................................................94 Section Two: Methods..................................................................................................................95 4.2.1. Matrix-assisted laser desorption/ionisation (MALDI)-imaging mass spectrometry (IMS) .......... 95 4.2.2. Data analysis ................................................................................................................................ 95 4.2.3. Lipid assignments ......................................................................................................................... 96 Section Three: Results..................................................................................................................97 4.3.1. Lipids in the postmortem human hippocampus .......................................................................... 97 4.3.1.1. Liquid-chromatography data ............................................................................................... 97 4.3.1.2. Matrix-assisted laser desorption/ionisation (MALDI)–imaging mass spectrometry (IMS) data ...................................................................................................................................... 99 4.3.1.3.Lipid profiles of different hippocampal sub-fields ............................................................... 105 4.3.2. Differentially expressed negatively charged lipids in Alzheimer’s disease................................ 108 4.3.2.1. Whole hippocampus ........................................................................................................... 108 4.3.2.2. White matter ...................................................................................................................... 108 4.3.2.3. Grey matter ........................................................................................................................ 108 4.3.2.4. Cornu Ammonis regions ..................................................................................................... 108 4.3.2.5. Dentate gyrus ..................................................................................................................... 110 4.3.2.6. Summary ............................................................................................................................ 110 4.3.3. Differentially expressed positively charged lipids in Alzheimer’s disease ................................. 113 4.3.3.1. Whole hippocampus ........................................................................................................... 113 4.3.3.2. White matter ...................................................................................................................... 113 4.3.3.3. Grey matter ........................................................................................................................ 113 4.3.3.4. Cornu Ammonis regions ..................................................................................................... 113 4.3.3.5. Dentate gyrus ..................................................................................................................... 115 4.3.3.6. Summary ............................................................................................................................ 115 Section Four: Discussion...............................................................................................................118 4.4.1. General discussion ..................................................................................................................... 118 4.4.2. Summary of findings .................................................................................................................. 123

The results from this chapter have been peer-reviewed and published: Mendis, L. H. S., Grey, A. C., Faull, R. L. M. and Curtis, M. A. (2016), Hippocampal lipid differences in Alzheimer's disease: a human brain study using matrix-assisted laser desorption/ionizationimaging mass spectrometry. Brain and Behavior, 6: 1–17. e00517, doi: 10.1002/brb3.517

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

Section One

Introduction

The hippocampus, which is a bilaminar structure consisting of the Cornu Ammonis (CA) and the dentate gyrus (DG), is part of the limbic lobe, and plays a central role in memory and spatial navigation (Scoville and Milner, 1957, O'Keefe and Dostrovsky, 1971, Anderson et al., 2007). Severe atrophy of the hippocampal formation in Alzheimer’s disease (AD), leads to the deficit in recent memory that is the hallmark of cognitive dissonance (Costa et al., 1997). Pathologically, the deposition of beta amyloid (Aβ) occurs at Stage II (Thal et al., 2002) in the hippocampus, and formation of neurofibrillary tangles (NFTs) are present by Stage III and IV described by Braak and Braak (1991). Although the presence of lipid aberrations has been previously outlined in the human brain in AD, only a few studies have focussed specifically on the human hippocampus. These report a decrease in phosphatidylethanolamines (PE) and ethanolamine plasmalogens (pPE), and specific reduction in phosphatidylcholines (PC) and phosphatidylinositols (PI) in the the synapatosome plasma (Wells et al., 1995, Guan et al., 1999). Only three studies report sphingolipid differences in AD, and these have specifically focussed on ganglioside (GM) changes. GM1 was increased in grey matter, with a decrease in b-series GM (Svennerholm et al., 1994, Valdes-Gonzalez et al., 2011). Further, HiranoSakamaki et al. (2015) described a specific decrease in the ratio of GM1 d20:1/C18:0 to GM1 d18:0/C18:0 in the molecular layer of the DG. Thus, more work is still required to understand how other sphingolipid species might be changed in the hippocampus in AD. The extraction techniques used for the majority of studies are not conducive for mapping the distribution of these changes. Further, only Veloso et al. (2011b) and Hirano-Sakamaki et al. (2015) have imaged the distribution of lipids in the human hippocampus using matrix-assisted laser desorption/ionisation (MALDI)-imaging mass spectrometry (IMS). Therefore, the overall aim of the current chapter was to image the distribution of lipids in the hippocampus and determine the differential expression of lipids in AD in precisely-delineated hippocampal sub-fields.

94 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

Section Two

Methods

4.2.1. Matrix-assisted

laser

desorption/ionisation–imaging

mass

spectrometry (MALDI–IMS) As outlined in Table 2.1, three control and AD cases were used to acquire pilot data, and six age- and sex-matched pairs were used to analyse differential lipid expression. The sections were prepared and analysed as outlined in Section 2.1.2. Pilot data were acquired using a step-size of 150 µm, while the rest of the datasets were acquired using a step-size of 100 µm.

4.2.2. Data analysis 4.2.2.1. Segmentation The spatial segmentation tool in SCiLS lab 2015b software (SCiLS GmbH, Bremen, Germany) was used, with the edge-preserving image denoising, as a first step in data mining. This analysis was based on bisecting k-means, which is a top-down clustering method that statistically groups spectra into two clusters that are maximally different (Steinbach et al., 2000, Trede et al., 2012). Each hippocampal region was analysed separately.

4.2.2.2. Analysis workflow Data were analysed using the workflow outlined in Chapter Three (Figure 3.12). However, the segmentation tool, described in the previous section, was first used to differentiate the white matter and grey matter regions. Next, a co-registered high-resolution scan of the section, stained with haematoxylin and eosin (H&E) and luxol fast blue (LFB), was used to trace out the other grey matter regions of interest (ROI; CA1, CA2/3, CA4 and DG regions) based on their histological appearance. Distribution maps of these selected m/z values were generated using SCiLS lab 2015b, with automatic hotspot removal and edge-preserving weak image denoising. The relative intensity change of these selected m/z values in AD was calculated as a percentage (%) change from control, and graphed for each region using GraphPad Prism 6 software.

95 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

4.2.3. Lipid assignments Lipid identifications were made in comparison with published mammalian lipid identifications where possible. For others, on-tissue MALDI-tandem mass spectrometry (MS/MS), which was performed using a UltrafleXtreme MALDI-time of flight (TOF)/TOF mass spectrometer (Bruker Daltonics Gmbh, Bremen, Germany) and analysed using the LIPID MAPS database (Fahy et al., 2007), was used. However, given the mass resolution of the MALDI-TOF and the relatively wide precursor ion selection window for MALDI-TOF/TOF analysis, product ion peaks of isobaric lipids were also present within some of the MS/MS spectra. Thus, liquid-chromatography (LC)-MS/MS, which was performed using a Thermo Finnigan LTQ-FT (Linear Ion Trap-Fourier Transform) mass spectrometer (Thermo Electron, Bremen, Germany) and analysed using LipidSearch software (Thermo Fisher Scientific, MA, USA), was used to validate MALDI-MS/MS results where possible. The results of the MALDI-MS/MS and LCMS/MS analysis are included as supplementary data. While structural assignments for some lipids could not be made due to a low abundance, those detected in the higher m/z range in positive ion mode could not be identified accurately using the LIPID MAPS database as the library is not yet complete and is being continuously updated (Han, 2016). Overall, putative lipid assignments have been made for 22 of the 43 lipids that were differentially expressed in AD.

96 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

Section Three

Results

4.3.1. Lipids in the postmortem human hippocampus 4.3.1.1. Liquid-chromatography data A lipid profile similar to that detected in the medial temporal gyrus (MTG) region was also detected in the hippocampus. Of all the lipid classes detected, the abundance of fatty acids, diglycerides, triglycerides, glycerophospholipids, and sphingolipids is seen in Figure 4.1. This figure illustrates data from one trial using a mixture of lipids extracted together from all six control and all six AD cases, and combines data obtained in both negative and positive ion modes. Both control and AD tissue showed a similar distribution of all lipid classes. Like in the MTG region, PC was the most abundant glycerophospholipid. However, unlike in the MTG region, PC showed a trend of increasing in AD. The levels of phosphatidic acid (PA), which was again the least abundant glycerophospholipid, also showed a trend of increasing in AD, which again is in contrast with MTG results. Like in the MTG, however, phosphatidylglycerol (PG), PI, and phosphatidylserine (PS), all showed a trend of being increased in AD, with a moderate increase in PE too. The abundance of sphingolipids was lower than those of glycerophospholipids. However, like in the MTG, ceramide-1-phosphate (Cer1P) and neutral glycosphingolipid 1 (CerG1), were the most abundant sphingolipids in the hippocampus (Figure 4.1B). The abundance of Cer, Cer1P, and CerG1, all showed a trend of increasing in AD, but sphingomyelin (SM) showed a decrease. Finally, like in the MTG, triglycerides were more abundant than fatty acids and diglycerides. The abundance of diglycerides showed a trend of decreasing in AD, while triglycerides were increased (Figure 4.1C). However, since the complete hippocampus region was used to acquire this LC-MS data, it does not allow us to determine the lipid spectral profile of different sub-regions.

97 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

Figure 4.1: Abundance (as area under the curve measurement) of different lipid classes detected in the hippocampus in the liquid chromatography-mass spectrometry (LC-MS) trial. (A) Abundance of all glycerophospholipid classes detected in the hippocampus. An enlarged view of phosphatidic acid (PA) and phosphatidylglycerol (PG) levels are shown inset. (B) Abundance of all sphingolipid classes detected in the hippocampus. (C) Abundance of fatty acids, diglycerides, and triglycerides. Data from the

98 | P a g e

control cohort is shown in blue, while data from the Alzheimer’s disease (AD) cohort is shown in red. A.U., arbitrary unit; AUC, area under the curve; Cer, ceramide; Cer1P, ceramide 1 phosphate; CerG1 and CerG2, simple glycosphingolipids; DG, diglycerides; FA, fatty acids; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol; PS, phosphatidylserine; SM, sphingomyelin; TG, triglycerides

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

4.3.1.2. Matrix-assisted laser desorption/ionisation–imaging mass spectrometry (MALDI–IMS) data Three age- and sex-matched control and AD pairs were first used to generate pilot data to validate the use of the MALDI-IMS protocol optimised in the previous chapter. Figure 4.2 illustrates the distribution of selected lipids in negative ion mode in the control hippocampus. Using this ion mode I was able to detect many PE, PI, and sulfatide (SF), species. Visually, the distribution patterns made grey and white matter easily distinguishable. The two pPE species detected, m/z 726 (pPE 36:2-H-) and m/z 728 (pPE 36:1-H-), were abundant in white matter in both control and AD. PE m/z 790 (PE 40:6-H-) was more abundant in grey matter. Both PI species, m/z 857 (PI 36:4-H-) and m/z 885 (PI 38:4-H-), were abundant in grey matter, with PI 38:4 particularly highly expressed in the DG. Of the twelve SFs examined, three species, m/z 862 (SF 22:1-H-), m/z 876 (SF 23:0-H-), and m/z 888 (SF 24:1-H-), were more highly expressed in grey matter, while the rest were abundant in white matter. The distributions of m/z 726 (pPE 36:2-H-), m/z 728 (pPE 36:1-H-), m/z 857 (PI 36:4-H-), m/z 862 (SF 22:1-H-), m/z 876 (SF 23:0-H-), m/z 890 (SF 24:0-H-), m/z 892 (SF 23:0 (OH)-H-), m/z 904 (SF 24:1 (OH)H-), and m/z 906 (SF 24:0 (OH)-H-), were all comparable to those illustrated by Veloso et al. (2011b). Only the distributions of m/z 790 (PE 40:6-H-) and m/z 888 (SF 24:1-H-) did not match. Although Yuki et al. (2011) only studied the distribution of SFs in the cortex, my hippocampal results indicating the higher expression of SFs in white matter are comparable to their findings. Next I compared the distribution of the same negatively-charged lipid species in the hippocampus between control and AD cases. As illustrated in Figure 4.3, each lipid showed a similar pattern of distribution in AD to that seen in the control case. However, some lipids, such as m/z 888 (SF 24:1-H-), were expressed at slightly different levels in AD.

99 | P a g e

Chapter 4: Lipid Changes in the postmortem Human Hippocampus in AD

Figure 4.2: Spatial distribution of selected lipid detected in negative ion mode in the control human hippocampus. Representative control human hippocampus sister-section stained with haematoxylin and eosin and luxol fast blue (top-left corner) showing the Cornu Ammonis (CA), dentate gyrus (DG) and white matter (WM) regions. The rainbow colour-bar (0–100%) indicates the relative intensity, and each image shows the distribution of a specific m/z value. The proposed lipid assignments are based on previous publications (Veloso et al.,

100 | P a g e

2011b, Yuki et al., 2011). Data were acquired in reflector negative ion mode, with a spacing of 150 μm between consecutive spectra. The scale bar corresponds to 10 mm. Images were generated using BioMap software (Novartis, Switzerland). m/z, mass-to-charge ratio; PE, phosphatidylethanolamine; PI, phosphatidylinositol; pPE, ethanolamine plasmalogen; SF, sulfatide

Chapter 4: Lipid Changes in the postmortem human hippocampus in AD

Figure 4.3: Spatial distribution of selected lipids detected in negative ion mode in the Alzheimer’s disease (AD) human hippocampus. Representative AD human hippocampus sister-section stained with haematoxylin, eosin and luxol fast blue (top-left corner) showing the Cornu Ammonis (CA), dentate gyrus (DG) and white matter (WM) regions. The rainbow colour-bar (0–100%) indicates the relative intensity, and each image shows the distribution of a specific m/z value. The proposed lipid assignments are based on previous publications (Veloso et al.,

2011b, Yuki et al., 2011). Data were acquired in reflector negative ion mode, with a spacing of 150 μm between consecutive spectra. The scale bar corresponds to 10 mm. Images were generated using BioMap software (Novartis, Switzerland). m/z, mass-to-charge ratio; PE, phosphatidylethanolamine; PI, phosphatidylinositol; pPE, ethanolamine plasmalogen; SF, sulfatide

101 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD Figure 4.4 illustrates the distribution of lipids detected in positive ion mode in the control hippocampus. The same distribution pattern was observed for both control and AD tissue, for all cases that were analysed. Three lipids, m/z 729 (SM 36:2+H+), m/z 786 (PC 32:2+H+), and m/z 804 (PC 34:4+Na+), showed a homogeneous distribution. In contrast, m/z 697 (SM 32:1+Na+), m/z 731 (SM 36:1+H+), m/z 732 (PC 32:1+H+), m/z 758 (PC 34:1+H+), and m/z 813 (SM 42:2+H+), were abundant in grey matter. The following lipid peaks were highly expressed in white matter: m/z 790 (PC 36:0+H+), m/z 792 (PE 40:6+H+), and m/z 832 (PC 38:4+Na+). The distribution of m/z 732 (PC 32:1+H+), m/z 758 (PC 34:1+H+), m/z 790 (PC 36:0+H+), m/z 792 (PE 40:6+H+), and m/z 832 (PC 38:4+Na+), were comparable to those reported by Veloso et al. (2011b). The distribution of these positively charged lipid species in the AD hippocampus is illustrated in Figure 4.5. Again, the pattern of distribution of these lipids remained unchanged in AD. However, some lipid signals, such as m/z 790 (PC 36:0+H+) and m/z 792 (PE 40:6+H+) appeared to be lower in AD. Nonetheless, the variation in data observed in Chapter Three was also observed in this pilot data cohort, thus, when statistically compared, the intensity of all lipids (in both ion modes) was not significantly different between control and AD cases. Further, the sections for the pilot data were taken from various levels of the hippocampus, which could have also contributed to this variation. Hence, I decided to double the sample size, more carefully matching the anatomical levels of the hippocampi, before implementing the analysis workflow developed in Chapter Three to detect lipids that were differentially expressed.

102 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

Figure 4.4: Spatial distribution of selected lipids detected in positive ion mode in the control human hippocampus. Representative control human hippocampus sister-section stained with haematoxylin, eosin and luxol fast blue (top-left corner) showing the Cornu Ammonis (CA), dentate gyrus (DG) and white matter (WM) regions. The m/z, rainbow colour-bar (0–100%) indicates the relative intensity, and each image shows the distribution of a specific m/z value. The proposed lipid assignments

are based on a previous publication (Veloso et al., 2011b). Data were acquired in reflector positive ion mode, with a spacing of 150 μm between consecutive spectra. The scale bar corresponds to 10 mm. Images were generated using BioMap software (Novartis, Switzerland). mass-to-charge ratio; PC, phosphatidylcholine; PE, phosphatidylethanolamine; SM, sphingomyelin

103 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

Figure 4.5: Spatial distribution of selected lipids detected in positive ion mode in Alzheimer’s disease (AD). Representative control human hippocampus sister-section stained with haematoxylin, eosin and luxol fast blue (top-left corner) showing the Cornu Ammonis (CA), dentate gyrus (DG) and white matter (WM) regions. The rainbow colour-bar (0–100%) indicates the relative intensity, and each image shows the distribution of a specific m/z value. The proposed

104 | P a g e

lipid assignments are based on a previous publication (Veloso et al., 2011b). Data were acquired in reflector positive ion mode, with a spacing of 150 μm between consecutive spectra. The scale bar corresponds to 10 mm. The scale bar corresponds to 10 mm. Images were generated using BioMap software (Novartis, Switzerland). m/z, mass-to-charge ratio; PC, phosphatidylcholine; PE, phosphatidylethanolamine; SM, sphingomyelin

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

4.3.1.3. Lipid profiles of different hippocampal sub-fields Figure 4.6 illustrates the segmentation results following the use of the bisecting k-means method as a top-down clustering method. The same segmentation pattern was seen with all datasets, but only data from one age- and sex-matched pair have been shown here for clarity. As the figure illustrates, the main spectral clusters that were maximally different were those associated with white matter and grey matter, which was subsequently confirmed using histological staining. This reflects the pattern of lipid distribution that was visually evident in the pilot data too.

A

B

Figure 4.6: Segmentation data. (A) Representative Alzheimer’s disease (AD; AZ80) and control (H152) sections stained with haematoxylin and eosin (H&E) and luxol fast blue (LFB), with an indication of different hippocampal sub-fields. (B) Segmentation map generated for the same representative AD and control cases. The segmentation was performed in SCiLS Lab 2015b software (SCiLS GmbH, Germany), using bisecting k-means, which statistically

determined the similarity between spectra in a given region and grouped similar spectra into one cluster. At the first level of segmentation, the data were grouped into grey matter and white matter regions, which was validated by overlaying H&E and LFB staining. CA, Cornu Ammonis

105 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD Next, I compared overview spectra detected in white matter and grey matter, which were based on the segmentation data, and in the hippocampal sub-fields, which were delineated using a coregistered image of the same section stained with H&E and LFB. Data were acquired from m/z 500 to m/z 1800, since this is the spectral range where lipids are normally detected. The intensities were averaged across all six datasets. The results are shown in Figure 4.7. In positive ion mode, a similar lipid profile was acquired from each area, and was similar to the lipid profile seen in the MTG too. The most abundant lipid was m/z 760 (PC 34:1+H+). For some peaks, such as m/z 734 (PC 32:0+H+), there were differences in intensity between grey matter and white matter, reflecting a variation in the abundance of these lipids in each region. The lipid profiles acquired in negative ion mode, for grey matter and white matter, reflected those acquired from the MTG. In grey matter, the most abundant lipid was m/z 885 (PI 38:4-H-), while in white matter, the most abundant lipid was m/z 888 (SF 24:1-H-). Figure 4.7A and B illustrates the unique lipid profile detected in each region. For instance, m/z 744 (PE 36:1-H-) and m/z 834 (PS 40:6H-) were abundant in grey matter, while m/z 726 (pPE 36:2-H-) and m/z 788 (PS 36:1-H-) were abundant in white matter. Given that the three CA regions and the DG are all grey matter regions, it is not surprising that the most abundant lipid detected in each of these regions was m/z 885 (PI 38:4-H-). There were, however, variations in the intensity of each lipid between sub-fields. The most distinct examples are the high abundance of m/z 790 (PE 40:6-H-) in the CA1 region and the lower abundance of m/z 904 (SF 24:1 (OH)-H-) in the DG region. I then used the analysis workflow developed in Chapter Three to detect lipids that were differentially expressed in each region in AD.

106 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

Figure 4.7: Overview of lipid spectra detected in the control hippocampus using matrix-assisted laser desorption/ionisation (MALDI). This figure shows the average overview spectra, acquired between m/z 500 to 1800, from: (A) grey matter, (B) white matter, (C) the CA1 region, (D) the CA2/3 region, (E) the CA4 region, and (F) the dentate gyrus (DG). Each spectrum shows data averaged from all six control sections. The relative intensity for each m/z value was calculated

using the most abundant peak in each dataset. Spectra acquired using MALDI reflector positive mode are shown in red (above the x-axis), while spectra using MALDI reflector negative mode are shown in black (below the x-axis). The most abundant peaks for each region are labelled. CA, Cornu Ammonis; m/z, mass-to-charge ratio

107 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

4.3.2. Differentially

expressed

negatively

charged

lipids

in

Alzheimer’s disease Of all the m/z peaks detected in negative ion mode, 26 m/z peaks (including some isotopes) were differentially expressed either across the whole hippocampus region or in at least one of its subfields (Figure 4.8A).

4.3.2.1. Whole hippocampus There were ten lipids that were differentially expressed in AD across the whole hippocampus, of which the majority showed a relative decrease (Figure 4.8B). These were m/z 767.6, m/z 778.6 (pPE 40:4-H-), m/z 779.6 (PG 37:6-H-), m/z 791.6, m/z 794.6 (PE 40:4-H-), and m/z 795.7 (PG 38:5-H-/PA 44:10-H-). In contrast, I detected a relative increase in m/z 739.2, m/z 786.6 (PS 36:2-H-), m/z 810.6 (PS 38:4-H-), and m/z 843.7.

4.3.2.2. White matter While three lipid species m/z 797.7 (PG 38:4-H-), m/z 917.7, and m/z 918.7 (SF 26:0 (0H)-H-), were increased in white matter in AD, m/z 778.6 (pPE 40:4-H-) was the only lipid species that was decreased in this region (Figure 4.8C).

4.3.2.3 Grey matter In contrast to the four lipids that were differentially expressed in white matter, fourteen lipids were differentially expressed in grey matter, of which half were increased and half were decreased (Figure 4.8D). The lipid species that were increased in grey matter were m/z 739.2, m/z 786.6 (PS 36:2-H-), m/z 789.6 (SM 36:1-H-), m/z 797.7 (PG 38:4-H-), m/z 810.6 (PS 38:4-H-), m/z 813.7, and m/z 843.7. In contrast, m/z 767.7, m/z 774.6 (pPE 40:6-H-), m/z 778.6 (pPE 40:4-H-), m/z 779.6 (PG 37:6-H-), m/z 791.6, m/z 794.6 (PE 40:4-H-), and m/z 795.7 (PA 44:10-H-) were decreased.

4.3.2.4. Cornu Ammonis regions Of the CA1, CA2/3, and CA4 regions, the CA1 region had the highest number of lipid differences in AD. There were eight lipids that were decreased in the CA1 region (Figure 4.8E), of which six reflected the change seen in grey matter. The lipids that were decreased were m/z 600.5 (Cer 39:4-H-), m/z 767.6, m/z 778.6 (pPE 40:4-H-), m/z 779.6 (PG 37:6-H-), m/z 791.6, m/z 794.6 (PE 40:4-H-), m/z 795.7 (PA 44:10-H-), and m/z 796.6. There were six lipids that were increased in the CA1 region, again reflecting the change seen in grey matter. These were m/z 739.2, m/z 786.6 (PS 36:2-H),

m/z

789.6

108 | P a g e

(SM

36:1-H-),

m/z

810.6

(PS

38:4-H-),

m/z

813.7,

and

m/z

843.7.

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

Figure 4.8: Mean intensity difference of natively charged lipids in Alzheimer’s disease (AD; as a % change from control). (A) The regions of interest analysed for this study, in a control (H152) and an AD case (AZ80). (B-H) Graphs showing the mean intensity difference of selected m/z values in AD, relative to intensity in control human hippocampus, in: (B) the whole hippocampus, (C) white matter, (D) grey

matter, (E) CA1, (F) CA2/3, (G) CA4, and (H) dentate gyrus regions. m/z values were selected using pairwise receiver operating characteristic comparisons for each anatomical region. Black arrows indicate the lipid species that were consistently changed in all AD cases. m/z, mass-to-charge ratio

109 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD There was only one lipid that was differentially expressed in the CA2/3 region in AD, and this was m/z 843.7, which was increased (Figure 4.8F). The three lipids that were differentially expressed in the CA4 region, m/z 600.5 (Cer 29:4-H-), m/z 789.6 (SM 36:1-H-), and m/z 843.7, were all increased (Figure 4.8G).

4.3.2.5. Dentate gyrus There were sixteen lipids that were differentially expressed in the DG in AD, of which only eight were specifically changed in this region (Figure 4.8H). There were nine lipids that were increased. These were m/z 600.5 (Cer 39:4-H-), m/z 646.5 (Cer 42:2-H-), m/z 647.5 (PA 32:0-H-), m/z 719.6, m/z 786.6 (PS 36:2-H-), m/z 813.7, m/z 843.7, m/z 885.6 (PI 38:4-H-), and m/z 886.7. The seven lipids that were decreased were m/z 791.6, m/z 794.6 (PE 40:4-H-), m/z 795.7 (PA 44:10-H-), m/z 905.7, m/z 906.7 (SF 24:0 (0H)-H-), m/z 918.7 (SF 26:0 (0H)-H-), and m/z 1574.0 (GM1 38:1-H-).

4.3.2.6. Summary The differential expression of these lipids in AD, in the sub-fields of the hippocampus, has been summarised in Table 4.1. I was able to putatively identify 19 of the 26 m/z values that were detected by the analysis, of which the majority (six) were PE species. While there are some discrepancies, which are outlined below, generally PE and PG showed a decrease, while PS, PI, SF, and SM, were increased. Cer was increased in the CA4 and DG, but decreased in the CA1, while the only ganglioside that could be identified was decreased in the DG. The distribution images in Figure 4.9 clearly show the region-specific differential expression of these lipids and their relative increase or decrease in AD. All the Cer, GM1, PA, PE, PI, and PS species that were identified were more abundant in grey matter. However, many of these lipids were not homogeneously distributed in this region. Many lipids, such as m/z 778.6 (pPE 40:4-H-), m/z 794.6 (PE 40:4-H-), and m/z 885.6 (PI 38:4-H-), were highly expressed in the CA1 region. Lipids that were abundantly expressed in the CA1 region were also usually expressed in the DG and CA4 region. However, others such as m/z 646.5 (Cer 42:2-H-), m/z 719.6, and m/z 786.6 (PS 36:2-H-), were highly expressed in the DG region alone, at least in AD. Six out of the 26 lipids detected in this analysis showed an inverse pattern of distribution, in that they were highly expressed in white matter. These were m/z 797.7 (PG 38:4-H-), m/z 813.7, m/z 905.7, m/z 906.7 (SF 24:0 (0H)-H-), m/z 917.7, and m/z 918.7 (SF 26:0 (0H)-H-). The presence of m/z 797.7 (PG 38:4-H-) is a departure from the distribution trend seen by the other two PG species that were identified, i.e. m/z 779.6 (PG 27:6-H-) and m/z 795.7 (PA 44:10-H-), which were more abundant in grey matter.

110 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

111 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

Figure 4.9: Spatial distribution of selected lipids detected in reflector negative ion mode in the human hippocampus. Representative hippocampus sections (AZ80 and H152) stained with haematoxylin and eosin (H&E) and luxol fast blue (LFB; top-left corner) showing white matter (green), and the grey matter regions, Cornu Ammonis 1 (CA1; yellow), CA2/3 (purple), CA4 (blue) and the dentate gyrus (orange). Edge-preserving image denoising and automatic hotspot removal (see

112 | P a g e

rainbow intensity colour-bar) was applied using SCiLS Lab 2015b software. The spatial distance between consecutive spectra is 100 μm. Each image shows the distribution of a specific m/z value. Cer, ceramide; GM, ganglioside; m/z, mass-to-charge ratio; PA, phosphatidic acid; PE, phosphatidylethanolamine; PG, phosphatidylglycerol; PI, phosphatidylinositol; PS, phosphatidylserine; SF, sulfatide; SM, sphingomyelin

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

4.3.3. Differentially

expressed

positively

charged

lipids

in

Alzheimer’s disease Analysis of the positive ion mode datasets yielded 17 lipids that were differentially expressed, in at least one of the sub-fields of the hippocampus (Figure 4.10A) in AD. The same lipid species, when differentially expressed in more than one sub-field, generally showed the same relative change (i.e. an increase or decrease), albeit to varying degrees. I was only able to assign putative lipid identifications to four of the 17 lipids that were detected.

4.3.3.1. Whole hippocampus The only lipid that was decreased across the whole hippocampus was m/z 496.5 (Figure 4.10B). In contrast, there was an increase in 11 lipids. These were m/z 718.6 (PE 34:1+H+), m/z 720.6 (PE 34:0+H+), m/z 721.6, m/z 730.5, m/z 753.7 (SM 36:1+Na+), m/z 1463.1, m/z 1478.1, m/z 1479.2, m/z 1481.1, m/z 1483.1.

4.3.3.2. White matter While m/z 522.4 was the only lipid that was decreased in white matter in AD, there were six lipids that were increased (Figure 4.10C). These were m/z 718.6 (PE 34:1+H+), m/z 720.6 (PE 34:0+H+), m/z 730.5, m/z 1463.1, m/z 1479.2, and m/z 1481.1.

4.3.3.3. Grey matter There were 14 lipids that were differentially expressed in AD in grey matter (Figure 4.10D). The lipid detected at m/z 496.5 was the only one that showed a decrease. In contrast, m/z 718.6 (PE 34:1+H+), m/z 720.6 (PE 34:0+H+), m/z 721.6, m/z 730.5, m/z 1463.1, m/z 1464.2, m/z 1467.2, m/z 1478.1, m/z 1479.2, m/z 1481.1, m/z 1483.1, m/z 1495.1, and m/z 1507.3 were increased in grey matter in AD.

4.3.3.4. Cornu Ammonis regions Of the 12 lipids that were differentially expressed in the CA1 region in AD, m/z 496.5 was the only one that showed a decrease (Figure 4.10E). The lipids that were increased were m/z 718.6 (PE 34:1+H+), m/z 720.6 (PE 34:0+H+), m/z 721.6, m/z 730.5, m/z 1463.1, m/z 1464.2, m/z 1467.2, m/z 1478.1, m/z 1479.2, m/z 1481.1, and m/z 1483.1. All eight lipids that were differentially expressed in the CA2/3 region were increased in AD (Figure 4.10F). These were m/z 718.6 (PE 34:1+H+), m/z 720.6 (PE 34:0+H+), m/z 721.6, m/z 730.5, m/z 1463.1, m/z 1464.2, m/z 1467.2, m/z 1478.1, m/z 1479.2, m/z 1481.1, and m/z 1483.1.

113 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

Figure 4.10: Mean intensity difference of positively charged lipids in Alzheimer’s disease (AD; as a % change from control). (A) The regions of interest in the postmortem human hippocampus analysed in this study, in a control (H152) and AD case (AZ80). (B-H) Graphs showing the mean intensity difference of selected m/z values in AD, relative to intensity in the control human hippocampus, in: (B) the whole

114 | P a g e

hippocampus, (C) white matter, (D) grey matter, (E) CA1, (F) CA2/3, (G) CA4, and (H) dentate gyrus. m/z values were selected using pair-wise receiver operating characteristic comparisons for each anatomical region. Black arrows indicate the lipid species that were consistently changed across the AD cohort. CA, Cornu Ammonis; m/z; mass-to-charge ratio

Chapter 4: Lipid changes in the postmortem human hippocampus in AD The CA4 region showed a similar pattern, with an increase in eight lipids in AD (Figure 4.10G). The lipids that were increased in this region were m/z 718.6 (PE 34:1+H+), m/z 1463.1, m/z 1464.2, m/z 1478.1, m/z 1479.2, m/z 1481.1, m/z 1483.1, and m/z 1507.3.

4.3.3.5. Dentate gyrus There were nine lipids that were increased in the DG in AD (Figure 4.10H). These were m/z 706.5 (PC 30:0+H+), m/z 718.6 (PE 34:1+H+), m/z 720.6 (PE 34:0+H+), m/z 721.6, m/z 730.5, m/z 1463.1, m/z 1464.2, m/z 1467.2, m/z 1478.1, m/z 1479.2, m/z 1481.1, m/z 1483.1, m/z 1495.1, and m/z 1507.3.

4.3.3.6. Summary Table 4.2 summarises the differential expression of the lipids, detected in positive ion mode, in AD, across the different hippocampal sub-fields. Of the four lipids that could be identified, two were PE species, one was a PC species, and one was the sodiated form of SM 36:1, which was also detected in negative mode at m/z 789.6, according to the lipid assignment reported by Samhan-Arias et al. (2012). The majority of positively-charged lipid species that were differentially expressed in AD were between m/z 1460 and m/z 1510. Although they could not be identified through MS/MS and have not been previously reported in the human brain before, given the proximity of their masses to gangliosides that have been reported in LIPID MAPS, some of these lipids might be ganglioside species. Figure 4.11 illustrates the distribution of each lipid, with an indication of its heterogeneous expression within the same hippocampus, and its differential expression in AD. Generally, lipids were either expressed abundantly in grey matter, e.g. m/z 496.5, m/z 1467.2, and m/z 1495.1, or in white matter, e.g. m/z 1463.1, m/z 1464.2, and m/z 1507.3. Like those lipids detected in negative ion mode, the lipids that were abundant in grey matter showed a heterogeneous distribution in this region. For instance, m/z 496.5 and m/z 522.4, which were more abundant in AD, were particularly highly expressed in the CA4 and DG region. Unlike in negative ion mode though, there were more positively charged lipids, such as m/z 721.6, m/z 1478.1, m/z 1481.1, and m/z 1483.1, which showed a homogeneous expression pattern through the hippocampus, with some hotspots in AD.

115 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

116 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

Figure 4.11: Spatial distribution of selected lipids detected in positive ion mode in the human hippocampus. Representative hippocampus sections (AZ80 and H152) stained with haematoxylin and eosin (H&E; top-left corner) showing white matter (green), and the grey matter regions, Cornu Ammonis 1 (CA1; yellow), CA2/3 (purple), CA4 (blue) and the dentate gyrus (DG; orange). Edge-preserving image denoising

and automatic hotspot removal (see rainbow intensity color-bar) has been applied. The spatial distance between adjacent spectra is 100 μm. Each image shows the distribution of a specific m/z value and was generated using SCiLS lab 2015b software (SCiLS Gmbh, Bremen, Germany). m/z, mass-to-charge ratio; PC, phosphatidylcholine; PE, phosphatidylethanolamine; SM, sphingomyelin

117 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

Section Four

Discussion

4.4.1. General discussion First, the use of LC-MS provided an overall understanding of the various lipid groups that make up the human hippocampus. The ratios between different lipid groups were the same as the MTG region. However, there were discrepancies in the relative change that some lipid groups showed in AD. Specifically these were the increase in PC and Cer1P and the decrease in SM and triglycerides. The increase in PC does not reflect previous findings. Prasad et al. (1998) did not find a change in PC in the subiculum or hippocampal gyrus, while Wells et al. (1995) reported a decrease in PC in the hippocampus, at least in the synapatosome plasma membrane. While the elevation of Cer in AD is well-documented (Han et al., 2002, Cutler et al., 2004, Bandaru et al., 2009, He et al., 2010), at least in the cortex, less is known about Cer1P levels, especially in the human hippocampus. However, Cer is phosphorylated by ceramide kinase to form Cer1P, which then inhibits apoptosis and induces cell survival (Gómez-Muñoz et al., 2004, Gangoiti et al., 2010). Thus, the increase in Cer1P might reflect a defence mechanism in response to the cell loss, and thus, the severe atrophy, that occurs in the hippocampus in AD. The decrease in SM reflects results from a study conducted by He et al. (2010) who used grey matter from the frontotemporal cortex. Finally, given the association between triglyceride elevation and memory dysfunction in humans, at least in those with non-insulindependent diabetes mellitus (Helkala et al., 1995), the decrease in triglycerides seen in AD is unexpected. The discrepancy in the abundance of PCs and triglycerides could be attributed to a variation in the composition of the tissue used, for instance in the grey to white matter ratio, from which the lipids were extracted, for this study. These findings highlighted the importance of analysing lipid differences in precisely delineated anatomical regions. Using MALDI-IMS, I first attempted to replicate previously published findings that used the same technique to map the distribution of lipids in the human brain (Veloso et al., 2011a, Veloso et al., 2011b, Yuki et al., 2011). Using a higher raster step-size of 150 µm for the pilot data, I was able to generate clearer distribution images in contrast to Veloso et al. (2011b) and Yuki et al. (2011), who acquired their images at 200 µm and 500 µm, respectively. Further, I used DAN as a matrix, instead of 2-mercaptobenzothiazole or 9-aminoacridine, which were used by Veloso et al (2011) and Yuki et al. (2011), respectively. Nonetheless, I was able to successfully replicate their findings. The results showed that the pattern of distribution of lipids in the hippocampus was conserved in AD. Although

118 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD there were some lipids that visually seemed to be expressed at a lower amount in AD, these relative changes were not statistically significant. Since there were large variations in the control and AD cohorts used to acquire pilot data, for the rest of the analysis, I doubled the sample size and better matched the anatomical level of each section in the age- and sex-matched pairs. I then applied the workflow developed in Chapter Three to analyse the whole dataset, and selected pair-wise lipid changes that were consistently present in all six datasets, in negative and positive ion modes, separately. I was able to successfully implement the analysis workflow to detect lipids that were differentially expressed in the anatomical sub-fields of the human hippocampus in AD. The detection of lipids that were both increased and decreased confirmed that the analysis technique was not biased by atrophy (and thus, the smaller area) of the hippocampus in AD. Due to the mass resolution limitations of the MALDI-TOF, peaks could not always be assigned a single lipid identity (e.g. m/z 795.7 (PA 44:10-H-)), and may contain more than one lipid species. To confirm the identities of each lipid detected in the study, high mass resolution imaging mass spectrometry equipment would be required. Nonetheless, this discussion focusses on the lipid assignments that I was able to make putatively using MS/MS and previously published results on the mammalian lipidome. With the exception of PC 30:0+H+, PE 34:1+H+, PE 30:4+H+, and PG 38:4-H-, all lipid species belonging to the same class of lipid showed the same change in AD. For example, in AD all the PS species detected in negative ion mode were increased. Additionally, if the same lipid species were changed in more than one hippocampal sub-field, it was either increased or decreased across all sub-fields. Cer 39:4-H- and SF 26:0 (OH)-H- were the only lipid species that show a departure from this trend, as the former was decreased in the CA1 region, but increased in the CA4 and DG, while the latter was decreased in the DG but increased in white matter. It is however important to note that the extent to which the same lipid species was changed, which is indicated by the percentage change in Tables 4.1 and 4.2, differed in each sub-field. This could largely be attributed to the variation in the abundance of these lipids in these sub-fields. Generally, there was an increase in Cer, PA, PC, PS, PI, SM, SF, and GM, with a decrease in PG and PE (except those detected in positive ion mode). With the exception of a few lipids, most of the results show trends that are consistent with previous work that have largely used traditional mass spectrometry and chromatography techniques to investigate lipid changes in AD (Ellison et al., 1987, Söderberg et al., 1991, Jolles et al., 1992, Nitsch et al., 1992, Jope et al., 1994, Svennerholm and Gottfries, 1994, Wells et al., 1995, Guan et al., 1999, Han et al., 2001, Pettegrew et al., 2001, Han et al., 2002, Mulder et al., 2003, Cutler et al., 2004, Landman et al., 2006, Lange et al., 2008, He et al., 2010, Martın et al., 2010, Valdes-Gonzalez et al., 2011, Chan et al., 2012, 119 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD Pernber et al., 2012, Cheng et al., 2013). However, this thesis adds to the body of literature in extending our understanding of how lipids change specifically in each hippocampal sub-field in AD. Of the 43 lipids that were changed in at least one hippocampal sub-field in AD, only approximately half were detected when the hippocampus was analysed as a whole. This highlights the merit of analysing various sub-fields of a region of interest separately, using MALDI-IMS. Additionally, with the exception of some lipids detected in positive ion mode that showed a homogeneous distribution, most lipids were abundantly expressed either in grey matter, or in white matter (Figure 4.9 and Figure 4.11). There is also considerable heterogeneity in the expression of these lipids in the grey matter alone. This variation in lipid expression can be linked to the unique anatomical composition of each sub-field. Since each hippocampal sub-field has a distinct function, it is important to consider how lipid expression changes in each field, in order to link this change to AD pathogenesis. Myelin is rich in SM and GM1 (O'Brien and Sampson, 1965a, Svennerholm and Vanier, 1973, Pernber et al., 2012, Schnaar et al., 2014), which they are particularly concentrated in the outer leaflet of the membrane and are key constituents of lipids rafts (Farooqui et al., 2000, Ramstedt and Slotte, 2002). An increase in SM has been previously reported in AD (Wells et al., 1995, Pettegrew et al., 2001, Bandaru et al., 2009). Further, a significant decrease in ganglio-series gangliosides (GT1b, GD1b, GD1a, GM1) in the frontal and temporal cortex, and an elevation in simple gangliosides like GM2, GM3, and GM4, have been reported (Kracun et al., 1989, Kalanj et al., 1990). The increase in SM 36:1 and the decrease in GM1 in this study reflect these trends. Given the evidence that these changes occur very early during AD, accompanied by biophysical alterations and differential recruitment of amyloidogenic proteins to lipid rafts (Fabelo et al., 2014, Díaz et al., 2015), it is possible that these lipid changes may drive AD pathogenesis. Further, while I was unable to accurately identify lipids detected in the m/z 1400-1500 region in positive ion mode, given their proximity to expected m/z values, some of lipids could be GMs. There is evidence that the aggregation of GM1 into clusters is accelerated in a cholesterol-rich environment, like that present in the human brain during normal aging or with apolipoprotein E4 (apoE4) expression (Yanagisawa, 2007, Ariga et al., 2008, Pernber et al., 2012). GM1 clusters bind Aβ1-42, altering its conformation from random coils to more ordered structures, with increased β-sheet content, which in turn leads to Aβ aggregation (Yanagisawa et al., 1995, Kakio et al., 2002, Yanagisawa, 2007). However, since I detected the depletion of only one GM1 species, i.e. GM1 d20:0/18:0, in the DG alone, further work specifically investigating other accurately identified GMs in the human hippocampus is needed to confirm its pathogenic role in AD. SFs are also specifically expressed in myelin (Han et al., 2002, Yuki et al., 2011). The analysis of SFs in AD has produced conflicting results. Some groups report an increase in the average SF level in the AD brain (Majocha et al., 1989) and cerebrospinal fluid in vascular dementia (Fredman et al., 1992). 120 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD Others have reported a significant depletion of up to 58% of SFs, even at the earliest clinical stage of AD that they investigated (Clinical Dementia Rating 0.5;Han et al., 2002), with no change in the cortical compositional distribution of hydroxylated and non-hydroxylated SFs between AD and control (Yuki et al., 2011). However, there is no indication of how SF expression changes in the hippocampus specifically. My results indicated a 20% increase in SF 26:0 (OH)-H- in white matter in AD, with a depletion of SFs in other sub-fields, namely the DG. There is evidence that alterations in apoE-mediated SF trafficking lead to this change in SF expression in AD (Han, 2010). The change in SF expression will affect its normal function, which includes myelin formation and maintenance, oligodendrocyte differentiation, myelin-associated axon outgrowth, and glial-axon signalling (Han et al., 2002, Takahashi and Suzuki, 2012). The most abundant types of glycerophospholipids found in neural membranes are PC, PE, PS, and PI. PG and PA, the main precursors of all neural membrane glycerophospholipids, are also found here, albeit with lower abundance (Farooqui et al., 2000). When grey matter was analysed separately, many of the species detected that were changed in AD belonged to these classes of lipids. The lipids that were changed in grey matter were often specifically changed in just the CA1 region too. Additionally, of all CA fields, the CA1 region yielded the majority of lipid changes. The CA1 region is considered the main output of the hippocampus, by the way of the alveus and then the fimbria (Duvernoy et al., 2013). Thus, it is a vital link in the intrahippocampal circuitry, which is crucial for memory formation (Duvernoy et al., 2013). In AD, with the exception of PE 34:1+H+, which was increased in the CA1 region, there was a decrease in PEs, including PE 40:6-H-, whose identification has been confirmed in the human brain before (Han et al., 2001). This PE decrease reflects previous findings (Ellison et al., 1987, Wells et al., 1995, Han et al., 2001, Pettegrew et al., 2001, Martın et al., 2010, Chan et al., 2012, Kosicek and Hecimovic, 2013). PEs are often linked to polyunsaturated fatty acids (PUFA), especially docosahexaenoic acid (DHA), which are highly oxidisable (Butterfield and Lauderback, 2002, Hartmann et al., 2007, González-Domínguez et al., 2014). Thus, in AD, where there is increased oxidative stress, PUFAs serve as substrates for lipid peroxidation (Butterfield and Lauderback, 2002). In addition to leading to membrane destabilisation, this process also generates aldehydes such as 4-hydroxynonenal, which in turn can oxidise other proteins and inhibit glycolysis, driving AD pathogenesis (González-Domínguez et al., 2014). In contrast to the decrease in PEs, other lipids were increased in the CA1 region in AD. PS 36:2-Hshowed the greatest change with a 75% increase. The PS content of grey matter increases from birth to the age of 80 (Vance and Tasseva, 2013, Glade and Smith, 2015), and under normal physiological conditions it is located in the inner-cytosolic leaflet of the membrane. However, during the early stages of apoptosis, PS is translocated to the outer layer of the membrane, where it can serve as an

121 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD active signal for phagocytosis (Vance and Tasseva, 2013, Glade and Smith, 2015). The elevated abundance of PS 36:2-H- may indicate an increase in apoptosis in the CA1 region in AD, however, further work is required to confirm this. Finally, in contrast to the other sub-fields, there were a number of lipid species that were changed in the DG alone. Of these lipids, the identification of PI 38:4-H-, SF 24:0 (0H)-H-, and Cer N24:1-H-, have been reported previously (Han et al., 2002, Dill et al., 2010, Veloso et al., 2011b, Yuki et al., 2011). The 20% increase in PI 38:4-H- in the DG does not reflect previously reported decreases in PI in the AD human brain (Stokes and Hawthorne, 1987). Nonetheless, it may be a change that is specific to the DG, although remains to be confirmed. I speculate that it may be a reflection of the reduction in PI kinases (Jolles et al., 1992) and the impaired hydrolysis of phosphoinositides, i.e. the phosphorylated form of PIs, in AD (Jope et al., 1994). The role phosphoinositides play in regulation and membrane dynamics, which will be affected in AD, have been previously extensively reviewed by Di Paolo and De Camilli (2006) and Frere et al (2012). While I did not analyse the compositional difference between hydroxylated and non-hydroxylated SF species like Yuki et al (2011), this analysis indicated several hydroxylated SFs that were decreased in AD in the DG alone, consistent with previous findings (Han et al., 2002, He et al., 2010). Since hydroxylated SFs are highly expressed in oligodendrocytes in grey matter, as mentioned previously, this decrease may affect oligodendrocyte differentiation, myelin formation and maintenance, myelinassociated axon outgrowth, and glial-axon signalling (Han et al., 2002, Takahashi and Suzuki, 2012). The decrease in SFs also consequently elevates Cer levels in AD (Han et al., 2002, Puglielli et al., 2003a, He et al., 2010, Hejazi et al., 2011), which is reflected by my results. Cer is thought to drive AD pathogenesis by playing a role in the execution of apoptosis (Hannun and Obeid, 2008, Hejazi et al., 2011), and stabilising the β-site amyloid precursor protein cleaving enzyme 1 (BACE1), consequently promoting Aβ biogenesis (Puglielli et al., 2003a). However, since work focusing on lipid changes in the postmortem human DG in AD is scarce, further work needs to be done to elucidate whether these lipids play a specific role in driving DG function. Additionally, the effect their changed expression in AD has on the DG also remains to be evaluated.

122 | P a g e

Chapter 4: Lipid changes in the postmortem human hippocampus in AD

4.4.2. Summary of findings The use of MALDI-IMS, in conjunction with the analysis workflow developed in the previous chapter, provided a novel method to investigate the differential lipid expression in the sub-fields of the hippocampus. The advantage of this approach was the ability to precisely delineate each sub-field post-hoc, thus offering the versatility of analysing global differences in grey matter and white matter regions, in addition to each specific anatomical sub-field. PE was the major lipid class that was differentially expressed in AD, reflecting my findings in the cortex. Of the different hippocampal sub-regions, the majority of lipids were differentially expressed in the CA1 region, which is the primary output of the hippocampus. However, interestingly, there were many lipids that were differentially expressed in the DG region alone. The DG, and the granule cell layer (GCL) in particular, plays an integral role in the production of episodic memory, since it is the first point that begins processing glutamatergic input from the entorhinal cortex (Insausti and Amaral, 2004, Amaral et al., 2007, Duvernoy et al., 2013). Further, the subgranular zone of the GCL is a neurogenic niche in the adult human brain (Eriksson et al., 1998). Thus, analysing the lipid composition of the DG, and its differences in AD, could help further elucidate the pathogenic process that occurs in this region, and is the focus of Chapter Six. Now that I have identified certain lipid classes that show a relative change in the human AD hippocampus, the targeted analysis of these classes will allow more accurate quantification of these changes. Better understanding how different lipid classes change in the hippocampus in AD will help expand the knowledge of their potential role in AD pathogenesis, which will allow us to better model the disease and develop potential therapeutics.

123 | P a g e

Chapter Five

Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Section One: Introduction ..........................................................................................................126 Seection Two: Methods ..............................................................................................................127 5.2.1 Generation of lipid-spiked tissue homogenate standards......................................................... 127 5.2.1.1 5.2.1.2 5.2.1.3 5.2.1.4 5.2.1.5 5.2.1.6 5.2.1.7

Lipid standards................................................................................................................................. 127 Tissue sample used for brain homogenates..................................................................................... 127 Incorporation of lipid standard in brain homogenates .................................................................... 127 Data acquisition ............................................................................................................................... 129 Data analysis ................................................................................................................................... 129 Standard curves ............................................................................................................................... 129 Histological staining ........................................................................................................................ 129

5.2.2 Quantifying phosphatidylethanolamine in the middle temporal gyrus .................................... 129 5.2.2.1 5.2.2.2 5.2.2.3 5.2.2.4 5.2.2.5

Sample preparation and data acquisition ....................................................................................... 129 Standard curves ............................................................................................................................... 130 Determining m/z values to analyse ................................................................................................. 130 Calculation of the amount of lipid in the MTG ................................................................................. 130 Statistical Analysis ........................................................................................................................... 130

Section Three: Results................................................................................................................131 5.3.1. Quality assessment of lipid-spiked tissue standards ................................................................. 131 5.3.1.1. Analysis of lipid standards using matrix-assisted laser desorption/ionisation (MALDI) .... 131 5.3.1.2. Morphology of tissue homogenates................................................................................... 131 5.3.1.3. Analysis of lipid standards using imaging mass spectrometry (IMS) ................................. 131 5.3.1.4. Standard Curves ................................................................................................................. 135 5.3.2. Lipid Quantification in the Middle Temporal Gyrus .................................................................. 137 5.3.2.1 Phosphatidylinositol quantification.................................................................................... 137 5.3.2.2 Phosphatidylethanolamine quantification ......................................................................... 140 5.3.2.3 Phosphatidylethanolamine differences in Alzheimer’s disease ................................................. 141 Section Four: Discussion ............................................................................................................146 5.4.1. General Discussion ..................................................................................................................... 146 5.4.1.1 Use of lipid-spiked tissue homogenates for quantification .............................................................. 146 5.4.1.2 Phosphatidylethanolamine changes in the grey matter in the middle temporal gyrus in Alzheimer’s disease ............................................................................................................................................. 147 5.4.1.3 Considerations for future use ........................................................................................................... 148

5.4.2. Summary of Findings.................................................................................................................. 149

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Section One

Introduction

Hankin and Murphy (2010) demonstrated the direct correlation between the intensity of m/z peaks measured using matrix-assisted laser desorption/ionisation (MALDI)-imaging mass spectrometry (IMS) and the quantity of lipid molecules in the tissue sample, at least for phosphatidylcholines (PC) in the rat brain. Thus, the use of intensity differences between control and Alzheimer’s disease (AD), to detect the differential lipid expression in previous chapters can be justified. However, since the lipids intensities in the middle temporal gyrus (MTG) and hippocampus in AD were presented in relation to control cases, this approach is at best, semi-quantitative. Thus, methods such as liquidchromatography-mass spectrometry (LC-MS) have been used for the absolute quantification of lipids in the postmortem human brain (Han et al., 2001, Han et al., 2002). Nonetheless, the use of MALDI-IMS for quantification has been demonstrated previously, mainly in pharmaceutical studies. It has been used to quantify drugs such as octreotide, olanzapine, propranolol, and reserpine in rodents (Hamm et al., 2012, Shahidi-Latham et al., 2012, Takai et al., 2013), ipratropium in the human lung (Marko-Varga et al., 2011), and cocaine in a postmortem human nucleus accumbens (Pirman et al., 2012). One approach to quantify biomolecules of interest using MALDI-IMS is to spot calibration standards on or beneath the tissue (Pirman et al., 2012). Another is to spot a series of matrix/calibration dilutions to create a standard curve (Shahidi-Latham et al., 2012). However, this does not account for tissue effects on the ionisation efficiency of the biomolecule of interest. Thus, Groseclose and Castellino (2013) introduced a tissue mimetic approach where the calibration standards were incorporated in tissue homogenates, which has recently been used for the absolute quantification of lipids, namely PC, in the rat brain (Jadoul et al., 2015). The absolute quantification of lipids in spatially delineated regions in the postmortem human AD brain will extend our knowledge about local changes. This will enable us to better model the disease and help target specific pathways to rectify lipid imbalances. Thus, the main aim of this chapter is to use the approach outlined by Jadoul et al. (2015) to quantify PC and phosphatidylethanolamine (PE) lipids, which were abundantly expressed in the MTG and differentially expressed in AD (Chapter Three). Further, this chapter also aims to quantify phosphatidylinositol (PI) lipid species, in an attempt to trial its use to quantify PI in the dentate gyrus (the focus of Chapter Six) where it is highly expressed. This is the first known attempt at quantifying lipids in the postmortem human brain, using MALDI-IMS. In this chapter, the assessment of the quality of the lipid-spiked tissue homogenates is first outlined. Next, the concentrations of PE species are quantified in the MTG. Finally, differences between the concentrations of PE species in the MTG are compared between control and AD cases. 126 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Section Two

Methods

5.2.1 Generation of Lipid-Spiked Tissue Homogenate Standards 5.2.1.1 Lipid Standards PC e16:0/18:1 (1-O-Hexadecyl-2-Oleoyl-sn-Glycero-3-Phosphocholine), PE 16:0/18:1 (1-palmitoyl-2oleoyl-sn-glycerol-3-phosphoethanolamine), and PI 16:0/18:1 (1-palmitoyl-2-oleoyl-sn-glycero-3phosphoinositol (ammonium salt), were purchased from Avanti Polar Lipids, Inc. (Alabaster, USA). Lipid standards were used without further purification. To validate peak detection, each lipid standard, which was made up at a concentration of 25 mg/mL, was spotted on a gold MALDI plate (AB Sciex Ltd, Ontario, Canada) at a 1:1 ratio with 10mg/mL of 1,5-diaminonapthalene (DAN). Given my previous analysis, I expected to detect the PE and PI lipid peaks in negative ion mode and the PC peak in positive ion mode. Stock solutions were prepared using a methanol/chloroform mixture (50/50 v/v), as outlined by Jadoul et al. (2015). For the PC standard, a stock solution was made at a concentration of 25 mg/mL. For the PE standard, stock solutions were made at concentrations of 250 mg/mL and 25 mg/mL. For the PI standard, stock solutions were made at concentrations of 25 mg/mL and 5 mg/mL.

5.2.1.2 Tissue sample used for brain homogenates Human brain tissue from the visual cortex of an 80-year-old female (30 hour postmortem delay) with a clinical diagnosis of AD, but with pathology more consistent with a diagnosis of frontotemporal dementia, was used to make the lipid-spiked tissue standards. On average, 500 (± 0.3) mg of tissue was used per lipid-spiked standard. To ensure optimum homogenisation, however, the tissue required for each concentration standard was prepared in two Eppendorf tubes, with half the amount of tissue in each tube.

5.2.1.3 Incorporation of lipid standard in brain homogenates A protocol described by Hare et al. (2013), which had been previously successfully used in the lab to generate matrix-matched standards for laser ablation inductively-coupled plasma-MS, was used to generate the lipid-spiked tissue standards. First, approximately 6 g of visual cortex tissue was finely diced with a polytetrafluoroethylene-coated blade (GEM, VA, USA), on a parafilm-coated surface to avoid possible contamination. Then, 250 (± 0.3) mg of homogenous diced tissue was transferred into a 2.0 mL tube (Axygen, CA, USA). The tissue in each tube was then spiked with different concentrations of the lipid stock solutions as outlined in Table 5.1, to a total added volume of 15 µL per tube, and centrifuged. Then, a few 2-mm 127 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS Table 5.1: Volumes of lipid standard stock solutions added to each 250 mg tissue homogenate to create lipid-spiked tissue standards. Stock solutions of PE 16:0/18:1 were made using methanol (MeOH)/chloroform (CHCl3) mixture (50/50 v/v), at concentrations of 250 mg/mL and 25 mg/mL. The final three tissue homogenates, spiked to a final PE

Final lipid Concentration (µg/g)

MeOH/CHCl3 solution (µL)

concentrations of 2500 µg/g, 5000 µg/g and 10 000 µg/g, were made following analysis of the first trial. PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol

250 mg/mL

25 mg/mL

25 mg/mL

25 mg/mL

5 mg/mL PI

PE solution

PE solution

PC solution

PI solution

solution

(µL)

(µL)

(µL)

(µL)

(µL)

0 (Blank)

15

-

-

-

-

-

50

11.5

-

0.5

0.5

-

2.5

100

8

-

1

1

-

5

250

7.5

-

2.5

2.5

-

2.5

500

-

5

-

5

-

5

1000

4

1

-

-

-

10

2500

12.5

2.5

-

-

-

-

5000

10

5

-

-

-

-

10 000

5

10

-

-

-

-

zirconium oxide homogenisation beads (Next Advance, NY, USA) were added to each tube. Tissue was then homogenised for 2 x 5 minutes at speed 8, using a Bullet Blender homogeniser (Next Advance, NY, USA), with brief centrifuging between cycles to ensure complete homogenisation. Each tube was then snap-frozen using CO2 ‘snow’ and then briefly warmed in the hand to extract the homogenised tissue pellet. Pellets that were spiked with the same lipid concentrations, i.e. two pellets per concentration, were transferred into a 1 cm2 histology mould (Tissue-Tek CryoMold, Sakura Finetek, CA, USA). The zirconium oxide beads were removed from the homogenised tissue as it defrosted. The tissue homogenate was then snap-frozen in the moulds using CO2 snow. Thereafter, the homogenate standard was removed from the mould and stored at -80°C until further use. When required, 12-µm thick sections were cut using a Leica CM3050 Cryostat (Leica Microsystems, Wetzlar, Germany).

128 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

5.2.1.4 Data acquisition Three serial sections of each lipid-spiked tissue standard were cut at 12 µm and prepared using the protocol outlined in Section 2.1.2.1. DAN matrix was sublimated and MALDI-IMS data were acquired using the parameters outlined in Section 2.1.2.4, using a spacing of 50 µm between consecutive spectra.

5.2.1.5 Data analysis Spectra were re-aligned and imported to SCiLS Lab 2015b software (SCiLS GmbH, Germany) as outlined Section 3.2.3.3. This software was used produce images of the distribution of each lipid standard and measure the intensity of lipid standard at each concentration.

5.2.1.6 Standard curves Data for each lipid standard peak, at each concentration (at a maximum of 1000 µg/g), were then exported from SCiLS Lab 2015b software (SCiLS GmbH, Bremen, Germany) to GraphPad Prism 6 software for statistical analysis. A linear regression algorithm, which takes the variation of the ion intensity at each concentration into account, was used to model the relationship between the ion intensity values and the quantity of the lipid standard that was spiked in each tissue homogenate.

5.2.1.7 Histological staining The lipid-spiked standards were stained with haematoxylin and eosin (H&E) and luxol fast blue (LFB) as outlined in Section 2.1.3.1 and imaged using a Nikon Eclipse Ni microscope (Nikon, NY, USA).

5.2.2 Quantifying phosphatidylethanolamine in the middle temporal gyrus 5.2.2.1 Sample preparation and data acquisition A 12-µm thick section from each concentration of the PE lipid-spiked tissue homogenates was cut and included on the same slide as the two postmortem MTG sections from each age- and sexmatched control and AD pair. Following tissue preparation with the sublimation of DAN, data from each lipid standard were acquired prior to data from the MTG region of the control and AD tissue. A spatial distance of 50 µm between spectra was used for both lipid standards and the MTG region.

129 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

5.2.2.2 Standard curves As outlined in Section 5.2.1.6, a linear regression algorithm was used to generate standard curves for the PE 16:0/18:1 lipid standard for each dataset. The linear regressions generated from each of the six datasets were also statistically compared, to determine if one regression model could be used to quantify the amount of lipids in all datasets.

5.2.2.3 Determining m/z values to analyse A list of m/z values for all PE species was generated using the LIPID MAPS database(Fahy et al., 2007). This list was then manually cross-checked for m/z peaks detected, in negative ion mode, in the postmortem human brain, to determine putative PE lipids for quantification.

5.2.2.4 Calculation of the amount of lipid in the MTG The amount of each PE species (based on the list generated in Section 5.2.2.3) in each MTG dataset was calculated using the equation for the standard curve that was generated for that individual dataset (based on the C13 peak). The following calculation was used: 𝐴𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑃𝐸 (µ𝑔/𝑔) =

Peak Intensity − y intercept value Slope of curve

5.2.2.5 Statistical analysis Differences in the amount of each PE species in control and AD cases were statistically analysed using GraphPad Prism 6 software. Given the limited sample size, the normality of the spread of data could not be confirmed, and hence, differences in the mean concentration of PE were compared using the non-parametric Mann-Whitney U test.

130 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Section three

Results

5.3.1. Quality assessment of lipid-spiked tissue standards 5.3.1.1. Analysis

of

lipid

standards

using

matrix-assisted

laser

desorption/ionisation (MALDI) Figure 5.1 outlines the m/z peaks generated from each lipid standard. Both the PE and PI lipid standard were detected in negative ion mode, while the PC standard was detected in positive ion mode. A peak was detected at m/z 716.6 for the PE 16:0/18:1-H- lipid standard, with isotopic peaks at m/z 717.6 and m/z 718.6. For the PI 16:0/18:1-H- lipid standard, the most abundant peak was detected at m/z 835.7, with isotopic peaks at m/z 836.7 and m/z 837.7. For the PC e16:0/18:1+H+ lipid standard, which was detected in positive ion mode, an abundant peak was detected at m/z 746.7, with isotopic peaks at m/z 747.7 and m/z 748.7. The most abundant peaks can be attributed to the

12

C isotope in each lipid, while the other two lower abundance peaks can be

attributed to the 13C isotopes in each lipid. This analysis confirmed the detection of each lipid at their predicted m/z values.

5.3.1.2. Morphology of tissue homogenates Figure 5.2 shows the morphology of a lipid-spiked tissue homogenate section (Figure 5.2A) and grey matter in the postmortem human MTG, in a control case (H152; Figure 5.2B). While there was a difference in H&E and LFB staining intensity, both sections showed an even spread of cell nuclei throughout the sample. However, although an abundance of intact whole cells (black arrows) were observed in the MTG section, none were seen in the tissue homogenate section. Finally, in comparison with the whole MTG section, the lipid-spiked tissue homogenate sections showed evidence of freezing artefacts.

5.3.1.3. Analysis of lipid standards using imaging mass spectrometry (IMS) Next, I imaged the distribution of each lipid standard, in triplicate, for each concentration, using MALDI-IMS. Figure 5.3 shows the results for one replicate. The blank tissue homogenate with 0 µg/g of the PE 16:0/18:1 lipid standard still showed an abundance of the lipid, indicating the presence of endogenous signal at m/z 716.6. However, the concentration-dependent intensity increase of the PE 16:0/18:1 lipid standard was visually evident by the colour shift towards the red end of the colour intensity spectrum with increasing concentrations.

131 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Figure 5.1: MALDI analysis of lipid standards. (A) Overlaid mass spectra showing peaks detected for the phosphatidylethanolamine (PE) 16:0/18:1 and phosphatidylinositol (PI) 16:0/18:1 lipid standards (in blue) and the phosphatidylcholine (PC) e16:0/18:1 lipid standard (in red). The PE and PI lipid standards

132 | P a g e

were detected in negative ion mode and the PC lipid standard was detected in positive ion mode. Enlarged views of the isotopic peaks generated for the (B) PE, (C) PC, and (D) PI lipid standards are also shown. m/z, mass-to-charge ratio

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Figure 5.2: Morphology of lipid-spiked tissue homogenate. (A) Micrograph of a 12-µm human brain homogenate section stained with haematoxylin and eosin (H&E) and luxol fast blue (LFB). (B) Micrograph of a 12-µm section from grey matter

in the middle temporal gyrus (MTG) of a control case (H152). Black arrows indicate intact whole cells. Images were acquired using a x20 objective on a Nikon Eclipse Ni microscope (Nikon, NY, USA), using NIS Elements v.4.30 software (Nikon, NY, USA).

133 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Figure 5.3: MALDI-imaging mass spectrometry (IMS) of tissue standards spiked with different concentrations of lipids. Figure showing the distribution of the phosphatidylethanolamine (PE), phosphatidylinositol (PI), and phosphatidylcholine (PC) lipid standards in tissue homogenates spiked at the indicated concentration PE and PI were spiked with the standard at a concentration between 0 µg/g and 1000 µg/g, and imaged in negative ion mode. PC was spiked at a concentration between 0 µg/g and 500 µg/g, and imaged in positive ion mode. Data

were to total ion current. Edge-preserving image denoising and automatic hotspot removal (see rainbow intensity colour-bar) were applied using SCiLS Lab 2015b software (SCiLS GmbH, Bremen, Germany). The spatial distance between adjacent spectra is 50 μm. Only images from one representative replicate have been shown here for clarity. Max; maximum; Min, minimum; m/z, mass-to-charge ratio

The PI 16:0/18:1 lipid standard showed a similar trend, with an increase in the intensity with increasing spiked concentrations, ranging from 0 µg/g to 1000 µg/g. Again this was visually evident by the shift in colour towards the red end of the colour intensity spectrum. There was a lower abundance of an endogenous signal at m/z 835.7, which is evident from the blank tissue homogenate (i.e. 0 µg/g). The PC e16:0/18:0 lipid standard also showed a concentration-dependent intensity increase (and colour shift). However, unlike the other two lipid standards, the blank tissue homogenate did not contain any endogenous levels at m/z 746.7.

134 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS The lipid distribution in some tissue homogenates showed heterogeneity, as evidenced by a range of colours appearing within a single imaged homogenate. For instance, in the tissue homogenate spiked with 100 µg/g of the PI 16:0/18:1 lipid standard and in that spiked with 500 µg/g of the PE 16:0/18:1 lipid standard. It is unclear if these hotspots can be attributed to a preparation artefact or to variations in the distribution of the endogenous lipids. In contrast, given its low endogenous expression as evidenced with the blank tissue homogenate (0 µg/g), the heterogeneous distribution of the PC e16:0/18:1 could be attributed to a preparation artefact, indicating a need to optimise the homogenisation technique. Nonetheless, next I compared the standard curves generated for each lipid standard.

5.3.1.4. Standard curves Figure 5.4 shows results from the linear regression analysis of each lipid standard. The data point at each concentration indicates the mean intensity (± standard deviation) across triplicates. All three lipid standards showed a positive linear trend, i.e. a concentration-dependant increase in intensity, reflecting the imaging results seen in Figure 5.3. The calculated slope varied from 0.00024 (for the PC e16:0/18:1 lipid standard) to 0.002638 (for the PI 16:0/18:1 lipid standard). This indicates the variation seen in the absolute intensity measured from tissue homogenates that were spiked with the same concentration of the three different lipid standards. For instance, the average intensity measured in the tissue homogenate spiked with 500 µg/g of the PE 16:0/18:1 standard was 4.14 (3 significant figures; sf), while the average intensity measured with the same concentration of PI 16:0/18:1 was 0.159 (3sf). The three standard curves also had different y-intercepts. The standard curve for PE 16:0/18:1 had the highest y-intercept (Figure 5.4A), reflecting the high abundance of the endogenous lipid species, while the standard curve for PC e16:0/18:1 (Figure 5.4C) had the lowest y-intercept, reflecting the absence of an endogenous signal. These results reflect the imaging results seen in Figure 5.3. The R2 value, which quantifies the goodness of fit, varied from 0.8988 (for the PE 16:0/18:1 lipid standard) to 0.9721 (for the PC e16:0/18:1 lipid standard). Thus, good standard curves were generated for all three lipid standards. Nonetheless, given that the absolute intensity acquired for the highest concentration of PC e16:0/18:1 standard was very low in comparison to the intensities acquired for endogenous PC species in the postmortem human brain sections, only the quantification of PE and PI was attempted in the next section.

135 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Figure 5.4: Linear regression analysis of lipid standards. The standard curves generated for: (A) Phosphatidylethanolamine (PE) 16:0/18:0, (B) Phosphatidylinositol (PI) 16:0/18:0, and (C) Phosphatidylcholine (PC) e16:0/18:1. The data point at each concentration shows the mean intensity (± standard deviation). The standard deviation indicates the variation between triplicates.

136 | P a g e

The slope and y-intercept of the fitted regression line 2 for each lipid standard is also given, with the R value indicating the ‘goodness of fit’. This figure was prepared using GraphPad Prism 6 software (CA, USA). a.u., arbitrary unit

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

5.3.2. Lipid quantification in the middle temporal gyrus 5.3.2.1. Phosphatidylinositol quantification Figure 5.5 shows results of the linear regression analysis of the PI 16:0/18:1 lipid standard that were generated for each age- and sex-matched pair. The concentration-dependent intensity increase was evident in each dataset. However, as indicated by the maximum value on the y-axis, the intensities measured for each dataset were highly variable. Nonetheless, given that some variation was evident in the previous dataset (see Figure 3.9), I went on to evaluate the R2 value to determine how well each curve fitted the data. All the R2 values were below 0.75, with a particularly low value of 0.0619 (3sf) for the AZ45/H169 dataset. However, the R2 value alone cannot be used as the main criterion to determine if the fitted line is reasonable (Motulsky and Christopoulos, 2003). Instead, a residual data-plot that measures the distance of each data point from the fitted line was generated for each dataset, which is shown in Figure 5.6. The residual data-plot can be used to determine if the regression model can be improved. If the appropriate model has been picked, then it is expected that the data will be randomly distributed around the fitted line (Motulsky and Christopoulos, 2003). While the limited number of data points makes it difficult to determine if the appropriate model was used, some clustering is evident in datasets like AZ45/H169 and AZ90/H180. Given this result, and the concern about the high variability between different datasets and the low R2 values, I did not proceed with the quantification of PI in the MTG.

137 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Figure 5.5: Linear regression analysis of the phosphatidylinositol (PI) lipid standard for each dataset. The standard curves generated for the PI 16:0/18:0 lipid standard for each age- and sexmatched control and Alzheimer’s disease case pair are shown as follows: (A) AZ32/H137, (B) AZ45/H169, (C) AZ71/H238, (D) AZ72/H190, (E) AZ80/H152, and (F) AZ90/H180. The data point at each concentration shows the mean intensity (± standard deviation). The standard deviation measures the ion intensity

138 | P a g e

variation between spectra acquired to generate the distribution image of each homogenate. The slope and y-intercept of the fitted regression line for each 2 lipid standard is also given, with the R value indicating the ‘goodness of fit’. This figure was prepared using GraphPad Prism 6 software (CA, USA). a.u., arbitrary unit

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Figure 5.6: Residual plots of each linear regression analysis of the phosphatidylinositol (PI) lipid standard curves shown in Figure 5.5. The residual plot, which shows the absolute deviation (y-axis) of each data point from the calculated linear fit, is shown for each age- and sex-matched control and

Alzheimer’s disease case pair as follows: (A) AZ32/H137, (B) AZ45/H169, (C) AZ71/H238, (D) AZ72/H190, (E) AZ80/H152, and (F) AZ90/H180. This figure was prepared using GraphPad Prism 6 software (CA, USA).

139 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

5.3.2.2 Phosphatidylethanolamine quantification For this section, the range of final concentrations of PE in the spiked tissue homogenates was extended up to ten-fold, and hence, included intensity values for concentrations at 2500 µg/g, 5000 µg/g, and 10 000 µg/g, as well. Figure 5.7 illustrates the presence of an endogenous signal at m/z 716.6 (in black). Thus, only putative PE species with signals that were much higher than the endogenous m/z 717.6 signal, i.e. four lipids, could be quantified. In contrast, given that the endogenous m/z 717.6 peak was so low, when this peak was used to generate the standard curves, many more putative PE species, i.e. 12 species (excluding isotopes), could be quantified. Thus, I used the m/z 717.6 peak to generate the standard curves illustrated in Figure 5.8, and subsequently quantify putative PE species in all the datasets. In contrast to the standard curves generated using PI 16:0/18:1, the absolute intensities measured with PE 16:1/18:1 were within the same range for each dataset. Thus, I then statistically evaluated if one standard curve could be used for all datasets. However, given that the p-value was below 0.0001, the standard curve for each dataset had different gradients and y-intercepts. Thus, the null hypothesis that one standard curve could be used to model all datasets had to be rejected. Next, the R2 value was used to evaluate the quality of fitted line for each dataset. With the exception of the AZ80/H152 dataset, the R2 values for the fitted lines were all higher than 0.75. Figure 5.9 shows the residual data-plots for each dataset. While some clustering is evident, especially at lower concentrations, there was no evidence of any definite trends. Thus, short of evaluating more data points, the regression model could not be improved, and so I thought it reasonable to use these standard curves to quantify putative PE species, which were selected as outlined in Section 5.2.2.3. Since PE is more abundantly expressed in the grey matter in the MTG, putative PE species in the white matter could not be quantified as their concentration was below the level of detection. Table 5.2 shows the quantification of the different PE species in the grey matter in MTG, in both control and AD cases. The lipid assignment was based on the LIPID MAPS database (Fahy et al., 2007) and a previous publication by Han et al. (2001). The isotopes of some lipids, which could be quantified, were also included in this analysis. Overall, of the 12 PE species that were quantified (excluding isotopes of the same species) only half had a concentration within the range used to spike the tissue homogenates. The rest had to be extrapolated up to 60 times the highest concentration, i.e. 10 000 µg/g of spiked homogenates.

140 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Figure 5.7: Unprocessed MALDI data from tissue homogenate sections spiked with 0 µg/g (black) and 10,000 µg/g (red) PE 16:0/18:1 lipid standard. Enlarged MALDI mass spectrum, ranging from m/z 12 715.5 to m/z 719.0, indicating the C peak at m/z 13 716.6 and C peak at m/z 717.6 for the PE 16:0/18:1 lipid standard (red), and the presence of endogenous

m/z peaks at these values (black) in the blank tissue homogenate section, i.e. 0 µg/g PE 16:0/18:1 lipid standard concentration. a.u., arbitrary unit; m/z, mass-to-charge ratio; PE, phosphatidylethanolamine

5.3.2.3 Phosphatidylethanolamine differences in Alzheimer’s disease While Table 5.2 highlights differences in the concentration of each PE species in the grey matter in the MTG, between control and AD cases, only four were significantly different in AD. The concentrations of these PE species, which were significantly decreased in AD, are illustrated in Figure 5.10A. The four PE species are m/z 766 (PE 38:4-H-; p = 0.0260), m/z 778 (pPE 40:4-H-; p = 0.0043), and m/z 790 (PE 40:6-H-; p = 0.0087) and its isotope (m/z 791; p = 0.0043). Figure 5.10B shows the distribution of each of these PE species in the MTG. While m/z 790 (PE 40:6-H-) and its isotope (m/z 791) were the most abundant of these four PE species, all four were intensely expressed in the grey matter alone. Figure 5.10B visually demonstrates the depletion of these PE species in AD.

141 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Figure 5.8: Linear regression analysis of the phosphatidylethanolamine (PE) lipid standard for each dataset. The standard curves generated for the PE 16:0/18:0 lipid standard for each age- and sexmatched control and Alzheimer’s disease case pair are shown as follows: (A) AZ32/H137, (B) AZ45/H169, (C) AZ71/H238, (D) AZ72/H190, (E) AZ80/H152, and (F) AZ90/H180. The data point at each concentration shows the mean intensity (± standard deviation). The standard deviation measures the variation between

142 | P a g e

spectra acquired to generate the distribution image of each homogenate. The slope and y-intercept of the fitted regression line for each lipid standard is also 2 given, with the R value indicating the ‘goodness of fit’. This figure was prepared using GraphPad Prism 6 software (CA, USA). a.u., arbitrary unit

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Figure 5.9: Residual plots of each linear regression analysis of the phosphatidylethanolamine (PE) lipid standard curves shown in Figure 5.8. The residual plot, which shows the absolute deviation (y-axis) of each data point from the calculated linear fit, is shown for each age- and sex-matched control and

Alzheimer’s disease case pair as follows: (A) AZ32/H137, (B) AZ45/H169, (C) AZ71/H238, (D) AZ72/H190, (E) AZ80/H152, and (F) AZ90/H180. This figure was prepared using GraphPad Prism 6 software (CA, USA).

143 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS Table 5.2: Phosphatidylethanolamine (PE) quantification in grey matter in the middle temporal gyrus (MTG) in control and Alzheimer’s disease (AD). The standard curve generated for each dataset using a PE 16:0/18:1 lipid standard was used for calculations. The average amount (± standard deviation, SD) of each PE species, rounded to the nearest integer, in control and AD cohorts (n=6) are shown. Putative lipid assignments are based on the

m/z

Lipid Assignment

LIPID MAPS database (Fahy et al., 2007) and previous publications (Han et al., 2001; 2005; Milne et al., 2003). * p < 0.05; ** p < 0.01 m/z, mass-to-charge; pPE, ethanolamine plasmalogen

Control

Alzheimer’s disease

(average amount ± SD)

(average amount ± SD)

(µg/g)

(µg/g)

700

pPE 34:1-H-

1769 ± 1878

2605 ± 2007

716

PE 34:1-H-

15328 ± 3445

13352 ± 4592

718

PE 34:0-H-

15328 ± 2143

3726 ± 2613

744

PE 36:1-H-

30580 ± 10558

23007 ± 10655

746

pPE 38:6-H-

4532 ± 2063

2658 ± 2201

748

pPE 38:5-H-

3277 ± 1591

3654 ± 1788

766

PE 38:4-H-

16817 ± 5941

7324 ± 5509*

774

pPE 40:6-H-

17693 ± 5070

11786 ± 6474

778

pPE 40:4-H-

5605 ± 2233

1881 ± 1189**

790

PE 40:6-H-

40850 ± 9234

21469 ± 9417**

836

PE 43:4-H-

1876 ± 1602

2239 ± 1360

886

PE 46:0-H-

36727 ± 14173

31284 ± 10957

144 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Figure 5.10: Concentration and distribution of selected PE species in the grey matter of the middle temporal gyrus (MTG) in control and Alzheimer’s disease (AD) cases. (A) The concentration of PE species detected at m/z 766, 778, 790, and 791, which was significantly decreased in the grey matter of the MTG in AD. The average PE concentration calculated for control cases (± standard deviation, SD) is shown in black, while the average PE concentration calculated for AD cases (± standard deviation) is shown in orange. Individual data points for each control and AD case are also shown (see key). (B) The distribution of the PE species in the control and AD

MTG. All data was normalised to total ion current and was processed using a weak denoising filter and automatic hotspot removal. The intensity scale with the adjusted maximum (indicated by the black arrow) is shown for each image. The spatial between spectra was 50 µm and the scale bar is 0.5 mm. This figure was prepared using GraphPad Prism 6 software (CA, USA) and SCiLS Lab 2015b software (SCiLS GmbH, Bremen, Germany). * p < 0.05; ** p < 0.01 m/z, mass-to-charge; PE, phosphatidylethanolamine

145 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

Section Four

Discussion

5.4.1. General discussion

While MALDI-IMS is now widely used as a qualitative tool to image the distribution of biomolecules of interest, its use in absolute quantification has been mostly limited to pharmaceuticals (MarkoVarga et al., 2011, Hamm et al., 2012, Pirman et al., 2012, Shahidi-Latham et al., 2012, Takai et al., 2013). Recently, however, a study by Jadoul et al. (2015) demonstrated the use of MALDI-IMS to quantify PC species in the rat brain, using lipid-spiked porcine brain tissue homogenates. The aim of this chapter was to use a similar approach to quantify three lipid groups, i.e. PC, PE, and PI, in the postmortem human MTG.

5.4.1.1. Use of lipid-spiked tissue homogenates for quantification The presence of multiple molecules, such as proteins, lipids, oligonucleotides, carbohydrates, and other small organic molecules, make biological tissue complex to analyse using MALDI-IMS, because each molecule class can affect the desorption and ionisation of others. This phenomenon, which is known as ion suppression, must be taken into account when designing a quantitative MALDI-IMS study (Chughtai and Heeren, 2010). Thus, rather than the use of pure lipid standards spotted alongside the sample, the incorporation of lipid standards into tissue homogenates ensures that the standard and analyte have a similar molecular microenvironment. Jadoul et al. (2015) outlined five different approaches that can be used to create the lipid-spiked tissue homogenates. Since the same instruments were not available, I used an approach described by Hare et al. (2013), who used cortical sheep tissue to generate tissue homogenate standards to quantify trace elements in the mouse brain using laser-ablated inductively coupled plasma-MS. Further, in contrast to the use of porcine brain tissue as outlined by Jadoul et al. (2015), I used postmortem human brain tissue to prepare the tissue homogenates to limit cross-species variation and postmortem treatment variation. Morphologically, the tissue homogenates showed an even spread of cell nuclei throughout the sample, similar to grey matter, but did not show any whole cells. Further, the tissue homogenate sections showed damages that could be attributed to the formation of ice crystals during freezing. These artefacts may have contributed to the high degree of ion intensity variability observed in the MALDI images of the spiked homogenates (Figure 5.3). Thus, in future, the use of another coolant, such as isopentane or liquid nitrogen, could be used to accelerate the freezing process and minimise the formation of ice crystals (Barnard, 1987). In human erythrocyte plasma, lipids belonging to different classes, but containing acyl chains that ranged from 14–20 carbons in length, showed nearly identical ionisation efficiencies (Han and Gross, 146 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS 1994). However, this finding has not been confirmed using postmortem human brain tissue, or for lipids that contain acyl chains of other lengths. Thus, I chose to use one lipid standard for each lipid class of interest. The PE and PI lipid standards were only detected in negative ion mode, while the PC lipid standard was detected in positive ion mode. Further, the three lipid standards generated different m/z peaks. Hence, all three standards were incorporated in the same tissue homogenate, at each concentration. The concentration-dependent intensity increase was evident for all three lipid standards (Figure 5.3). However, given the low absolute intensity of the PC standard and the variation seen across datasets for the PI standard, only the PE lipid standard proved to be useful for quantification.

5.4.1.2. Phosphatidylethanolamine changes in the grey matter in the middle temporal gyrus in Alzheimer’s disease Table 5.2 shows the quantification of PE species, in the grey matter in the MTG, of which four were statistically significantly decreased in AD. These were m/z 766 (PE 38:4-H-), 778 (pPE 40:4-H-), and 790 (PE 40:6-H-) and its isotope (m/z 791). It is interesting to note that the loss of PE 39:5-H- and PE 40:6-H- in the MTG was previously detected in the MTG using the analysis workflow (Chapter Three). Further, the depletion of all three lipids was also detected in the hippocampus (Chapter Four). The changes detected in this study, however, were not as extensive as those reported by Han et al. (2001), who used the liquid-chromatography mass spectrometry (LC-MS) method to quantify PE species in the human cortex. Although the trends in my results are comparable to their study, since they calculated the amount of PE as nanomoles per gram of protein the values in this thesis could not be directly compared to theirs. The results in this thesis were calculated as a concentration (in micrograms per gram of tissue) as outlined previously (Jadoul et al., 2015). Further, Jadoul et al. (2015) quantified the absolute amount of each PE species in micrograms. However, given the incomplete sample ablation seen in my work, the scale was not sensitive enough to detect the mass of the sample that was measure. In future, the use a technique such as electron microscopy might allow me to accurately measure the area of sample ablation, and thus, calculate the mass of the sample that was ablated. However, this remains to be confirmed. Finally, since this was the first use of a lipid-spiked tissue homogenate approach to quantify PE in the postmortem human brain, these results remain to be further validated using a method such as LCMS.

147 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

5.4.1.3. Considerations for future use A limitation of this quantification approach is that since the spiked standard is not part of the membrane, the physical properties of the homogenate are not identical to intact tissue slices. Nonetheless, it is the closest approximation that is currently available. This chapter highlights a few considerations that need to be taken into account when applying this approach in future. Firstly, the standard curve concentration range did not cover the endogenous levels of all PE species since published data on the PE concentration in the MTG is unavailable. For example, when quantifying the concentration of the 12 PE species (i.e. excluding the isotopes of the same species) in the MTG, only half had a concentration within the range used to spike the tissue homogenates. Further, of the PE species that were significantly depleted in AD, only the concentration of m/z 778 (pPE 40:4-H-) was within the range used to spike the tissue homogenates. This required the extrapolation of data, which could have caused errors. In future, LC-MS, or a series of diluted lipid concentrations that are spotted on the section, could be used to determine a more appropriate concentration range. Secondly, the presence of the chosen lipid standard, led me to use the quantification of the

13

C lipid peak for the

12

C peaks of putative PE species. This approach has not been previously

reported in the literature and could have also led to errors. The use of a deuterated lipid standard can be used in future to circumvent this challenge. Deuterated synthetic lipids consist of deuterium (heavy hydrogen) instead of a hydrogen atom, at the alpha carbon at the C2 position, for example. This increases their mass weight, thus, shifting the m/z value of the observed lipid peak to the right. Apart from this mass shift, deuterated synthetic lipids have similar chemical characteristics to the target analyte, and thus, will ionise in a comparable manner (Pirman et al., 2012). Provided that no other endogenous signal is present at the expected m/z for the deuterated lipid standard, and that the concentration range is large enough, this approach will provide a more accurate quantification of PE in the postmortem human brain.

148 | P a g e

Chapter 5: Quantifying lipids in the middle temporal gyrus using MALDI-IMS

5.4.2. Summary of findings The primary aim of this chapter was to quantify lipid species belonging to the PC and PE classes, as these were abundantly expressed in the MTG and showed the greatest change in AD. The quantification of PI was also attempted as it is one of the most abundant lipid classes in the DG, a region in the hippocampus, which is the focus of the next chapter. The accurate quantification of lipids could provide more insight into region-specific lipid imbalances that could be addressed in the future. For quantification I used a lipid-spiked tissue homogenate approach to quantify lipids belonging to the PC, PE, and PI, lipid classes. Following the validation process, only standard curves generated using the PE 16:0/18:1 lipid standard could be used for quantification. Given the low abundance of PE in white matter, only PE species in the grey matter in the MTG were quantified. Further, only PE species with a concentration higher than the endogenous level of PE 16:0/18:1 could be quantified. Nonetheless, three PE species, i.e. PE 38:4-H-, PE 39:5-H-, and PE 40:6-H- were significantly depleted in AD, of which two were detected using the analysis workflow developed in Chapter Three. This study validated the use of lipid-spiked tissue homogenates for the quantification of glycerophospholipids using MALDI-IMS. However, since it is the first to quantify PE in the postmortem human brain using this approach, the results need to be corroborated and there are several considerations that need to be taken into account before this approach can be used routinely. The use of a deuterated lipid standard, a large enough concentration range, and the validation of results using LC-MS, will address the current limitations of the methodology outlined here.

149 | P a g e

Chapter Six

Lipid distribution and quantification in the dentate gyrus

Section One: Introduction ..........................................................................................................152 Section Two: Methods................................................................................................................153 6.2.1 High-resolution imaging of the dentate gyrus ........................................................................... 153 6.2.1.1 6.2.1.2 6.2.1.3 6.2.1.4

Cases used and tissue preparation .................................................................................................. 153 Data acquisition ............................................................................................................................... 153 Data processing ............................................................................................................................... 153 Histological staining ........................................................................................................................ 155

6.2.2 Phosphatidylethanolamine quantification in the dentate gyrus ............................................... 155 6.2.2.1 Standard curves ............................................................................................................................... 155 6.2.2.2 Calculation of phosphatidylethanolamine concentration in the dentate gyrus ............................... 155 6.2.2.3 Statistical analysis............................................................................................................................ 155

Section Three: Results ................................................................................................................156 6.3.1 High-resolution imaging of lipid distribution in the dentate gyrus ........................................... 156 6.3.1.1 Mismatch between the defined area for data acquisition and the actual measured area 156 6.3.1.2 Lipid distribution in the dentate gyrus in control sections ................................................. 156 6.3.1.3 Distribution of lipids that are differentially expressed in Alzheimer’s disease ................... 158 6.3.2 Phosphatidylethanolamine quantification in the dentate gyrus ............................................... 161 6.3.2.1 Standard curves .................................................................................................................. 161 6.3.2.2 Concentration of phosphatidylethanolamine in the dentate gyrus ................................... 165 Section Four: Discussion .............................................................................................................167 6.4.1. General discussion ..................................................................................................................... 167 6.4.1.1 High-resolution matrix-assisted laser desorption (MALDI)-imaging mass spectrometry (IMS) ....... 167 6.4.1.2 Lipid distribution in postmortem human dentate gyrus .................................................................. 167 6.4.1.3 Quantification of the phosphatidylethanolamine concentration in the dentate gyrus ................... 169

6.4.2. Summary of findings .................................................................................................................. 169

Chapter 6: Lipid distribution and quantification in the dentate gyrus

Section One

Introduction

The trilaminar structure of the dentate gyrus (DG) consists of: 1) the molecular layer (ML), 2) the granule cell layer (GCL), and 3) the polymorphic layer (PL), and is conserved across rodents and humans (Anderson et al., 2007, Duvernoy et al., 2013). The GCL, the principal cell layer in the DG, which contains small, round densely-packed granule cells, is flanked by the ML and PL (Amaral et al., 2007, Duvernoy et al., 2013). The DG, and the GCL in particular, play an integral role in the production of episodic memory, since it is the first point that begins processing glutamatergic input from the entorhinal cortex (Insausti and Amaral, 2004, Amaral et al., 2007, Duvernoy et al., 2013). The specific roles of the dorsal DG, in separating spatial and contextual patterns, and the ventral DG, in separating odour patterns, have been previously reviewed extensively by Kesner (2013). Further, the persistence of neurogenesis in the DG in adulthood, from a population of progenitor cells in the subgranular zone (SGZ; Eriksson et al., 1998), is postulated to enhance pattern separation, at least in the rodent brain (Dieni et al., 2016). In Alzheimer’s disease (AD), the DG is largely resistant to the formation of hyperphosphorylated tau tangles and Aβ plaques until later stages of the disease (Braak and Braak, 1991). However, the dendritic loss of granule cells and the decline in the synaptic connections in the inner ML occurs much earlier, affecting memory function (De Ruiter and Uylings, 1987, Scheff and Price, 1998). While neurogenesis continually produces new cells in the SGZ, even in AD, these cells fail to mature, and thus, cannot reverse hippocampal atrophy (Li et al., 2008). The anatomical and neurochemical changes that occur in the DG in AD have been comprehensively reviewed by Ohm (2007). However, only one study, which was conducted by Hirano-Sakamaki et al. (2015), has addressed lipid aberrations in the DG in AD. They reported a specific decrease in the GM1 d20:1/C18:0 in relation to GM1 d18:0/C18:0 in AD, in the ML of the dentate gyrus. Thus, nothing is known about the distribution of other lipid classes in the postmortem human DG. Therefore the primary aim of this chapter is to visualise the distribution of lipids in the DG using high resolution matrix-assisted laser desorption/ionisation (MALDI)-imaging mass spectrometry (IMS). While Hirano-Sakamaki et al. (2015), only focussed on the distribution of four gangliosides, I focussed on all the lipids that were previously identified as differentially expressed in AD in the DG alone (Chapter Four). The second aim of this chapter is to use the lipid-spiked tissue homogenate approach,

outlined

in

Chapter

Five,

to

quantify

the

concentration

of

different

phosphatidylethanolamine (PE) lipids in the DG in control sections, and analyse these changes in AD.

152 | P a g e

Chapter 6: Lipid distribution and quantification in the dentate gyrus

Section Two

Methods

6.2.1 High-resolution imaging of the dentate gyrus 6.2.1.1 Cases used and tissue preparation Given the limited tissue availability, I was only able to image the DG from five control and five AD cases, as outlined in Table 6.1. The protocol outlined in Section 2.1.2.1 was used to prepare tissue sections. Each slide contained hippocampus sections from one age- and gender-matched control and AD pair. Sections from the tissue homogenates that were spiked with lipids ranging from a concentration of 0 µg/g to 10 000 µg/g (see Section 5.2.1) were also mounted on each slide. Sections were then washed with 50 mM ammonium formate and 1,5-diaminonapthalene (DAN) was applied using the protocol outlined in Section 2.1.2.3.

6.2.1.2 Data acquisition Data for this chapter was acquired using an UltrafleXtreme MALDI TOF/TOF mass spectrometer (Bruker, Bremen, Germany), equipped with a 2 kHz Smartbeam IITM UV MALDI laser. The instrument was set up as outlined in Section 2.1.2.4. All datasets were acquired in negative ion mode alone. The DG was imaged using a raster step-size of 20 µm, with the laser beam size set to 'minimum’, i.e. approximately 10 µm, while the lipid-spiked tissue homogenates were imaged using a raster step-size of 50 µm, with the laser beam size set to ‘small’, i.e. approximately 40 µm. A total of 75 laser shots per spectrum were accumulated at each point. All data were collected in the mass range of m/z 400 to 2000. Prior to data collection, the three-point teaching procedure on FlexImaging (version 4.1) software (Bruker Daltonics GmbH, Bremen, Germany) was used to accurately match the pre-scanned optical image of the tissue section and the actual target position in the instrument. Teaching points were set as close as possible to the imaged area. An imaging dataset of the lipid-spiked tissue homogenate standards were acquired before each DG imaging acquisition run.

6.2.1.3 Data processing Spectra were then re-aligned, as outlined in Section 3.2.3.4, imported to SCiLS Lab 2015b software (SCiLS GmbH, Bremen, Germany), and normalised to total ion count (TIC). SCiLS Lab 2015b software was used to generate images illustrating the distribution of lipids.

153 | P a g e

Chapter 6: Lipid distribution and quantification in the dentate gyrus presented. The extent of atrophy (A; 0-3); neurofibrillary tangles (B; 0-3), and neuritic plaques (C; 0-3) for these cases are also given. Age- and gender-matched cases have been listed consecutively in each section.

Table 6.1: Cases used for Chapter Six. Summary of age, gender, postmortem delay, cause of death, and pathology, for control (n=5) and Alzheimer’s disease (AD; n=5) cases used in this study. The control cases used in this study did not show significant histological abnormalities. The Braak and Braak stage and the Consortium to Establish a Registry for Alzheimer's disease (CERAD) score for the Alzheimer’s disease cases used in this study are

Control

Alzheimer’s disease

154 | P a g e

HP, hippocampus

Gender

Postmortem delay (hrs)

Hippocampus block used in study

Pathology

77

F

21

HP1

No significant histological abnormalities

H152

79

M

18

HP2

No significant histological abnormalities

H180

73

M

33

HP3

No significant histological abnormalities

H190

72

F

19

HP2

No significant histological abnormalities

H238

63

F

16

HP3

No significant histological abnormalities

Case

Age (years)

H137

AZ32

75

F

3

HP2

AZ80

77

M

4.5

HP2

AZ90

73

M

4

HP2

AZ72

70

F

7

HP2

AZ71

61

F

6

HP2

Braak: Unknown; CERAD: Probable Alzheimer’s A2 B1 C2 Braak: VI; CERAD: Definitive Alzheimer’s A3 B3 C3 Braak: IV; CERAD: Definitive Alzheimer’s A3 B3 C3 Braak: V; CERAD: Indicative of Alzheimer’s A0 B1 C3 Braak: VI; CERAD: Definitive Alzheimer’s A2 B3 C3

Chapter 6: Lipid distribution and quantification in the dentate gyrus

6.2.1.4 Histological staining Since most of the DG was ablated following data acquisition, histological staining of the MALDI imaged sections could not be performed. Therefore, sister-sections of each case were stained with haematoxylin and eosin (H&E) and luxol fast blue (LFB), as outlined in Section 2.1.3, and imaged using a Nikon Eclipse Ni microscope.

6.2.2 Phosphatidylethanolamine quantification in the dentate gyrus 6.2.2.1 Standard curves Data for the C13 peak of the PE 16:0/18:1 lipid standard was exported from SCiLS Lab 2015b software to GraphPad Prism 6 software (CA, USA). A linear regression algorithm was used to model the relationship between the ion intensity values and the quantity of the lipid standard that was spiked in each tissue homogenate, which ranged from 0 µg/g to 10 000 µg/g. The generated curves were then statistically compared to determine if one regression model could be used to quantify the amount of lipids in all datasets.

6.2.2.2 Calculation of phosphatidylethanolamine concentration in the dentate gyrus The PE concentration was quantified using the appropriate standard curve, i.e. one that was generated for the dataset acquired prior to the particular DG image acquisition run. GraphPad Prism 6 software was used to interpolate these values using the following calculation: 𝐴𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑃𝐸 (µ𝑔/𝑔) =

Peak intensity − y intercept value Slope of curve

6.2.2.3 Statistical analysis Differences in the mean concentration of each PE species in control and AD cases were statistically analysed using GraphPad Prism 6 software. Some values could not be quantified as the concentration of those PE species were below the limit of detection. Thus, the normality of the spread of data could not be confirmed. Hence, differences between control and AD cases were compared using the nonparametric Mann-Whitney U test.

155 | P a g e

Chapter 6: Lipid distribution and quantification in the dentate gyrus

Section Three

Results

6.3.1 High-resolution imaging of lipid distribution in the dentate gyrus 6.3.1.1 Mismatch between the defined area for data acquisition and the actual measured area Figure 6.1A shows a sister section from an AD case (AZ71) that was stained with H&E and LFB. The GCL is outlined in grey, and a higher-resolution image of the three layers of the DG, i.e. the PL, the GCL, and the ML, is shown inset. The area defined before data acquisition, which was based on a high-resolution scan of the section prepared for MALDI, encompassed all three layers. However, when the data was analysed, there was a slight mismatch in the expected distribution of m/z 1544 (GM1 d18:1/C18:0), which is absent in the GCL in the dentate gyrus, in the murine (Colsch et al., 2011) and human brain (Hirano-Sakamaki et al., 2015). Thus, the area that was actually used to acquired data was corrected using the distribution of the m/z 1544 (GM1 d18:1/C18:0) peak, as shown in Figure 6.1C. The slight mismatch between the two areas is clearly illustrated in Figure 6.1D. Given this mismatch, the area from which data was actually acquired did not contain comparablysized ML and PL regions, either side of the DG. Thus, a detailed investigation on the differential lipid expression in the three layers of the DG could not be conducted. Nonetheless, the higher-resolution imaging of the DG provided valuable information to understand how lipids were distributed across the three layers in the DG.

6.3.1.2 Lipid distribution in the dentate gyrus in control sections Table 6.2 outlines the lipids, which were detected as being differentially expressed in AD, in the dentate gyrus alone, in Chapter Four (Section 4.3.2.5). Figure 6.2 illustrates the distribution of these lipids in the dentate gyrus in a control (H238) section, at a higher resolution (20 µm) than previously imaged. Figure 6.2A indicates the adjusted region, clearly showing the area from which data was acquired for this case, i.e. H238. The GCL is seen as a bright band within the DG, and has also been outlined in the sister section, which was stained with H&E and LFB (Figure 6.2B). A higher-resolution image of the three layers of the DG, i.e. the PL, the GCL, and the ML, is also shown inset. Given the mismatch between the defined area and that from which data was acquired, Figure 6.2C-J mainly show the distribution of lipids in the GCL and ML, with some indication of lipid abundance in the PL at each end.

156 | P a g e

Chapter 6: Lipid distribution and quantification in the dentate gyrus

Figure 6.1: Mismatch between user-defined area and actual area used for data acquisition. (A) Sister section from an Alzheimer’s disease case (AZ71) stained with haematoxylin and eosin (H&E) and luxol fast blue (LFB) illustrating the anatomy of the dentate gyrus (DG). (B) Distribution of m/z 1544 (GM1 d18:0/C18:0) overlaid on the area that was defined prior to data acquisition. The granule cell layer (GCL) is highlighted in white. Since m/z 1544 is reported as

being absent in this layer, this overlay shows the slight mismatch. (C) Actual area of measurement. (D) Overlay of the user-defined area (yellow) and the actual measured area (red), portraying the mismatch between the two. CA, Cornu Ammonis; GM, ganglioside; ML, molecular layer; m/z, mass-to-charge ratio; PL, polymorphic layer; SGZ, subgranular zone; SL, stratum lacunosum; SR, stratum radiatum

157 | P a g e

Chapter 6: Lipid distribution and quantification in the dentate gyrus Table 6.2: Lipids that were differentially expressed in the dentate gyrus (DG) alone. Table indicating the the relative change (i.e. an increase ↑ or decrease ↓; mean percentage change given in brackets) in the mean intensity of selected negatively-charged lipids in the DG in Alzheimer’s disease, as presented in

Chapter Four. Putative lipids assignments were based on MS/MS data (see Appendix; Figure A1; Table A2) and previous publications as indicated. Cer, ceramide; GM, ganglioside; PA, phosphatidic acid; PI, Phosphatidylinositol

Observed

Lipid

Relative intensity difference in

m/z

assignment [Ref]

the DG

646.5

Cer 42:2-H [1]

-

↑ (26.1)

-

647.5

PA 32:0-H *

↑ (23.9)

719.6 885.6

↑ (20.6) -

PI 38:4-H [2,3,4]

↑ (20.1)

886.7

↑ (17.8)

905.7

↓ (-37.2) -

906.7

SF 24:0 (0H)-H [2,3,4,5]

↓ (-36.8)

1574.0

GM1 d20:1/18:0-H [6,7,8]

-

↓ (-37.1)

* Putative lipid assignment based on MALDI-TOF-TOF MS/MS data References: 1. 2. 3. 4.

Hsu and Turk (2002) Dill et al. (2010) Veloso et al. (2011b) Jackson et al. (2005)

5. 6. 7. 8.

Yuki et al. (2011) Ariga et al. (1982) Chan et al. (2009) Whitehead et al. (2011)

m/z 646.5 (Cer 42:2-H-), m/z 647.5 (PA 32:0-H-), and m/z 719.6, were abundant in both the GCL and ML. All three were homogenously distributed along the DG, with the latter lipid species showing some hotspots in the GCL. m/z 885.6 (PI 38:4-H-), and its probable isotope m/z 886.7, were also particularly abundant in the GCL. The abundance of m/z 905.7 and m/z 906.7 (SF 24:0 (OH)-H-), however, were low in the GCL. m/z 1574.0 (GM1 d20:1/18:1-H-), like m/z 1544 (GM1 d18:1/C18:0) was also completely absent in this layer.

6.3.1.3 Distribution of lipids that are differentially expressed in Alzheimer’s disease Figure 6.3 shows the distribution of the same lipids in the AD dentate gyrus in a representative section, i.e. AZ71. The selected lipids showed a similar pattern of distribution in comparison to that seen in control cases. Further, the differential expression of these lipids, which were detected in Chapter Four, outlined in Table 6.3, was clearly evident. m/z 646.5 (Cer 42:2-H-), m/z 647.5 (PA 32:0-H-), and m/z 719.6, were increased across the DG. m/z 885.6 (PI 38:4-H-), and its isotope m/z 886.7, was also increased in AD, particularly, in the GCL. The decreased level of m/z 905.7, 906.7 (SF 24:0 (OH)-H-), and m/z 1547.0 (GM1 d20:1/18:1-H-), was also clearly evident.

158 | P a g e

Chapter 6: Lipid distribution and quantification in the dentate gyrus

Figure 6.2: Distribution of selected lipids in the dentate gyrus (DG). (A) Pre-acquisition scan of control (H238) section. (B) Sister section stained with haematoxylin and eosin (H&E) and luxol fast blue (LFB). The granule cell layer (GCL) has been traced. Inset: Enlarged DG area. (C)–(J) Distribution of selected lipids. Edge-preserving image denoising and automatic hotspot removal (see rainbow intensity colour-bar) have been applied. The spatial distance between consecutive spectra is 20 μm.

Lipid distribution images were generated using SCiLS lab 2015b software (SCiLS Gmbh, Bremen, Germany). CA, Cornu Ammonis; Cer, ceramide; GM, ganglioside; ML, molecular layer; m/z, mass-to-charge ratio; PA, phosphatidic acid, PI, phosphatidylinositol; PL, polymorphic layer; SF, sulfatides; SL, stratum lacunosum; SR, stratum radiatum

159 | P a g e

Chapter 6: Lipid distribution and quantification in the dentate gyrus

Figure 6.3: Distribution of selected lipids in the dentate gyrus in Alzheimer’s disease (AD). (A) Preacquisition scan of AD (AZ71) section. (B) Sister section stained with haematoxylin and eosin (H&E) and luxol fast blue (LFB). The granule cell layer (GCL) has been traced. Inset: Enlarged DG area. (C)–(J) Distribution of selected lipids. Edge-preserving image denoising and automatic hotspot removal (see rainbow intensity colour-bar) have been applied. The

160 | P a g e

spatial distance between adjacent spectra is 20 μm. Lipid distribution images were generated using SCiLS lab 2015b software (SCiLS Gmbh, Bremen, Germany). CA, Cornu Ammonis; Cer, ceramide; GCL, granule cell layer; GM, ganglioside; ML, molecular layer; PA, phosphatidic acid; m/z, mass-to-charge ratio; PI, phosphatidylinositol; PL, polymorphic layer; SF, sulfatide; SL, stratum lacunosum; SR, stratum radiatum

Chapter 6: Lipid distribution and quantification in the dentate gyrus

6.3.2 Phosphatidylethanolamine quantification in the dentate gyrus 6.3.2.1 Standard curves Given that only the PE 16:0/18:1 lipid standard proved to be useful for quantification in the previous chapter, the same protocol was used to quantify PE in the dentate gyrus. Figure 6.4 shows the results of the linear regression analysis of the m/z 717.6 peak (the 13C isotope of PE 16:0/18:0), which were generated for each control dataset, while Figure 6.5 shows the linear regression analysis generated for the AD datasets. The standard deviation of each data point shows the variation of the mean intensity recorded across the area of the standard that was imaged. In the control cohort, the standard curve for H238 showed the highest y-intercept (0.2673). The R2 values of the standard curves generated for H137, H190, and H152, were all above 0.7. However, those generated for H238 and H180, were 0.5871, and 0.5272, respectively, which is quite low. In the AD cohort (Figure 6.5), the y-intercept ranged from 0.1217, for AZ32, to 0.2746, for H238. While most of the standard curves showed R2 values close to or greater than 0.7, the curve generated for the AZ90 dataset was very low at 0.4853. I generated the residual plots of each standard curve to explore if the regression model could be improved for any of the datasets (Figure 6.6). Generally, most of the residual data plots show random scatter. However, data for H180 (Figure 6.6E) and AZ90 (Figure 6.6J), for the concentrations between 500 µg/g and 10,000 µg/g show a parabolic trend, suggesting the need for a better regression model. However, given that not all points follow the classic parabolic shape, more data points are required to make this decision definitively and select a more appropriate regression model. Finally, the fit of the standard curves generated from all the datasets were compared to assess if one standard curve could be used for all datasets using an extra sum of squares F-test. However, given that the p-value for this test was