(PET) with the Second Generation Translocator Protein (TSPO) Ligand [18F]GE-180

Quantitative analysis of positron emission tomography (PET) with the second generation translocator protein (TSPO) ligand [18F]GE-180

A thesis submitted to the University of Manchester for the degree of Doctor of Philosophy (PhD) in the Faculty of Biology, Medicine and Health

2016

Sujata Sridharan

School of Health Sciences

Contents

Contents ...... 2 List of tables ...... 7 List of figures ...... 8 Abstract...... 10 Declaration ...... 11 Copyright statement ...... 11 Acknowledgements...... 12 Abbreviations ...... 14 1 Introduction ...... 16 1.1 Rationale for the alternative format ...... 16 1.1.1 Aims ...... 16 1.1.2 Structure of the thesis ...... 16 1.2 Role of the author ...... 17 1.3 Inflammation and microglia ...... 19 1.4 The 18 kDa translocator protein ...... 21 1.5 PET imaging ...... 23 1.5.1 [11C]-(R)-PK11195 ...... 26 1.5.2 Specific and non-specific binding ...... 26 1.5.3 The TSPO polymorphism ...... 27 1.6 TSPO expression in neurological disease ...... 28 1.6.1 Multiple sclerosis ...... 28 1.6.2 Preclinical studies ...... 31 1.7 Plasma and metabolites of [11C]-(R)-PK11195 ...... 32 1.8 Kinetic analysis and modelling of TSPO data ...... 34 1.8.1 Overview of techniques ...... 35 1.8.2 Plasma input functions ...... 36 1.8.2.1 Compartment models ...... 36 1.8.2.2 Graphical analysis – Logan plots...... 38 1.8.2.3 Graphical analysis – Patlak plots ...... 39 1.8.2.4 Spectral analysis ...... 39 1.8.2.5 Key results from studies using plasma input functions ...... 41 1.8.2.6 Summary of plasma input approaches ...... 43 1.8.3 ROI analysis vs. parametric mapping ...... 44 1.8.4 Image-derived input functions ...... 44

1.8.5 Reference tissue input functions ...... 45 1.8.5.1 The simplified reference tissue model ...... 46 1.8.5.2 Reference tissue graphical analysis ...... 47 1.8.5.3 Reference tissue input parametric images ...... 49 1.8.5.4 Key results from studies using reference input functions .... 50 1.8.5.5 Summary of reference tissue input approaches...... 52 1.8.6 Cluster analysis ...... 52 1.8.6.1 Unsupervised clustering ...... 53 1.8.6.2 Supervised clustering ...... 55 1.8.6.3 Key results from studies using cluster analysis ...... 57 1.8.6.4 Summary of cluster analysis approaches ...... 59 1.8.7 Summary of kinetic analysis approaches ...... 59 1.9 Summary and rationale for study ...... 62 2 Comparative evaluation of three TSPO PET radiotracers in a LPS-induced model of mild neuroinflammation in rats ...... 63 2.1 Abstract ...... 63 2.2 Introduction ...... 65 2.3 Materials and methods ...... 68 2.3.1 Tracer synthesis ...... 68 2.3.2 Animals ...... 68 2.3.3 LPS and AMPA administration ...... 68 2.3.4 Scanning protocol ...... 69 2.3.5 Image analysis ...... 70 2.3.6 Data-driven method for the extraction of reference tissue kinetics71 2.3.7 PET data modelling ...... 72 2.3.8 Immunohistochemistry ...... 72 2.3.9 Autoradiography ...... 73 2.3.10 Statistical analysis...... 74 2.4 Results ...... 75 2.4.1 PET imaging ...... 75 2.4.1.1 Non-paired scans ...... 77 2.4.1.2 Dual scans...... 77 2.4.2 Kinetic analysis ...... 77 2.4.2.1 SRTM with contralateral reference input ...... 78 2.4.2.2 SRTM with cerebellar reference input ...... 78 2.4.2.3 SRTM with data-driven clustering reference input ...... 78 2.4.3 Autoradiography ...... 79 3

2.4.4 Immunohistochemistry ...... 79 2.5 Discussion ...... 82 2.5.1 PET imaging and ex vivo observations ...... 82 2.5.2 PET data modelling ...... 84 2.6 Conclusions ...... 86 Supplementary materials ...... 88 3 Characterisation of a blood sampling system for generating [18F]DPA-714 TSPO-PET plasma input functions in the rat brain ...... 90 3.1 Abstract ...... 90 3.2 Introduction ...... 92 3.3 Materials and methods ...... 95 3.3.1 Animals ...... 95 3.3.2 PET data acquisition ...... 95 3.3.3 Arterial sampling...... 95 3.3.4 Dispersion correction ...... 97 3.3.5 PET image analysis ...... 98 3.3.6 Kinetic modelling ...... 99 3.4 Results ...... 101 3.4.1 Blood data ...... 101 3.4.2 PET imaging ...... 102 3.4.2.1 Plasma input modelling with dispersion correction ...... 103 3.4.2.2 Reference tissue modelling – contralateral input...... 105 3.4.2.3 Reference tissue modelling – population-based input ...... 105 3.5 Discussion ...... 108 3.5.1 Characterisation of blood sampling system ...... 109 3.5.2 Kinetic modelling comparisons ...... 110 3.6 Conclusion ...... 112 Supplementary materials ...... 113 3.7 Results without dispersion correction ...... 113 4 Initial evaluation of [18F]GE-180 as a TSPO-PET imaging biomarker in multiple sclerosis ...... 115 4.1 Abstract ...... 115 4.2 Introduction ...... 117 4.3 Materials and methods ...... 120 4.3.1 Subjects ...... 120 4.3.2 Study timeline...... 120 4.3.3 MRI scanning...... 121

4.3.4 PET scanning ...... 121 4.3.5 Measurement of arterial plasma ...... 122 4.3.6 Image processing ...... 123 4.3.6.1 PET images ...... 123 4.3.6.2 MR images ...... 123 4.3.7 Kinetic analysis ...... 123 4.3.8 Model comparison and statistical tests ...... 125 4.4 Results ...... 126 4.4.1 MRI findings ...... 126 4.4.2 Blood data ...... 127 4.4.3 Kinetic analysis of PET data ...... 128 4.4.3.1 Regional rank order of uptake ...... 131 4.4.3.2 HAB vs. MAB ...... 132 4.4.3.3 MS vs. HV ...... 132 4.4.3.4 Lesion quantification ...... 134 4.4.4 Subject demographics and observations from PET ...... 138 4.5 Discussion ...... 140 4.5.1 Blood activity ...... 140 4.5.2 Plasma input modelling ...... 142 4.5.3 Simplified reference region modelling ...... 144 4.5.4 Standardised uptake values ...... 146 4.5.5 HAB/MAB differences ...... 147 4.5.6 Summary...... 147 Supplementary materials ...... 148 5 Therapy response monitoring with [18F]GE-180 in a cohort of relapsing- remitting multiple sclerosis patients ...... 151 5.1 Abstract ...... 151 5.2 Introduction ...... 153 5.3 Materials and methods ...... 155 5.3.1 Subjects ...... 155 5.3.2 Study timeline...... 155 5.3.3 PET scanning ...... 156 5.3.4 Kinetic analysis ...... 157 5.4 Results ...... 159 5.4.1 Blood data ...... 159 5.4.2 Clinical observations ...... 159 5.4.3 Imaging data ...... 160

5.5 Discussion ...... 164 5.6 Conclusion ...... 168 6 Plasma-protein binding of the 18 kDa TSPO-PET tracer [18F]GE-180 ...... 169 6.1 Abstract ...... 169 6.2 Introduction ...... 170 6.3 Materials and methods ...... 173 6.3.1 Genotyping assays ...... 173 6.3.1.1 Blood preparation and DNA extraction ...... 173 6.3.1.2 Genotyping ...... 174 6.3.2 Plasma-protein binding displacement assay ...... 174 6.4 Results ...... 176 6.4.1 Genotyping ...... 176 6.4.2 Plasma-protein binding ...... 176 6.4.3 Repeat experiment ...... 178 6.4.3.1 Method adjustments ...... 178 6.4.3.2 Results ...... 179 6.5 Discussion ...... 181 Supplementary materials ...... 183 7 Summary and conclusion ...... 184 7.1 Recapitulation of research aims ...... 184 7.2 Preclinical studies ...... 184 7.2.1 The LPS model of inflammation – Chapter 2 ...... 184 7.2.2 Quantification of [18F]DPA-714 brain data – Chapter 3 ...... 187 7.3 Clinical studies ...... 188 7.3.1 Baseline RRMS and healthy data – Chapter 4 ...... 188 7.3.2 Monitoring treatment in RRMS patients – Chapter 5 ...... 194 7.3.3 Plasma-protein binding of [18F]GE-180 – Chapter 6 ...... 196 7.4 The polymerisation of TSPO and other considerations ...... 198 7.5 Final comments ...... 200 Appendix 1 Definitions of macroparameters ...... 203 Appendix 2 Posters and conference presentations ...... 206 References ...... 208 Word count = 58, 900

List of tables Table 1.1 Common radioisotopes used in PET imaging...... 23 Table 1.2 Summary of plasma and reference tissue approaches...... 61 Table 2.1 Regional uptakes expressed in %ID.cm-³ and core/contralateral ratio for each group (40–60 minute sum image)...... 76 Table 2.2 푩푷푵푫 values from the simplified reference tissue model in the core ROI with different reference tissues: dual scans...... 81 Table 3.1 Summary of 푲ퟏ (ml.cm-3.min-1), 푽푻 (ml.cm-3) and 푽풃 results from the one and two tissue compartment models for individual animals in core and contralateral regions and for the population derived kinetic...... 104 Table 3.2 Summary of regional kinetic modelling results across all LPS animals using blood (2TCM) and reference inputs ...... 106 Table 4.1 Summary of MRI acquisition parameters used for MS and HV subjects...... 121 Table 4.2 Summary of MS subject demographics...... 126 Table 4.3 Average 푽푻 values and standard errors (S.E.) from 1 and 2 tissue compartment modelling for HAB MS and HV subjects...... 129 Table 4.4 Average 푽푻 values and standard errors (S.E.) from 1 and 2 tissue compartment modelling for MAB MS and HV subjects...... 129 Table 4.5 Summary of individual lesion SUV_corr and 푽푻 results (ml.cm-3) for MS subjects from 1TCM, with corresponding lesion volumes (cm3)...... 136 Table 5.1 Summary of results from baseline and follow up clinical assessment for all subjects...... 160 Table 6.1 Summary of free fraction of tracer in plasma (풇풑) in healthy subjects for some common TSPO-PET tracers...... 172 Table 6.2 Cold spike concentrations for displacement of hot [18F]GE-180 in plasma samples...... 175 Table 6.3 Volumes of hot [18F]GE-180 spikes for each conditions (hot tracer only, hot tracer with excess cold GE-180 and hot tracer with excess cold PK11195)...... 175 Table 6.4 Individual subject results for hot tracer only and with excess cold GE-180 and PK11195...... 183 Table 6.5 Volumes and spike concentrations of cold and hot tracer used in blood samples...... 178 Table 6.6 Average percentage of unbound [18F]GE-180 from whole blood incubated with tracer and cold spikes before plasma extraction...... 179

List of figures Figure 1.1 Schematic summarising the process of microglial activation...... 20 Figure 1.2 Schematic summarising the partial volume effect...... 25 Figure 1.3 Metabolic decay pathways of [11C]-(R)-PK11195...... 33 Figure 1.4 The relationship between data, inputs, models and results...... 35 Figure 1.5 Examples of compartment models commonly used with arterial input functions to quantify dynamic PET data...... 37 Figure 1.6 Example of a spectral analysis results plot for [11C]-(R)-PK11195...... 41 Figure 1.7 The two compartment reference tissue model...... 46 Figure 1.8 Flow diagram describing how a clustering algorithm proceeds...... 54 Figure 2.1 Chemical structure of radiotracers used in the study; [11C]-(R)-PK11195, [18F]GE-180 and [18F]DPA-714...... 68 Figure 2.2 Average core and contralateral tissue time activity curves (TACs) of LPS- injected animals that underwent dual scans with each tracer...... 75 Figure 2.3 Average TACs for cerebellum (black) and contralateral (grey) reference regions for [11C]-(R)-PK11195, [18F]GE-180 and [18F]DPA-714 in LPS animals...... 78 Figure 2.4 Co-registered (sum 40-60 minutes) PET-CT images of representative animals...... 80 Figure 2.5 Immunohistochemistry slides performed on sections in LPS and AMPA- injected rodent brains...... 81 Figure 3.1 Photograph of the experimental setup for arterial blood sampling of rats...... 96 Figure 3.2 Experimental setup of dispersion correction apparatus...... 98 Figure 3.3 Example HPLC chromatograms showing the metabolism over time of [18F]DPA-714...... 101 Figure 3.4 Plasma-over-whole blood (POB) ratios and parent fraction of tracer for individual animals...... 102 Figure 3.5 Calibrated sum 40-60 minute [18F]DPA-714 PET images with co-registered CT images...... 103 Figure 3.6 Plasma input function fits for [18F]DPA-714 with the 1 and 2 tissue compartment models...... 105 Figure 3.7 Bland-Altman plots comparing all regional results from plasma/reference input in all animals...... 107 Figure 4.1 Examples of types of MS pathology observed in study cohort...... 126 Figure 4.2 Average plasma-over-blood ratios for HAB HV, MAB HV, HAB MS and MAB MS subjects...... 127

Figure 4.3 Correlation plot of time to peak concentration of [18F]GE-180 in whole blood vs. height of that peak for HV and MS subjects...... 128 Figure 4.4 Co-registered T1 pre-contrast MR with sum 60-90 minute PET images...... 128 Figure 4.5 Example fits for HV and MS subjects with 1TCM and 2TCM...... 130 Figure 4.6 Akaike information criterion and Bland-Altman plots comparing 1TCM and 2TCM in HV and MS subjects...... 131 Figure 4.7 푽푻, SUV and SUV corrected for regional blood activity contribution (c) in all standard regions...... 133 Figure 4.8 푽푻, corrected SUV and SUVR in lesions > 100 mm3 for individual MS subjects...... 137 Figure 5.1 Diagram summarising the timeline for patient visits in natalizumab study ...156 Figure 5.2 Average parent fraction in plasma for baseline and post-treatment scans. ....159 Figure 5.3 Sum 60-90 minute [18F]GE-180 PET images with co-registered T1 pre-contrast images (top) and (below) corresponding post-contrast T1 images...... 161 Figure 5.4 푽푻 estimates in standard regions at baseline (blue) and after 10 weeks of treatment with natalizumab (red)...... 162 Figure 5.5 푽푻 and SUV_corr in individual lesions for each patient at baseline (blue) and after 10 weeks of natalizumab treatment (red)...... 163 Figure 6.1 Allelic discrimination plot for genotyping of the TSPO single nucleotide polymorphism rs6971 in five volunteer subjects...... 177 Figure 7.1 TACs showing uptake in whole blood (venous sinus) and Gd-enhancing lesion...... 190 Figure 7.2 Representation of the binding of a TSPO ligand to the protein...... 199

Abstract Quantitative analysis of positron emission tomography (PET) with the second generation translocator protein (TSPO) ligand [18F]GE-180 A thesis submitted to the University of Manchester for the degree of Doctor of Philosophy (PhD) by Sujata Sridharan, August 2016 Background: The 18 kDa translocator protein (TSPO), expressed at a low level in the healthy human central nervous system (CNS), is upregulated in inflammatory brain diseases by activated microglia and other immune cells. Using positron emission tomography (PET) radioligands targeting TSPO, it is possible to localise this signal and map the course of microglial activation and its effects on disease progression. Here, a newly developed second generation TSPO PET ligand, [18F]GE-180, was evaluated in different models of preclinical and clinical neuroinflammatory disease. Methods: A preclinical model of low-level inflammation with lipopolysaccharide (LPS) was designed. Rats were scanned with the first generation TSPO ligand [11C]- (R)-PK11195 and either [18F]GE-180 or [18F]DPA-714, with dual scanning enabling the direct comparison of second generation tracers with [11C]-(R)-PK11195. An arterial blood sampling system for rodent imaging with [18F]DPA-714 was set up and characterised. The performance of [18F]GE-180 was assessed in a clinical study in nine relapsing-remitting multiple sclerosis patients (RRMS) and ten healthy volunteers (HV). A comparison of kinetic modelling approaches for [18F]GE-180 human brain PET data was performed, as well as a longitudinal analysis with intervention using the disease-modifying treatment, natalizumab to evaluate the potential of [18F]GE-180 as a biomarker for therapy monitoring in MS subjects. Finally, the plasma-protein binding behaviour of [18F]GE-180 was evaluated in vitro using ultrafiltration. Results: In LPS animals, [18F]GE-180 produced a significantly higher ipsi- to 11 contralateral uptake ratio and binding potential (퐵푃푁퐷) than [ C]-(R)-PK11195 (p = 0.03), but [18F]DPA-714 did not. There was no significant difference between animals scanned with [18F]GE-180 and [18F]DPA-714, suggesting no overall superiority of the former. Characterisation of an arterial sampling system for rodent studies with [18F]DPA-714 allowed correction for dispersion effects. A comparison of reference regions showed that a novel externally derived tissue estimated 퐵푃푁퐷 with lower bias than a contralateral reference region. In human [18F]GE-180 brain PET data, the unconstrained two-tissue compartment model (2TCM) best described tracer behaviour in RRMS and HV subjects. Normal appearing white matter (NAWM) in patients was elevated over that of HVs. Standardised uptake values (SUVs) for the tracer in rodents were 0.28±0.12 and 0.84±0.31 in healthy tissue and LPS lesions respectively, and in humans were 0.36±0.04 (HV) and 0.58 (in a gadolinium- enhancing MS lesion). [18F]GE-180 uptake was also significantly reduced in the brains of RRMS subjects treated with natalizumab, correlating with clinically-identified improvement. [18F]GE-180 has a free fraction of between 1 and 8%. Conclusions: [18F]GE-180 shows good brain uptake in the rodent brain and produces superior signal to [11C]-(R)-PK11195, but not to [18F]DPA-714. The 2TCM fits human [18F]GE-180 PET data well, and the tracer is able to identify an elevated signal in RRMS patients compared to healthy subjects. [18F]GE-180 shows a large fraction of non-displaceable binding in human blood, thus further optimisation of kinetic modelling approaches is suggested.

Declaration I declare that no portion of the work referred to in this thesis has been submitted in support of an application for another degree or qualification at this or any other university or other institute of learning.

Copyright statement i. The author of this thesis (including any appendices and/or schedules to this thesis) owns certain copyright or related rights in it (the “Copyright”) and s/he has given The University of Manchester certain rights to use such Copyright, including for administrative purposes. ii. Copies of this thesis, either in full or in extracts and whether in hard or electronic copy, may be made only in accordance with the Copyright, Designs and Patents Act 1988 (as amended) and regulations issued under it or, where appropriate, in accordance with licensing agreements which the University has from time to time. This page must form part of any such copies made. iii. The ownership of certain Copyright, patents, designs, trade marks and other intellectual property (the “Intellectual Property”) and any reproductions of copyright works in the thesis, for example graphs and tables (“Reproductions”), which may be described in this thesis, may not be owned by the author and may be owned by third parties. Such Intellectual Property and Reproductions cannot and must not be made available for use without the prior written permission of the owner(s) of the relevant Intellectual Property and/or Reproductions. iv. Further information on the conditions under which disclosure, publication and commercialisation of this thesis, the Copyright and any Intellectual Property University IP Policy (see http://documents.manchester.ac.uk/display.aspx?DocID=24420), in any relevant Thesis restriction declarations deposited in the University Library, The University Library’s regulations (see http://www.library.manchester.ac.uk/about/regulations/) and in The University’s policy on Presentation of Theses

Acknowledgements The funding for this PhD project came primarily from the Engineering and Physical Sciences Research Council (EPSRC) and partly from GE Healthcare, to both of whom I am very grateful.

During the course of this project, I have had the great pleasure of being able to work alongside a dedicated and enthusiastic team of researchers and scientists at the Wolfson Molecular Imaging Centre (WMIC). Of these, my foremost thanks must go to my supervisory team: Rainer, Hervé, Steve and Alex. Their scientific expertise and technical and professional support (as well as the odd spare chocolate bar from the trusty vending machine!) throughout my time at the WMIC have been invaluable and I am extremely grateful for their encouragement and patience. Thank you to Prof. Federico Turkheimer for provision of the SuperPK software, and to Michael Kassiou for precursor materials. Special thanks must also go to the team at GE Healthcare at the Grove Centre in Amersham, especially Hina, Christina, Bern, Jan, Adrian, Steve, Paul, Rabia, Jindy, the Marks and the rest of building 22. Chris Buckley and Will Trigg made sure that I not only gained a balanced perspective of research in industry and was able to work productively alongside their team, but also that I thoroughly enjoyed every visit to the south! Under their guidance, I was able to set up a new collaboration with Imperial College London; here I must also thank Joel and Richard for their specialised and crucial help in understanding the disease processes and pathology of MS, as well as their professional support and willingness to share their data with me.

Within the WMIC, I have been lucky enough to meet a great number of people who have consistently brightened the sometimes overly heavy cloud of PhD-ing with their friendship. Liv, Fiona and Amaia – I could not imagine having reached this stage without the snacks, the odd lunchtime crossword break and your irrepressible humour. Georgios and Sophie – thanks for always answering my sometimes trivial science questions and not laughing (too much) when I forgot how to ride a bicycle in Amsterdam. Phil and Joe – thanks for making my first few weeks at the WMIC so welcoming. Natale, Katy, Toby, David, Stephen, Jo, José, Silke, Evan, Iggy and all of cake club – thanks for being there every Wednesday with cake and a short break from thesis-writing! Without the help of the Mikes, Carrie, Lizzie, Carol, Gemma, Anton and the rest of the basement folk, not only I but almost all of the first floor in the WMIC would be unable to conduct PET studies, and thanks also to Sarah and Ally, as well as the radiography team, Ben and Eleanor, for going above and beyond in this department. Thank you to all the IT staff, especially Dave, for wise words and technical help! The second floor also cannot be

12 forgotten; without help from FX, Leonie, Lidan, Duncan and Aisling, as well as Ibrahim, I would not have learnt so much about preclinical imaging so quickly! Catherine and Jane at reception – I am so grateful for your patience and efficiency at dealing with my relentless Egencia bookings, as well as your always friendly words! PET is a truly multi- disciplinary field and without the commitment of all the people in the building, no single one of us would be able to produce the high standard of research which we do. Undoubtedly I have failed to mention some people – perhaps it is better to say a huge thank you to everyone in the building for making it such a tremendous academic centre at which to work; I am proud to have shared it with you since 2013.

It goes without saying that in getting to this stage, I have also relied heavily on the support of my oldest friends. To Lucy, Andrew, Alex, Catherine, Naomi, Livvy, Hannah, Sophia and Johnny, Ra’eesa, Ghazal, Sara, Marie, Kate, Annabel, Jade, Adam, Laurence and of course Matt – thank you for always being on my side, for your love, advice, support and endless patience with my flakiness and mood swings – know that it will forever be appreciated, as will your determination that I have some fun when I did take a break!

Last, but of course not at all least, I would like to dedicate this thesis to my parents and sister, Aarti, without whose love, support and motivation I doubt I would have got this far in life, let alone in the last three and a half years. Thank you for all your patience and encouragement even when I talked about nothing but work for days on end!

Abbreviations (1,2)TCM (One, two) tissue compartment model (A, P, H)D (Alzheimer’s, Parkinson’s, Huntington’s) disease (ATP)BC Adenosine triphosphate binding cassette (transporter) (D, PS)IR (Double, phase sensitive) inversion recovery (ID)IF (Image derived) input function (NA)WM (Normal appearing) white matter (RR, SP)MS (Relapsing remitting, secondary progressive) multiple sclerosis (RS-E)SA Rank-shaping exponential spectral analysis (S, M)RTM (Simplified, multilinear) reference tissue model (S, R)PM (Statistical, reference) parametric mapping (W)NLLS (Weighted) non–linear least squares 푐표푛푡푟표,푝표푝 퐵푃푁퐷 Non-displaceable binding potential with contralateral or population based reference 푉푇 Total volume of distribution 푓푝 Free fraction of tracer in plasma 9HPT 9-hole peg test AGC Automatic gamma counter AIC Akaike Information Criterion AMPA Alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid ANOVA Analysis of variance BBB Blood brain barrier BCRP Breast cancer resistance protein CD11b Cluster of differentiation molecule 11b CIS Clinically isolated syndrome CNS Central nervous system COV Coefficient of variation CSF Cerebrospinal fluid CT Computed tomography DTI Diffusion tensor imaging DVR Distribution volume ratio EAE Experimental autoimmune encephalitis EDSS Expanded disability status scale EDTA Ethylenediaminetetraacetic acid FBP Filtered back projection FLAIR Fluid attenuation inversion recovery FOV Field of view FSL Functional MRI brain software library FWHM Full width at half-maximum Gd Gadolinium contrast agent GFAP Glial fibrillary acidic protein GM Grey matter H/M/LABS High, mixed, low affinity binders HPLC High performance liquid chromatography HSA Human serum albumin HV Healthy volunteers ICC Intra-class correlation coefficient IgG Immunoglobulin G IL Interleukin IRF Impulse response function LMA Local means analysis LPS Lipopolysaccharide MCAO Middle cerebral artery occlusion

MCI Mild cognitive impairment MNI Montreal neurological institute MR(I) Magnetic resonance (imaging) MSFC Multiple sclerosis functional composite NeuN Neuronal nuclear antigen NI Neuroinflammation NNLS Non-negative least squares OSEM Ordered subset expectation-maximisation P(TF)E Poly(tetrafluoro)ethylene PBR Peripheral benzodiazepine receptor PBS Phosphate buffered saline PEI Polyethlyeneimine PET Positron emission tomography P-gp P-glycoprotein PMT Photomultiplier tube POB Plasma over (whole) blood PSL Photostimulated luminescence PV(E,C) Partial volume (effect, correction) ROI Region of interest S(E, D) Standard (error, deviation) SA Spectral analysis SDMT Symbol digits modality test SNP Single nucleotide polymorphism SNR Signal to noise ratio SPE Solid phase extraction SPGR Spoiled gradient echo SPMS Secondary progressive multiple sclerosis SUV(R, _corr) Standardised uptake value (ratio, corrected for vascular contribution) SVCA Supervised cluster analysis T25FW Time to walk 25 feet TAC Time activity curve TLC Thin layer chromatography TNF Tumour necrosis factor TRT Test-retest TSPO 18 kDa translocator protein

1 Introduction

1.1 Rationale for the alternative format

This thesis presents work related to quantification of the translocator protein (TSPO) positron emission tomography (PET) radiotracers [11C]-(R)-PK11195, [18F]GE- 180 and [18F]DPA-714 in different models of neuroinflammatory disease.

The work in this thesis is presented in the alternative format, which allows incorporation of sections that have or will be submitted to peer-reviewed journals for publication, but must otherwise conform to the same standards as a traditional format thesis. In this instance, an alternative format thesis was chosen due to the nature of the work, which separates in parallel into papers relating to a) different disease models, b) different quantitative analysis approaches and c) clinical and preclinical studies.

1.1.1 Aims Primary aims of the study were:

1. Quantification of second generation TSPO radiotracers for PET imaging in preclinical models of neuroinflammation. 2. Establishment and validation of the imaging methodology for quantitative analysis of [18F]GE-180 in clinical brain imaging studies.

1.1.2 Structure of the thesis The work in this thesis is presented as original journal articles, either submitted for publication or in appropriate format for submission. All chapters are written by the candidate. The rest of Chapter 1 provides a summary of relevant literature in the field of TSPO-PET imaging and outlines the context of the research aims. All papers were reviewed by the co-authors.

Chapter 2: ‘Comparative evaluation of three TSPO PET radiotracers in a LPS- induced model of mild neuroinflammation in rats’ is an original research article recently published in the journal Molecular Imaging and Biology (MIB). This paper details a study performed in a rodent brain model of inflammation induced by lipopolysaccharide (LPS).

Chapter 3: ‘Characterisation of a blood sampling system for generating [18F]DPA- 714 TSPO-PET plasma input functions in the rat brain’, is an original research article prepared for publication. This study was also performed in an LPS model and aimed to characterise an arterial blood sampling system for use of [18F]DPA-714 in rodents at the Wolfson Molecular Imaging Centre (WMIC). The work also comparedarterial input

16 function kinetic modelling with reference tissue methods, with the ultimate aim of avoiding the need for the invasive blood sampling process.

Chapter 4: ‘Initial evaluation of [18F]GE-180 as a TSPO-PET imaging biomarker in multiple sclerosis’ is an original research article prepared for publication. This paper contains an analysis on human [18F]GE-180 data from a cohort of relapsing-remitting multiple sclerosis (RRMS) patients and healthy control subjects.

Chapter 5: ‘Therapy response monitoring with [18F]GE-180 in a cohort of relapsing-remitting multiple sclerosis patients’ is an original research article prepared for publication. The work in this chapter is related to a longitudinal aspect of the same study as Chapter 4; detailing the potential of [18F]GE-180 PET to visualise changes in the MS brain as a response to natalizumab therapy.

Chapter 6: ‘Plasma-protein binding of the 18 kDa TSPO-PET tracer [18F]GE-180’ is a short investigational study describing an experimental quantification of the plasma- protein bound fraction of [18F]GE-180 in human blood.

Chapter 7: ‘Summary and conclusion’ summarises the key findings of the project and outlines important future directions which may be taken.

1.2 Role of the author

This study was designed with the support of General Electric (GE) Healthcare to evaluate the performance of their newly developed tracer, [18F]GE-180, which targets TSPO, upregulated in inflammatory disease.

The two preclinical studies presented in Chapters 2 and 3 were conceptualised by Dr. Hervé Boutin and I. Dr. Rainer Hinz and I provided methodological amendments to the study design. I attended all animal scans, assisting with collection and reconstruction of the CT and PET images. I designed, optimised and performed all image analysis, including the development of novel analysis methods, with support from Dr. Hinz. I cryosectioned frozen rat brains, performed all ex vivo immunohistochemical staining and photographed the resulting slides. I also performed autoradiography on these sections with the support of Dr. Boutin. I also assisted with setup and performance of dispersion correction experiments described in Chapter 3.

Clinical studies with [18F]GE-180 were in developmental stages when I commenced my PhD project. At the Wolfson Molecular Imaging Centre (WMIC), these included a dual tracer study in ischaemic stroke patients with both [18F]GE-180 and [11C]- (R)-PK11195, and a study in patients with the movement disorder multiple system

17 atrophy (MSA). Unfortunately, due to a combination of technical and administrative impediments, neither of these studies was able to recruit or start scanning subjects until very recently (November 2015). Once this became apparent, I worked to source clinical data from collaborators at GE Healthcare and, more recently but to a deeper extent, I began to work closely with researchers at Imperial College London. The study described in Chapters 4 and 5 was designed by Dr. Richard Nicholas and Dr. David Brooks at Imperial, with input from Dr. William Trigg (GE Healthcare). All recruitment, scanning and reconstruction took place within the Imperial College NHS Healthcare Trust. After receiving raw image data from my collaborators, I designed a methodological approach to quantification of the scans, and subsequently processed and analysed all the data after close discussion with Dr. Joel Raffel (clinical research fellow) and Dr. Nicholas regarding the pathology of multiple sclerosis. Dr. Raffel kindly provided MR lesion maps and demographic data for all analysed patients.

I contributed intellectually to the design of the work presented in Chapter 6, with support from Dr.s Boutin, Hinz and Trigg, Ms. Elizabeth Barnett and Ms. Carrie-Anne Mellor (WMIC). I carried out all blood preparation and DNA extraction and assisted with genotyping (Christopher Clark at the Paterson Institute for Cancer Research at the University of Manchester). Ms. Barnett and Ms. Mellor, as qualified radiation safety workers, performed plasma-protein binding work.

Cumulatively, my role in the work described in this thesis has been to contribute to the intellectual design of the studies and to the acquisition and interpretation of data. I have performed all data analysis and written each of the following chapters as first author. Over the duration of my PhD project, I have also attended several national and international conferences (see Appendix 2), for which I have written and presented posters and oral presentations. In those areas to which I was not able to contribute myself (for example, animal/radiotracer handling), I would like to thank my colleagues for their invaluable support.

1.3 Inflammation and microglia

A key role of glial cells in the central nervous system (CNS) is to monitor the health of the brain. One of the first signs of distress in the brain is inflammation. An acute inflammatory response occurs when the brain undergoes some insult. The primary aim of this response is to contain the damage, and then to promote healing in the affected area. It is widely observed, however, that acute inflammation is detrimental in the CNS and that inhibition of the response may be beneficial [1-2]. This can also be extended to chronic inflammation, which can be maintained, or in some cases exacerbated and does not resolve, thereby preventing tissue repair and regeneration.

The CNS itself comprises many different cells; neurons, astrocytes, oligodendrocytes and microglia, of which the latter are largely, though not solely, responsible for driving the inflammatory processes in the brain. Microglia are non- neural cells belonging to a family of mononuclear phagocytes, including macrophages, which monitor the health of the CNS. Their true origin in the adult brain is still somewhat contentious [3-4], but it has been shown that the major contribution to the adult microglial population is derived from primitive haematopoiesis in the foetal yolk sac, taking up ‘residence in the brain during early foetal development’ [5-7]. This is in contrast to tissue macrophages and mononuclear phagocytes, which are generally accepted to be of definitive bone-marrow lineage (for review, see [8-9]). Microglia constitute approximately 10% of all healthy adult brain cells [10-11]. In the healthy brain, they are said to be in a so-called ‘quiescent’, or resting, state. In the last decade, however, this definition has been adapted to incorporate the fact that microglia are actually constantly ramified, scanning their local environment against potential brain insult/invasion with a series of thin, branching processes and exhibiting continuous anti-inflammatory features [12-14], see Figure 1.1.

Microglia become activated when the brain is subject to injury. When activated, they begin to change their morphology, becoming more macrophage-like (amoeboid) and releasing pro-inflammatory cytokines such as interleukins (IL), in particular IL-1α and IL-1β. Microglia play a dual role by potentially producing both pro- and anti-inflammatory cytokines, with different microglial phenotypes working to either aggravate or aid repair after insult to the CNS [15]. The exact nature of the relationship between phenotypes and neuro-protective or destructive effects is extremely complex and outside the scope of this work. Figure 1.1 summarises the function of these phenotypes according to one currently suggested theory. In short, in

19 the figure, the blue activated microglia are of M1-phenotype, while brown are M2. Supposedly, the presence of M1 microglia is associated with the production of pro- inflammatory cytokines such as IL-1 and has been linked with detrimental effects, while M2 have associated with the production of anti-inflammatory cytokines such as IL-4 and IL-10 and is commonly found at sites of tissue repair (see [15-16] for a more detailed review). It should be emphasised, however, that M1 and M2 may not be two discrete states; rather, as recently suggested [17], classification of microglia in this way has primarily been driven by conjecture rather than true scientific evidence. Furthermore, the author points out that microglia do not also necessarily exist in a continuum, but can co-express markers representative of the two polarisation states [18]. Clearly, understanding this complex interplay is key to determining how to, or indeed whether we should at all, manipulate ‘inflammation’ in the broad sense and in the context of determining disease therapies.

Figure 1.1 Schematic summarising the process of microglial activation. The morphology of microglia changes from their primed resting statetending towards a macrophage-like appearance. Many researchers consider that there are two different polarised phenotypes of microglia, M1 and M2, however, there is no clear scientific evidence supporting this theory.

Microglia are not the only CNS defence cells involved in the inflammatory response; astrocytes are also known to be present at the site of inflammation (for review, see [19]). The work in this thesis is related to the detection of inflammation in the brain predominantly involving microglial activation. The extent and location of this inflammation may be indicative of the disease state and pathology of a patient; intuitively, (activated) microglia can be used as an early indicator of changes in brain pathology [20]. Specifically, most biological processes can be associated with a

20 particular biomarker. For example, several neuroinflammatory conditions are known to be hallmarked by expression of the 18 kDa translocator protein (TSPO) [21-22].

1.4 The 18 kDa translocator protein

The peripheral benzodiazepine receptor (PBR), or 18 kDa translocator protein (TSPO) as it is now more commonly known [23], is a protein with five transmembrane domains [24-25]. TSPO belongs to the family of tryptophan-rich sensory proteins and in humans is encoded by the TSPO gene [26-27]. The protein is primarily located in the membranes of mitochondria [23], but small amounts can be found in other non-mitochondrial subcellular locations [28-30], including in red blood cells [31] and in glial cells within the CNS [32]. The previous understanding of the TSPO binding complex included the concept of it containing three channels: the 18 kDa (molecular mass) for isoquinolines; the 32 kDa voltage-dependent anion channel (VDAC) for benzodiazepines; and the 30 kDa adenine nucleotide carrier, also for benzodiazepines [33]. Perhaps one of the most significant recent findings in this arena was that of Korkhov et al., who describe, in a bacterial homologue of TSPO, a pair of monomers, tightly bound in a symmetrical dimer and each with 5 trans-membrane α- helices [25, 34].

Much recent work has been undertaken to characterise the structure of isolated mammalian TSPO ([35], see [36-37] for review), as well as the complex which the protein forms with TSPO ligands. Nuclear magnetic resonance spectroscopy has been used to probe the structure of TSPO [35]; but the authors experienced difficulties in isolating the signal for the protein, suggesting a dynamic nature in its free state [38]. When (R)-PK11195 (see section 1.5.1) was added to free mammalian (mouse) TSPO, however, its stabilising effect allowed identification of five transmembrane helices (termed TM1 to TM5 by the authors [35]), with a ‘binding pocket’ formed by these helices. In addition, the presence of a stabilising loop formed between helices TM1 and TM2 is an important consideration with respect to the fact that PK11195 is not susceptible to the rs6971 genetic polymorphism in TSPO which affects other TSPO radioligands [39] (see section 1.5.3). A further finding involved the fact that TSPO is known to play a key role in cholesterol transport between cytoplasm and mitochondria; the authors describe the binding site for cholesterol as existing on the outside of TSPO, suggesting that PK11195 binding should not interfere with cholesterol binding. Simultaneously, however, the location of this site on TSPO and the dimerising ability of cholesterol indicate that the protein may oligomerise in the

21 presence of cholesterol, a feature which itself may affect both the binding of TSPO ligands and cholesterol synthesis [40] (see section 7.4).

Clearly questions remain about the effects of polymerisation on TSPO, as well as the exact location, crystallography and binding sites of the protein, which need to be addressed. Nevertheless and importantly, the stabilising effect of PK11195 on the 3D structure of TSPO implies that the ligand could have potential use in increasing cholesterol transport [41-42]. Other functions of TSPO include haem synthesis and cell proliferation ([23, 43], see [44] for review), although the former has recently been shown to be unaffected by reductions in TSPO, suggesting that the protein is not critical for the process [45].

There are several endogenous ligands of TSPO, including the diazepam binding inhibitor [46], which is known to interact with TSPO to stimulate steroidogenesis [47] and movement of protoporphyrin [48-49], but which also functions as an effective anxiolytic. The full function of the TSPO-ligand complex is still being studied; due to their involvement in cholesterol transport, TSPO ligands may also be able to modulate neurosteroidogenesis [50-51] and apoptosis via activation of the VDAC channel (see [52-53] for review). It has also been recently observed that TSPO, rather than being a purely immunological marker, is also involved in oxidative stress regulation and cell mitophagy [54-55]. Details of all these functions are outside the scope of this thesis, but it should be noted that our understanding of TSPO and its protein-protein, as well as ligand-protein, interactions is by no means exhaustive.

TSPO is also expressed by activated microglia and is therefore upregulated when inflammation is present in the brain. It is not currently clear in what way TSPO itself is affected during the upregulation process; there are changes in the binding site density (as well as its polymerisation) which may alter its interaction with ligands compared to resting microglia [25]. These effects are thought to be dependent on the degree of microglial activation and the disease state. In general, there is low expression of TSPO in the non-inflamed CNS, but assuming that its background presence is negligible may not be accurate. Of course, the very fact that the protein shows binding sites for protoporphyrin and cholesterol [56-57] implies that TSPO is present even in the physiologically normal brain; [58] also gives a short review of evidence for TSPO expression in resting state microglia. Nevertheless, the copious (ex vivo) data confirming TSPO expression by activated microglia in neurological disease has naturally bred interest in targeting the receptor for imaging. This has

22 consequently led to the development of several positron emission tomography (PET) radioligands for TSPO.

1.5 PET imaging

PET is an extremely valuable tool in medicine and medical research. While other imaging modalities such as magnetic resonance (MR), computed tomography (CT), ultrasound and optical imaging offer detailed structural information, PET allows visualisation of functionality on an organ level.

PET has been used routinely in the clinic for diagnosis, staging and monitoring of disease for over four decades. A radiolabelled tracer, or molecular probe, is administered to the subject at the start of the PET scan. As the radioisotope label starts to undergo 훽+ decay, it emits positrons, as shown in equation 1.1:

퐴 퐴 + 푍푋 → 푍−1푋 + 푒 + 휗푒 Eq. 1.1 where 푋 is the isotope undergoing decay, 퐴 and 푍 are respectively the atomic mass + and atomic number of 푋, 푒 is a positron and 휗푒 is an electron neutrino. Positrons travel a short distance through the body, losing kinetic energy, before interacting with a local electron to form positronium. This highly unstable configuration soon results in annihilation of the positron-electron pair, forming two 511 keV photons, which are emitted at approximately 180° to one another; accounting for any residual momentum of the positron immediately prior to annihilation. Below, a description of the behaviour of these annihilation photons in the PET camera is given.

A transmission scan using an external photon source is often performed before administration of the tracer; in this way, correction for attenuation and scatter in the scanner can be performed. An emission scan may last from a few minutes up to several hours depending on the kinetics of the tracer and the half-life of the radioisotope label. Table 1.1 shows the half-lives of the some commonly used radioisotopes in PET.

Table 1.1 Common radioisotopes used in PET imaging. Radioisotope Example tracer Half-life (minutes)

11C [11C]-(R)-PK11195 20.38

18F [18F]-FDG 109.77

15 15 O [ O]-H2O 2.03

The most commonly used radioisotopes in PET brain imaging are 11C, 18F and 15O. Advantageously, 11C-labelling can mimic the exact endogenous function of 12C- containing ligands. Short half-lives of the radioisotopes do mean that use of these radioligands is heavily time-dependent, but the same feature is also an advantage since the radiation dose to the patient is minimised. The pharmacokinetics of the ligand also need to be considered; accounting for the clearance of the tracer from blood and tissue, as well as its metabolism and the association and dissociation rates of the tracer in relation to the target receptor.

The photons produced by positron annihilation travel to the edges of the PET camera, where dedicated detectors are able to detect them. Every clinical PET detector varies structurally according to requirements (for example, whole body scanners are constructed differently to dedicated brain scanners), but all types include coincidence detectors, either in a ring around the inner circumference, or in the form of a series of ‘heads’. Most simply, the heads comprise scintillator material, photomultiplier tubes (PMTs) and underlying analogue electronics. A ‘true’ coincidence event is defined when two photons interact with a detector almost simultaneously (or within a short time-window), accounting for such corrections as Compton scatter, dead time and attenuation.Each coincident even is then assigned a line of response (LOR) between these detectors, allowing mapping of the activity and physical location of the annihilation associated with that event. PET dynamic images can then be reconstructed from the data using either filtered back projection (FBP) or iterative methods such as ordered subset expectation-maximisation (OSEM) algorithms [59-61]. PET images require processing and correction for a number of effects, including those mentioned above, but in addition to these, the partial volume effect (PVE) may need to be taken into account.

PVE is a term given to the phenomena of apparently misplaced activity in a PET image. It can introduce discrepancies in raw PET data of the same order as those introduced by attenuation [62]. The effect is most pronounced when the region of interest is of approximately the same size (or smaller) than the resolution of the image. The term PVE broadly encompasses two different phenomena: firstly, ‘3D blurring’ of the reconstructed PET image, which is a combined result of the finite spatial resolution of the scanner [63] and the reconstruction process. Ideally, the concentration of tracer in a region of the subject should be proportional to the concentration seen in the same region on the image; the PVE occurs when the activity in an object that does not full occupy the sensitive volume of the detectors is greater than that in the surrounding areas. The result is that activity within a given area ‘spills

24 out’ into regions which are not being sampled (see Figure 1.2); in the reconstructed image, the tracer concentration is underestimated in that region. The reverse can also happen – i.e. the activity from neighbouring regions can ‘spill in’ to the region of interest. Unfortunately, the two effects do not always cancel one another out.

The second cause of the PVE is image sampling; that is, since the distribution of a tracer is mapped on a voxel grid in PET imaging, and the voxels are not a true representation of the boundaries of tracer distribution, some voxels may contain more than one type of tissue [62]. The result is that the intensity measured from a specific voxel is actually an average of all the tissues within that voxel. This is known as the ‘tissue fraction effect’. While there are methods employed to correct for the PVE [62- 63], they are scanner, tracer and disease-specific [64-65] and therefore there is little consensus over which approach to use. Nevertheless, it is necessary to be aware of the limitations of PET data due to the PVE and to interpret images with this in mind.

Figure 1.2 Schematic summarising the partial volume effect. The diagram is a two dimensional simplification representing the PVE, which is due to the limited resolution of the PET detector. If the anatomical region being imaged is less than twice the full width half maximum of the detector in the x, y or z plane (i.e. smaller than the sensitive volume of the detector), the measured activity in that region is underestimated (green curve). Meanwhile, in surrounding adjacent voxels, the activity is apparently increased (‘spillover’, red dotted area). PVE can be corrected for using a number of methods, including deconvolution, which acts as a type of de-blurring to keep activity in the relevant areas.

The primary advantage and success of PET is its sensitivity, which, theoretically, allows clinicians and researchers to detect functional changes within organs and requires only low tracer doses of nano- to pico-moles. Unfortunately, the low temporal resolution of PET, combined with the limits on spatial resolution

25 imposed by the PVE and the scanner, mean that for anatomical localisation of group differences in uptake, information from structural scans such as MR or CT is required.

1.5.1 [11C]-(R)-PK11195 PK11195 is an isoquinoline ligand which binds specifically to TSPO and thus, in the inflamed CNS, primarily to activated microglia (although also to activated astrocytes [66-67]); an increase in expression of the ligand is taken as an indication of inflammation [20]. Synthesis of the PET radioligand, [11C]-(R)-PK11195, usually involves attaching a [11C]-methyliodide compound to a precursor which is a close analogue of PK11195 (usually PK11195 without a methyl group). This solution, by default, also includes a number of impurities (chemical and/or radiochemical), a small amount of precursor material and cold (12C) PK11195. A racemic mixture of [11C]- PK11195 was used in PET studies until the 1990s, when it was discovered that the (R)-enantiomer had a significantly higher affinity and showed higher uptake (between two and five-fold) and retention in areas of TSPO expression. Non-specific binding does not appear to depend on the enantiomer [68].

1.5.2 Specific and non-specific binding In addition to an increase in expression and as mentioned, microglial activation also corresponds to a change in the structure of TSPO and an increase in binding site density, i.e. the number of binding sites, 퐵푚푎푥, as opposed to an increase in affinity of [11C]-(R)-PK11195 for them [69], among other effects. The intensity and complexity of these effects is heavily dependent on the extent and duration of microglial activation, which in turn is dependent on the disease state [25] and thus requires an in-depth understanding in order to truly quantify PET data. Section 7.4 deals with these features of TSPO in more detail.

In binding studies, parameters of interest include the total ligand concentration (i.e. the amount of radiolabelled ligand in the injected dose) and the total binding (i.e. the amount of ligand which binds to the tissue of interest). Additionally, because it is non-saturable and non-displaceable, nonspecific binding (i.e. binding to non-target receptors, e.g. cell membranes or other proteins) can be measured as the amount of binding which is left after displacement by a competing, pharmacologically similar compound. Conversely, specific binding is saturable and displaceable, but is only measurable by deducting the non-specifically bound tracer from the total binding. This concept applies in most reference models in the quantification of PET image data; non-specific binding in a reference region is subtracted from the total binding in the target region to give a measure of the specific

26 binding in the latter (see section 1.8.5.1), with the assumption that non-specific binding is the same in the target and reference regions. The Michaelis-Menten principle for the kinetic between an enzyme and a substrate and can also be used to 푅 .퐿 describe receptor-ligand kinetics. The equation [푅퐿] = 푡표푡푎푙 , where [푅퐿] is the 퐾푑 + 퐿 receptor-ligand complex, 푅푡표푡푎푙 is the total number of receptors available and 퐾푑 is the dissociation constant for the receptor-ligand complex, describes how total binding is dependent on the concentration of ligand, i.e. the higher the concentration, the more binds to the receptor, up to 퐵푚푎푥 (see Appendix 1), which is fixed for a tissue at a given time. At this point, increasing the concentration of the tracer will result in non- specific binding increasing linearly, while the saturable binding will remain at the (constant) maximum (i.e. the signal to noise ratio, SNR, will dramatically decrease). [11C]-(R)-PK11195 shows relatively high non-specific binding and consequently a poor SNR, both of which are undesirable characteristics. Conversely, some newer ‘second generation’ TSPO tracers, such as [11C]-PBR28 and [18F]DPA-714, show a higher level of specific binding that renders them potentially more effective than [11C]- (R)-PK11195 at detecting inflammation.

1.5.3 The TSPO polymorphism In addition to the levels of specific and non-specific binding of a radioligand, their affinity for the target receptor must also be considered. A recently developed second generation tracer [11C]-PBR28, N-acetyl-N-(2-[11C]methoxybenzyl)-2-phenoxy- 5-pyridinamine [70] showed significantly increased specific uptake in monkey brains compared to [11C]-(R)-PK11195 (90% vs. 50%, without peripheral clearance corrections) [71-72]. More recently, however, human in vivo studies have observed that the affinity of [11C]-PBR28 varies between subjects [73-75]. The phenomenon had not been observed in any subjects during previous [11C]-(R)-PK11195 studies. Kreisl et al. [73] performed whole-body PET imaging of 10 healthy volunteers (5 binders and 5 non-binders) with [11C]-PBR28 and [11C]-(R)-PK11195 to try to establish why [11C]- PBR28 exhibits non-binding where [11C]-(R)-PK11195 does not. The results showed that uptake of [11C]-PBR28 in five organs [76] with the highest TSPO density (lungs, brain, heart, kidney and spleen) was a marked 50 – 75% lower in non-binders than in binders. In contrast, [11C]-(R)-PK11195 showed a significant difference in uptake between binders and non-binders only in the heart (p = 0.002) and lungs (p = 0.009). Indeed, it became apparent that TSPO in non-binders showed a ‘10-fold lower affinity for PBR28 than…in binders’ [73]. In vitro, the difference between the two groups is ~50-fold [75]. This effect was later discovered to be attributable to the rs6971 single

27 nucleotide polymorphism (SNP) in the TSPO gene (Ala147Thr), wherein human subjects exhibited two high affinity binding sites for TSPO (high affinity binders, HABs), one high and one low affinity binding site (mixed affinity binders, MABs), or two low affinity binding sites (low affinity binders, LABs, or non-binders) [77]. For [11C]-(R)-PK11195, however, (in vitro) the different structure of the TSPO polymorph has recently been shown not to alter the binding of this tracer [39].

It has also been noted that the existence of HAB, LAB and MAB sites indicates that the lack of [11C]-PBR28 binding in some subjects should not be interpreted as (solely) being due to a reduction in TSPO density. This relatively recently observed phenomenon must now be investigated in all new tracers, including [18F]GE-180. The TSPO polymorphism has not been described or observed in animal models.

1.6 TSPO expression in neurological disease

1.6.1 Multiple sclerosis Microglia are activated in cases of acute neuroinflammation, but TSPO is also upregulated in many chronic conditions such as Parkinson’s disease (PD) [78-79] and diseases involving phases of both acute and chronic inflammation, such as multiple sclerosis (MS) [20]. MS is an inflammatory disease, primarily characterised by demyelination of axons and white matter pathology, with lesions developing throughout the course of the disease. The clinical classification of MS can be broadly sub-categorised into two forms: relapsing-remitting (RRMS), where a patient may suffer an acute attack followed by a period of remission; or progressive, where a patient’s clinical symptoms gradually worsen, with little to no recovery period [80]. The aetiology of MS is not known; though it is generally thought to stem from a combination of genetic and environmental factors [81]. Approximately 80% of patients eventually diagnosed with the condition initially present with an acute episode of demyelination in at least one location; this is known as clinically isolated syndrome (CIS). The chance of this developing into RRMS or progressive forms of the disease in ensuing episodes increases from 50% over 2 years, to 82% in 20 years [82].

The exact role of inflammation and activated microglia in MS is not fully understood; it is known that inflammatory episodes are present during the early stages of the disease, with CD4+ T-lymphocytes driven by expression of IL-23 causing disruption to the blood brain barrier (BBB); thereby enabling destructive cell types to damage CNS neurons and ultimately leading to demyelination and development of neuronal plaques [81, 83-84]. In later stages and/or during phases of remission,

28 however, inflammation may not be as prevalent, and likely involves different cell populations to those at the onset of disease [84]. Primary progressive MS patients, for example, do not appear to show any extensive intrusion by cells associated with inflammation (for review, see [85]). Additionally, while anti-inflammatory treatment does reduce central inflammation and the frequency/intensity of subsequent relapses, it fails to fully halt demyelination and brain atrophy [86-87]. Pathology also shows that axonal damage is present in otherwise normal appearing white matter (NAWM) and inactive plaques, a phenomenon apparently separate to that seen in active lesions [88-90]. This has led to MS being proposed as a primarily neurodegenerative disease, rather than a purely autoimmune one as previously thought [91]. Notably, there has been no definitive study ruling out the presence of a small, diffuse inflammatory component in NAWM, nor in inactive lesions, which could itself be the underlying force behind neurodegeneration [90, 92-93]. Frischer et al. [94] performed a detailed ex vivo investigation, using autopsy samples from RRMS and progressive patients and healthy controls, into the relationship between inflammation and neurodegeneration. In their cohort, neurodegeneration was consistently found only in patients with ‘pronounced inflammation in the brain’. Furthermore, inflammation appeared to decline in the late, or progressive, stages of MS, as did neurodegeneration, to the levels seen in age-matched healthy controls. In the progressive phase, it is likely that inflammation becomes ‘compartmentalised’ locally to the CNS ([95-96], see [97] for review), with peripheral levels subsiding to those seen in healthy controls. It is therefore possible that current anti-inflammatory therapeutics for MS are not best tailored to target this specific compartment in progressive MS. It should also be noted that activated microglia release neurotoxins that can in many cases promote neurodegeneration [98]. Clearly, understanding the contribution and effects of inflammation and its link with neurodegeneration are essential to deciding on appropriate therapeutic strategies for patient care.

MRI is commonly used in the diagnosis and staging of MS, with white matter lesions visible as hyper-intensities on T2 FLAIR sequences, or as hypo-intensities in T1-weighted images. Contrast-enhanced MRI with gadolinium-based (Gd) agents can provide additional information relating to the integrity of the BBB at the site of a lesion; new pathology characteristically appears concurrently with BBB leakage, thus post-contrast images can be used to detect new ‘active’ lesions. Clinical staging of MS, based on various physical measures, including impairment to movement and the visual system, is traditionally performed using the expanded disability status scale (EDSS) or the multiple sclerosis functional composite (MSFC) [99-100]. In early

29 stages, the primary MR outcome measure of disease, the total volume of lesions visible in T2 MRI (lesion load), is known to be well correlated with these scales; with an increase being linked to the worsening of physical symptoms and development of the disease from CIS to MS [101-102]. At later stages, however, previous studies have shown a poor correlation between these clinical measures of disease, disease progression, and the lesion load as measured by MRI [103-105].

The discrepancy between MRI measures and clinical status based on physical symptoms indicates a gap in our understanding of the specific pathophysiology of MS; in this case, the sensitivity and functional nature of PET may offer the opportunity to glean further information. There have been several PET studies using [11C]-(R)- PK11195 in scanning of patients with MS [106-108]. Such studies have noted, via ex vivo staining, that it is primarily activated microglia and not reactive astrocytes, which are the cause of increased [11C]-(R)-PK11195 binding [20]. The findings of this latter study indicated that TSPO expression increases along white matter tracts, which do not necessarily also show increased numbers of T-cells. The authors suggest that this may be evidence that local activated microglia, still releasing cytokines but a stage before developing macrophage-like properties, play a key role in early demyelination [109-110]. Activated microglia may then be assumed to migrate preferentially along white matter tracts ‘that are ready for demyelination’; that is, areas that correspond to ‘subtle or imminent brain damage’ [20]. Thus TSPO expression can be observed in areas remote to the site of primary neuroinflammation. For example, a cross-sectional study in 10 patients with secondary progressive multiple sclerosis (SPMS) and 8 age- matched healthy controls recently found increased [11C]-(R)-PK11195 binding in NAWM, but also in the perilesional area of 57% of chronic T1 lesions [108].

Second generation TSPO tracers have also been employed in studies of MS patients. A study using [11C]PBR-28 in 11 subjects with MS and 7 healthy volunteers found focally increased binding in areas of gadolinium-enhancement in T1 MRI [111]. In a subset of patients (n = 6), 4-month follow-up scans revealed that increased tracer binding in the first scan was related to the appearance of new contrast-enhancing lesions, suggesting that inflammation may play a role in the development of MS- related pathology. A more recent test-retest (TRT) study in RRMS patients and healthy controls also found that [11C]PBR-28 had good reproducibility of the outcome measure

(total volume of distribution, 푉푇, see section 1.8.2 and Appendix 1) with a TRT value of between 7 and 9%. Other tracers have also been successful at identifying increased TSPO signal in lesional and perilesional areas, as well as NAWM [106, 112].

While TSPO-PET is thus seemingly able to identify microglial activation in MS, the true link between neuroinflammation and disease severity and progression is still unknown. A therapy response study with a sensitive second generation tracer (with superior SNR to [11C]-(R)-PK11195) in a cohort of active MS patients might offer further illumination in this area (see Chapter 5). Additionally, the relatively well- described neuroinflammatory component of MS offers a robust model for characterisation of a new TSPO tracer.

As mentioned earlier, TSPO is expressed by microglia and astrocytes, but the relationship between which cells are activated during which stage of disease is also currently unclear. Additionally, differentiation of pro- and anti- inflammatory phenotypes of microglia may help to clarify when inflammation is destructive during a disease process, and where it may in fact help to resolve damage. There is therefore considerable interest in developing a PET tracer which can specifically differentiate these cell types.

1.6.2 Preclinical studies Using animal models poses several opportunities and advantages when investigating the behaviour of new tracers. Importantly, this includes the possibility of tailoring preclinical models of neuroinflammation to mimic clinical diseases. Ex vivo work is also more straightforward in animals, allowing researchers to establish co- localisation of cell types with a PET signal using immunohistochemistry and quantifying the specificity of the binding of tracers using autoradiography. In order to directly compare the performance of two tracers, it is necessary to scan the same subject with each in close succession. In humans, this is especially challenging in studies of disease where the clinical symptoms of a patient and radiation dose restraints limit their tolerance for the PET scan. Animals, however, can be scanned multiple times on the same day (with short half-life tracers).

Animal models of neuroinflammatory disease are often used in the initial characterisation of new TSPO tracers. The middle cerebral artery occlusion (MCAO) ischaemic stroke model [113-116], as well as transgenic mouse models of AD (see [117] for review) and experimental autoimmune encephalomyelitis (EAE), which is a preclinical model of MS [118-119], have relatively well-characterised inflammatory components, allowing for assessment of tracer performance in disease states with varying levels of inflammation present. There is evidence that direct induction of inflammation in the CNS, for example using the bacterial endotoxin lipopolysaccharide (LPS), may lead to chronic neuropathology; notably this allows the development of

31 models of Parkinson’s disease (PD, for review, see [120]) and AD [121]. LPS models typically use high doses of the endotoxin [122-124], resulting in neuronal damage and high levels of microglial and astrocytic activation. While such models are relevant to clinical disease involving dramatic and acute inflammation, the smaller changes that are observed in the early stages of AD and in MS are typically more subtle and self- resolving [125].

In models involving mild levels of inflammation, there arises a challenge in the quantification approach. Evidently, it is important to quantify preclinical data as accurately as possible, in order to obtain a representative understanding of the performance of a tracer. Unfortunately, kinetic analysis in preclinical PET studies has also historically been challenging due to the terminal process of arterial sampling in rodents. Unilateral models of focal neuroinflammation, such as ischaemic stroke and LPS injection, have typically used a region within the contralateral hemisphere as reference input [115, 122]. As with human data, however, reference region models (see section 1.8.5) are contentious since there are no truly TSPO-free anatomical structures in the brain. Quantification is further complicated in cases where the level of inflammation is low, such as in models of AD. The emergence of second generation tracers has led to an increase in the number of preclinical studies to fully characterise their behaviour. Few of these models, however, have elected to compare the performance of different second generation TSPO tracers with one another (rather than with [11C]-(R)-PK11195). It would thus be of great value to develop a model of subtle inflammation in which to investigate quantification approaches further.

1.7 Plasma and metabolites of [11C]-(R)-PK11195

Plasma input functions are used in kinetic analysis of PET data as a measure of the radioactivity in arterial blood. They are most commonly derived from the validated ‘gold-standard’ of direct arterial blood sampling during performance of the PET scan [126-127]. Blood samples are also periodically collected to calibrate the online blood curves and to correct for metabolites; the radioligand starts to metabolise in blood as soon as the patient is injected and this transformation into radioactive metabolites must be taken into account for a fully quantitative plasma input curve. Blood detectors may be on or offline; the former holds the advantage of reducing the loss of radioactivity from the samples through decay, but they also have low sensitivity due to the high flow rate. Offline systems comprising an automatic gamma counter (AGC) of sodium iodide (NaI) detectors, tend to exhibit a lower

32 resolution than their online counterparts, but conversely, they are not subject to the temporal and volume restrictions of the latter.

The radioactivity present in arterial blood is a combination of parent compound and radiolabelled metabolites. Since only the radioactivity of [11C]-(R)- PK11195 is relevant for kinetic modelling, the radioactivity of these metabolites must be distinguished from that of the parent compound; more specifically, it is necessary to determine whether these metabolites form in plasma (and the rate at which they do this) and whether, if at all, they can cross the BBB to contribute to the measured PET signal from the TSPO binding sites. The speed of this determination and quantification of metabolites, which can also vary between patients, is critical, as measurement and processing of several blood samples must be completed before a plasma input function for the radioligand can be generated [128]. Possible metabolites of the parent compound [11C]-(R)-PK11195 are shown in Figure 1.3.

Figure 1.3 Metabolic decay pathways of [11C]-(R)-PK11195. Schematic showing the two major possible pathways of metabolic decay for [11C]-(R)- PK11195; demethylation and amide hydrolysis, producing 11C formaldehyde and 11C-N-Methyl- sec-butylamine respectively [129-130], of which the former is non-polar (can penetrate the BBB). The other enantiomers of [11C]-PK11195 are not thought to exhibit difference metabolic profiles, although the ratio of radioactivity in whole blood to that in plasma is lower in rats for the (S)-enantiomer for the first 15 minutes after injection [68], suggesting that more tracer partitions into the red cells and therefore that it may be more likely to cross the BBB by passive diffusion.

The two commonly used methods of metabolite separation are high performance liquid chromatography (HPLC) and thin layer chromatography (TLC). The former is often used in combination with solid phase extraction (SPE) to optimise the separation process. There are several variations on these techniques [128, 130- 132], but broadly, SPE-HPLC involves separation of the parent tracer and its non-polar metabolites according to their retention times on an analytical HPLC column. TLC, conversely, involves application of plasma supernatant samples to gel plates, which can be phosphor imaged to identify parent and metabolite fractions via their retardation factor (푅푓), which is a measure of the distance travelled by the substances [128, 130]. While TLC is less expensive than SPE-HPLC, it is limited by the number of plasma samples which can be processed concurrently and separation between the metabolites is less clear. Results from the two methods tend to agree well, with the generally accepted number of polar metabolites of [11C]-(R)-PK11195 being two [130].

Although there is some inconsistency in the extent to which metabolites can cross the BBB, contribute to the radioactive signal from the brain and bind preferentially to TSPO sites, all in vivo PET studies with arterial sampling agree that there is a definite need for close metabolite analysis of plasma samples in order to generate an exact input function for quantification [130].

1.8 Kinetic analysis and modelling of TSPO data

Basic tracer kinetic analysis of [11C]-(R)-PK11195 brain studies, including the use of plasma input (from direct arterial sampling or derived from images) and reference tissue models, has been in use since the tracer first became widely used in PET imaging in 1986 [133]. Historically, standardised uptake values (SUVs) have also been used as a semi-quantitative, but direct measure of regional tracer uptake:

푐푡푖푠푠푢푒(푡) 푆푈푉 = 푖푛푗푒푐푡푒푑 푑표푠푒 Eq. 1.2 ⁄푏표푑푦 푤푒푖푔ℎ푡 where 푐푡푖푠푠푢푒(푡) is the concentration in a given region at time 푡. Evidently, this measurement makes no prior assumptions about the compartmental tracer kinetics in plasma and tissue, so while it is advantageous in terms of speed and simplicity, it cannot offer a fully quantitative measure of tracer pharmacokinetics in vivo.

The kinetics of a tracer must be quantified with respect to such factors as its metabolism in blood, tissue clearance and perfusion [98]. As discussed previously, it is difficult to quantitatively analyse [11C]-(R)-PK11195 in dynamic PET brain studies

34 because of the low levels of specific binding in the brain. In addition, there is no easily definable reference region in the brain; that is, since glial cells are present universally, there is no truly microglia-free region (with no PK11195 binding sites) to use as a reference against areas of interest. Thus, in recent years, there has been an increase in efforts to develop new methodologies for extraction of the kinetics and vascular activity for reference tissues using supervised and unsupervised cluster analysis [20, 134-136].

The theory behind kinetic modelling is extensive and details of some key techniques which are commonly used, as well as many used in this project, are given in the following sections.

1.8.1 Overview of techniques All methods of systems analysis require an input function. This provides the ‘impulse’ behind the system and if it is incorrect or non-representative, even a perfect model will not deliver accurate results. When selecting the ‘correct’ model which best fits a data set, one must consider the bias (the difference between the parameter estimate as calculated by the model and the true parameter value) and the variance (a measure of the intrinsic standard error). Both the bias and the coefficient of variation (COV) are affected by the noise level within the data, but often a good indication of whether the chosen model fits the data correctly is the behaviour of the parameter estimates and the residual sum of squares at low noise levels. Figure 1.4 summarises this relationship.

Figure 1.4 The relationship between data, inputs, models and results. In the context of modelling real-world situations, having the correct input function with accurately measured data is key. As highlighted in the diagram, if the data are noisy and/or the input function derived from it is flawed, the results will be negatively affected even if the perfect model is designed. Equally, having the correct input function with ‘good’ data is not enough to obtain accurate results if the model selection is incorrect.

In PET imaging, input functions have conventionally been derived using arterial blood sampling. Although these so called ‘plasma input functions’ are considered the gold-standard, acquiring the blood samples involves cannulating the subject for the duration of the scan, which is uncomfortable, carrying risks of complication for them, and the subsequent metabolite analysis, which itself is technically demanding and error-prone, must be performed swiftly in a fully equipped blood laboratory on site (see section 1.7). Thus, efforts have been made to develop non-invasive methods of quantifying PET data. These methods can be grouped under the heading ‘reference tissue models’.

1.8.2 Plasma input functions In general, samples from the arterial whole blood line are transferred continuously to a blood trolley for activity monitoring. Simultaneously, discrete arterial blood samples are sent to a blood lab for metabolite analysis, as described in section 1.7. By assessing the retention times and corresponding polarity of these metabolites, it is then possible to deduce whether they are likely to cross the BBB. Most groups consider functions representing both the total activity concentration in plasma and the activity in plasma due to the parent compound ([11C]-(R)-PK11195) to generate a plasma input function [135-136]. The plasma over blood ratio (POB) is derived from the discrete blood samples and fitted to a straight line model function, as shown in equation 1.3.

푃푂퐵(푡) = 푝0 + 푝1. 푡 Eq. 1.3 where 푝0 and 푝1 are parameters of the linear model and 푡 is time. The POB is then multiplied with the whole-blood activity obtained from continuous sampling to produce a total plasma activity curve. A plasma input function (IF) can be generated by combining this curve with discrete plasma activity concentration measurements taken during the scan.

Once the IF has been generated, a model must be chosen to ‘process’ the data; in the context of PET, this model describes how [11C]-(R)-PK11195 binds to the tissues within the brain. Many different models have been designed for use with plasma input functions. One of the most popular for [11C]-(R)-PK11195 data has traditionally been compartmental models.

1.8.2.1 Compartment models Provided the correct one is selected for analysis, compartment models can comprehensively describe the kinetics of a tracer in the brain, also accounting for the

36 delay between arrival of the bolus injection of [11C]-(R)-PK11195 in the brain and at the online blood detector [134, 137]. A compartment model essentially breaks down tissue response to the tracer into as few ‘compartments’ as possible to accurately represent the binding of the tracer. These are represented as boxes in model diagrams, with individual rate constants describing the exchange of the tracer between them. The models are based on the principle that the uptake of a tracer in different materials varies, (see Figure 1.5). In 2001, a general model of this form was created to describe the behaviour of a tracer [138]. The group involved concluded that a general compartmental model leading to a set of first-order linear differential equations can be used to derive the ‘total tissue radioactivity in terms of a plasma input or reference tissue model’. The behaviour and binding of the tracer in individual compartments can also be determined by deriving and quantifying certain parameters using tissue time-activity curves (TACs) and (weighted) non-linear least squares fitting (WNLLS). The binding potential (퐵푃) is the most commonly used outcome measure of specific binding of a radioligand to a receptor (see Appendix 1 for a definition and derivation of binding potential and other parameters of interest).

Reversible models such as a) and c) in Figure 1.5 can be used to obtain values for distribution volume, 푉푇, non-displaceable binding potential, 퐵푃푁퐷, and 퐾1, 푘2, 푘3 and 푘4, while irreversible models such as that depicted in b) output values of 퐾퐼. Appendix 1 gives a full explanation of the physical meanings of these parameters.

Figure 1.5 Examples of compartment models commonly used with arterial input functions to quantify dynamic PET data. Kropholler et al. [134] compared three different compartment models for analysis of [11C]-(R)- PK11195 brain data: a) a one-tissue reversible model (1TCM), b) a two-tissue irreversible

37 model and c) a two-tissue reversible compartment (2TCM) model. The term ‘reversible’ indicates a radioligand which appears to dissociate from the receptor during the course of the

PET scan. 푉푏 is the fractional blood volume and 푓푝is the free fraction of parent tracer available to pass into tissue (i.e. unbound by plasma-proteins). 퐾1, 푘2, 푘3 and 푘4 are the rate constants for transfer of the tracer between compartments (see Appendix 1). 퐶푝푙푎푠푚푎 indicates the total concentration of tracer in plasma (as measured during metabolite correction), 퐶푓푟푒푒 is the concentration of free tracer in the target tissue and 퐶푠푝 is the concentration of specifically bound tracer in the target tissue. 퐶푛푠 is the concentration of non-specifically bound tracer in the target tissue. The grey box indicates the components of the models which are relevant for regional kinetic quantification.

Usually two input functions are required for compartmental analysis; the whole blood concentration and the metabolite corrected plasma (tissue input) concentration [134]. WNLLS fitting is used to fit the data with the chosen compartment model [139]. The number of compartments in the model is dependent on the biology and kinetics of the tracer. Theoretically, any number of compartments can be defined and, as with increasing the order of a polynomial when curve fitting, more compartments appear to fit the data better. Somewhat obviously, however, increasing the number of compartments (and thus the number of parameters) also increases the complexity of parameter estimation and may lead to ‘numerically unidentifiable parameter estimates’ [140]. Additionally, (W)NLLS fitting methods are hampered by the fact that they are prone to detecting local (rather than global) minima in cost functions.

The target region and the reference region are often best described by different compartment models; since an ideal reference region should be devoid of specific binding, it is expected to be best represented by a 1TCM [135]. The results of some studies using the compartment model approach to kinetic analysis are described in section 1.8.2.5.

1.8.2.2 Graphical analysis – Logan plots Graphical methods, such as the Logan plot for reversibly binding radioligands, were initially developed to surmount the problem that compartment models are bound by; namely that they require prior knowledge of a system and the results are therefore not model-independent (Logan et al. 1990). Logan graphical analysis, which is a linear operation, is much simpler than traditional non-linear compartment models.

푇 ( ) 푇 ( ) ∫0 퐶푡푖푠푠푢푒 푡 푑푡 ∫0 퐶푝푙푎푠푚푎 푡 푑푡 = 푉푇. + 푖푛푡 Eq. 1.4 퐶푡푖푠푠푢푒(푇) 퐶푡푖푠푠푢푒(푇)

As described in equation 1.4, the theory detailed by Logan et al. [141], using an

푇 푇 ∫ 퐶 (푡)푑푡 ∫ 퐶 (푡)푑푡 arterial plasma input function, is that a plot of 0 푝푙푎푠푚푎 against 0 푡푖푠푠푢푒 where 퐶푡푖푠푠푢푒(푇) 퐶푡푖푠푠푢푒(푇) 퐶 denotes the concentration of tracer in plasma or tissue of interest respectively and 푡, 푇 denote time, will reach a time point, 푡∗, after which the graph becomes linear, indicating that the tracer has reached equilibrium. It is assumed that at this time point, the time dependence of the intercept becomes sufficiently small that it can be ignored in terms of parameter estimation. This theory was formally proven thirty years ago [142] for irreversibly binding ligands. Logan et al. [141] showed that for the reversibly binding ligand ([11C]-cocaine), after time 푡 = 푡∗, the aforementioned plot will also become linear, with the slope of the line of best fit for the remaining data corresponding to the distribution volume, 푉푇 (see Appendix 1).

Although a good alternative to compartmental modelling, the Logan plot does have a tendency to underestimate 푉푇 in the presence of noise [143], introducing a statistical bias [144] which is not seen when using compartmental models.

1.8.2.3 Graphical analysis – Patlak plots Although [11C]-(R)-PK11195 is generally considered to bind reversibly (see section 1.8.2.5) to receptors, it is also worth mentioning a comparable linear method for irreversibly binding tracers with the values of 퐾1 from model b) in Figure 1.5. As 푇 ∫ 퐶 (푡)푑푡 with Logan plots, the theory behind Patlak plots indicates that a graph of 0 푝푙푎푠푚푎 퐶푝푙푎푠푚푎(푇) 퐶 (푡) against 푡푖푠푠푢푒 will reach a time point after which the tracer kinetics have reached 퐶푝푙푎푠푚푎(푇) equilibrium and the plot becomes linear with a slope equivalent to 퐾1. It should be noted that choosing the incorrect model for the tracer in question (i.e. selecting a Patlak graphical analysis for a reversibly binding tracer) will result in incorrect parameter estimations.

1.8.2.4 Spectral analysis Another attempt to bypass the need for prior knowledge regarding the number of compartments or the kinetics of the tracer was proposed in 1993 for dynamic PET studies [145]. Spectral analysis (SA) requires a plasma input function, which is convolved with the unit impulse response function (IRF) for a given tissue to produce a best fit curve for the data. The tissue concentration, 퐶푡푖푠푠푢푒(푡), is considered as a convolution of the plasma input function, 퐶푝푙푎푠푚푎(푡), with ‘a sum of k exponential terms’, as described in equation 1.5.

푖=푘 퐶푡푖푠푠푢푒(푡푗) = ∑푖=1 훼푖. 퐶푝푙푎푠푚푎(푡) ∗ exp (훽푖. 푡) Eq. 1.5 where 훼푖 and 훽푖 are parameters which are ≥ 0. 푘 is the maximum number of exponential terms to be included in the sum and is usually a large number of order 100.

As shown in equation 1.5, the model is linear with respect to 훼푖 and values for this parameter can be obtained either using the ‘simplex method’ [146], which essentially minimises the weighted sum of errors using 훼 values as variables, or, more commonly, a non-negative least squares method (NNLS) [147], which is often weighted in inverse proportion to the variance at each data point, as an optimisation algorithm. Both return a vector of optimal 훼 values, but the latter is subject to the constraint that 훼푖 ≥ 0. The values of 훽풊 are selected and fixed before the optimisation algorithm is run; they are chosen by considering ‘the slowest possible loss of radioactivity from the tissue (i.e. that associated with the radioactive half-life of the isotope) up to a value appropriate to transient phenomena (e.g. the passage of activity through the tissue vasculature)’ [145]. For 11C, with a decay constant of 0.0005663 s-1, this range is typically 5×10-4 to 1 s-1. The resulting values of 훼푖 are plotted against the logarithm of 훽풊 values, producing a spectrum of the kinetic components of the radioligand, which is representative of the response of the tissue of interest to the plasma IF.

The peaks produced by SA can be used to obtain the same parameters as are output by compartmental modelling and graphical analyses. The values of 훽풊 are discrete, meaning that the peaks of the spectra comprise one or more α values, which are then combined in subsequent analysis. The high frequency components, which are usually coincident with the upper limit of 훽 (i.e. 훽 = 1 or 푙표푔훽 = 0), are taken to be representative of the blood volume [145]. Meanwhile, the intermediate and low frequency components represent the extravascular behaviour of the ligand.

The number of intermediate components of the spectrum is generally taken as a good indication of the number of reversible compartments in the tissue of interest.

The combined height of the intermediate peaks is equivalent to 푉푇, or the integral of the decay-corrected unit IRF. When the lowest frequency peak is coincident with the lower limit on 훽 (i.e. the decay constant of the radio-isotope), the peak is taken to represent an irreversible component of the tracer’s kinetic behaviour; thus SA does not require any prior knowledge of whether a tracer binds reversibly or irreversibly. In fact, Cunningham and Jones [145] point out that, in the case of an irreversible tracer, the slowest component (i.e. the irreversible one), can be objectively quantified

40 with SA, while with a Patlak plot, a time after which the slow component dominates tracer behaviour must be defined. In this study, SA identified four peaks for [11C]- diprenorphine in the inferior frontal cortex. The lowest of these is the irreversible component, while the three intermediate peaks, two of which combine to form one, can be taken as an indication of the tracer being well represented by the 2TCM in this

ROI. The blood volume (푉푏) peak is furthest to the right.

Figure 1.6 Example of a spectral analysis results plot for [11C]-(R)-PK11195. The figure shows the results in the unsegmented cerebellum of an elderly healthy subject (data re-analysed from that previously published [135, 148]). The fractional blood volume peak is furthest to the right (pink), with intermediate peaks summing to 푉푇.

It should be noted that SA is usually performed on data without a decay correction, allowing the 훽 parameters to fall within a certain predictable range [145].

1.8.2.5 Key results from studies using plasma input functions Kropholler et al. [134] performed a study to deduce the optimum compartmental model for [11C]-(R)-PK11195 brain kinetics. Thirteen subjects were used: 2 with AD, 5 with MCI and 6 healthy controls. Using metabolite corrected and whole blood input functions (as described in sections 1.7, 1.8.2), the authors compared a 1TCM, 2TCM and irreversible 2TCM, as depicted in Figure 1.5, fitting each 퐾 to tissue TACs. In the case of the 2TCM, the authors chose to fix the ratio 1 (the ratio 푘2 of forward/backward transport across the BBB) and 푘4 to that of the whole cortex for stability. They also fixed 푉푏 and weighted the TACs with the square of the duration of each frame in the scan, dividing by the number of trues in each frame. They examined

41 the effects of varying the ROI size on the noise levels (through Monte Carlo

퐾1 simulations) and investigated ‘the sensitivity of [퐵푃푁퐷] and [푉푇] to fixing 퐾1, , 푘4 and 푘2

푉푏 to incorrect values’.

In order to directly compare the models, the authors used a weighted version of the Akaike information criterion (AIC), corrected for small sample sizes [149-150], ultimately concluding that the two-tissue reversible compartmental model with fixed 퐾 whole-cortex values of 1 best described [11C]-(R)-PK11195 tracer kinetics when 푘2 using a plasma input function. They did also note that the estimates of binding 퐾 potential obtained by fixing 1 were much higher than those obtained using the SRTM. 푘2 The authors suggest that this discrepancy may be due to the plasma input model not accounting for the exchange of the tracer between non-specifically and specifically

11 bound states, leading to an overestimation of 퐵푃푁퐷, especially since [ C]-(R)- PK11195 shows high non-specific binding [98].

Turkheimer et al. [135] identified a slow kinetic component of [11C]-(R)- PK11195 in the vasculature which suggested the presence of specific binding in the blood pool. This would make the two-tissue compartment model more complex than currently described by Kropholler et al. [134] and the subsequent estimation of 퐵푃푁퐷 values less robust. The authors suggest that rank-shaping exponential spectral analysis (RS-ESA, see section 1.8.5.3) in combination with a supervised clustering extracted reference region (see section 1.8.6) can overcome this problem.

Another issue with the use of plasma input function modelling is the large variability in direct measures such as 푉푇. Although it is a commonly recognised feature when using these approaches, this variability has not been widely reported [151]. These discrepancies are thought to be driven by differences in the plasma behaviour of TSPO tracers such as [11C]PBR-28 [152] in different subject cohorts. There is currently no consensus as to what extent 푉푇 (and other blood-derived measures) are affected, nor over what drives these differences in patients vs. controls. For instance, the latter study reported (unexpectedly) reduced 푉푇 in subjects with schizophrenia compared to healthy controls [152]. As such, Bloomfield et al. and others have elected to normalise 푉푇 to some reference region, for example the whole brain, giving a distribution volume ratio which accounts for the plasma behaviour of the tracer in individual subjects and thus is more stable. Hence, while plasma input modelling is still widely accepted as the reference standard, such effects can complicate interpretation of results, especially where the blood kinetic of a tracer is affected by

42 disease state, and appropriate considerations should be made with respect to quantification.

Characterisation of the optimum choice for kinetic modelling of new tracers must also be performed. This is further complicated by the necessary interpretation of the outcome measures depending on the binding status of the subjects. For example, Owen et al. recently performed a blocking study using [11C]PBR-28 and the TSPO ligand XBD173 in 26 healthy volunteers (16 HABs and 10 MABs) [153], determining that 퐵푃푁퐷 for HABs was approximately double that for MABs. By blocking with

XBD173 in 6 subjects, the authors were also able to calculate a global population 푉푁퐷 from occupancy plots of 1.98. A modelling study of [11C]PBR-28 recently found that incorporating an irreversible component into the 2TCM (2TCM-1K) to account for slow binding to the endothelium produced improved AIC values in 94% of regions

[154]. Additionally, the new model showed less time sensitivity and the 푉푇 distribution showed improved correlation with the known pattern of TSPO gene expression across cerebral structures.

[18F]DPA-714 has also been evaluated in 6 healthy humans and 10 AD subjects [155]. While the results of this study suggested that the tracer was not sensitive to group differences in TSPO expression, they did conclude that the 2TCM with free 푉푏 parameter was the preferred model for analysis, with good correlation to 퐵푃푁퐷 from the SRTM with cerebellar grey matter reference input. Importantly, this study did not correct for differences in binding due to the TSPO polymorphism.

1.8.2.6 Summary of plasma input approaches Kinetic modelling with plasma input functions is often quoted as the ‘gold standard’ approach to quantification of PET data. Indeed, when evaluating a novel tracer, it is highly desirable to perform compartmental modelling with a plasma input function first to provide a reference point for parameter estimates for comparison with other quantification approaches. These input functions are typically implemented with compartment models, but advantageously can be used with a variety of different models to bypass the need to know of the number of compartments for a complete description of tracer behaviour; for example with graphical analysis approaches. In reality, however, the process of obtaining the blood samples is not only traumatic for the subject, but also technically challenging and prone to error. The results from plasma input analysis therefore often show high variability. In addition, it is increasingly recognised that inter-subject variability in the behaviour of tracers in plasma, possibly related to disease state but currently not fully

43 understood, can cause large discrepancies in the results from such modelling approaches [151-152]. As such, other techniques to derive input functions for kinetic modelling have been developed, with variable success.

1.8.3 ROI analysis vs. parametric mapping While traditionally, analysis of PET brain data is performed on a regional basis, manual and automated versions of this type of ROI analysis are (somewhat obviously) subject to intra-observer variability. Additionally, due to the nature of ROI definition, signals in unexpected and therefore undefined areas can be overlooked [156]. One of the assumptions of ROI approaches is that the tissue within the region behaves in a kinetically homogenous manner; this is not always the case, particularly where grey and white matter cannot be sufficiently resolved due to the limits on spatial resolution of a PET scanner (typically 5 mm full width at half maximum, FWHM, at most) [157]. Unfortunately, other methods that analyse the data voxel-by- voxel may be too sensitive to noise to be used in [11C]-(R)-PK11195 studies. Consequently, these methods use smoothing on the raw PET images to minimise the effects of noise, thus also reducing the grey/white matter contrast. In this case, ROI analysis may indeed be preferable. Nevertheless, a type of voxel–based analysis called parametric mapping, linearised for use on [11C]-(R)-PK11195 data, has been used with some success [79, 158].

ROI analysis involves deriving a tissue TAC from data relating to a particular ROI; a model is then selected and applied to this data to produce parameter estimates in vector form. Parametric mapping (PM), on the other hand, applies the selected model to the TACs from each voxel within the PET image, producing a parameter matrix (or ‘map’), which can then be averaged to a parameter vector.

PM techniques are compatible with Logan (and Patlak) graphical analysis, as well as SA, but, since they require definition of tissue compartments within an ROI, compartmental models cannot be applied on a voxel-by-voxel basis.

1.8.4 Image-derived input functions As mentioned earlier, the arterial cannulation required to generate plasma input functions is distressing for the patient and also carries a risk of them developing arterial thrombosis [159]; the longer the exposure of the patient to the cannula, the higher the risk. Thus, recent efforts have sought to develop less invasive methods of quantifying an input function. Initial use of arterialised venous blood (heating the arm in a warm bath to arterialise the venous blood before cannulating a vein, which causes

44 less discomfort than sampling from an artery), were not successful as the venous blood was not always completely arterialised (i.e. did not represent the tracer kinetics in arterial blood).

Image derived input functions (IDIFs) were originally validated for use in cardiac PET studies in the 1980s [160]. The method involves measuring arterial input functions from the PET images themselves. It has been successful when used in cardiac studies because the volume of blood in the heart is much higher than that in the brain, which substantially decreases the PVE. The internal carotid artery, which is the largest in the brain, conversely, has a small diameter (with regards to the PVE; the diameter is comparatively quite large for a blood vessel) of 4-6 mm [161] and suffers from the PVE much more (see section 1.3). Correction of the input function for metabolites is more difficult with IDIFs, while patient movement can introduce further limitations of the method [162]. Co-registered MR and PET images can be used to achieve better resolution when defining ROIs and obtaining IDIFs [163], but due to variability in patient position between scans etc., inter-modality co-registration is also not a straightforward procedure.

Although variable, the results of studies using IDIFs on the whole suggest that blood sampling (and thus determination of radiolabelled metabolites) is necessary for optimisation of the method [159, 164]; clearly this is not beneficial to the underlying aim of eliminating the need for arterial sampling.

1.8.5 Reference tissue input functions The limited success of IDIFs coincided with the development of so-called reference tissue models (RTMs) in the noninvasive quantitation of PET brain data. Besides sparing the patient the discomfort of arterial cannulation, another advantage of this method is that it does not require correction for metabolites of the parent radioisotope. The method involves selecting a reference tissue devoid of specific uptake of the tracer. Somewhat obviously, the choice of this reference region depends on the disease being studied [20] and so has varied between authors; some have chosen the cerebellum [106, 165-166], but historically there has been some contention over this choice, as specific binding is not wholly absent in any region of the brain. Additionally, the cerebellum exhibits greater vasculature to the cerebrum; since TSPO is known to be expressed in the vessel walls of even healthy subjects [135], use of the cerebellum as a reference region could be questionable. This has led to the development of an algorithm for the ‘supervised clustering’ selection of grey matter voxels on the PET image; in the cerebellum and in the cerebral cortex [135].

Clustering algorithms are covered in more detail in section 1.8.6. After selecting a reference region, either the simplified [167] or full [168] reference tissue models (SRTM or FRTM) or uptake normalised to this reference region [165, 169-170] can be applied to calculate binding potential and other parameters of interest. Use of the reference tissue approach needs validation by comparison with the plasma input function; this has only been achieved relatively recently for [11C]-(R)-PK11195 [134], while data from other TSPO tracers are analysed using ‘pseudo’-reference regions [171-172] and there is still some debate over the use of this method.

1.8.5.1 The simplified reference tissue model The SRTM is actually a variation of a more complete, or full, reference model, which in itself is the reference tissue equivalent of the plasma input compartment model. The full model uses equations 1.6 – 1.8 to obtain parameter estimates for

퐾1 퐵푃푁퐷, 푘2, 푘3 and 푅1. The latter is equal to ′ or the ratio of unidirectional clearance of 퐾1 the radioligand from plasma to tissue for the ROI to that from the reference region [173].

Figure 1.7 represents the behaviour of a tracer according to the reference tissue model.

Figure 1.7 The two compartment reference tissue model. Here, the ROI behaves as in the 2TCM, while the reference region is described by a 1TCM; i.e. free of specific binding. The model assumes that only the parent compound can cross the BBB.

푑퐶 (푡) 푟푒푓 = 퐾′퐶 (푡) − 푘′ 퐶 (푡) Eq. 1.6 푑푡 1 푝푙푎푠푚푎 2 푟푒푓

푑퐶 (푡) 푓푟푒푒 = 퐾 퐶 (푡) − 푘 퐶 (푡) − 푘 퐶 (푡) + 푘 퐶 (푡) Eq. 1.7 푑푡 1 푝푙푎푠푚푎 2 푓푟푒푒 3 푓푟푒푒 4 푠푝

푑퐶 (푡) 푠푝 = 푘 퐶 (푡) − 푘 퐶 (푡) Eq. 1.8 푑푡 3 푓푟푒푒 4 푠푝

-1 ′ All symbols have the usual meanings (units of 퐶푝푙푎푠푚푎,푠푝,푓푟푒푒 are kBq.ml ). 퐾1 is the rate constant for transfer from plasma to the reference compartment and ′ 푘2 describes transfer from the reference compartment to plasma. Although 퐵푃푁퐷 itself is robust, the other parameters are prone to large errors [167] and this led to the development of an adaptation to this model which contains only three parameters

(퐵푃푁퐷, 푘2 and 푅1) and is applicable for tracers which exhibit sufficiently rapid transfer between the free and specifically bound compartments. Both models operate under two assumptions: 1) that the non-specifically bound component of the volume of distribution of non-specifically bound tracer (푉푁푆) is the same in both the reference ′ 퐾1 퐾1 and target regions (i.e. = ′ ) and 2) that the reference tissue is not affected by any 푘2 푘2 pathology typical of the disease. The SRTM additionally assumes 3) that the kinetics of the tracer in the reference tissue can be described by a single compartment.

Lammertsma and Hume [167], performing a study using [11C]-raclopride on healthy controls, patients with PD and healthy controls pre-dosed with a neuroleptic

(to cover a wide range of clinically relevant 퐵푃푁퐷 values), showed that not only did the SRTM give 퐵푃푁퐷 values which were comparable with those obtained from the FRTM, but it was also more insensitive to initial estimates, produced smaller errors for the other two parameters and converged more rapidly than the full model.

Although the SRTM is usually successful when using ROI analysis, voxel-level analysis using this method is much slower and more sensitive to noise. Parametric images may, however, be obtained using a basis function implementation of the SRTM [140] which changes the NLLS equations used in the latter into a set of linear basis functions. This method, termed reference parametric mapping (RPM) was shown by

11 Gunn et al. [140] to produce 퐵푃푁퐷 estimates for [ C]-raclopride which were in good agreement with those obtained by NLLS estimates.

1.8.5.2 Reference tissue graphical analysis Both Logan and Patlak graphical analysis can also be performed using a reference tissue input function. The advantage of these methods is that they do not assume that the reference tissue chosen is represented by a 1TCM. While the Patlak technique remains much the same as with a plasma input function, Logan plots with a reference tissue input can be used to derive parameter estimates for the distribution volume ratio (DVR, see Appendix 1) instead of 푉푇. Equation 1.4 becomes:

푇 퐶푟푒푓(푇) 푇 ∫ 퐶 (푡)푑푡+ ∫ 퐶 (푡)푑푡 0 푟푒푓 ′ 0 푡푖푠푠푢푒 = 퐷푉푅. 푘2 + 푖푛푡 Eq. 1.9 퐶푡푖푠푠푢푒(푇) 퐶푡푖푠푠푢푒(푇)

47 such that the slope of the plot after time point 푡∗, when the tracer kinetics have reached equilibrium, is equal to the DVR. There is now, however, also a dependence

′ 퐶푟푒푓(푇) on 푘2 introduced by the ′ term. This suggests that obtaining estimates for the 푘2 ′ DVR now requires some further correction for the variability of 푘2 as compared with Logan graphical analysis using a plasma input function, which does not include this term. This effect is (somewhat obviously) tracer dependent; in order to assess the ′ ′ sensitivity to the choice of 푘2, Logan et al. [174] used simulated data, varying 푘2 by ± 25 and ± 50% for the PET tracer [11C]-dMP (d-threo-methylphenidate), and found that parameter estimates for the DVR did not change by more than 3-4%, which was within the accepted range for this tracer, as described by the COV on the DVR. It should also be noted that those ROIs with a lower receptor density will be subject to

퐶푟푒푓 lower errors, since the ratio reaches equilibrium in a shorter time than for high- 퐶푡푖푠푠푢푒 density regions.

′ In a more detailed investigation into whether the 푘2 parameter could be ignored altogether, or whether a population average or subject-specific values should be used, one group [175] used three reversible ligands with varying equilibration rates in tissue (the D1 dopamine receptor binding ligand [11C]-SCH23900, the type 2 monoamine vesicular transporter ligand [11C]-dihydrotetrabenzene and the dopamine uptake transporter ligand [11C]-methylphenidate). The study cohort consisted of 10 healthy control subjects and arterial blood sampling was used both to derive subject- ′ ′ specific 푘2 values and for assessment of the significance of the choice of 푘2 by comparison with the reference tissue Logan results for regional DVRs.

′ Ignoring the 푘2 term reduces the precision of DVR values obtained in graphical analysis with a reference tissue input since the time range over which the slope can be calculated is constrained. Since the main advantage of reference input Logan analysis is that the burden of metabolite analysis and the discomfort of arterial sampling are ′ not necessary, using subject-specific 푘2 values is not favoured or practical. Using an ′ average ‘population’ value of 푘2, on the other hand, could result in variability in DVR ′ values due to variability in 푘2 amongst the study cohort. The group concluded that two ′ of their three tracers could be used with population average 푘2 values to derive DVR values in good agreement with those derived by plasma input modelling (subject- ′ specific 푘2) and the chosen reference tissues could most likely be described by a one compartment model. They do point out that the latter assumption may not be true, but

퐶푟푒푓(푇) the fact that the ratio became constant over the time during which the slope 퐶푡푖푠푠푢푒(푇)

48 was estimated meant that the kinetic behaviour of the tracers in the reference compartment became less relevant. The third tracer, [11C]-methylphenidate, however, ′ appeared to have a much larger sensitivity to the choice of 푘2. The authors believe this sensitivity is an effect of the tracer equilibrating more slowly than the others and suggest that care should be taken to establish the tracer properties in tissue before using the reference tissue Logan approach.

Parametric maps can also be obtained using Logan and Patlak analysis with a reference tissue input. To overcome noisiness, the Logan plot decreases the parameter estimates on the parametric maps, producing ‘regularized maps with good noise levels’ [176] which are prone to a large bias. An alternative method, termed ‘multi- resolution Bayesian regression’ [177] was developed to reduce noise with no effects on bias and proved successful when performed on [18F]-FDG, a tracer of glucose metabolism. This technique is only applicable to irreversible tracers (i.e. Patlak graphical analysis).

1.8.5.3 Reference tissue input parametric images Although ‘traditional’ SA can only be performed when a plasma input function is used, there have been some attempts at adaptations on the method to generate parametric maps for use with a reference tissue input function that have met with varying levels of success. The main reason that SA cannot be used with a reference tissue input is that it typically uses an NNLS approach to optimise parameter estimates (see section 1.8.2.4); reference tissues cannot be constrained to a set of positive values in this way and still present an accurate input function.

Exponential spectral analysis (ESA) was developed in an effort to simplify the non-linear methods of optimisation used in traditional SA. ESA instead uses linear estimates of the coefficients of a ‘preselected set of exponential basis functions’ [176]. Turkheimer et al. [176] designed a variation on this method which does not rely on the same non-negative constraints as ESA. RS-ESA obtains a linear least squares solution, shrunk to a degree determined by (an estimate of) the SNR. The reader is referred to the reference for full details of the method. Comparing results from real and simulated data for fast and slowly equilibrating tracers ([11C]-raclopride, [11C]- flumazenil and [11C]-MDL100907), the authors found that RS-ESA performed as well as Logan graphical analysis at low noise levels and exhibited a much lower bias than Logan (8% vs. 20%) in the presence of more noise.

An alternative approach to producing parametric maps using a reference tissue input was proposed in 2002 [178]. This ‘basis pursuit strategy’ operates

49 similarly to the plasma input compartment model; determining and characterising the IRF by defining a few basis functions, using a technique known as basis pursuit de- noising [179] instead of an NNLS algorithm to obtain a ‘data-driven estimation of parametric images based on compartmental theory’, or DEPICT. As the name suggests, the method requires no a priori knowledge of the kinetic behaviour of the tracer; it is a ‘data-driven’ method that returns the number of compartments using a basis function approach [140, 145, 180]. The technique can be applied to reversible and irreversible tracers, can also be used with plasma input functions and, similar to RS-ESA, does not constrain the coefficients to be positive.

1.8.5.4 Key results from studies using reference input functions Another study by Kropholler et al. [181] aimed to establish the best reference tissue model to use in kinetic analysis of [11C]-(R)-PK11195 data. The group studied 13 subjects; 8 healthy controls and 5 with traumatic brain injury and tested the accuracy and reproducibility of 9 variations on the reference tissue model, using the plasma input model as a standard. ROIs were fitted to each of the models and to the 2TCM as per the group’s previous findings [134]. Their results indicated that the SRTM was the best choice when only a reference tissue input was available, while a variation on SRTM, SRTMpl_corr, corrected for bias due to non-specific binding, performed best when a metabolite-corrected plasma input function was used.

Another study [182] on patients with AD and MCI found that the most appropriate anatomical reference region for these diseases is the bilateral cerebellum. This study also indicated that the cluster analysis technique to group reference region voxels fails to work in approximately 10% of cases. It is possible that the disease state of this 10% of patients was so severe that there was not a high enough percentage of healthy tissue to serve as a set of reference voxels [20]. A further discussion of this study is given in section 1.8.6.

It has generally been observed [134-135, 167, 181] that the 퐵푃푁퐷 estimates obtained from plasma input modelling (even corrected for non-specific binding) are considerably higher than those obtained via (reference tissue) SRTM, implying that there is a significant fraction of specific binding present in the so-called ‘reference’ region.

Schuitemaker et al. [156] performed a study to establish whether 퐵푃푁퐷 or

11 푉푇 was more sensitive to changes in binding of [ C]-(R)-PK11195 between young and elderly healthy patients [183]. The group generated a metabolite corrected plasma

50 input function for the Logan graphical analysis and a bilateral cerebellum TAC as the reference tissue input for an RPM analysis. The group corrected for ‘intersubject

퐾1 differences in the ratio, i.e. differences in [푉푇] of the ‘free tracer compartment in 푘2 tissue’ in the Logan graphical method by applying proportional scaling, a technique to increase the sensitivity of SPM to detection of local variations [184]. Their results indicated that RPM analysis, leading to parametric maps of specific binding potential, was the most sensitive method to changes in [11C]-(R)-PK11195 binding between young and elderly healthy subjects. The main issue with this method, however, is that it again relies on the choice of a suitable reference tissue (here, the cerebellum), devoid of specific binding of the tracer; since such a region is not easily definable for [11C]-(R)-PK11195, as discussed, this could potentially lead to an underestimation of binding potential.

Another study by this group [158] established that Logan graphical analysis (with a metabolite corrected arterial input function) was the best method for obtaining parametric maps of 푉푇 at the voxel level. The study also showed that basis function implementations of SRTM, Ichise’s multilinear regression [185], RPM and Logan analysis with a reference tissue input (here, the cerebellum) give reasonably accurate results of estimates of binding potential (when there are reduced flow rates or large variations in 푉푏, RPM is the best choice as it shows less bias in these conditions).

A recent study in healthy rodents aimed to validate SUV as a measure of inflammation when using [11C]PBR-28. Toth et al. [186] performed arterial sampling in 8 rats (n = 4 retest) and 4 C57BL/6J mice. Comparing the 푉푇 and SUV, they found significant correlation and good TRTs (8 – 20%, 7 – 23% in rats and mice respectively), suggesting that SUV is a good surrogate measure of [11C]PBR-28 uptake in the healthy rodent brain. A retrospective study in baboons, however, found variable correlation between 푉푇 calculated using Logan graphical analysis and SUV between healthy animals and those injected with LPS [187]. In a large cohort of 25 AD subjects,

11 with mild cognitive impairment (MCI) and 21 healthy controls, 푉푇 from the 2TCM correlated highly with SUVR (SUV ratio) using the cerebellum as a pseudo reference region [172]. Indeed, results from this study suggest that the SUVR method may offer greater sensitivity to disease-related binding, as it was able to detect increases in [11C]PBR-28 signal in an extra region than absolute quantification with the 2TCM.

1.8.5.5 Summary of reference tissue input approaches Reference tissue input functions are an attractive alternative to plasma input functions, obviating the need for painful and error-prone arterial sampling. The input functions from anatomical reference tissues are used in conjunction with reference tissue models such as the SRTM, but can also be used with graphical approaches. The main issue with using the kinetics of an anatomical tissue as an input function is that there is no region in the brain which is truly devoid of binding sites for [11C]-(R)- PK11195 and other tracers of neuroinflammation which target TSPO. Therefore, as mentioned above, there has been increasing interest in the development of alternative data-driven techniques to derive an input function different to an anatomical reference tissue, but still non-invasive.

1.8.6 Cluster analysis The theory of cluster analysis existed in mathematics many years before it was applied in PET tracer kinetics. The term ‘cluster analysis’ derives from the inherent aim of the method to separate the objects in question into a given number of groups (or clusters), such that the members of any group are more similar to each other than to the members of any other group. Conceptually, defining a ‘cluster’ is quite a challenging task; where does one cluster start and another end? How many clusters are needed to fully describe a given set of data? Loosely, clustering can be divided into two types; hard clustering, where an object has a probability of either 1 or 0 of belonging to a certain cluster, and soft clustering, where an object has a non-integer likelihood (between 1 and 0) of belonging to any cluster.

The term ‘cluster analysis’ is very broad and encompasses several algorithms that have each been designed to try to answer these questions. One commonly used method is centroid-based analysis, of which Lloyd’s algorithm, or ‘k-means clustering’, is particularly well-known. This method separates 푛 members of a data set into 푘 clusters such that a) each member belongs to the cluster which has the closest mean and b) the within-cluster sum of squares (the square of the Euclidean distance between the point and the centroid of the cluster) is minimised. The underlying algorithm is a variation on the expectation-maximisation algorithm. It is iterative and can be split into two steps; an ‘assignment’ (or expectation) step, where every 푛 is assigned to the 푘 which has the closest mean to its own value, and an ‘update’ (or maximisation) step, where the new means of the (new) clusters are set as their centroids [188-189]. When the assignment of each object stops changing, the algorithm has converged. One of the main drawbacks of the k-means clustering

52 method is the fact that it requires specification of the number of clusters, 푘, as an input to the algorithm, which is not always immediately obvious for a given data set [190]. It is thus recommended that the optimum 푘 for the data set in question is chosen by means of ‘trial and error’ diagnostic checks before full analysis is carried out.

K-means clustering is computationally demanding in Euclidean space, but the algorithm does perform well on large data sets, such as those obtained in PET, and has thus become a favourable method for cluster analysis. When applied to PET data, the aim of clustering is to group similar tissue TACs. It is a data-driven technique, which gives it an advantage over compartmental modelling as it requires no a priori knowledge of the behaviour of the tracer or the target/reference tissues. In recent years, two different forms of cluster analysis have emerged; supervised and unsupervised.

1.8.6.1 Unsupervised clustering Cluster analysis was first applied to dynamic PET data in 1996 [191]. The ultimate aim of this work was ‘to partition the data according to its probability of belonging to each of 푘 clusters…based purely on the shape of the pixel vectors’ [191]. The term ‘unsupervised’ refers to the fact that the centroids of the clusters are not defined prior to implementation of the clustering algorithm.

Figure 1.8 summarises the operation of the algorithm; essentially, calculation of the log likelihood of a voxel belonging to a particular underlying TAC is repeated iteratively until it stops changing. Once the data have been partitioned, selection of the cluster representing a reference tissue must be performed by manual inspection. The authors adapted their clustering algorithm from an earlier ‘mixture model’ [192] in order to account for subtle variations in TAC shapes. They also assumed that noise across the time dimension of the image is constant (which may not be the case in FBP- reconstructed images). A key issue with this simple implementation is that the assumption that there are only a limited number of underlying tissue TACs with discrete values of their parameters is not strictly accurate. The continuum of parameter values that is more characteristic of PET is the result of the PVE and spill in, as well as patient movement, and subsequent smoothing corrections can distort the distributions of the data so that defining the correct number of clusters to search for within the data is more difficult.

Figure 1.8 Flow diagram describing how a clustering algorithm proceeds. The diagram is reproduced from [191]. PET data is considered on a voxel-by-voxel basis. The log likelihood of a voxel belonging to a particular kinetic class of TAC is calculated iteratively until it stops increasing.

The first work using cluster analysis on [11C]-(R)-PK11195 data was published in 1999 [193]. As mentioned earlier, there is no immediately and easily definable reference region completely free of a specific signal for this tracer. The fact that regional TACs in the healthy brain have similar shapes to one another, however, implies ‘consistent behaviour of the ligand in non-pathological tissue’; this suggests that the cluster analysis approach of segmenting data into distinct clusters dependent on their tissue kinetics could be a practical choice for [11C]- (R)-PK11195. Myers et al. [193] found that two main clusters (out of ten that are commonly generated in unsupervised cluster analysis) represent normal tissue kinetics in diseased brains (when compared with the TACs from control subjects). They concluded that these two clusters were suitable for use as an input function to the SRTM to produce parametric maps of relevant parameters.

To put use of this method into context, one group, mentioned earlier, [20] performed a study on a cohort of 11 patients with MS and 9 healthy control subjects, using an ECAT 953B scanner 35 (CTI/Siemens). Due to the pathology of MS, it is often impossible to define an anatomical reference region for input to the SRTM. The authors therefore elected to separate the voxels in the raw dynamic data into 10 clusters ‘distinguished by the shape of their [TACs]’. They found that ~90% of the

54 voxels were separable into two clusters; one representing the TAC from the skull and scalp, and another, the TAC from the cerebral cortex, which was selected as the reference TAC. The suitability of this reference TAC was assessed by comparison with a healthy grey matter kinetic previously derived, also by cluster analysis, from a healthy population. The study involved comparison of the TACs from the diseased subjects with those from the healthy controls, with dissimilarity assessed using a 휒2 test (p < 0.05). Although normal white and grey matter can contain small amounts of specifically bound [11C]-(R)-PK11195, leading to underestimation of the binding potential, the common reference input TAC meant that any reduction in measurement of the true binding was consistent across all subjects. For one MS patient, however, no reference TAC could be derived; the authors suggest that this was due to the fact that, in advanced stages of disease, the volume of healthy reference tissue is too low to serve as an accurate input function. The analysis identified increased [11C]-(R)- PK11195 expression areas of pathology corresponding to those seen on T1 and T2- weighted MR scans, but also in normal-appearing anatomy, which in many cases was found to relate to ‘neuronal projection areas’. The findings also showed that, although patients with high disability scores showed increased local and/or global [11C]-(R)- PK11195 binding, this did not always correlate with increased clinical disability measured using standardised clinical assessment scales (EDSS) [20, 194]. This highlights the limitations of using clinical scales when examining inflammatory neurodegenerative conditions such as MS. The authors note that using [11C]-(R)- PK11195 binding as a marker of inflammation may be more apt for measuring certain parameters such as the speed of disease progression than for ‘cumulative measures of longstanding and recent disability’; i.e. the actual status of relapsing patients who show increased local [11C]-(R)-PK11195 binding may not be immediately derivable from their EDSS scores. A further discussion of implementations of unsupervised clustering is given in section 1.8.6.3.

1.8.6.2 Supervised clustering While unsupervised clustering theoretically requires no information on whether two given voxels are likely to be linked, in practice there are many situations where such knowledge is available, dependent on established pathology [195]. In such cases, unsupervised clustering, which is purely data-dependent, may not full describe the relevant physiology [135]. Supervised clustering (SVCA) incorporates this knowledge by allowing the user to pre-define the centroids of the clusters. Turkheimer et al. [135] aimed to develop a new methodology which allowed extraction of a ‘proper’ reference region (that is, one that resulted in robust and

55 reproducible 퐵푃푁퐷 estimates which were comparable to those derived from plasma input modelling). The reference class of voxels was defined as that which exhibited kinetic behaviour closest to that of grey matter in healthy controls. The algorithm essentially comprised 3 components:

• ‘An input normalisation procedure to scale each volume of the dynamic sequence’

• ‘A set of predefined kinetic classes’

• ‘A regression procedure to calculate the contribution of each kinetic class to each pixel’s kinetic’.

Normalisation was achieved by subtracting the mean of each frame from the frame itself and dividing by its standard deviation (SD) to create an input [191]. For stage two, six kinetic classes (SVCA6) were defined from a historical database of 12 healthy control subjects, 4 AD patients and 3 patients with Huntington’s Disease (HD). They were: ‘non-specific grey matter, non-specific white matter, pathologic PBR binding, blood pool, skull and muscle’. The kinetic of each pixel was ultimately modelled as a ‘weighted linear combination of [its] class kinetics’. The result of this process was a volumetric map of weights for each class, with the reference region TAC a weighted average of the un-normalised pixel TACs [135]. The reference voxels excluded regions with specific binding that are typically located ‘in the venous and arterial circulation’. In this way, the relative differences between TACs for each class can be examined. The method was validated by comparison with a plasma input model and the reliability assessed using SRTM. RS-ESA was used rather than traditional compartmental modelling, and SRTM was chosen due to the low SNR of [11C]-(R)-PK11195. The authors concluded that their supervised clustering method resulted in reproducible and reliable parametric maps for extraction of binding potential estimates that correlate better with plasma input models than the pure reference region input SRTM.

The number of predefined kinetic classes can be reduced; for example, the bone and soft tissue regions can be removed from the PET scan prior to analysis [196- 197]. This is done using a brain mask to filter out the skull and scalp on the MR image, leaving 4 kinetic classes (SVCA4). Boellaard et al. [196] found that this technique improved precision, probably due to the elimination of noise from the skull and soft tissue (i.e. fewer parameters for fitting). The authors used SVCA4 in conjunction with a modified version of the SRTM with a correction for blood volume, SRTMV (see

56 section 1.8.5.4) and noted better differentiation between AD and control subjects [197].

The latter group [197] proposed a study to investigate the optimum number of kinetic classes for a cohort of patients including young controls, older controls, patients with mild cognitive impairment (MCI) and patients with AD. The thalamus was again chosen as an ROI due to its relatively high specific binding and its variable uptake as described by other studies [20, 198]. A cerebellar ROI was chosen as a reference tissue TAC for comparison with those extracted via SVCA4. The 2TCM (with an additional blood volume correction parameter) was used for analysis of the extracted reference tissue TACs and these were used as input for SRTM, the standard 2TCM, the basis function implementation of SRTM [140] and SRTMV. The findings indicated that the incorporation of blood volume in the model improved contrast in grey and white matter segmentation and 퐵푃푁퐷 contrast between groups of subjects. Tomasi et al. [199] suggested that this was because [11C]-(R)-PK11195 may bind specifically to the blood vessel walls; SRTMV claims to correct for this effect without the need for arterial sampling by including a linear term in both the target and reference regions corresponding to the image-derived blood component of tracer activity (퐶푏(푡)). The SVCA methods performed better than using the pure cerebellar reference input. SVCA4 resulted in input curves with lower 푉푇 and 푉푏 fractions than SVCA6, and both SVCA methods showed significant differences in specific binding between groups, while the pure cerebellar reference tissue approach did not. They conclude that SVCA4 in conjunction with SRTMV may be the best approach for quantification of dynamic PET brain data. The authors do note, however, that SVCA methods may be dependent on the PET scanner and image reconstruction techniques [196], as well as the scanning protocol, and thus may need to be modified and validated at each site. Additionally, the fact that SVCA4 requires use of a brain mask may limit its reliability as it can result in incorrect removal of brain tissue.

1.8.6.3 Key results from studies using cluster analysis A study on MCI and AD patients [182] used a total of 39 subjects (10 young and 10 elderly healthy controls, 10 with MCI and 9 with probable AD). 퐵푃푁퐷 values were calculated using a plasma input function (BPPLASMA), a target reference region (bilateral cerebellum) and SRTM (BPSRTM), and the latter corrected for non-specific binding (BPSRTM_CORR). In their clinical evaluation, the group used a clustering algorithm to extract tissue TACs; in total, they derived 10 clusters and chose the TAC with the lowest BPPLASMA as the reference region cluster. The results showed that the BPSRTM and

BPSRTM_CORR values obtained by cluster analysis were in good agreement with those obtained through the anatomical reference region of the cerebellum. They also note that the cerebellum is only affected in later stages of AD as compared with the cerebrum and should therefore be preferred as an anatomical reference region. As mentioned in section 1.8.5.4, however, approximately 10% (4 of 39) of scans were rejected because the algorithm ‘did not result in physiologically meaningful clusters’ for these patients.

Another group compared the results from using the cerebellum as a reference region and from using cluster analysis to extract a reference TAC [200]. Using [11C]- (R)-PK11195 and the SRTM for quantitation of the PET data, the study aimed to deduce whether there is an age-related increase in activated microglia in the normal healthy brain. The parameters of interest were the regional mean 퐵푃푁퐷 and the ‘lesion load’, defined here as ‘the fraction of voxels with significantly increased binding’ and calculated through volume of interest analysis of parametric binding potential maps. T1-weighted MRI and PET images were co-registered for 14 healthy subjects between the ages of 32 and 80 years. The study did not use arterial blood sampling. Unsupervised cluster analysis involved segmentation of the raw data into 10 clusters based on the shape of their TAC. They also found that ~90% of the voxels could be separated into two clusters representing the skull and scalp and the cerebral cortex respectively. The results showed an increase in the mean 퐵푃푁퐷 values in older patients. In addition, the authors noted a correlation between the age and the mean

퐵푃푁퐷 value for the right and left thalamus (which were chosen as ROIs due to their dense connections with most cortical areas; i.e. those associated with age-related microglial expression) independently when cluster analysis was used, but a significantly lower correlation coefficient was calculated when the subjects’ individual cerebellar TACs were used as a reference input. Minimal binding was observed in the cerebral cortex, possibly due to the low sensitivity of [11C]-(R)-PK11195 for ‘very low- grade cortical pathology in “normal” aging brains’. The thalamus appeared not to show any significant changes due to ageing; the authors thus hypothesise that the increase in microglial activation observed in the ageing brain may be a secondary effect of more widespread cortical changes. They conclude that (unsupervised) cluster analysis is a more sensitive approach than a cerebellar reference input for quantitation of a reference input in the case of normal ageing human brains.

Parametric mapping techniques have been adapted considerably in the last decade and establishing which technique is most suitable for each application is difficult. One group [201] performed a study in patients with AD (to assess reliability

58 of the techniques) and HD (to assess sensitivity of each technique), using a reference region input derived by supervised clustering. The two diseases are known to show different binding of [11C]-(R)-PK11195: low and diffuse in AD [170, 183] and higher and more central in HD [202]. The group aimed to assess parametric maps based on the ‘reproducibility and sensitivity to microglia activation’ and minimise noise. Five kinetic modelling techniques were analysed: SRTM, DEPICT [178, 203], a simple target-to-reference ratio or RATIO [204], Logan graphical analysis or LOGAN and a wavelet-based Logan plot or WAVE [205]. Comparison of these methods using the intra-class correlation coefficient (ICC, a measure of reliability) and z-scores (a measure of sensitivity) showed that DEPICT, WAVE and SRTM had good reliability (though DEPICT had a slightly reduced sensitivity, despite the fact that this method includes more biological compartments), while RATIO and LOGAN had higher noise levels.

1.8.6.4 Summary of cluster analysis approaches Unsupervised cluster analysis is a robust method for generating reference tissue input TACs for use with SRTM and other quantitative models. Reference TACs may not be derivable from patients in the late stages of diseases such as MS, while the increased microglial activation observed in the thalamus of ageing patients may be a secondary effect of cortical changes. Supervised cluster analysis is now far more commonly used than its unsupervised predecessor, since incorporating knowledge of the kinetic behaviour of a tracer in a given tissue gives a better representation of brain physiology. It is also robust and reproducible, but its usefulness may be limited if the method requires validation depending on the PET scanner and protocol used at each site. Additionally, there is no current evidence to suggest that even reference voxels obtained in these ways are completely devoid of specific binding.

1.8.7 Summary of kinetic analysis approaches Table 1.2 summarises the uses and relative advantages and disadvantages of plasma, reference and cluster input functions. The variation in results when using different input function and model combinations is a confounding factor in PET kinetic analysis. While detailed efforts have been made to establish and optimise the modelling methodology for [11C]-(R)-PK11195, second generation tracers are more challenging, with problems introduced by variable binding affinities and the lack of a true TSPO-free reference region. The optimum choice needs to account for practical factors such as scanner resolution and length of scan, as well as the disease state of

59 the subjects and tracer kinetics, which may also be affected by disease-related pathology.

Table 1.2 Summary of plasma and reference tissue approaches.

Input function Model name ROI Parametric Optimisation When to use? Notes Analysis Mapping

PLASMA Compartment   NLLS When a plasma input function is available.

Graphical   LLS To obtain voxel-level maps of Subject to bias at high noise levels, may

(Logan, Patlak) parameters of interest (푉푇 or 퐾퐼), best require pre-smoothing. when a plasma input function is available.

SA   NNLS Can only be used in original form with a plasma input function.

REFERENCE SRTM   NLLS/basis When only a reference tissue input is SRTMV can be used in patients with vascular function available, there are reduced flow rates characteristics. (n.b. reference or large variations in 푉 . region selected 푇 anatomically or Graphical   LLS Produces relatively robust estimates but is through (Logan, Patlak) better with a plasma input function. supervised/ RS-ESA, DEPICT   LLS, basis When only a reference tissue input is DEPICT shows slightly reduced sensitivity but unsupervised pursuit available (although can also be used performs better than reference Logan in high clustering) with a plasma IF). noise conditions.

NLLS = Non-Linear Least Squares, LLS = Linear Least Squares, SA = Spectral Analysis, NNLS = Non-Negative Least Squares, RS-ESA = Rank Shaping Exponential Spectral Analysis, DEPICT = Data-driven Estimation of Parametric Images based on Compartmental Theory

1.9 Summary and rationale for study

PET is an invaluable tool in the diagnosis and staging of a variety of medical conditions. The field of neuroinflammation is no exception; the extensive amount of research into how best to image inflammation using PET over the last four decades is evidence of the clinical relevance and requirement for a reliable and sensitive tracer. Going hand in hand with this need is the necessity to accurately quantify the data from these tracers and to understand the potential limitations of the information derived from using them. Above is a summary of some of the methodological aspects related to quantification of PET images, with specific emphasis on TSPO imaging. [11C]-(R)- PK11195, the most widely used TSPO-PET tracer in research, is known to suffer from a high level of non-specific binding and troublesome radiochemistry. In the interests of addressing these issues, several so-called second generation tracers have been developed, with some exhibiting varying levels of improved SNR and non-specific binding compared to [11C]-(R)-PK11195 in humans. Each new tracer requires an in depth characterisation in a variety of different models of disease, ideally reflecting different levels of TSPO expression, before it can be considered for widespread use, either in research or in the clinic.

This project, designed in conjunction with GE Healthcare, aimed to characterise the behaviour of a new second generation tracer for TSPO, [18F]GE-180, in different models of neuroinflammation. In order to assess the relative efficacy of this new tracer with respect to existing ones, this work involved the collection and analysis of TSPO-PET brain data using second generation tracers, comparing them, where possible, to [11C]-(R)-PK11195, in different preclinical and clinical models of neuroinflammatory disease. A parallel aim of the project was to assess different methods of quantification in these disease models, ultimately proposing a suggested methodological approach to analysis of [18F]GE-180 PET brain data. The hope is that such work will enable the utilisation of such tracers in research and beyond, to produce a reliable and meaningful measure of inflammation in the brain.

2 Comparative evaluation of three TSPO PET radiotracers in a LPS-induced model of mild neuroinflammation in rats

Sujata Sridharan1, Francois-Xavier Lepelletier1, William Trigg2, Samuel Banister3, Tristan Reekie3, Michael Kassiou3,4, Alexander Gerhard1, Rainer Hinz1, Hervé Boutin1

1 Wolfson Molecular Imaging Centre, University of Manchester, Manchester, UK

2 GE Healthcare, The Grove Centre, Amersham, Buckinghamshire, UK

3 School of Chemistry, University of Sydney, NSW 2006 Australia

4 Faculty of Health Sciences, University of Sydney, NSW 2006 Australia

2.1 Abstract

Purpose: Over the past 20 years, the role neuroinflammation (NI) plays in many neurodegenerative diseases, including Alzheimer’s disease, has become increasingly recognised. As such, being able to image NI non-invasively in patients is critical to monitor pathological processes and potential therapies targeting neuroinflammation. The translocator protein (TSPO) has proven a reliable NI biomarker for positron emission tomography (PET) imaging. However, if TSPO imaging in acute conditions such as stroke provides strong and reliable signals, TSPO imaging in neurodegenerative diseases has proven more challenging. Here, we report results comparing the recently developed TSPO tracers [18F]GE-180 and [18F]DPA-714 with [11C]-(R)-PK11195 in a rodent model of subtle focal inflammation. Procedures: Adult male Wistar rats were stereotactically injected with 1μg lipopolysaccharide (LPS) in the right striatum. Three days later, animals underwent a 60-minute PET scan with [11C]-(R)-PK11195 and [18F]GE-180 (n = 6) or [18F]DPA-714 (n = 6). Ten animals were scanned with either [18F]GE-180 (n = 5) or [18F]DPA-714 (n = 5) only. Kinetic analysis of PET data was performed using the simplified reference tissue model (SRTM) with a contralateral reference region or a novel data-driven input to estimate binding potential 퐵푃푁퐷. Autoradiography and immunohistochemistry were performed to confirm in vivo results. Results: At 40-60 minutes post-injection, [18F]GE-180 dual- scanned animals showed a significantly increased core/contralateral uptake ratio vs. the same animals scanned with [11C]-(R)-PK11195 (3.41±1.09 vs. 2.43±0.39, p = 0.03); the same measure in [18]DPA-714dual-scanned animals was not significantly higher than [11C]-(R)-PK11195 (2.80±0.69 vs. 2.26±0.41). There was no significant difference between [18F]GE-180 and [18F]DPA-714 scanned animals (p=0.47). Kinetic modelling with a contralateral reference region identified significantly higher 퐵푃푁퐷 in the core of 63 the LPS injection site with [18F]GE-180 but not with [18F]DPA-714 vs. [11C]-(R)- PK11195. A cerebellar reference region and novel data-driven input to the SRTM were unable to distinguish differences in tracer 퐵푃푁퐷. Conclusions: Second generation TSPO-PET tracers are able to accurately detect mild-level NI. In this model, [18F]GE-

11 180 showed a higher core/contralateral ratio and 퐵푃푁퐷when compared to [ C]-(R)- PK11195, while [18F]DPA-714 did not.

2.2 Introduction

Inflammation is known to play a role in the development and progression of a wide variety of neuropathological conditions, including stroke [166], multiple sclerosis [20] and neurodegenerative diseases [11, 78]. Sensitive and spatially accurate imaging of neuroinflammation (NI) in vivo, however, has long been a challenge. Microglia, which are resident immune cells of the brain, are in a state of ramification and constantly probing against potential insult or injury, even in the healthy brain [12-14]. Once activated, they begin to alter their morphology, becoming increasingly macrophage-like and releasing pro-inflammatory cytokines such as interleukin-1 (IL-1) and tumour necrosis factor-α (TNF- α). Whether this cytokine cascade exacerbates damage or helps to restore equilibrium conditions varies depending on the nature of the insult and with time [15]. The 18 kDa translocator protein (TSPO), a mitochondrial protein which is primarily located in the membranes of mitochondria [23], is expressed at a low level in the healthy mammalian brain. Although the true role of TSPO in NI remains to be fully understood, an increase in expression of the protein is well established as a correlate with the level of microglial activation, proliferation and/or level of infiltrated macrophages, and to some extent astrogliosis [20, 44, 66, 69].

Positron emission tomography (PET) studies of neuroinflammation have been conducted for over 30 years. The original and still most widely used radioligand for TSPO in clinical imaging is the 11C-labelled R-enantiomer of 1-(2-chlorophenyl)-N- methyl-N-(1-methylpropyl)-3-isoquinoline carboxamide, [11C]-(R)-PK11195. Despite its application to a wide range of diseases in the clinic, the sensitivity and quantification of this tracer has been hampered by its high level of non-specific binding and low free fraction in plasma in humans, which lead to a poor signal to noise ratio [206]. Additionally, labelling of the ligand with 11C requires an on-site cyclotron and its radiosynthesis is well-known to be troublesome. In recent years, therefore, efforts have been made to develop and characterise so-called ‘second generation’ TSPO ligands with lower non-specific binding, which have exhibited varying levels of success in clinical and preclinical studies. Unfortunately, and unlike [11C]-(R)- PK11195, second generation TSPO ligands are affected by a single-nucleotide polymorphism (rs6971, Ala147Thr) in the TSPO gene which results in variable binding affinity in humans [77] but which has never been reported in animals. One such ligand is [18F]DPA-714 [207], or N,N-diethyl-2-[4-(2-fluoroethoxy)phenyl]-5,7- dimethyl-pyrazolo[1,5-a]pyrimidine-3-acetamide, which was initially evaluated in

65 vivo in a rodent model of acute neuroinflammation using alpha-amino-3-hydroxy-5- methyl-4-isoxazolepropionic acid (AMPA) [208] and then in models of stroke and Herpes encephalitis [113, 209]. Compared to [11C]-(R)-PK11195, the tracer showed lower non-specific binding and improved bioavailability in brain tissue in the AMPA and stroke models, leading to a better signal to noise ratio than [11C]-(R)-PK11195; however, this was not the case in the Herpes encephalitis model. Moreover, a more recent preclinical study using [18F]DPA-714 in a longitudinal model of HIV-1 in transgenic rats showed no significant group differences in tracer uptake between diseased and wild-type animals [210]. Altogether, these reports support the idea that lower and diffuse NI as observed in neurodegenerative diseases is more variable and difficult to detect [113, 208-209]. Nevertheless, the initial evaluation in humans also suggests that this tracer is a promising TSPO ligand [211] and recently a large clinical study in AD patients showed that neuroinflammation was present in both prodromal and confirmed AD patients [212]. Another second generation tracer, [18F]GE-180 (S- N,N-diethyl-9-[2-18F-fluoroethyl]-5-methoxy 2,3,4,9-tetrahydro-1H-carbazole-4- carboxamide), has been used in in vitro, ex vivo and in vivo studies on the detection of microglial activation in acute models of NI [115, 122]. Both studies found an improved signal to noise ratio of [18F]GE-180 over [11C]-(R)-PK11195, but both also used models that induced strong and acute NI (i.e. high dose of lipopolysaccharide (LPS) and stroke) and not the more challenging low/mild NI. Another study in a transgenic AD mouse model, however, did identify increased binding in aged wild-type and AD animals, although this result was semi-quantitative and did not directly compare the behaviour of [18F]GE-180 with other tracers [213].

As mentioned above, most of the previous preclinical studies testing second generation TSPO tracers have primarily used either acute excitotoxic lesion models or models of stroke relevant to the strong levels of NI observed in stroke or brain trauma. In neurodegenerative diseases and models of neurodegeneration however, the level of TSPO expression is much lower and more widespread than in focal lesions [183, 214-215], and is thus more difficult to detect and quantify. Hence, there is a need to develop an easy-to-perform, robust preclinical model, with low levels of TSPO expression that are clinically relevant to those observed in neurodegeneration. To this end, we performed a comparative evaluation of [11C]-(R)-PK11195, [18F]DPA-714 and [18F]GE-180 in rodents injected in the striatum with a low dose (1µg) of lipopolysaccharide (LPS) that induce a lower level of neuroinflammation when compared to previously published dose (50µg [123]) and animals were scanned 3

66 days post-injection which is after the peak of neuroinflammation induced by LPS between 6-12h post-injection [122-123].

Kinetic analysis and modelling of preclinical PET data is difficult; as well as being technically challenging, arterial sampling of rodents is terminal, while reference tissue models using the contralateral hemisphere [208, 216] as input are hampered by the fact that the region may not be devoid of specific binding in certain disease models. Moreover, Folkersma et al. [217] suggest that PET data from any disease model where BBB damage is present cannot be quantified using a reference tissue approach with anatomical reference input. Many studies have resorted to semi- quantitation, including use of standardised uptake values (SUVs) [218] and lesion-to- background ratios [119]. There is no consensus over tracer or model-specific quantification approaches, but previous studies involving a unilateral effect have used the cerebellum [219] or a contralateral reference region [115, 122] together with the simplified reference tissue model (SRTM) or MRTM (multilinear reference tissue model) [124]. However, there is, to date, no such evaluation in a model of low-level inflammation where the target vs. the contralateral reference region is lower and the identification and quantification of the lesion is more challenging.

Indeed, due to the presence of potentially widespread brain inflammation, these anatomically defined reference regions may also contain TSPO in disease. Therefore, data-driven methods have previously been employed in clinical [11C]-(R)- PK11195 brain PET studies to extract reference tissue kinetics in place of anatomically defined regions (see section 2.3.6 for more details and [220] for a review). Here, we sought to adapt one such approach [197] for the analysis of rodent PET brain scans in our model of modified LPS injection with the three different TSPO tracers.

Thus, the overall aims of this study were to i) assess the performance of second generation TSPO tracers [18F]GE-180 and [18F]DPA-714, directly compared to [11C]-(R)-PK11195 in a modified LPS model of mild-level neuroinflammation and ii) test various quantification approaches in this model of NI.

2.3 Materials and methods

2.3.1 Tracer synthesis [11C]-(R)-PK11195, [18F]DPA-714 and [18F]GE-180 (Figure 2.1) were synthesised as described elsewhere [115, 207, 221-223]. Briefly, [11C]-(R)-PK11195 was formed via N-[11C]methylation of an (R)-demethyl precursor [221, 224]; [18F]GE- 180 was synthesised on a FASTlabTM platform by nucleophilic fluorination of an S- enantiomer mesylate precursor [223, 225]; and [18F]DPA-714 by nucleophilic aliphatic substitution of a tosylate precursor [207].

Figure 2.1 Chemical structure of radiotracers used in the study; [11C]-(R)-PK11195, [18F]GE-180 and [18F]DPA-714.

2.3.2 Animals 37 adult male Wistar rats (396±52g, Charles River, Margate, Kent, UK) were used in this study: 22 received an LPS injection, 2 received an AMPA injection, 5 naïve (no stereotactic injection) rats were scanned at baseline as controls with [18F]GE-180 (data re-analysed from that previously published [115]), 4 with [18F]DPA-714 and 4 with [11C]-(R)-PK11195. All procedures were carried out in accordance with the Animals (Scientific Procedures) Act 1986 and the project was approved specifically by the UK Home Office. Animals were kept under a 12 hour light-dark cycle with free access to food and water.

2.3.3 LPS and AMPA administration For all procedures, animals were anaesthetised by isoflurane inhalation

(induction 5% and thereafter 2-2.5%) in O2/NO2 (30%/70%). Animals underwent a stereotactic injection of 1µg LPS (Lipopolysaccharide from Escherichia coli 055:B5, Sigma, ref. L2880, Lot 102M4017V) or 7.5nmol (0.5µl of a 15mM solution) of AMPA (n = 2) in the right striatum via a craniotomy (Bregma: +0.7mm, lateral: -3.0mm, depth: 5.5mm from the surface of the brain) using a 2µl Neuros™ microsyringe (Hamilton,

USA) and micropump (injection rate: 0.5µl/min, UltraMicroPump II® and Micro4® Controller, WPI Inc., USA). Animals were maintained normothermic (body temperature: 36.5±0.5°C, mean±SD) during the surgery through the use of a heating blanket (Homeothermic Blanket Control Unit, Harvard Apparatus Limited®, Edenbridge, Kent, UK).

2.3.4 Scanning protocol Supplementary figure 2.1 presents a summary of study design; full details of the tracer injected doses, specific activities and purities can be found in Supplementary table 2.1. Animals were split into 4 groups: those scanned sequentially with [11C]-(R)-PK11195 then [18F]GE-180 (group 1, n = 6) or [11C]-(R)-PK11195 then [18F]DPA-714(group 2, n=6) and those scanned with only [18F]GE-180 (group 3, n = 5) or [18F]DPA-714 (group 4, n = 5). In groups 1 and 2, animals were scanned on the same day with [11C]-(R)-PK11195 in the morning and an 18F-labelled tracer in the afternoon. Supplementary figure 2.2 is a correlation plot between specific activity and % injected dose of tracer in the inflammatory core at 40-60 minutes. Although the range of specific activities was quite broad, especially for [11C]-(R)-PK11195, this did not significantly affect brain uptake (Figure 2.2, Figure 2.3 and Figure 2.4).

All animals were anaesthetised by isoflurane inhalation (induction 5% and thereafter 2-2.5%) in O2/NO2 (30%/70%) at 3 days post-LPS injection. All tracers were injected intravenously as a bolus through a tail vein. They were then scanned on a Siemens Inveon® small animal PET-CT as described previously [115]. The following data acquisition protocol was used: a CT scan was performed immediately prior to the PET scan for each animal to acquire attenuation correction factors. The time coincidence window was set to 3.432 ns and levels of energy discrimination to 350 keV and 650 keV. List mode data from emission scans were histogrammed into 16 dynamic frames (5 × 1 min; 5 × 2 mins; 3 × 5 mins and 3 × 10 mins) and emission sinograms were normalised, corrected for attenuation, scattering and radioactivity decay and reconstructed using OSEM3D (16 subsets, 4 iterations) into 128 × 128 × 159 images with 0.776 × 0.776 × 0.796 mm3 voxel size. Respiration and temperature were monitored throughout the scans using a pressure sensitive pad and rectal probe (BioVet, m2m imaging corp, USA). Body temperature was maintained through the use of a heating and fan module controlled by the rectal probe via the interface of the BioVet system.

At the end of the PET scan, rats were rapidly decapitated and brains removed and frozen immediately in isopentane in dry ice. Brains were stored at -80°C before

69 being cut with a cryomicrotome into adjacent 20µm coronal slices. Sections were then stored again at -80°C until immunohistochemistry and autoradiography were performed.

2.3.5 Image analysis Images were segmented automatically using local means analysis (LMA) in the BrainVisa/Anatomist framework (http://brainvisa.info, [226-228]). Automatic segmentation was preferred to manually drawn regions of interest (ROIs), which can introduce a user-dependent bias. In short, the LMA algorithm first extracts the whole body of the rodent from the background, before identifying voxels in the core of organs based on the notion that their variation in PET signal should be lower than those voxels at the organ borders, which are more subject to variations driven by physiological movement and spillover from neighbouring organs/vessels. Neighbouring voxels are then identified based on the similarity of their kinetics to these organ ‘cores’. Ultimately, the implementation in BrainVisa segments the whole body image into 200 regions based on the similarity of kinetics of the voxels contained within them. Within the brain (delineated using the CT images), segmented ROIs were then manually selected and labelled under the following scheme: 1) core of the LPS lesion (covering the right striatum and/or the region with the highest uptake), 2) edge 1 (ROI around the core with the 2nd highest uptake of tracer), 3) edge 2 (ROI with the 3rd highest uptake), 4) a contralateral region (ROI with the lowest uptake, used as a reference region in kinetic analysis), 5) cerebellum and 6) skull edges (voxels on the edges of the skull taken to be noise and/or spillover from regions outside the brain, not included in the data analysis) (see top-right corner inset Figure 2.4). Where applicable, a region of 4th highest uptake (edge 3) was also included in ROI segmentation. In the control animals, regions of high and low brain uptakes, as well as cerebellum and skull edges were defined. While using a contralateral region defined by mirroring region 1) might reduce inter-animal variability, the approach does not give any information about the extent of healthy tissue vs. the heterogeneous nature of tissue response to the LPS injection. The method described above allows visualisation of all voxels in the contralateral hemisphere which can be classed as non-pathological. Approximate regional volumes for all tracers were, in order of regions 1) to 5), 86±20mm3, 202±3mm3, 498±114mm3, 268±31mm3 and 217±16mm3.

Previous work published by our group has presented results with partial volume correction (PVC). Although in our previous study with high amplitude response in TSPO expression, group differences were not affected by PVC being

70 applied or not [115], for low level NI models PVC might actually reduce the signal to background ratio and therefore damp the PET signal. This is due to PVC inherently amplifying noisy, high frequency signals to reduce spill-over effects. Therefore, in our model of low/mild NI, applying PVC might be somewhat counter-productive. Having analysed results with and without PVC, here we selected to present raw (uncorrected) data, which showed consistent group differences between tracers.

2.3.6 Data-driven method for the extraction of reference tissue kinetics The method by Yaqub et al. (2012) [197] is a successor to that by Turkheimer et al. (2007) [135], using a pre-defined brain mask to segment all masked voxel PET time activity curves (TACs) into 4 classes (grey matter with specific binding, grey matter without specific binding, white matter and blood). Subsequently, in dynamic PET data, the reference tissue is selected as the tissue class with the kinetic most similar to that of grey matter without specific binding. This method requires segmented MR data for the generation of a (camera-specific) population database. Since the spatial resolution of the Inveon® small animal PET-CT camera is between 1- 2mm, it is not possible to resolve distinct grey and white matter kinetics in the rat brain. Therefore, we replaced the MR-guided tissue segmentation [135, 197] with an automated kinetic data segmentation approach developed for preclinical PET data (LMA) [226-228]. The LMA method, implemented in BrainVisa, produced 3 classes of sufficient size (over 10% of total brain voxels) in the rat brain and with voxels which were located in spatial proximity to each other. We additionally attempted to define more classes (as used in clinical supervised clustering [135, 197]); however, no more than three kinetically distinct classes were visible in the stroke rodent brain (identifiable by TAC shape), suggesting that this was the optimum number for classes definition. This was consistent for [11C]-(R)-PK11195, [18F]GE-180 and [18F]DPA-714.

In our implementation, a population of adult male Wistar rats in a model of focal cerebral ischaemia were used for class definition. Six animals underwent transient right middle cerebral artery occlusion (MCAO) for 60 minutes (data published previously and re-analysed [115]) and were dual scanned with [11C]-(R)- PK11195 and [18F]-GE-180 5 or 6 days later. In addition, 4 animals with MCAO were scanned with [11C]-(R)-PK11195 and [18F]-DPA-714 (data re-analysed from [113]). All data were reconstructed and pre-processed as described above. Brain masks [197] were extracted from the individual CT images for each animal. The classes were defined within the brain as: ‘activated tissue’ represented by the core of the infarct;

‘normal tissue’ from the contralateral ROI; and ‘tissue with intermediate binding’, represented by cerebral tissue located between the core and contralateral tissue and in the cerebellum. As described above, these regions were defined for each animal using the BrainVISA LMA segmentation, and TACs were normalised (by subtracting from each frame its mean and dividing it by its standard deviation, SD, to create a unit input) and averaged to produce a population database. The algorithm was then implemented on the dynamic images using the Matlab code, SuperPK Software (Imperial Innovations, Imperial College London, UK [135, 196]), developed for clinical studies and adapted here for preclinical class definitions, from the LPS animals to produce cluster maps for normal, intermediate and activated tissue, as well as a reference kinetic, based on the class of voxel most kinetically similar to normal tissue identified in the training data set, for input to the SRTM.

2.3.7 PET data modelling Kinetic analysis of data was performed using the SRTM [167] in the PMOD software package (version 3.6, PMOD Technologies Ltd.) for ROIs 1) to 5). The model assumes a 1 tissue compartment arrangement in the target region (lesion) and the reference region. Input functions were derived using the contralateral hemisphere

ROI (region 4), cerebellum (region 5) and the supervised clustering approach. R1 (ratio of tracer delivery in the target ROI vs. that in all respective reference regions) and 퐵푃푁퐷 were calculated for all ROIs.

2.3.8 Immunohistochemistry Brain sections including the striatum from 5 LPS animals which were also scanned in vivo were stained for CD11b and glial fibrillary acidic protein (GFAP) to confirm the presence of microglia and astrocytes respectively. Staining for neuronal nuclei was also completed using an anti-NeuN antibody. Sections from 4 animals were also stained to investigate BBB integrity: visualising tight junctions with claudin-5 and large protein diffusion with IgG.

For all procedures described below, phosphate buffered saline (PBS) at 100 mM was used. Rat brain sections were snap frozen in isopentane and stored at -80°C. Sections were then post-fixed in paraformaldehyde (4% in PBS) for 10 min and washed (3 × 5 min) in PBS. Sections were incubated in 1 mg/ml sodium borohydride in PBS (3 × 5 min) to reduce auto-fluorescence and washed again in PBS (3 × 5 min). They were then permeabilised with 2 hours of incubation in 0.1% Triton X-100 containing 3% normal donkey serum in PBS to block non-specific binding. Without further washing, sections were incubated overnight at 4°C with primary antibodies in

3% normal donkey serum/0.1% Triton X-100 in PBS. Double immunohistochemical staining was performed against glial fibrillary acidic protein (GFAP) with rabbit anti- cow GFAP (Dako, 1:400) and CD11b (OX42) with mouse anti-rat CD11b (Serotec, 1:1000). Adjacent sections from the same animals were stained with mouse anti- mouse NeuN (Chemicon, 1:1000). Sections were then washed (3  10 min) in PBS/Triton and incubated for 2h at room temperature with secondary antibodies (AlexaFluor® 488nm donkey anti-rabbit for GFAP, AlexaFluor® 594nm donkey anti- mouse for CD11b, or AlexaFluor® 594nm donkey anti-mouse for NeuN (Molecular Probes, Invitrogen)), all 1:500 in 3% normal donkey serum/0.1% Triton X-100 in PBS and then washed again (3  10 min) in PBS.

Following the same protocol, brain sections from 2 animals injected with LPS from the in vivo study and 2 animals injected with AMPA (positive control) were incubated with monoclonal rabbit anti-Claudin-5 (Sigma-Aldrich, 1:100) primary antibody and AlexaFluor® 488nm donkey anti-rabbit. To stain for endogenous rat IgG infiltration in the brain parenchyma, adjacent sections were incubated with AlexaFluor® 594nm donkey anti-rat. AMPA-injected animals were of the same strain and comparative age/weight to LPS animals. Supplementary table 2.2 provides a summary of primary and secondary antibodies used during immunohistochemistry.

All sections were mounted with a Prolong® Gold Antifade kit (Molecular Probes, Invitrogen). Images were collected on an Olympus BX51 upright microscope using a 4×/0.13, 10×/0.30 or 40×/0.50 UPlanFLN objectives and captured using a Q- Capture Retiga 6000 camera through Q-Capture Pro Software (Molecular Devices). Specific band pass filter sets were used to prevent bleed through from one channel to the next. Images were processed and analysed using ImageJ (http://rsb.info.nih.gov/ij).

2.3.9 Autoradiography In order to confirm the presence of specific TSPO binding, [18F]DPA-714 autoradiography was performed with 20 μm brain sections of LPS animals (adjacent to those used for immunohistochemistry). Sections were first incubated for 5 minutes in cold Tris buffer (Trizma crystals (Sigma, UK) 50mM adjusted to pH 7.4 at 4°C with 120mM NaCl) and then incubated for 1 hour at room temperature with either [18F]DPA-714 (total binding) (98.7GBq/μmol; 5nM) or [18F]DPA-714 with excess PK11195 (5μM) to assess non-specific binding. Sections were rinsed twice for 2 minutes with cold buffer and then once quickly with cold distilled water, before being dried and exposed overnight to a Phosphor-Imager screen (Fujifilm BAS-1800II,

Raytek Scientific Ltd., Sheffield, UK). Autoradiographs were visualised and analysed using AIDA software (Raytest GmbH, Germany). ROIs were drawn manually on the lesion and the contralateral area and binding is expressed as the intensity of photostimulated luminescence (PSL) per pixel.

2.3.10 Statistical analysis All data are expressed as mean±SD. Paired Wilcoxon signed rank tests were

11 18 used to compare [ C]-(R)-PK11195 and the corresponding F tracer uptake or 퐵푃푁퐷 values obtained in the same animal (dual tracer scans). Non-paired [11C]-(R)- PK11195, [18F]GE-180 and [18F]DPA-714 results were compared using the Mann- Whitney U-test. Non-paired comparisons of tracers were performed as [11C]-(R)-

PK11195[18F]GE-180 (group 1) vs. [18F]DPA-714only (group 4) and [11C]-(R)-

PK11195[18F]DPA-714 (group 2) vs. [18F]GE-180only (group 3) (see Supplementary table 2.1). Comparisons between all animals scanned with [18F]GE-180 and [18F]DPA-714 (n = 11 vs. n = 11) were also performed. Statistics were performed in GraphPad Prism for Windows (v6.04, GraphPad Software, Inc., San Diego, California USA).

2.4 Results

2.4.1 PET imaging All PET images in LPS animals (groups 1 – 4) showed localised increase in tracer uptake (%ID.cm-3) in the core ROI (Figure 2.2, and Table 2.1). Furthermore, the uptake in edges 1, 2 and 3 was lower compared to the core and not statistically different to that in the contralateral side, indicating a gradual decrease in neuroinflammation with distance from the injection site.

Figure 2.2 Average core and contralateral tissue time activity curves (TACs) of LPS- injected animals that underwent dual scans with each tracer.

11 18 18 11 TACs are for (a) [ C]-(R)-PK11195[ F]GE-180 and (b), [ F]GE-180 and for (c) [ C]-(R)-

18 18 PK11195[ F]DPA-714 and (d) [ F]DPA-714 over 60 minute duration of scans. Core/contralateral ratios indicated on the figures are calculated over the 40-60 minute time frame. Dual scan [18F]GE-180 core/contralateral ratios were significantly higher than [11C]-(R)-PK11195 (p = 0.03) (e), while [18F]DPA-714 (f) values were not (Wilcoxon signed rank, p<0.05). Individual animals are represented by different symbols, which correspond between [11C]-(R)-PK11195 and the respective 18F tracer. Data are presented as mean±SD.

Table 2.1 Regional uptakes expressed in %ID.cm-³ and core/contralateral ratio for each group (40–60 minute sum image). Data expressed as mean±SD.

Group Tracer Core Contralateral Core/contra. Cerebellum Edge 1 Edge 2 Edge 3 ratio 11 18 1 [ C]-(R)-PK11195([ F]GE-180) 0.19±0.03 0.08±0.02 2.43±0.39 0.12±0.02 0.14±0.03 0.11±0.02 0.08±0.02 18 [ F]GE-180 0.24±0.08 0.08±0.03 3.41±1.09* 0.13±0.04 0.16±0.04 0.11±0.03 0.08±0.03 2 [11C]-(R)-PK11195( 18 0.20±0.05 0.09±0.03 2.26±0.41 0.14±0.04 0.16±0.05 0.13±0.05 0.11±0.04 [ F]DPA-714) [18F]DPA-714 0.19±0.03 0.07±0.03 2.80±0.69 0.14±0.03 0.14±0.03 0.12±0.05 0.09±0.05

18 3 [ F]GE-180only 0.19±0.08 0.07±0.02 2.76±0.48 0.15±0.04 0.15±0.05 0.10±0.03 0.09±0.03

18 4 [ F]DPA-714only 0.18±0.04 0.06±0.01 2.92±0.43 0.12±0.02 0.13±0.03 0.09±0.02 0.09±0.06

11 *indicates 퐵푃푁퐷 values significantly different from [ C]-(R)-PK11195 (Paired Wilcoxon signed rank test, P<0.05).

2.4.1.1 Non-paired scans For non-paired scans, there was no significant difference in 40-60 minute core or contralateral uptakes between tracers. There were no significant differences between the core/contralateral uptake ratios at 40-60 minutes between [11C]-(R)- PK11195 and either 18F tracer when considering non-paired groups of animals ([11C]-

(R)-PK11195[18F]DPA-714 (2.26±0.41, n = 6) vs. [18F]GE-180only (2.76±0.48, n = 5) or [11C]-

(R)-PK11195[18F]GE-180 (2.43±0.39, n = 6) vs. [18F]DPA-714only (2.92±0.43, n = 5). The unpaired comparison of animals scanned with [18F]GE-180 (n = 11) and animals scanned with [18F]DPA-714 (n = 11) showed no significant difference in the core/contralateral uptake at 40-60 minutes (p = 0.47). Additionally, there were no significant differences between animals scanned with [18F]GE-180only (n = 5) and

[18F]GE-180 (n = 6), nor with [18F]DPA-714only (n = 5) and [18F]DPA-714 (n = 6) (Table 2.1). There was also no significant difference between the contralateral ROI and the edges 1, 2, 3 and cerebellar ROIs (Table 2.1).

For comparison with the control (naïve) animals, all the LPS-injected animals were pooled for each tracer. Average contralateral uptakes in all LPS animals at 40-60 minutes (0.09±0.02%, 0.07±0.03% and 0.07±0.02% for [11C]-(R)-PK11195 (n = 12), [18F]GE-180 (n = 11) and [18F]DPA-714 (n = 11) respectively) were not significantly different from the low uptake regions in control animals (0.09±0.01% (n = 4), 0.08±0.01% (n = 5) and 0.05±0.01% (n = 4) for [11C]-(R)-PK11195, [18F]GE-180 and [18F]DPA-714 respectively), supporting the use of the contralateral ROI as reference region.

2.4.1.2 Dual scans For dual scans, there was no significant difference between all the ROIs studied for all tracers. However, core/contralateral ratios were significantly higher (+40±36%) for [18F]GE-180 than [11C]-(R)-PK11195 (3.41±1.09 vs. 2.43±0.39, p = 0.03), while [18F]DPA-714 values were not (2.80±0.69 vs. 2.26±0.41, p = 0.09, +25±25%; Table 2.1 and Figure 2.2).

2.4.2 Kinetic analysis Previous work [113, 115] has shown that only paired scans can truly account for inter-individual variability; thus, for the kinetic analysis, only dual scan data were considered.

2.4.2.1 SRTM with contralateral reference input

Table 2.2 shows 퐵푃푁퐷 results for dual scanned animals. Average values calculated with a contralateral reference input for animals scanned with [11C]-(R)- PK11195 then [18F]GE-180 were 1.25±0.29 and 1.94±0.75 respectively (+57±57%, p =

11 18 0.03), while [ C]-(R)-PK11195 vs. [ F]DPA-714 dual scan 퐵푃푁퐷 values were 1.11±0.31 and 1.58±0.54 respectively (+43±34%, p = 0.06).

2.4.2.2 SRTM with cerebellar reference input

Overall, using a cerebellar reference input returned lower 퐵푃푁퐷 values than using a contralateral reference input (p = 0.001 for all tracers) and 퐵푃푁퐷 results were not significantly different between tracers (0.60±0.26 vs. 0.84±0.46 for [11C]-(R)- PK11195 vs. [18F]GE-180; 0.46±0.17 vs. 0.43±0.26 for [11C]-(R)-PK11195 vs. [18F]DPA- 714, Table 2.2).

Figure 2.3 Average TACs for cerebellum (black) and contralateral (grey) reference regions for [11C]-(R)-PK11195, [18F]GE-180 and [18F]DPA-714 in LPS animals. At 40-60 minutes, the cerebellar uptake (%ID.cm-3) was significantly higher for all three tracers (p = 0.001) than the contralateral uptake.

2.4.2.3 SRTM with data-driven clustering reference input

퐵푃푁퐷 values with a data-driven reference input function were overall lower than those from a contralateral reference region (p = 0.001 for [11C]-(R)-PK11195 and [18F]DPA-714, and p = 0.002 for [18F]GE-180), but were highly correlated (Spearman’s

78 correlation coefficient ρ = 0.83 and 0.90 respectively for dual scans). For dual scanned animals, [11C]-(R)-PK11195 and [18F]GE-180 values were 0.83±0.22 and 0.63±0.43

11 18 respectively (p = 0.31, n.s.), while [ C]-(R)-PK11195 and [ F]DPA-714 퐵푃푁퐷 values were 0.75±0.20 and 0.49±0.20 respectively (p = 0.03, Table 2.2).

2.4.3 Autoradiography Autoradiographs confirmed the presence of specific [18F]DPA-714 binding in the ipsilateral striatum compared to the contralateral striatum. Quantification of the autoradiography revealed a 5.62±0.72 fold increase in specific binding in the right striatum (ipsilateral) compared to the left (inset bottom right Figure 2.4).

2.4.4 Immunohistochemistry Anti-rat IgG immunohistochemistry will label rat IgG in blood vessels which, as a large protein, should not diffuse into the brain if the BBB is intact. Staining indicated that rat IgG were still localised in blood vessels in the LPS model hence supporting the presence of an intact BBB, whereas in the AMPA model the staining was diffuse in the brain parenchyma and lost altogether in the blood vessels, indicating BBB leakage. This was confirmed by immunohistochemistry of the tight-junction protein claudin-5, as LPS animals showed similar intact vessels in the ipsilateral and contralateral area, while in AMPA-challenged animals, the vessels were clearly damaged (Figure 2.5 a) in the ipsilateral side, indicating a compromised BBB.

Activated microglial cells (CD11b) were observed in large numbers in the core of the inflamed area compared to the contralateral hemisphere, as were astrocytes (GFAP) (Figure 2.5 b). NeuN stained nuclei showed no obvious difference between contralateral and ipsilateral striatum, supporting that there was no neuronal loss induced by the LPS injection (Figure 2.5 c).

Figure 2.4 Co-registered (sum 40-60 minutes) PET-CT images of representative animals. Animals were dual-scanned with [11C]-(R)-PK11195 followed by either [18F]GE-180 (top) or [18F]DPA-714 (bottom). Inset top: example of regions defined by automatic segmentation. Inset bottom: [18F]DPA-714 autoradiographic image showing increased specific uptake in the right striatum.

Figure 2.5 Immunohistochemistry slides performed on sections in LPS and AMPA- injected rodent brains. (a) Claudin-5 (60× magnification) and IgG (20× magnification) staining in ipsilateral and contralateral striatum for LPS and positive control AMPA-challenged animals. Claudin-5 staining revealed no differences between contralateral and ipsilateral sides in the LPS- challenged animals, while for AMPA-challenged animals, there was visible damage to vessels in the ipsilateral side and some disruption in the contralateral hemisphere. IgG appears normal in ipsilateral and contralateral sides of LPS animals but shows signs of leakage in the ipsilateral side of AMPA animals. (b): Staining for microglia (CD11b, red) and astrocytes (GFAP, green), merged and separate at 10× and 20× (white rectangles) magnification in ipsi and contralateral regions; (c) NeuN staining at 20× magnification in core and contralateral striatum.

Table 2.2 푩푷푵푫 values from the simplified reference tissue model in the core ROI with different reference tissues: dual scans. Reference [11C]-(R)- [18F]GE-180 [11C]-(R)- [18F]DPA-714

18 18 region PK11195[ F]GE-180 PK11195[ F]DPA-714 Contralateral 1.25±0.29 1.94±0.75* 1.11±0.31 1.58±0.54

Cerebellum 0.60±0.26 0.84±0.46 0.46±0.17 0.43±0.26

Supervised 0.83±0.22 0.64±0.42 0.75±0.20 0.49±0.20* clustering 11 Data expressed as mean±SD. *indicates 퐵푃푁퐷 values significantly different from [ C]-(R)- PK11195.

2.5 Discussion

2.5.1 PET imaging and ex vivo observations Although results from rodents are not directly transferable to humans due to inter-species differences, preclinical studies are still valuable for assessment and characterisation of tracer behaviour, especially where models can be developed to mimic clinically relevant levels of inflammation. The primary aim of this study was to quantitatively compare the recently-developed second generation TSPO radioligand [18F]GE-180 with the more well-established [18F]DPA-714 and the original [11C]-(R)- PK11195 in a rat model of low/mild focal inflammation. Recently published work has shown that both 18F-labelled tracers have an improved SNR over [11C]-(R)-PK11195 in models of acute inflammation [113, 115, 122-123], but to date, there is no reported investigation comparing these new tracers, nor looking at their performance in a model producing a lower level of inflammation (~2 fold increase in TSPO PET signal), which is more similar to the levels observed in neurodegenerative diseases than stroke [113, 115], excitotoxic lesions [208, 229] or high doses of LPS [123-124] (~4-6 fold increase in TSPO PET signal). The second aim of this study was to assess the kinetic quantitation methodology via use of the SRTM with different reference input functions. Non-terminal preclinical kinetic modelling has so far been limited to semi- quantitation (SUV) or use of an anatomical reference region in unilateral models. In order to evaluate methods that could be used in neurodegenerative diseases, such as AD transgenic models, which display low levels of neuroinflammation and where a contralateral reference tissue does not exist, we have compared here the 퐵푃푁퐷 values obtained with the SRTM and three different reference input functions: 1) a contralateral region, 2) a cerebellar region, 3) a reference kinetic derived from a data- driven supervised clustering approach.

Previous preclinical studies using an LPS model have typically used larger doses of the endotoxin, inducing high-level inflammation [122-123]. Moreover, previous publications did not compare [11C]-(R)-PK11195 with second generation TSPO tracers [123-124] in the more challenging setting of low/mild neuroinflammation and animals were not dual scanned with the tracers [122]. Finally, the comparison between [11C]-(R)-PK11195 and [18F]GE-180 performed by Dickens et al. only used the contralateral reference region as input for modelling.

18 Our study showed that [ F]GE-180 uptake and 퐵푃푁퐷 were significantly higher than those of [11C]-(R)-PK11195, while [18F]DPA-714 were not. Figure 2.2shows the core/contralateral uptake ratios for dual-scanned animals; it is inferable from Figure

2.2 e) that the significant difference seen between [18F]GE-180 and [11C]-(R)-PK11195 may be driven by three particular animals of the six scanned with both [11C]-(R)- PK11195 and [18F]GE-180. These animals are also likely to drive the increased variability observed with [18F]GE-180 over the other tracers. We also performed a comparison in ‘non-paired’ animals individually scanned with different tracers ([11C]-

(R)-PK11195[18F]DPA-714 vs. [18F]GE-180, [11C]-(R)-PK11195[18F]GE-180 vs. [18F]DPA-714, Table 2.1). Interestingly, core/contralateral uptake ratios between 40 and 60 minutes post-injection for unpaired groups were not significantly higher for either [18F]GE-180 nor [18F]DPA-714 when compared to [11C]-(R)-PK11195, nor compared to one another. Altogether, the lack of differences between [18F]DPA-714 and [18F]GE-180 suggest that both tracers provide similar readouts of neuroinflammation.

These observations highlight i) that the differences in sensitivity (if any) between the 3 tracers used in this study are rather small, ii) that, as hypothesised, the inherent variability within an experiment or populations makes quantification more challenging when measuring low/mild level of neuroinflammation and iii) further supports the importance of dual scanning in the same subject to truly compare the performance of tracers both in preclinical and clinical settings.

Ex vivo experiments were performed in order to confirm the in vivo observations. Firstly, immunohistochemistry was performed with CD11b to assess microglial presence in the core of the LPS injection site, GFAP staining for astrocytic activation, NeuN to evaluate neuronal loss and claudin-5 and IgG to assess BBB disruption. Results supported the PET findings; both activated microglia and astrocytes were significantly present in the right striatum compared to the left (contralateral) side. Autoradiography also confirmed the presence of TSPO specific binding in the ipsilateral striatum while the NeuN staining showed no neuronal loss in the right striatum. Finally, claudin-5 staining showed no BBB disruption compared to a positive control AMPA model (Figure 2.5 a).

Although this study was designed to directly compare the three tracers in vivo, comparison of [18F]GE-180 vs. [18F]DPA-714 in the same animal was not practically feasible due to the rapid progression of TSPO expression over 24h as induced by LPS injection, requiring the scans to be conducted within hours, and the constraints imposed by the half-life of 18F and the tracer production. Similarly, considering the difference in tracer pharmacokinetics, the comparison of these 2 tracers by in vitro autoradiography would not have yielded relevant information regarding their in vivo behaviour as PET tracers, hence our comparison of unpaired scans for these 2 tracers.

2.5.2 PET data modelling In the case of a pathologically damaged BBB, the tracer gains access to the brain parenchyma; thus the extent of binding to TSPO is affected non-trivially and should be accounted for in kinetic modelling. Here, ex vivo investigation of the BBB disruption in LPS-challenged animals vs. positive control AMPA-challenged animals showed no significant BBB disruption in the inflamed area with this low dose of LPS when compared to the AMPA treated animals (Figure 2.5 a). Additionally, R1 values calculated using the contralateral reference region were close to unity in the core of the inflamed area for all animals, suggesting that tracer diffusion to the contralateral (reference) and ipsilateral regions was similar. This is not the case in strong excitotoxic or stroke models, which can affect quantification [115].

Compared to the contralateral region, both a cerebellar input and a supervised clustering derived input produced significantly lower 퐵푃푁퐷 values for all tracers. Using a cerebellar input showed no group differences between [11C]-(R)-PK11195 and second generation tracers in paired comparisons, while a cluster-derived input

18 identified significantly lower 퐵푃푁퐷 values with [ F]DPA-714 in animals dual-scanned with [11C]-(R)-PK11195. The blood perfusion of the cerebellum in rodents is different to that in the brain and leads to a different tracer pharmacokinetic (i.e. a higher peak and uptake than in the normal brain/contralateral ROI, Figure 2.3). It is also possible that the cerebellum contains some specific TSPO binding. Together, these two factors would lower the 퐵푃푁퐷 values calculated in the core of the inflamed area by over- estimating the input of the reference tissue.

By the same rationale, a cluster-derived reference region could contain some specific binding, therefore introducing a bias; the MCAO model may not be optimal to define a reference cluster, as the ‘healthy’ tissue could itself contain a compartment of specifically bound tracer, further complicating the derivation of TSPO-free reference voxels in the LPS animal brains. However, this is unlikely since, in a comparison of the region of lowest uptake in control animals to the contralateral reference region in LPS animals, no significant difference in uptake was observed, supporting the use of the contralateral ROI as healthy reference tissue in this model. We applied this approach by reducing and reassigning the number of partitionable classes compared to the clinically established method, producing significantly lower 퐵푃푁퐷 values for all tracers than a contralateral reference input to the SRTM. Clinical studies have used 4 or 6 class supervised clustering; having also attempted to define four classes in the rodent brain (data not shown), we saw no obvious improvement to the 퐵푃푁퐷 results from the LPS animals. This could be due to a number of factors: i) unlike in the stroke model,

84 the lower level of neuroinflammation induced in this LPS model may inherently not be separable into three or more kinetically distinct classes and/or ii) the supervised clustering algorithm as it stands is not fit to use with preclinical images, which have a lower number of voxels to be used to identify classes.

18 Furthermore, [ F]DPA-714 퐵푃푁퐷 values were significantly lower than those from [11C]-(R)-PK11195, while [18F]GE-180 results were not (Table 2.2). We speculate that this difference could be the result of voxels exhibiting pathology-related signal in LPS animals being mistakenly classed as ‘normal’ tissue by the supervised clustering algorithm. In defining classes in the training data set, we were unable to identify a blood pool class such as that used in clinical supervised clustering [135], as the venous sinus in the rodent brain is smaller in diameter than the resolution of the PET camera, and external derivation (from, for example, the left ventricle of the heart) was not possible due to field of view limitations. The intermediate binding class defined instead, while reliably identified by the LMA brain segmentation, was kinetically not orthogonal to the ‘normal’ tissue class. It is therefore mathematically probable that the supervised clustering algorithm mis-identified some ‘intermediate’ binding voxels as normal tissue. In particular, if [18F]DPA-714 is more sensitive to these effects (i.e. if the voxels classed as ‘normal’ tissue are more contaminated by TSPO-related signal, through inherently increased levels of microglial activation in these animals and/or higher specificity for TSPO of [18F]DPA-714 compared to the other tracers, it may explain the significant difference seen when estimating 퐵푃푁퐷 with the supervised clustering approach. Indeed, the fact that the core/contralateral ratio of dual-scanned [18F]DPA-714 and [11C]-(R)-PK11195 animals was not significant (while for [18F]GE- 180 it was) supports this hypothesis, since a higher signal in the contralateral (reference) region could also be due to higher specificity of the tracer for TSPO. In order to assess specificity of binding, we would need to perform autoradiography studies using [18F]GE-180 and [11C]-(R)-PK11195 for comparison with [18F]DPA-714.

Overall, these results suggest that i) further work on the development and implementation of this method for preclinical imaging should be performed and ii) the automatic segmentation of ROIs as implemented in the BrainVisa platform or PMOD® is currently the best approach, although unfortunately best suited for models with clearly identifiable rather than diffuse areas of neuroinflammation.

A recent study by Ory and colleagues using [18F]DPA-714 [124] validated the use of a reference tissue model with the contralateral striatum as input, producing

퐵푃푁퐷 values which correlated well with a two tissue compartment modelling

85 approach. Interestingly, using a cerebellar input function to the SRTM in their high- dose LPS model produced 퐵푃푁퐷 results which were also highly correlated with blood modelling, albeit with higher bias and producing 퐵푃푁퐷 values lower than any other reference region, similarly to what we observed in the present study. Discrepancies in these results, as well as those using the data-driven supervised clustering approach, which was unable to identify a suitable class of reference voxels, are most likely related to differences in the inflammatory responses induced by high (50µg) and low dose (1 µg) of LPS. However, high doses of LPS inducing a neuroinflammatory response similar to what is observed after acute neurodegeneration are not representative of pathological conditions with lower neuroinflammatory levels, such as neurodegenerative diseases. Interestingly, the study also used a 90 minute imaging window with this high-dose LPS injection; in the current study, we were able to show differences between tracers by the 60 minute time point, suggesting that it is not necessary to scan for longer time periods with a low-dose LPS model or in a stroke model [113, 115]. Furthermore, it is well described in the TSPO imaging literature [230-232] that models or clinical cases in which a strong neuroinflammation (such as in stroke) are much easier to model consistently, whereas in the case of low or subtle neuroinflammation levels as observed in neurodegenerative diseases, the modelling is far more challenging, leading to discrepancies between studies. Considering the failure of the data-driven supervised clustering approach presented here, it is clear that further work is needed to determine a reliable reference tissue allowing the accurate quantification of TSPO tracers in preclinical studies (without the need of performing arterial blood sampling) when models of neurodegeneration such as transgenic mice or rats are used.

2.6 Conclusions

The second generation TSPO-PET radiotracer [18F]GE-180 showed an increased signal to noise ratio over [11C]-(R)-PK11195 when considering dual scans whereas [18F]DPA-714 did not. However, the comparison of unpaired scans did not reveal such differences. The discrepancy between the statistical group results from paired and non-paired comparisons highlights the importance of performing dual scans for direct tracer comparison. Moreover, no significant differences were found between [18F]DPA-714 and [18F]GE-180. Altogether, these results suggest that when measuring low level of neuroinflammation, the differences between second generation tracers (between them or when compared to [11C]-(R)-PK11195) are small in terms of sensitivity, while they retain their attractiveness in terms of off-site use

86 due to the longer 18F half-life. Cerebellar and data-driven inputs to the SRTM were unable to produce comparable 퐵푃푁퐷 values to those using a contralateral reference region. The choice of the contralateral ROI as reference tissue was supported by the fact that comparison between tracer uptakes in the contralateral ROI and healthy animals showed no significant difference and that ex vivo staining showed that the contralateral ROI contained very few activated microglia. Our study however also highlighted the remaining issue that when a low level of neuroinflammation is present and no anatomically-defined reference region can be identified, there is still no well- established method to identify a reference tissue in preclinical studies.

Supplementary materials

Supplementary figure 2.1: Distribution of animal numbers per tracer and scanning group. Animals were dual-scanned with [11C]-(R)-PK11195 (n = 6) in the morning and [18F]GE-180 (n = 6) in the afternoon or [11C]-(R)-PK11195 (n = 6) in the morning and [18F]DPA-714 (n = 6) in the afternoon. A further n = 5 animals each were scanned with [18F]GE-180 or [18F]DPA-714 only.

Supplementary figure 2.2: Specific activity at injection time vs. % injected dose per cubic centimetre in the core of the LPS injection site. Figures shown are for a) [11C]-(R)-PK11195, b) [18F]GE-180 and c) [18F]DPA-714. r2 values were, respectively, 0.01, 0.00 and 0.34; showing no strong relationship between the injected specific activity of each tracer and the corresponding uptake in the core of the LPS site.

Supplementary table 2.1 Summary of injected doses, specific activities and injected amounts for each tracer (mean±SD).

Group Injected dose (MBq) Specific activity at injection (GBq/µmol) Injected amount (nmol) Mean Range Mean Range 11 18 1 [ C]-(R)-PK11195([ F]DPA-714) 33.96±6.20 113.95±66.7 2.84-192.04 1.70±3.48 0.18-8.80 [18F]GE-180 29.26±5.35 149.92±98.13 68.58-344.13 0.24±0.10 0.09-0.35

2 [11C]-(R)-PK11195 18 32.79±11.69 165.43±106.28 61.53-304.59 0.27±0.16 0.11-0.53 ([ F]GE-180) [18F]DPA-714 29.26±9.25 102.94±80.66 25.93-213.91 0.45±0.37 0.18-1.14

18 3 [ F]GE-180only 32.37±6.40 132.45±99.42 64.07-300.38 0.35±0.20 0.10-0.57

18 4 [ F]DPA-714only 34.77±6.91 58.57±40.98 13.41-109.53 1.00±0.82 0.34-2.22

Supplementary table 2.2 Primary and secondary antibodies used for immunohistochemistry Staining Primary antibody, source (concentration) Secondary antibody (concentration) CD11b Mouse anti-rat, AbD Serotec (1:1000) AlexaFluor® 594nm donkey anti-mouse (1:500) GFAP Rabbit anti-cow, Dako (1:400) AlexaFluor® 488nm donkey anti-rabbit (1:500) NeuN Mouse anti-mouse, Chemicon (1:1000) AlexaFluor® 594nm donkey anti-mouse (1:500) Claudin-5 Rabbit anti-Claudin-5, Sigma-Aldrich (1:100) AlexaFluor® 488nm donkey anti-rabbit (1:500)

3 Characterisation of a blood sampling system for generating [18F]DPA-714 TSPO-PET plasma input functions in the rat

brain

Sujata Sridharan1, Rainer Hinz1, Alexander Gerhard1, Michael Kassiou2,3, Elizabeth Barnett1, Hervé Boutin1

1 Wolfson Molecular Imaging Centre, The University of Manchester, Manchester U.K.

2 School of Chemistry, University of Sydney, NSW 2006 Australia

3 Faculty of Health Sciences, University of Sydney, NSW 2006 Australia

3.1 Abstract

Introduction: The 18 kDa protein (TSPO), expressed by activated microglia in the central nervous system, is a marker of inflammatory brain disease. Because arterial sampling is both technically difficult and a terminal procedure, preclinical kinetic modelling in TSPO positron emission tomography (PET) is challenging, especially in circumstances where inflammation is low or diffuse, and/or when no TSPO-free anatomical reference region exists in the affected brain. Here, we set up and characterise an arterial sampling system for preclinical rat studies and assess the potential use of an externally derived reference tissue from healthy animals in a model of low dose lipopolysaccharide (LPS) injected into the striatum. Methods: Adult male Wistar rats were stereotactically injected with 1µg LPS in the right striatum. Three days later, the femoral artery and vein were cannulated for blood sampling with a Twilite™ Swisstrace® for the duration of a 60-minute [18F]DPA-714 PET scan. Naïve animals of similar age and weight were also scanned. Regions of interest were defined, including the core of the injection site and a contralateral region in LPS animals and regions of low and high uptake in control animals. Low uptake regions were averaged and used to generate a dose and weight adjusted population-based ‘tissue’ time activity curve for each LPS animal for input to the simplified reference tissue model (SRTM). Non-displaceable binding potentials (퐵푃푁퐷) were compared to results from a contralateral reference ROI input into the SRTM and to results from plasma input compartment modelling. Results: The two tissue compartment model (2TCM) proved the optimum fit to all brain regions. 퐵푃푁퐷, calculated as the ratio of distribution volume between target and reference (contralateral and population-based) tissue (DVR - 1) in the core of the LPS injection site was 2.91±0.39 and 3.82±0.58 respectively. Reference tissue models with both

90 reference inputs produced significantly lower 퐵푃푁퐷 (1.47±0.30 for contralateral input and 1.92±0.23 for population-based input) than the 2TCM (p <0.01), but results were well correlated (ρ>0.78 in all cases). Additionally, 퐵푃푁퐷 from a population-based reference input showed lower bias and variability across all animals and brain regions than a contralateral input. Plasma input functions for rats were subject to large delay and dispersion effects, which were successfully corrected for. Full characterisation of the system is ongoing. Conclusions: Compared to the 2TCM 퐵푃푁퐷, an externally derived population-based reference tissue input to the SRTM produced 퐵푃푁퐷 with lower bias than a contralateral reference region and could be a promising alternative to the latter, particularly in cases where inflammation is low or diffuse.

3.2 Introduction

Inflammation is a normal response to insult or injury and is present in a wide range of clinical disease (see [233] for review). In the central nervous system (CNS), the inflammatory response is primarily mediated by microglia. These non-neural cells of haematopoietic lineage are, in the healthy brain, in a state of constant ramification, scanning their local environment for signs of disturbance [3-4, 12, 14, 234]. Being highly sensitive, microglia respond to even minor changes in the homeostasis of the brain, often before there is any histological change or loss of neuronal function [235]. When such variations are detected, microglia become ‘activated’, a broad term which encompasses a range of phenotypic changes. In the case of pathological changes being detected, this activation of microglia corresponds to their rapid proliferation and the concomitant release of an array of cytokines, chemokines and tissue growth factors such as interleukin (IL) 1β and tumour necrosis factor (TNF) α [236-237]. Simultaneously, the 18 kDa translocator protein (TSPO), previously known as the peripheral benzodiazepine receptor (PBR), is upregulated and overexpressed in activated microglia [238]. The otherwise low expression of TSPO in the healthy CNS offers the potential to study differences between subjects with and without microglial activation.

The 11C-labelled positron emission tomography (PET) radioligand of the isoquinoline carboxamide family, (R)-PK11195, has been used for nearly 3 decades as it binds with good selectivity to TSPO. Unfortunately, the tracer is hampered by its high level of non-specific binding. The broad applicability of a TSPO tracer in neuroinflammatory and neurodegenerative disease has thus fuelled the development of so called ‘second generation’ TSPO-PET ligands with improved pharmacological characteristics. The selective TSPO ligand [18F]DPA-714, first synthesised by James et al. in 2008 [207], showed good specificity for TSPO, and has since been used in several animal models of neuroinflammatory disease, including herpes encephalitis [209], cerebral ischaemia [113, 116] and traumatic brain injury [239], as well as focal inflammatory models using lipopolysaccharide (LPS) [123] and (R,S)-alpha-amino-3- hydroxy-5-methyl-4-isoxazolopropionique (AMPA) [208], where the tracer has shown high SNR. While [18F]DPA-714 and other second generation ligands appear to be promising alternatives to [11C]-(R)-PK11195, unlike [11C]-(R)-PK11195, they are affected by a single nucleotide polymorphism in the TSPO gene [77] which impacts binding of the tracer. In clinical studies, this leads to subjects having to be stratified into high and mixed affinity binding categories, as well as non-binders. While initial

92 results in healthy humans and unstratified subjects are encouraging [155, 240], at the time of writing, there has been only one study published in human subjects with inflammatory pathology using [18F]DPA-714 and incorporating their TSPO binding status [212]. A total of 64 patients with Alzheimer’s disease (AD) and 32 control subjects were dual scanned with the amyloid PET tracer Pittsburgh compound B ([11C]-PIB) and [18F]DPA-714. The authors were able to show a difference between high and mixed affinity binding AD patients with different sub-categories of disease (slow and fast decliners) and amyloid-negative controls. Interestingly, the results suggest more intense microglial activation at the prodromal/early stage of AD, rather than late. Furthermore, [18F]DPA-714 binding was significantly higher in the frontal cortex of prodromal AD and amyloid-positive ‘controls’, which was not directly correlated with amyloid in the same area. Importantly, the authors point out that, due to the lack of arterial sampling in their study, they were unable to fully validate their results from a standardised uptake value ratio (SUVR) with a grey matter cerebellar reference approach.

Quantification of TSPO-PET brain data is notoriously challenging (see [220] for review). Ideally, arterial sampling of subjects is performed to enable generation of a plasma input function for a compartmental or equivalent model. In rodents, this is particularly challenging due to the size of the animals and not feasible in longitudinal studies as femoral artery cannulation is a terminal procedure. Setup and characterisation of an arterial sampling system for rats has to date not been performed on-site in our preclinical imaging department at the Wolfson Molecular Imaging Centre (WMIC). Previously, semi-quantitative [186] or reference tissue model approaches [115, 122] have been employed in their place. It should be noted, however, that using a reference region methodology in TSPO-PET quantification is not appropriate in many cases, due to the universal expression of the protein, particularly in the diseased brain [135]. Ory et al. [124] recently published a preclinical study to validate the use of a reference region approach to quantify PET data from their model of acute focal microglial activation (using a high dose of LPS, 50 µg). The authors suggest that, in this model, the optimum scanning window for animals scanned with [18F]DPA-714 was 90 minutes. They also found that non-displaceable binding potential 퐵푃푁퐷 (calculated as distribution volume ratio DVR–1) from the two-tissue compartment model (2TCM) correlated well with 퐵푃푁퐷 from the simplified reference and multilinear reference tissue models (SRTM, MRTM) with a contralateral reference tissue as input.

Unfortunately, in the cases of AD, multiple sclerosis (MS) or experimental autoimmune encephalitis, for example, inflammation may be widespread, with commonly used reference regions such as the cerebellum also exhibiting upregulated TSPO (see [241-242] for reviews). This limits the applicability of anatomical reference region strategies for quantification of PET data. We have previously used a model of low-dose LPS (1 µg) administered in the right striatum of adult male Wistar rats (see

Chapter 2, [243]), showing that 퐵푃푁퐷 from a cerebellar reference region as input to the SRTM was not well-correlated to estimates obtained when using a contralateral reference tissue. This result was in contrast to the study by Ory et al. [124], who reported good correlation between 퐵푃푁퐷 using the two inputs, and with 퐵푃푁퐷 from the 2TCM. Results from reference tissue methods, however, were significantly lower than those using a blood input [124]. Ex vivo staining for microglia (CD11b) and activated astrocytes (GFAP) in our study also showed that both cell types were present not only at the site of the LPS injection, but also to some extent in the contralateral hemisphere. Although results from our study suggested that the contralateral region was suitable as a ‘pseudo-reference’ input to the SRTM in this model, variability of 퐵푃푁퐷 estimates could be further reduced with the use of an external reference tissue unaffected by the pathology present in diseases such as AD [212].

Our primary research aim was to characterise the setup of an arterial sampling system at the WMIC for use in rodents using [18F]DPA-714 with a population-based reference kinetic derived from a separate group of healthy animals. A secondary aim was to validate the contralateral reference input against the gold- standard of arterial sampling by systematically comparing the results of 퐵푃푁퐷 from both in this model of low-dose LPS in the rodent brain.

3.3 Materials and methods

3.3.1 Animals All animals were housed individually under a 12-hour light/dark cycle with free access to food and water. All surgical procedures were carried out according to the Animals (Scientific Procedures) Act 1986 under isofluorane anaesthesia

(induction 5% and maintained at 2-2.5% in a 30%/70% O2/NO2 mixture). Three days before PET scanning, four adult male Wistar rats (403±14g, mean±standard deviation) were injected sterotactically with 1µg lipopolysaccharide (LPS from Escherichia coli 055:B5, Sigma, ref. L2880, Lot 102M4017V) in the right striatum via a craniotomy (Bregma: +0.7mm, lateral: -3.0mm, depth: 5.5mm from the surface of the brain). Additionally, four naïve animals were scanned under the same conditions to serve as control for the derivation of an external population-based reference tissue.

3.3.2 PET data acquisition [18F]DPA-714 was synthesised as described elsewhere using nucleophilic aliphatic substitution of a tosylate precursor [207].

Animals were scanned on a Siemens Inveon® small animal PET-CT with the following acquisition protocol: a 10 minute CT scan was performed immediately prior to the 60-minute dynamic PET scan to acquire attenuation correction factors. [18F]DPA-714 was injected as a bolus through a tail vein (Jelco® 24G catheter) with an average injected dose of 34.82±6.86 MBq (mean±standard deviation, SD), range 25.01-40.18MBq for LPS animals and 33.37±5.73MBq, range 28.62-40.82MBq for naïve animals (not significant). List mode PET data were histogrammed into 26 frames (15 × 5s, 3 × 15s, 3 × 60s, 2 × 300s and 3 × 900s) and reconstructed using an OSEM 3D iterative algorithm (16 subsets, 4 iterations). Images were of matrix dimensions 128 × 128 × 159 with 0.776 × 0.776 × 0.796 mm3 voxels. Respiration and temperature were monitored throughout using a pressure-sensitive pad and rectal probe (BioVet, m2m imaging corp., USA). Body temperature was maintained through a heating/fan module connected to the rectal probe and controlled via the BioVet interface. Immediately after the PET scan, animals were rapidly sacrificed by decapitation.

3.3.3 Arterial sampling The femoral vein and artery of each animal were cannulated with a 26G catheter (Terumo® Surflo®-W) connected to fine-bore polyethylene tubing (inner diameter: 0.58mm, outer diameter: 0.96mm) and passing through a Twilite™

Swisstrace® blood sampler (swisstrace GmbH, Switzerland) in an arterio-venus shunt of 60cm length in total. This sampler contains a dual LYSO detector system (detector head), with tubing carrying arterially sampled blood running between the two crystals. Blood flow was maintained by a peristaltic pump (350 µl.min-1) sampled continuously from the femoral artery for the duration of the scan. Discrete whole blood samples were also taken over the duration of the scan. A foot switch was used to record the timings of the discrete blood samples. Figure 3.1 shows this experimental setup in more detail.

Figure 3.1 Photograph of the experimental setup for arterial blood sampling of rats. The figure shows key components including Twilite™ Swisstrace® detector system and femoral artery/vein shunt in a closed sampling loop.

After passing through the Twilite™ detector head, discrete blood samples of approximately 350 µl were collected at 2, 5, 10, 30 and 60 minute time points during the scan. These samples, collected in heparinised Eppendorf tubes, were placed immediately on ice while a 50 µl aliquot of whole blood was counted using a γ-counter (1470 Wizard Automatic Gamma Counter, Perkin Elmer, UK), before being centrifuged at 8050×g to obtain plasma. An aliquot of plasma (200 µl) was then mixed with ice cold acetonitrile (1:1 v/v) and centrifuged again at 8050×g to precipitate plasma proteins. Total free radioactivity was measured using an organic precipitation by centrifugation technique. In brief, an aliquot of plasma (150-200 µl post-precipitation) was injected manually into a HPLC ACE 5 Phenyl (150×4.6 mm) separation column (Advanced Chromatography Technologies Ltd., UK) with an isocratic method of 20 minute run time. In the mobile phase, 40% acetonitrile and 60% water was used at a flow rate of 1.2ml.min-1. Counts were measured using a bespoke in-house NaI well

96 detector with a 1ml loop. Ultraviolet (UV) absorbance was measured with a UV/vis detector (Shimadzu SPD-20A Prominence) at a wavelength of 254nm with a universal chromatography interface (UCI) to convert between the electronic signal and digital output. Peaks were manually identified on HPLC chromatograms and area under the curve was integrated in Shimadzu LC Solutions software, finally expressed as a percentage of total peak area. Cold reference standard was also spiked into some samples and used as an internal reference to confirm the relative retention time of the parent peak.

3.3.4 Dispersion correction Dispersion of tracers during continuous arterial sampling is a recognised effect which is the result of inhomogeneous behaviour of the tracer in the tubing and discrepancy between this and the blood vessels in vivo. The effect may also vary in magnitude depending on the lipophilicity or ‘stickiness’ of the tracer to the material of the tubing, as well as other physical properties of the tubing, including its diameter. The consequence of this dispersion is the ‘spreading’ of the continuously measured blood, resulting in an underestimated arterial input function. Consequently, parameters from compartmental modelling which are directly dependent on blood flow, most notably including 퐾1 and 푉푏, are overestimated to a degree determined by the level of dispersion. Given the novel experimental setup and unknown level of dispersion of [18F]DPA-714 through this tubing, we elected to also perform a dispersion correction for the arterial sampling system for the accurate calculation of an input function for kinetic modelling.

One dispersion correction method which limits the noise addition of other techniques [244-245] is that of Munk et al. [246]. This method, originally developed in the pig, involves collecting arterial samples before and after tracer administration to create a step function, which is then sampled continuously between the two separate beakers and changes in concentration measured over time (퐶표(푡)). The ‘true’ 퐶 (푡+푇) 훼푘 푡 undispersed blood activity is calculated as 퐶 (푡) = 표 − ∫ 퐶 (푡′ + 푖 1−훼 (1−훼)2 0 표

푘(푡−푡′) − 푇)푒 1−훼 푑푡′, where 푇 is the transit time, 푘 is a rate constant linked to the ‘stickiness’ of the catheter tubing and 훼 is the fraction of molecules which interact with the tubing (‘stagnant fraction’). Once 푘 and 훼 are determined, the undispersed arterial input function can be calculated. With the already large technical burden of arterial sampling in rodents, we elected to measure the dispersion in a different set of experiments (in vitro) using the blood of 5 naïve rats pooled together. The experimental setup is shown in Figure 3.2. Blood was taken by cardiac puncture 97 immediately before euthanasia and collected in 15ml heparinised falcons. The total volume of blood was then split into two equal volumes; into one of which [18F]DPA- 714 was added and mixed thoroughly to obtain a concentration of 300 kBq.ml-1. In order to measure the impact of the length of tubing, various lengths between 30 and 60 cm of PE tubing (which was also used in the in vivo setup was prepared as shown in Figure 3.1) were tested. Initially, tracer-free blood was run through the tubing with the peristaltic pump (350 µl.min-1) to the Swisstrace® measuring head and into a waste falcon. After approximately 30 seconds of measurement of tracer-free blood with the Twilite™ system, the tubing was quickly transferred to a falcon containing blood with tracer (300 kBq.ml-1), clamping the tubing to prevent air bubbles from forming, and sampled for approximately 90 seconds before being switched quickly back to tracer-free blood for a further 30 seconds. We elected not to use a three-way tap such as that used by Munk and colleagues, since this adds a large dead volume which was not present during the in vivo PET experiments, as well as an increased chance of introducing air bubbles to the system.

Figure 3.2 Experimental setup of dispersion correction apparatus. Whole blood from rats is collected in two falcon tubes, with [18F]DPA-714 added to one. Sampling is started from the falcon without tracer first, then quickly switched to the falcon with tracer for approximately 90 seconds, before being switched back to the tracer-free falcon. This produces a step function measured by the Swisstrace® measuring head and Twilite™ system.

3.3.5 PET image analysis Reconstructed PET images were co-registered with CT images for delineation of the skull contour. PET images were then automatically segmented with a local means analysis (LMA) algorithm in the BrainVisa/Anatomist framework

(http://brainvisa.info), which was preferred to manual definition of regions of interest (ROIs) for elimination of user-dependent bias. Segmented regions were labelled and quantified as follows: 1) core of the LPS inflammation (region of highest uptake) in the right striatum, 2) edge 1 (region of 2nd highest uptake, adjacent to core), 3) edge 2 (region of 3rd highest uptake), 4) a contralateral region (with lowest uptake), 5) cerebellum and 6) skull edges (voxels located at the edge of the skull, categorised as noise). The latter two regions were not included in data analysis.

Naïve animals (n = 4) were scanned under the same protocol but without blood sampling. Sham operated animals were not used since the effects of NaCl injection have been shown to be negligible [122, 124]. PET images from control animals were segmented and ROIs labelled under the following scheme: 1) high uptake region, 2) low uptake region, 3) cerebellum and 4) skull edges. The high uptake region typically encompassed the periventricular regions as previously reported with other TSPO tracers [186].

3.3.6 Kinetic modelling In order to generate an input function for modelling with the SRTM, the low uptake regions from naïve control animals were averaged and expressed as a percentage of injected dose and Kleiber-corrected weight [247-248] of each animal. An input function was then calculated for each LPS animal by correcting this population-based average low uptake time activity curve (TAC) for their Kleiber- weights and injected doses.

Arterial blood data from online sampling was decay corrected and calibrated for background counts using PMOD software (v3.6, PMOD Technologies Ltd., Switzerland). This was then scaled by the plasma over blood (POB) ratio, fitted to a straight line function of the form 푃푂퐵(푡) = 푝0(푡) + 푝1, and multiplied with the parent fraction of [18F]DPA-714 in discrete plasma samples to produce a continuous input function covering the duration of the scan. PET data was quantified on a regional level using the one and two-tissue compartment models (1TCM, 2TCM) with free blood volume 푉푏. Reference tissue modelling was also used (SRTM) with the contralateral region (contralateral) and population-based (population) input function. Parameters of interest were the 1TCM and 2TCM total volume of distribution, 푉푇, and non- displaceable binding potential, derived from plasma input models as 퐵푃푁퐷 = 푡푎푟푔푒푡 푉푇 푟푒푓푒푟푒푛푐푒 − 1 (i.e. distribution volume ratio DVR–1 [249]). 퐵푃푁퐷 was evaluated with 푉푇

푟푒푓푒푟푒푛푐푒 푐표푛푡푟표 푉푇 as the contralateral region for each animal (퐵푃푁퐷 ) and with the 푝표푝 population-based input function derived from control animals (퐵푃푁퐷 ).

Between-model comparisons were made using the Akaike Information Criterion (AIC) [149]. Since the sample size is small in this preclinical observational study, the non-parametric paired Wilcoxon sign rank test and Bland-Altman plots were used for comparison of results from individual kinetic approaches and the Mann-Whitney U-test was used for statistical comparisons between LPS and control animals and core and contralateral regions (significance at 5%). Statistical tests were performed in Graphpad Prism (v6.04, GraphPad Software, Inc., San Diego, California USA).

100

3.4 Results

3.4.1 Blood data Three metabolites of [18F]DPA-714 were detected in plasma after 60 minutes with respective retention times of 2 minutes, 4 minutes and 6 minutes, with the parent compound eluting at 9 minutes (Figure 3.4). All of these were more polar and all eluted earlier than the parent tracer. This is in agreement with previous reports [208] which also showed hydrophilic metabolites and support the suggestion that only the parent compound enters the brain.

Figure 3.3 Example HPLC chromatograms showing the metabolism over time of [18F]DPA-714. The figure depicts polar metabolites of [18F]DPA-714 at 2 minutes, 10 minutes and 60 minutes post injection of tracer. Metabolites 2 and 3 elute only after 2 minutes. All metabolites have shorter retention times than the parent tracer.

The parent fraction of [18F]DPA-714 in plasma decreased rapidly from approximately 97±2% at 2 minutes to 18±3% at 60 minutes. Figure 3.4 shows, for all animals, the plasma to whole blood (POB) ratios, fitted to a straight line model of the form 푝0(푡) = 푝1(푡) + 푎 and parent fractions fitted to a sigmoidal model of the form 1 푃푎푟푒푛푡(푡) = [250]. [1+(퐴푢)퐵]퐶

101

Figure 3.4 Plasma-over-whole blood (POB) ratios and parent fraction of tracer for individual animals. POB ratios are shown in a), and b) shows the corresponding parent fractions at discrete time points for individual animals. Dashed lines represent averaged curves for the four individual data points fitted to a) a straight line model and b) a sigmoidal model.

The delay (due to the length of arterial cannula tubing) between delivery to the tissue and measurement of arterial blood activity at the Swisstrace™ detector was calculated using the 2TCM as an average of 78.3±1.59 seconds, which was corrected for in kinetic modelling after correction for dispersion of the input function. In dispersion correction, the parameters 훼 and 푘 described in section 3.3.4 were 0.56 and 2.68 min-1 respectively for 60cm of PE tubing.

3.4.2 PET imaging All LPS animals showed localised uptake of tracer (% injected dose, ID.cm-3) at the site of injection (Figure 3.5). The total average core/contralateral uptake ratio at 40-60 minutes was 2.31±0.66 (29% variability). Core uptake was significantly higher than that in the contralateral regions (p = 0.03). Naïve control animals also showed low variability in the low uptake region at 40-60 minutes (19%), with average uptake being 0.05±0.01 %ID.cm-3. Contralateral uptake in LPS animals was not significantly different to the low uptake region in control animals (p = 0.49). Uptake at the core of the LPS injection site was also not significantly different to the high uptake regions of control animals (p = 0.06).

102

Figure 3.5 Calibrated sum 40-60 minute [18F]DPA-714 PET images with co-registered CT images. PET/CT images are shown for an example naïve control animal (left) and LPS injected animal (right). Focal uptake of tracer is seen in the region of the LPS injection site.

3.4.2.1 Plasma input modelling with dispersion correction After correction of the input function for dispersion effects via the method described in section 3.3.4, the average 푉푇 estimates at the core of the LPS injection site were 14.45±2.34 and 16.71±3.96 for the 1TCM and 2TCM respectively, with average standard errors of 18% and 11%. 푉푏 in the core was 13±3% with the 1TCM and 6±5% with the 2TCM. An example of uncorrected and corrected whole blood curves is given 푐표푛푡푟표 in Supplementary figure 3.1. Average 2TCM DVR-1 results were 퐵푃푁퐷 = 2.91±0.39 푝표푝 in the core and 퐵푃푁퐷 = 3.82±0.58.

In addition to having a visually superior fit, AIC was lower with the 2TCM (average of 29 compared to 56 for the 1TCM) indicating an overall preference for this model in terms of fitting, Average standard errors were approximately 10% for both core and contralateral regions. 푉푇 values from the 1TCM were also significantly lower than those from the 2TCM (p<0.01). The 2TCM was thus used as reference standard for comparison of 퐵푃푁퐷 from reference tissue models. Figure 3.6 shows, for an example LPS animal, the fits for 1TCM and 2TCM in core and contralateral regions.

Individual results for 퐾1, 푉푇 and 푉푏 in both the 1TCM and 2TCM are shown in Table 3.1 for all animals in the core of the LPS lesion, contralateral region and population- derived ‘region’.

103

-3 -1 -3 Table 3.1 Summary of 푲ퟏ (ml.cm .min ), 푽푻 (ml.cm ) and 푽풃 results from the one and two tissue compartment models for individual animals in core and contralateral regions and for the population derived kinetic. Results are presented as parameter ± % standard error.

1TCM 2TCM

Core Contralateral Population Core Contralateral Population

K1 VT Vb K1 VT Vb K1 VT Vb K1 VT V푏 K1 VT Vb K1 VT Vb 1 0.88±6 14.45±30 0.14±13 0.64±8 3.30±16 0.10±10 0.43±9 2.92±24 0.12±10 1.46±3 17.26±8 0.05±16 0.84±4 5.92±11 0.07±8 0.67±13 3.77±18 0.09±13 2 0.69±5 11.31±11 0.10±13 0.54±5 4.03±8 0.08±11 0.52±8 3.21±12 0.10±12 1.12±16 11.17±11 0.06±21 1.85±15 4.45±9 0.05±3 0.96±9 3.51±16 0.08±19 3 1.13±5 15.12±9 0.12±16 1.35±8 3.36±7 0.05±25 1.59±17 3.56±19 0.15±17 2.60±4 20.55±11 0.06±18 1.95±5 5.99±13 0.05±5 1.02±11 5.67±33 0.12±15 4 0.45±7 16.93±21 0.18±6 0.37±9 4.27±14 0.15±8 0.36±8 3.11±13 0.16±5 0.92±9 17.86±14 0.11±10 0.73±5 6.39±8 0.10±6 0.55±8 4.60±12 0.14±5

104

Figure 3.6 Plasma input function fits for [18F]DPA-714 with the 1 and 2 tissue compartment models. Dashed (1TCM) and solid (2TCM) lines respectively represent fits to uptake in the core (•) and contralateral (×) regions for an example LPS-injected animal.

3.4.2.2 Reference tissue modelling – contralateral input

퐵푃푁퐷 from the SRTM with contralateral input was 1.47±0.30 in the core with

13% standard error on average across animals. Across all regions and animals, 퐵푃푁퐷 푐표푛푡푟표 푝표푝 values were significantly lower than 퐵푃푁퐷 (p<0.01) and 퐵푃푁퐷 (p<0.01), see section 3.4.2.1.

3.4.2.3 Reference tissue modelling – population-based input

퐵푃푁퐷 using the population-based input function was, on average, 1.92±0.23 in the core of the LPS injection site, with 17% standard error across animals. Results were significantly smaller than both DVR-1 methods (p<0.01 in both cases). 퐵푃푁퐷 values using this population-based input were also significantly higher than those using the contralateral reference input. 퐵푃푁퐷 in the contralateral region was, on average, 0.16±0.14 across all animals.

Considering Bland-Altman plots illuminated these relationships further

(Figure 3.7). Specifically, the population-based 퐵푃푁퐷 values showed a very slight

105 tendency to be higher than those using the contralateral reference region as input to 푝표푝 the SRTM (b), while 퐵푃푁퐷 showed a strong negative relationship with 푐표푛푡푟표 푝표푝 푐표푛푡푟표 퐵푃푁퐷 ; 퐵푃푁퐷 estimates were proportionately higher than 퐵푃푁퐷 estimates when 푐표푛푡푟표 the absolute values were higher (a). Conversely, the relationship between 퐵푃푁퐷 or 푝표푝 퐵푃푁퐷 and 퐵푃푁퐷 using a contralateral or a population-based input to the SRTM showed that the latter methods underestimated values compared to the plasma input model (c) – f)). Quantifying the average percentage difference (bias) between the 푐표푛푡푟표 푝표푝 reference tissue methods and 퐵푃푁퐷 or 퐵푃푁퐷 showed that the contralateral region input to SRTM produced an average underestimation of 77±23% and 62±25% respectively, while the population-based input produced average underestimates of

57±29% and 48±30%. Regional 퐵푃푁퐷 and 푉푇 results are summarised in Table 3.2.

Given that the results of plasma input function modelling indicated that the 2TCM was the preferred fit in all rodent brain regions, the SRTM, which inherently assumes 1TCM behaviour in target and reference regions, may not be the appropriate choice for modelling. We thus also fitted the data to Watabe’s reference tissue model, which is based on the presence of two compartments in the reference tissue [251].

Average 퐵푃푁퐷 in the core with a contralateral input was 1.39±0.27 and with a population-based input was 2.26±0.58. Regional results in all animals from both were 푐표푛푡푟표 푝표푝 significantly lower than 퐵푃푁퐷 and 퐵푃푁퐷 and correlation was relatively poor (ρ<0.70). The method was thus discarded.

Table 3.2 Summary of regional kinetic modelling results across all LPS animals using blood (2TCM) and reference inputs 2TCM SRTM

퐜퐨퐧퐭퐫퐨 퐩퐨퐩 Region 퐁퐏퐍퐃 퐁퐏퐍퐃 Contralateral Population 퐁퐏퐍퐃 퐁퐏퐍퐃 Contralateral - 1.32±0.22 - 0.16±0.14 Core 2.91±0.31 3.82±0.58 1.47±0.30 1.92±0.23 Edge 1 2.19±0.15 2.89±0.42 0.95±0.06 1.20±0.34 Edge 2 1.48±0.08 1.96±0.34 0.42±0.03 0.56±0.27 Cerebellum 2.14±0.11 2.82±0.42 0.98±0.09 1.17±0.38

106

Figure 3.7 Bland-Altman plots comparing all regional results from plasma/reference input in all animals. 푐표푛푡푟표 Figures show all comparisons for DVR-1 methods with plasma input function: 퐵푃푁퐷 and 푝표푝 퐵푃푁퐷 , and from reference tissue modelling: contralateral and population.

107

3.5 Discussion

Preclinical PET imaging is a widely used tool for characterisation of novel tracers and is particularly of interest when the model in use is relevant in terms of clinical pathology. The rodent brain, for example, offers an opportunity to assess not only the inflammatory response to a chemical or physical stimulus, but also the pharmacokinetic behaviour and suitability of tracers such as [18F]DPA-714 as a biomarker for inflammation in this model. It is thus critical to ensure that results from such models are both quantitatively accurate and reliable. Importantly, this requires an appropriate methodological approach to quantitation, which is often hindered preclinically by the complex and terminal nature of arterial sampling in animals. Instead, reference tissue models are frequently used in unilateral lesion models. Unfortunately, there is no truly TSPO-free region in either the human or rodent brain, a fact which hampers quantification even more significantly in models of disease which involve a diffuse or systemic component of inflammation, for example AD (see [252] for review). In such situations, using an external healthy reference tissue could provide a measure of the level of TSPO binding in the diseased brain.

Here, we induced a mild neuroinflammatory challenge by administering a low dose of LPS (specifically in comparison to doses used previously [122-124]) intracerebrally to adult male Wistar rats. The aim of this model was to induce a mild level of inflammation in the rodent brain, which was more subtle than the excito-toxic lesions [123] and stroke infarct [113, 115] models commonly used preclinically, known to induce neuronal death. We have previously shown (Chapter 2, [243]) that a 1 µg dose of LPS triggers activation of microglia and astrocytes, both at the site of injection and in the contralateral hemisphere, but not neuronal death. Furthermore, the level of inflammation induced (core/contralateral uptake) is small compared to stroke. We thus use this as a model of clinical relevance in cases where neuroinflammation is more subtle and the ratio of uptake in affected vs. healthy tissue is lower. We first sought to characterise and set up a preclinical arterial sampling system for use in such studies. We simultaneously used an external reference tissue from healthy rats (population-based) as input to the SRTM and to compare results from this method to those from traditional plasma input functions, as well as the widely used contralateral reference region approach.

108

3.5.1 Characterisation of blood sampling system Since the technical demand of arterial sampling for the duration of a PET scan is high, and since the pharmacokinetic of [18F]DPA-714 has been shown to reach approximate equilibrium after 20 minutes [113, 115], we used a 60 minute scan length rather than the 90 minutes used by Ory and colleagues [124]. We found that metabolism of the tracer was much faster than reported in this study (97% of parent compound at 2 minutes to 18% at 60 minutes vs. 91% at 10 minutes to 56% at 120 minutes), but was in line with that reported in baboons [253] and rats [208]. This could be reflective of an inherent difference between the metabolic speed of different genders of rat (male vs. female, see [254-255] for review), or of radioanalytical laboratory techniques employed at different sites.

Kinetic modelling with a plasma input function is taken to be the preferred method of quantification of TSPO PET data, both in animals and in humans. Due to the invasive and terminal nature of sampling in small animals, the procedure is not commonly used at the WMIC. In order to validate the use of a reference tissue model for quantification of rodent PET data, however, it is necessary to set up a robust method for arterial sampling to derive model parameters for comparison. This has not previously been performed at the WMIC and we aimed to characterise such a sampling system for use in future studies. One of the key issues with continuous blood sampling is dispersion, the effects of which become most noticeable when using a tracer with very rapid kinetics, for example [15O]H2O [256-258]. The tracer kinetics of [18F]DPA-714 have been shown to be relatively slow [124, 207] in the blood and brain, but the level of dispersion of the tracer has not been characterised in a preclinical setup. In order to assess the effects on quantitation with plasma input function models, and given the technical demand in performing dispersion correction experiments to obtain the time constant associated with the phenomenon [246], we elected to perform this experiment in a separate group of rats. Given the definition

퐾1 = 퐹 퐸, where 퐹 is blood flow and 퐸 the extraction fraction and that 퐹 is directly related to the external measurement of arterial blood, large dispersion effects can lead to an overestimation of 퐾1 (i.e. unidirectional rate of transport of tracer from plasma to tissue), and correspondingly, 푉푇. In models with free blood volume fraction (푉푏), this phenomenon also leads to an overestimation of this parameter.

The average regional brain cerebral blood flow, 퐹, in adult rats has been approximated at between 0.89 ml.g-1.min-1 (pons) and 2.82 ml.g-1.min-1 (auditory cortex) [259]. There are, to our knowledge, no values of 퐾1 available in the literature for rats; however, given that the net extraction fraction, 퐸, of all TSPO tracers is 109

-3 -1 considerably lower than 1, one could expect 퐾1 values of less than 2.82 ml.cm .min .

-3 -1 Indeed, Golla et al. report 퐾1 estimates of approximately 0.20 ml.cm .min [155] in healthy humans (similar in AD subjects), suggesting an extraction fraction of approximately 0.4 (assuming a cerebral blood flow of 50 ml.100ml-1.min-1 [260]), although values in non-human primates were in the range of 0.79 ml.cm-3.min-1 [261]. In the latter study, there is no reference to correction for dispersion effects before calculation of the arterial input function, thus we here assume representative values

-3 -1 of 퐾1 of << 2.82 ml.cm .min in cases where no BBB disruption is present (Chapter 2).

Results without dispersion correction are presented in the Supplementary

Materials. Parameter estimates for 퐾1 appeared to be physiologically consistent with predicted values. 푉푏 estimates, however, exceeded 20% in several regions, which is 4 fold higher than the measured and accepted fractional blood volume in the healthy rat (~4.8% [262]), which indicates a large dispersion effect in the original experimental setup. After correction for dispersion, estimates of 퐾1 decreased to between 0.55 and 2.60 ml.cm-3.min-1 in individual animals across the core, contralateral and population- based ‘regions’ with the 2TCM. Correspondingly, the estimates of 푉푏 also decreased to within the expected physiological range. Together, these results suggest that the dispersion correction implemented here was effective at producing an input function which could be used to generate more accurate PET parameters than the uncorrected input function.

A parallel way to investigate the effects of dispersion on the input function would be to estimate the dispersion factor through comparison with an image derived input function from the ventricle of the heart. Although this is outside the scope of the current work, dispersion correction in these ways will be further investigated in the future, since the effects on physiological parameters such as 퐾1 and 푉푏 appear to be non-trivial in this sampling setup.

3.5.2 Kinetic modelling comparisons As well as producing a visually better fit, the 2TCM was AIC-preferred in all regions and produced 푉푇 estimates with lower standard errors than those from the 1TCM. This finding was in keeping with those from Ory and colleagues [124], who examined the use of reference tissue approaches in a high-dose LPS model. 푉푇 estimates in the core were lower than those reported by Ory, as expected considering the lower dose of LPS administered in our model and in spite of the presumed overestimation due to dispersion effects. Due to the terminal nature of arterial blood sampling, we were unable to perform a displacement experiment to assess the

110 specificity of in vivo binding of [18F]DPA-714. Previous displacement experiments, however, in an acute excito-toxic model of neuroinflammation, have shown a reduction of [18F]DPA-714 binding across the whole brain after administration of unlabelled DPA-714 or PK11195, indicating an element of specific binding across the whole rodent brain [208]. Combined with our previous findings of activated microglia in the contralateral hemisphere (Chapter 2, [243]), specific binding in the whole brain is thus also expected to be present in our low dose model of neuroinflammation.

Nonetheless, we investigated the use of different reference tissues to generate

퐵푃푁퐷 by the relationship DVR–1. The use of this definition (termed 퐵푃퐷 by the authors of [249]) inherently accounts for the fact that the reference tissue may contain a small 푐표푛푡푟표 component of displaceable binding. The contralateral region (퐵푃푁퐷 ) and the

푝표푝 푘3 population-based tissue (퐵푃푁퐷 ) were assessed. The calculation of 퐵푃푁퐷 = from 푘4 the 2TCM was significantly higher than either of the indirect DVR–1 plasma input

푘3 methods. Since the classical calculation of 퐵푃푁퐷 is directly derived from the 푘4 microparameters of compartmental analysis, values vary significantly with the estimates of 푘3 and 푘4, and thus should not be taken at their absolute values. 푘 Nevertheless, the fact that 3 was in all cases significantly higher than DVR-1 estimates 푘4 suggests that the latter methods somewhat underestimate the specific binding in target tissue, or more explicitly, confirms that there is specific binding in the reference

푘3 푐표푛푡푟표 푝표푝 tissues used. Furthermore, did not correlate well with 퐵푃푁퐷 or 퐵푃푁퐷 , and 푘4 variability in the former in the core of the lesion was higher than the latter two and for SRTM approaches when averaged across all animals (107% vs. 15% and 16% respectively), in keeping with previous findings [124, 155, 263] and presumably due to inter-individual and inter-regional differences in tracer influx and efflux. This variability suggests that the use of a reference approach, whether using 푉푇 ratios or a reference tissue model, may be more appropriate for group comparisons.

The general trend towards underestimation of 퐵푃푁퐷 by reference tissue methods has been widely reported ([124], see [264] for summary). Our results from reference tissue approaches were indeed significantly lower than plasma input results, but we found good correlation between both methods. There was good correlation between 퐵푃푁퐷 from both reference tissue inputs, however, the novel external reference tissue input from a population of healthy animals produced significantly higher 퐵푃푁퐷 than the contralateral input, particularly at the core of the LPS injection site, suggesting an equivalent or slightly superior ability of this method

111

18 to identify increased [ F]DPA-714 binding in this model. Quantifying 퐵푃푁퐷 in the contralateral region of LPS animals using the population-based input function also indicated a small specific uptake of tracer in the region relative to the theoretical population-based reference ‘tissue’.

In the interests of minimising use of animals, we designed this study as a pilot investigation, performing arterial sampling with n = 4 rats to assess the ability of a population-based reference tissue for quantification of [18F]DPA-714 binding. The use of a contralateral reference input to the SRTM has been previously validated in a high dose LPS model [124]. In our low-dose model, however, a contralateral reference region produced 퐵푃푁퐷 with higher bias and variability than an external reference tissue input to the SRTM. This suggests that the modelling approach in cases of more subtle neuroinflammation is indeed more complex than in excito-toxic models. The use of an external reference tissue holds promise in such models, since the level of specific binding in the healthy brain is very low [33]. Clearly this small sample size has some statistical limitations and lacks power. Given the promising results, however, further validation is suggested, both in a larger sample and with different disease models.

3.6 Conclusion

An arterial sampling system was set up at the WMIC for use in preclinical studies with rats. The effects of external tracer dispersion through this system have been qualitatively identified with [18F]DPA-714. In a model of low level inflammation, an externally derived population-based reference tissue was able to estimate 퐵푃푁퐷 values which were well correlated with those from the 2TCM and with the widely used contralateral reference region input to the SRTM, with lower bias than the latter. Further characterisation into minimising dispersion effects and kinetic validation are needed in a larger group of animals.

112

Supplementary materials

3.7 Results without dispersion correction

-3 푉푇 values from the 1TCM were 17.93±6.43ml.cm at the core of the LPS injection site and 3.61±0.73ml.cm-3 in the contralateral area. Average standard errors 푐표푛푡푟표 on 푉푇 values, however, were over 70% in both regions. Average 퐵푃푁퐷 was 푝표푝 4.69±1.02 in the core of the LPS lesion and 퐵푃푁퐷 was 7.36±2.14. For the 2TCM, 푉푇 was 15.58±9.24ml.cm-3 in the core and 4.52±2.38ml.cm-3 in the contralateral region, 푐표푛푡푟표 푝표푝 with corresponding 퐵푃푁퐷 of 3.35±0.70 and 퐵푃푁퐷 of 3.95±1.30 in the core. Across 푝표푝 푐표푛푡푟표 all regions, 퐵푃푁퐷 were larger than 퐵푃푁퐷 (p<0.01). Average 푉푏 from the 2TCM was 20±4% across regions 1) – 4), with the cerebellum showing the highest average fractional blood volume (27±9%).

In this analysis without dispersion correction, we also found 퐾1 values of between 0.40 and 3.65 ml.cm-3.min-1 with the 2TCM, which were highly variable in both contralateral and core regions of individual animals respectively (see Supplementary table 3.1). Thus, in comparison to the dispersion-corrected results presented in section 3.4.2.1, 푉푏 is overestimated, and while some 퐾1 values fall within the accepted physiological range, there was a large variability in these estimates. These discrepancies in the plasma input modelling suggest that there was indeed some dispersion of the measured blood through the sampling catheter, supporting the requirement for correction for these effects. An example of uncorrected and corrected whole blood curves is given in Supplementary figure 3.1.

Supplementary figure 3.1 Dispersion and delay-corrected (red) and original (uncorrected) whole blood curves as measured by the Twilite™ arterial sampling system for example animal.

113

-3 -1 -3 Supplementary table 3.1 Summary of non-dispersion corrected 푲ퟏ (ml.cm .min ), 푽푻 (ml.cm ) and 푽풃 results from the one and two tissue compartment models for individual animals in core and contralateral regions and for the population derived kinetic.

114

4 Initial evaluation of [18F]GE-180 as a TSPO-PET imaging biomarker in multiple sclerosis

Sujata Sridharan1, Rainer Hinz1, Hervé Boutin1, Alexander Gerhard1, William Trigg2, Christopher Buckley2, David Brooks3, 4, Richard Nicholas5, Joel Raffel5

1 Wolfson Molecular Imaging Centre, The University of Manchester, U.K.

2 GE Healthcare, The Grove Centre, Amersham, Buckinghamshire, U.K.

3 Division of Brain Sciences, Imperial College London, Hammersmith Hospital Campus, U.K.

4 Institute of Clinical Medicine, Aarhus University, Denmark

5 Faculty of Medicine, Department of Medicine, Imperial College London, U.K.

4.1 Abstract

Introduction: Multiple sclerosis (MS) is a progressive neuroinflammatory disorder characterised by demyelination and neuronal loss. PET imaging with ligands for the 18 kDa translocator protein (TSPO), which is upregulated by activated microglia in neuroinflammatory disease, may offer the ability to visualise inflammation in the brains of MS patients. The newly developed TSPO-PET ligand [18F]GE-180 was here used to image a cohort of relapsing-remitting MS (RRMS) patients and healthy control subjects (HV). Methods: Nine highly active RRMS patients (n = 5 high affinity binders, HAB, n = 4 mixed affinity binders, MAB) and ten HV subjects (n = 5 HAB, n = 5 MAB) underwent clinical and cognitive assessment, a series of MR sequences, including Gd- enhancing contrast images, and 90 minute dynamic PET acquisitions with arterial sampling. PET images were co-registered with T1 pre-contrast images and, in MS subjects, lesions were defined using a semi-automated thresholding approach. Anatomical regions, excluding lesions, were also assessed using automatic definition in PMOD. Compartmental modelling with the one and two-tissue compartment models (1/2TCM) was used to investigate the appropriate kinetic modelling methodology in both subject groups. Outcome measures were the total volume of distribution, 푉푇, and standardised uptake values (SUV). Statistical comparisons between RRMS and HV subjects were performed using two-way ANOVA. Results: The 2TCM was found to be the most appropriate fit to data in both HV and RRMS subjects. 푉푇 in normal appearing white matter (NAWM) was significantly elevated in RRMS subjects compared to HV (p = 0.01). There were no significant differences between groups in any other regions.

115

-3 Maximum 푉푇 in lesions was 1.16 ml.cm . In one subject, who exhibited a Gd- enhancing lesion, individual lesion uptake (푉푇 and SUV) was significantly higher than uptake in NAWM. There were no significant differences between HAB and MAB subjects in either RRMS or HV subjects. Conclusion: The 2TCM is recommended for kinetic modelling of RRMS and HV data with [18F]GE-180 in these cohorts. Increased uptake was identified in the NAWM of patients compared to controls, but lesion heterogeneity was also observed; only one patient exhibited lesions elevated in signal over their surrounding NAWM. There were no genotype differences in uptake observed in any anatomical regions in either HV or RRMS subjects.

116

4.2 Introduction

Multiple sclerosis (MS) is a chronic demyelinating disorder of the central nervous system (CNS) broadly characterised by periods of acute relapse followed by remission. Taking all forms into account, MS affects more than 2.5 million people worldwide [265], making it globally the most common neurological disease in young adults [266]. Initially, patients may present clinically with problems with vision and balance, fatigue and muscular pains. The large majority of patients diagnosed with MS are initially in the relapsing-remitting stage of the disease (RRMS), where periods of remission often result in partial to full recovery of clinical symptoms. During the secondary, progressive stage of disease, physical symptoms worsen and disability accumulates.

Pathologically, MS is primarily characterised by inflammatory plaques or lesions in white matter (WM), with a recognised component of cortical demyelination (for review, see [267]). In early stages of the disease, inflammation is closely associated with the development of lesions during acute relapse phases [268-269]. This is accompanied by local BBB breakdown and infiltration of T-lymphocytes and macrophages [81, 83-84] to the tissue, while activated microglia and astrocytes release an array of pro-inflammatory factors, ultimately building a demyelinated plaque of damaged CNS neurons. By the time of remission, however, inflammation has often at least partially resolved. Furthermore, in progressive stages, the extent of inflammation is much lower than during early relapses [94], although clinical status can still degrade. In fact, the nature of inflammation in late stages may change from active infiltration of inflammatory cells through a damaged BBB to a diffuse compartmentalised component behind a repaired BBB [93, 96, 269-270]. MS lesions may be categorised as acute or chronic, depending on the number of infiltrating macrophages and activated microglia; with the former involving dense populations of both, as well as myelin-degrading factors throughout the lesion area, and the latter more typically displaying an inactive core and a rim of activated microglia [94]. Inactive lesions, conversely, show considerably lower levels of leukocytes and activated microglia [271-273]. In fact, the heterogeneity of MS lesions, as well as their associated levels of inflammation, infiltrating macrophages, intensity of demyelination and neuronal damage have conventionally made selection of therapy, especially in the progressive phase, very difficult. Additionally, it is worth noting that targeted anti- inflammatory treatment in early-stage MS is not always successful at slowing the

117 evolution of pathology, including demyelination and neurodegeneration [86-87] and not all patients respond to treatment to the same extent.

Clinical staging of MS usually uses a symptoms-rating scale such as the Kurtzke expanded disability status scale (EDSS) [194], which categorises patients based on a variety of physical measures of disability. As well as laboratory findings, such as elevated immunoglobulin concentrations in cerebrospinal fluid, diagnosis of

MS is often supported by MRI findings, with lesions visible as T2 hyperintensities, T1 hypointensities and/or, in the case of new lesion activity, gadolinium (Gd) contrast enhancement. Although in RRMS patients during relapse, MR measures such as the T2 WM lesion load correlate well with clinical disability scores [101-102], this relationship dissipates in later stages of the disease [104]. Indeed, while histopathology shows the persistent presence of active demyelination [94] and symptoms continue, ‘old’ or inactive lesions appear to become less prominent in MR as MS progresses. Additionally, a component of diffuse inflammation with activated microglia and axonal injury is present in normal appearing white matter (NAWM) and cortical areas in all stages of the disease [92, 274-275]. It is clear, then, that the complex interplay between lesion formation and the role that inflammation plays, both in the development of MS pathology and its clinical progression, requires further elucidation.

Positron emission tomography (PET) may offer insights into the location of inflammation in the MS brain and how this correlates with lesions as identified in MRI. PET ligands for CNS inflammation have frequently targeted the 18 kDa translocator protein (TSPO), which is upregulated by activated microglia. [11C]-(R)-PK11195 is the most commonly used ligand for TSPO and it has been characterised extensively in a variety of neuroinflammatory diseases. So-called second generation TSPO radioligands have showed improved bioavailability and signal to noise ratios over [11C]-(R)-PK11195 in preclinical and clinical studies (see [230] for review). These tracers, however, are sensitive to a single nucleotide polymorphism in the TSPO gene and have thus raised their own challenges. With the identification of different clinical binding group subtypes, studies with these tracers require subject stratification into high, mixed and low affinity binders (HABs/MABs/LABs).

Previous clinical studies in MS subjects have used several of these TSPO tracers with varying results. [11C]-(R)-PK11195 is able to identify increased binding in areas of active disease [20]. Furthermore, in a recent study, the tracer was used to image a cohort of secondary progressive (SPMS) patients [108] and the authors found

118 increased binding in NAWM and periventricular area, suggesting that activated microglia are indeed the source of diffuse inflammation in whole WM. [11C]-PBR28 and [18F]PBR111 are second generation TSPO ligands which have shown good test- retest reproducibility in RRMS subjects and healthy controls [276] with focal increases in uptake correlating well with areas of active disease [112], but also in areas which later became Gd-enhancing in MR [111]. Other tracers such as [18F]FEDAA1106, however, have been unable to identify main effect differences between MS patients with active plaques and healthy controls [277].

[18F]GE-180 (S-N,N-diethyl-9-[2-18F-fluoroethyl]-5-methoxy 2,3,4,9- tetrahydro-1H-carbazole-4-carboxamide) is a recently developed TSPO ligand [223] which has shown a significantly improved signal to noise ratio in preclinical models of acute neuroinflammation [115, 122]. Two recently published papers using [18F]GE- 180 [278-279] have shown that the unconstrained 2TCM is preferred for compartmental modelling of tracer kinetics in healthy subjects. Our primary aim here was to establish the appropriate kinetic modelling approach for [18F]GE-180 brain data in a disease cohort. Secondary aims included assessing the ability of the tracer to show group differences in uptake which may be related to pathology in disease and investigating the use of other methods of quantification.

119

4.3 Materials and methods

4.3.1 Subjects All subjects were scanned according to a protocol which was approved by the Riverside Ethics Committee, London. Nine adult patients (average age at time of PET scan ± standard deviation, SD: 38.5±12.5 years, 7 female, 2 male) with clinically confirmed RRMS were recruited from the Imperial College Healthcare NHS Trust. In addition, 10 healthy subjects with no clinical history and who were not significantly different in age (41.0 ± 9.0 years, 3 female, 7 male) were recruited as controls. All subjects gave full consent prior to commencing the study.

4.3.2 Study timeline Subjects attended a preliminary screening at Hammersmith Hospital, Imperial College London, to assess suitability for the study. At this screening, all subjects underwent a basic medical assessment with screening blood tests to assess TSPO genotype (maximum 10 ml), red blood cell count and renal/bone/liver function (maximum 15 ml). Subjects who had not previously undergone an MR scan were also scanned at this screening. Female participants of child-bearing age were required to provide a urine sample to exclude the possibility of pregnancy.

MS subjects aged between 20 and 50 years with confirmed diagnosis according to the revised 2010 McDonald criteria for RRMS [280], who were not taking any medication which could affect results of the study, and who were otherwise clinically healthy (within one month prior to commencing the study) were considered eligible. Subjects had previously decided (with their clinicians) and been deemed eligible to commence natalizumab treatment, as per National Institute for Health and Care guidelines. These guidelines require the subject to be categorised as having either; ‘Rapidly Evolving Severe MS’ - having had at least two relapses in one year, and with one or more Gd-enhancing lesions (or a significant increase in T2 lesion load); or be part of a ‘Suboptimal therapy group’ - having failed to respond to a full course of interferon beta therapy, with at least one relapse whilst on therapy and having at least nine T2 lesions (or one T1 Gd-enhancing lesion). Subjects were also required to have adequate audio-visual and communication ability in order to give informed consent.

Healthy controls (HV) aged between 28 and 56 years, with no history of neurological or psychiatric disease or drug use and who were not taking any concurrent medication were also recruited. In the case of both MS subjects and healthy controls, LABs were excluded from recruitment.

120

At the second visit, all subjects underwent MRI and PET scans, blood sampling and clinical (e.g. EDSS) and cognitive assessment (including the symbol digits modality test, SDMT, now commonly used as a measure of cognitive impairment in MS [281]). Other measures of clinical disability included the 9-hole peg test (9HPT), which assesses patients’ manual dexterity; requiring them to lift and place, one by one, 9 pegs into 9 holes, before removing them (again one by one), whilst being timed. The test is repeated twice for each hand (dominant and non-dominant). Patients’ time-to-walk 25 ft (T25FW) was also measured (repeated three times).

4.3.3 MRI scanning All subjects underwent T2-weighted scans and T1-weighted sequences on a 3T Siemens Magnetom MR B19 scanner. Table 4.1 summarises the relevant parameters associated with these sequences.

Table 4.1 Summary of MRI acquisition parameters used for MS and HV subjects. MR sequence TR/TE (ms) Slice thickness FOV Voxel size Flip angle (mm) (mm) (mm3) (°)

FLAIR 9000/95 4.0 220 0.9×0.9×4.0 -

DIR 7500/325 1.3 256 1.3×1.3×1.3 -

Pre and post- 2300/2.98 1.0 256 1.0×1.0×1.0 9 Gd SPGR

DTI* 9500/103 2.0 256 2.0×2.0×2.0 -

PSIR* 10000/9.3 2.0 256 0.5×0.5×2.0 150 fMRI* at rest 2000/30 3.0 192 3.0×3.0×3.0 -

FLAIR – fluid-attenuated inversion recovery, DIR – double inversion recovery, SPGR – spoiled gradient echo, DTI – diffusion tensor imaging, PSIR – phase sensitive inversion recovery, fMRI – functional MRI, TR/TE – repetition time/echo time, FOV – field of view.

*DTI, SPIR and fMRI were performed for HV subjects only.

4.3.4 PET scanning [18F]GE-180 was synthesised on a FastLab™ platform as previously described [225]. The tracer was injected as an intravenous bolus over 30-40 seconds at a target dose of 185.0 MBq 30 seconds after the start of dynamic scanning. After 4.5 minute low dose CT scans for attenuation correction, 90 minute dynamic PET scans were performed on a Siemens Biograph 6 with a field of view of 168 × 168 × 148 mm3. List- mode data were histogrammed into 24 frames (6 × 90, 3 × 180, 5 × 600, 5 × 1500 and

121

5 × 3000 seconds) and reconstructed using both filtered back projection with a ramp filter and ordinary subset expectation-maximisation (OSEM) with 4 iterations and 8 subsets, a zoom of 2.6 and a 5 mm FWHM 3D Gaussian filter. The former were used in further analysis. Reconstructed voxel dimensions and spatial resolution were 1.57 × 1.57 × 1.92 mm and approximately 5 mm respectively.

4.3.5 Measurement of arterial plasma All subjects were cannulated under local anaesthetic prior to the start of the PET scan in their radial artery. At the start of the PET scan, a venous blood sample was also taken from each subject to assess blood biomarkers and cytokines. These included IL-6, TNF-α, HsCRP, monocyte chemotactic protein-1, interferon gamma- induced protein-10, macrophage inflammatory protein 1-β, S100β, CD56 natural killer cells, plasma chitotriosidase, anti-natalizumab antibody and anti-neurofilament antibody. At the time of writing, results from peripheral inflammatory markers were not available for analysis.

Continuous sampling was performed at a target rate of 2.5 ml/min for the first 15 minutes. Discrete blood samples were also taken at 4.5, 9.5, 14.5, 29.5, 49.5, 69.5 and 89.5 minutes for metabolite analysis. No more than 300 ml was withdrawn from each participant. Whole blood and plasma total tracer concentrations were measured using a well counter and radiometabolite analysis was performed using high performance liquid chromatography (HPLC). In short, samples were spun down to obtain plasma, which was then added to HPLC-grade acetonitrile to precipitate proteins. Samples were then centrifuged and the solvent layer was rotary evaporated to collect analytes, which were added to 5.5ml of 7% ethanol solution for reconstitution. 15mm syringe filters with a nylon membrane of 0.2µm pore-size were used for filtration of the reconstituted analyte solution. For analysis of consecutive samples, two HPLC systems were used (Agilent 1260 Infinity and Agilent 110 Series systems), both in isocratic mode with direct injection onto a 5ml sample loop comprising a guard column and a Waters µBondapak C18 10µm 125A 7.8×300mm column with a mobile phase of HPLC-grade acetonitrile:water (60:40) and a 3ml.min-1 flow rate. Online UV (230nm) and radioactivity detectors were also used. Individual radiochromatograms typically showed 4 metabolite peaks with approximate retention times of 11 minutes, while the parent peak eluted at approximately 12 minutes. Supplementary figure 4.1 shows a typical [18F]GE-180 HPLC chromatogram. [18F]GE-

180 has a relatively low lipophilicity (logD7.4 = 2.95 vs. 3.01 for [11C]PBR-28 and 4.58

122 for [11C]-(R)-PK11195) and these more polar metabolites are thus not expected to cross the BBB.

4.3.6 Image processing

4.3.6.1 PET images PET images were co-registered with T1 (pre-Gd) MRI using FLIRT within FSL, part of the Functional MRI Brain (FMRIB, Oxford, U.K., build 414). In short, MR images were skull stripped using FMRIB’s brain extraction tool and linearly co-registered to 60-90 minute sum-PET images to obtain a transformation matrix, which was then used in another linear co-registration to the dynamic PET image. SS performed the co- registration; quality of image matching was checked visually by JR and SS. MR images were segmented into grey and white matter maps and CSF in PMOD (v 3.6, PMOD Technologies Ltd., Switzerland) and regions of interest (ROIs) were defined in target space (T1) using the 83-region Hammers atlas [282]. Regions were inspected manually for overlap and edited where necessary to minimise spill-over from large- vessel vascular activity.

4.3.6.2 MR images For each MS subject, lesions were semi-automatically defined by a local thresholding technique, using T1, FLAIR, DIR and PSIR images in an in-house package (JIM, Imperial College London). DIR/PSIR sequences provide improved sensitivity of lesion detection, particularly in cortical areas, over FLAIR alone [283]. All lesions were defined in T1 space. Total lesion loads were also calculated.

4.3.7 Kinetic analysis All kinetic modelling was performed in PMOD. Calibrated continuous and discrete blood data were corrected for decay and the parent fraction in plasma was determined for discrete samples. Plasma over blood (POB) ratios were calculated using discrete whole blood and plasma samples. The parent fraction of tracer in plasma, calculated from metabolite analysis of the discrete blood samples, was fitted to a Hill function with the following equation:

1 푓 (푡) = 푓 . { 퐴푢퐵 } Eq. 4.1 푝푎푟푒푛푡 푝 1−( ) 푢퐵+퐶

This was then multiplied with the continuous whole blood data to produce a metabolite-corrected arterial plasma input function.

123

In order to minimise the effects of partial volume, individual lesions were defined based on a lower volume cut-off of 100 mm3, which is at the approximate limit for the spatial resolution of the Biograph (~5 mm). Brain ROIs were also selected from the Hammers atlas: left and right middle frontal gyrus (frontal lobe grey matter), left and right thalamus, left and right grey matter cerebellum and left and right cerebral whole WM. In the case of MS subjects, NAWM was also defined by first dilating all lesions by a factor of 3 to reduce partial volume effects, and then subtracting them from the Hammers-defined WM. A cortical region was selected to assess a possible signal related to inflammation in grey matter (GM) in MS subjects ([284], see [285] for review). The thalamus is commonly selected as an ROI as TSPO ligands have previously been shown to exhibit specific binding in this region, both in the healthy and diseased brain [240, 277, 286]. Cerebellar grey matter is frequently used as a ‘pseudo’ reference tissue in TSPO-PET studies, as the expression of the protein is low and many tracers show negligible specific binding there [156, 166, 172]. The cerebellum is, however, often subject to MS-related pathology [287-289], including inflammatory lesions in white and grey matter and atrophy (see [290] for review). A unilateral ROI approach was chosen due to the recognised asymmetrical nature of MS [291-292].

Preclinical work and previous pharmacokinetic analysis of [18F]GE-180 have suggested a reversible binding kinetic of the tracer (data unpublished; personal communication, William Trigg). Thus, the one and two-tissue compartment models

(1TCM, 2TCM) with correction for blood delay and free blood volume (푉푏) were fitted to dynamic PET data as per equations 4.2, 4.3 and 4.4 below, with the regional activity concentration of PET tracer measured by the scanner being 퐶푡(푡) = (1 − 푉푏)퐶푖(푡) +

푉푏퐶푏(푡), where 퐶푖(푡) is the total tissue concentration of tracer and 퐶푏(푡) the concentration of tracer in whole blood

푑퐶 (푡) 1 = 퐾 퐶 (푡) − 푘 퐶 (푡) Eq. 4.2 푑푡 1 0 2 1 for the 1TCM, where 퐶0,1 are the concentrations in the plasma and tissue (free, nonspecific and specific) compartments respectively and 퐾1, 푘2 are the rate constants from plasma to tissue and from tissue to plasma respectively.

푑퐶 (푡) 1 = 퐾 퐶 (푡) − (푘 + 푘 )퐶 (푡) + 푘 퐶 (푡) Eq. 4.3 푑푡 1 0 2 3 1 4 2

푑퐶 (푡) 2 = 푘 퐶 (푡) − 푘 퐶 (푡) Eq. 4.4 푑푡 3 1 4 2

124 for the 2TCM, where 퐶1 is now the concentration in the first compartment (free and nonspecifically bound tracer), 퐶2 is the concentration in the specific compartment and

푘3,4 are the uni-directional rate constants from the first to the second tissue compartment and back.

퐾1 The outcome measures were the total volume of distribution, 푉푇 ( for the 푘2

퐾1 푘3 푘3 1TCM and (1 + ) for the 2TCM) and non-displaceable binding potential, 퐵푃푁퐷 ( ) 푘2 푘4 푘4 for the 2TCM). Specifically, 푉푇 was considered since the measure is a more stable

퐾1 macroparameter due to the incorporation of , compared to 퐵푃푁퐷, which is often 푘2 affected by the large variability in the 푘3 and 푘4 microparameters.

60-90 minute standardised uptake value (SUV) images were also produced, with ROI analysis performed in the same regions as for dynamic data.

4.3.8 Model comparison and statistical tests Results from the 1TCM and 2TCM were compared by considering the Akaike information criterion, corrected for small sample size [149-150, 293], as shown in equation 4.5.

푆푆 2×푘×(푘+1) 퐴퐼퐶 = 푛 × 푙푛 + 2 × 푘 + Eq. 4.5 푐 푛 푛−푘−1 where 푛 is the number of data points, 푘 is the number of estimated parameters and 푆푆 is the sum of squares. Statistical group comparisons were performed using two-way ANOVA with genotype and disease state as factors, and significance set at 5%. Statistical tests were performed in GraphPad Prism (v. 6.04, GraphPad software, CA, USA, www.graphpad.com).

125

4.4 Results

4.4.1 MRI findings A summary of subject demographics and characteristics for MS patients are shown in Table 4.2. Across patients, only one Gd-enhancing lesion was observed (subject 7 in Table 4.2). The variety and location of non-enhancing lesions observed was broad, including T1 black holes and thin periventricular bands, indicating a range of disease activity (see Figure 4.1). Total lesion loads were between 2.0 and 21.0 cm3, while individual lesions (post cut-off) varied in volume between 0.1 cm3 and 6.2 cm3.

Table 4.2 Summary of MS subject demographics. Subject Sex EDSS Age at PET Time since MS WM lesion Genotype score scan (years) onset (years) volume (cm3)

1 F 3.0 26.1 2 2.52 HAB 2 F 6.5 46.3 2 0.17 HAB 3 M 4.0 31.2 15 9.82 MAB 4 F 4.0 57.4 42 20.11 MAB 5 M 5.5 33.1 3 2.64 HAB 6 F 3.5 33.8 9 20.71 MAB 7 F 3.0 20.8 6 9.37 HAB 8 F 3.5 45.2 15 3.20 HAB 9 F 5.0 47.0 2 14.83 MAB

Figure 4.1 Examples of types of MS pathology observed in study cohort. In the figure, white arrows highlight a) thin periventricular lesions (FLAIR), b) T1 black hole (T1 pre-contrast), c) FLAIR hyperintensity, d) Gd-enhancing (T1 post-contrast).

126

4.4.2 Blood data POB ratios were calculated at each of the 7 discrete time points and did not change significantly over time. On average over the duration of the scan, the POB ratio for HAB HV subjects was significantly higher than for HAB MS subjects (1.71 ± 0.02 vs. 1.67 ± 0.01, p < 0.01), with the same seen in MAB HV vs. MAB MS subjects (1.67 ± 0.01 vs. 1.64 ± 0.02, p < 0.01), see Figure 4.2). Within both subject groups, HABs had higher average POB ratios than MABs (p < 0.01 and p = 0.01 for HV and MS respectively).

Overall, there was a significant difference in the time it took for [18F]GE-180 to reach a peak in whole blood, with the tracer reaching an earlier peak in MS subjects (58.2s post-injection) compared to HVs (66.4 s post-injection), p = 0.01. When split for TSPO binding status, this difference remained significant only for MAB MS vs. MAB HV subjects. The peak also tended to be on average higher in MS than HV subjects, although this did not reach significance (p = 0.08) irrespective of binding group. Figure 4.3 represents, for all subject binding groups, this relationship between higher, earlier peaks, and lower, later ones.

Figure 4.2 Average plasma-over-blood ratios for HAB HV, MAB HV, HAB MS and MAB MS subjects. In the figure, HAB HV subjects are represented by • with linear fit to the POB ratio represented by a solid red line (y = -0.0002x + 1.7205), while MAB HVs are ○ with linear fit represented by a solid black line (y = 0.0001x + 1.6627). Equally, for MS subjects, ∎ represents HABs (dashed red line linear fit, y = 0.0001x + 1.6610) and ⧠ represents MABs (dashed black line, y = 0.0001x + 1.6357).

Whole brain uptake at 60-90 minutes was significantly lower (p = 0.03) in HAB HVs compared to MS subjects, but not in MABs. Average whole brain volume was not significantly different between groups and variability was low (HAB/MAB: 26/7% and 20/34% for HV and MS subjects respectively).

127

Figure 4.3 Correlation plot of time to peak concentration of [18F]GE-180 in whole blood vs. height of that peak for HV and MS subjects. In the figure, HV subjects are represented as circles and RRMS patients as squares. Data are also split into HAB (closed symbols) and MAB (open symbols) subjects. A trend of sharper, earlier peaks was seen in MS subjects compared to HV subjects (p = 0.08). The time for tracer to peak post-injection was significantly earlier for MAB MS subjects compared to HVs.

4.4.3 Kinetic analysis of PET data The average injected dose of [18F]GE-180 for RRMS subjects was 182.7±4.8 MBq and 182.5±2.9 MBq for healthy controls (not significant). Across all patients, bilateral volumes of ROIs were 14 ± 2 cm3 for the thalamus, 42 ± 3 cm3 for the middle frontal gyrus, 95 ± 8 cm3 for cerebellar grey matter and 324 ± 55 cm3 for NAWM. For HVs, equivalent values were 16 ± 1 cm3, 45 ± 5 cm3, 109 ± 10 cm3 and 409 ± 60 cm3. Figure 4.4 shows the 60 – 90 minute sum-PET images for a HV and age and genotype matched MS subjects with and without a Gd-enhancing lesion.

Figure 4.4 Co-registered T1 pre-contrast MR with sum 60-90 minute PET images. Images are for age and genotype-matched HV (left), MS subject with Gd-enhancing lesion (centre) and MS subject with no enhancing lesions (right).

128

Table 4.3 Average 푽푻 values and standard errors (S.E.) from 1 and 2 tissue compartment modelling for HAB MS and HV subjects. ‘White matter’ for MS subjects is whole WM, including lesions.

MS (n = 5) HV (n = 5) 1TCM 2TCM 1TCM 2TCM

Region 푽푻 % S.E. 푽푻 % S.E. 푽푻 % S.E. 푽푻 % S.E. Grey matter 0.15 28 0.17 32 0.17 16 0.26 79 left Grey matter 0.14 26 0.16 24 0.18 15 0.23 40 right Thalamus left 0.19 42 0.22 32 0.22 22 0.27 34 Thalamus 0.19 37 0.21 28 0.23 31 0.28 56 right Cerebellum 0.17 30 0.20 33 0.20 29 0.24 50 left Cerebellum 0.16 48 0.20 35 0.19 23 0.27 74 right White matter 0.15 31 0.18 28 0.18 15 0.22 26 left White matter 0.14 24 0.23 109 0.18 15 0.22 24 right

Table 4.4 Average 푽푻 values and standard errors (S.E.) from 1 and 2 tissue compartment modelling for MAB MS and HV subjects. ‘White matter’ for MS subjects is normal appearing white matter.

MS (n = 4) HV (n = 5) 1TCM 2TCM 1TCM 2TCM

Region 푽푻 % S.E. 푽푻 % S.E. 푽푻 % S.E. 푽푻 % S.E. Grey matter 0.18 12 0.33 67 0.17 10 0.25 50 left Grey matter 0.18 14 0.29 57 0.16 9 0.29 82 right Thalamus left 0.22 12 0.26 31 0.20 12 0.25 46 Thalamus 0.22 16 0.27 74 0.19 12 0.23 44 right Cerebellum 0.20 19 0.30 85 0.20 22 0.30 78 left Cerebellum 0.19 17 0.32 171 0.18 15 0.26 53 right White matter 0.18 15 0.27 50 0.18 12 0.30 60 left White matter 0.20 16 0.33 267 0.18 16 0.22 23 right

129

As summarised in Table 4.3 and Table 4.4, standard errors (S.E.) for 푉푇 values were generally higher for the 2TCM than the 1TCM for both MS subjects and HVs. This was particularly true in small, noisy ROIs (for example, individual lesions), where 2TCM S.E. were commonly above 100% and the 1TCM was Akaike-selected in 76% of lesions. Supplementary tables 4.1 and 4.2 show the rate constant estimates for RRMS and HV subjects.

Figure 4.5 shows a typical fit for the middle frontal gyrus (left) with the 1TCM and 2TCM in an example HV and MS subject. Akaike values were lower for the 1TCM (i.e. preferred) in 53% of standard anatomical regions across all MS subjects. Across standard regions in each subject, however, the 2TCM showed a lower Akaike value (Figure 4.6), suggesting a slight overall preference for this model. As visualised in

Bland-Altman plots, also shown in Figure 4.6, there was some agreement between 푉푇 from the 1TCM and 2TCM in both MS (Spearman correlation coefficient ρ = 0.80) and HV (ρ = 0.61) groups. However, 2TCM estimates were significantly higher than those from the 1TCM (p < 0.01 in both MS and HV). In spite of the apparent preference for the 1TCM in lesions, as a trade-off between variance and quality of fit (visual, residual sum of squares and AIC) in large, non-noisy regions in both MS and HV subjects, the 2TCM was selected as the preferred model for this data.

Figure 4.5 Example fits for HV and MS subjects with 1TCM and 2TCM. In the figures, dashed lines represent 1TCM and solid lines are for the 2TCM. Fits are in the left middle frontal gyrus (×) for an MS subject (left) and HV subject (right).

130

Figure 4.6 Akaike information criterion and Bland-Altman plots comparing 1TCM and 2TCM in HV and MS subjects. In the figures, a) shows the average regional Akaike information criterion for each subject. While standard errors were high for the 2TCM, mean Akaike across subjects was lower (i.e. preferred) in both MS and HV subjects. (b) and (c) show Bland-Altman plots depicting the relationship between 푽푻 estimates from the 1TCM and 2TCM, split into individual MS subjects (b) and HVs (c). Results from the 1TCM were typically lower than those from the 2TCM.

4.4.3.1 Regional rank order of uptake Of the ROIs under consideration, TACs were highest in the thalamus and grey matter cerebellum, then in the frontal lobe cortical region, with the lowest uptake being in whole white matter. This pattern was also observed in MS subjects, although standardised uptake in NAWM was closer to the level of cortical grey matter in MS compared to HV. This rank order applied to both HAB and MAB subjects.

When considering all brain regions, cortical grey matter TACs, particularly from the temporal, frontal and occipital lobes, frequently appeared highest, with the thalamus approximately middle in rank order and striatal regions (caudate nucleus, putamen and globus pallidus) were universally among the lowest TACs.

While arterial sampling was performed in this study, the technique is not generally favourable as cannulation is a painful and risky procedure for subjects to undergo, and the measurement of plasma samples can be error prone. Therefore, since the 1TCM provided a reasonable (if not preferred) fit in the majority of brain regions, which is an assumption of the simplified reference tissue model (SRTM), we also elected to use the striatum as a reference region for input to the SRTM to calculate 퐵푃푁퐷. We also calculated SUV ratio (SUVR) outcomes using the striatum as

131 reference to assess the capacity of these measures to differentiate between lesions and between subject groups.

4.4.3.2 HAB vs. MAB

Across standard regions in MS and HV subjects, 푉푇 was not significantly elevated in HAB subjects compared to MABs. There were also no significant

푘3 differences in 퐵푃푁퐷 ( ) between HABs and MABs in either MS or HV subjects. SUVs 푘4 were not significantly different in HAB vs. MAB MS subjects or HVs. Additionally, and presumably due to the highly variable estimates of 푘3 and 푘4, 퐵푃푁퐷 calculated directly from the 2TCM showed large standard errors (typically greater than 100%). We therefore do not report values here.

The SRTM with striatum as input failed to converge in several standard ROIs and produced very low 퐵푃푁퐷 values where it did (for example, 0.07±0.07 and 0.05±0.06 for HABs and MABs respectively in the left middle frontal gyrus of HVs and 0.07±0.09 and 0.07±0.05 for MS subjects). Standard errors exceeded 50% in the majority of regions. There were no statistically significant group differences between

HAB and MAB subjects based on 퐵푃푁퐷 outcomes. The highest SUVRs were identified in the GM cerebellum: for example, in HABs/MABs 1.20±0.12/1.21±0.12 for HVs vs. 1.26±0.11/1.19±0.05 for MS subjects in the left cerebellum and 1.24±0.15/1.26±0.09 vs. 1.28±0.15/1.23±0.11 in the right cerebellum. There were, however, no HAB/MAB differences identified using SUVR.

4.4.3.3 MS vs. HV

HAB MS and HAB HV subjects showed no significant difference in 푉푇 in any standard regions. In the same comparison in the MAB cohorts, 푉푇 in the NAWM of MS was significantly higher than that of HVs (Figure 4.7). Conversely, SUVs were significantly higher in HAB MS subjects than HAB HVs in the left and right thalamus and left cerebellum but there were no regional significant differences in MAB MS vs. HV subjects (Figure 4.7).

132

Figure 4.7 푽푻, SUV and SUV corrected for regional blood activity contribution (c) in all standard regions. In all plots, circles represent HV subjects and squares MS subjects, while HABs are represented by open symbols and MABs by closed symbols. * indicates significance (p = 0.05)

푠푡푎푛푑푎푟푑 푑푒푣푖푎푡푖표푛 In all standard regions, inter-subject variability (defined as ) 푎푣푒푟푎푔푒 푢푝푡푎푘푒 in HABs and MABs was below 20% for SUV. 푉푇 variability was higher, particularly in the middle frontal gyrus (e.g. 51% in MAB MS subjects) and white matter (e.g. 64% in

NAWM for MAB MS subjects). Overall, in all regions and in both MS and HV cohorts, 푉푇 did not correlate well with SUV (Pearson correlation coefficient, R = 0.29, R = 0.58 respectively). Furthermore, group differences between HV and MS subjects showed a

133 reversed relationship, with HVs having slightly (though not significantly) higher 푉푇 than MS subjects in many regions, but lower SUV. These discrepancies may be driven by differences in the shapes of blood curves between groups, with the high tracer concentration in blood resulting in misleadingly low 푉푇 for MS subjects.

푉푇 values were universally low, reflecting the fact that a large amount of tracer remains in the blood throughout the duration of the scan, contributing to the signal in the ROI. In this case, SUV60-90 would be unrepresentative of the true tissue uptake. In order to investigate this possibility further, we converted 60-90 minute whole blood activities to SUV and calculated the ‘corrected’ SUV for each region and subject, assuming a 5% 푉푏 in all regions, by the equation 푆푈푉푡표푡푎푙 = 푆푈푉푏푙표표푑 + 푆푈푉푡푖푠푠푢푒.

This analysis showed an approximate maximum contribution of blood to the SUV60-90 of 35%, with no significant difference in blood SUV between MS and HV subjects. When comparing HAB and MAB MS and HV subjects’ corrected SUV (SUV_corr), however, there were no group differences observed (Figure 4.7). We herein use SUV_corr in place of SUV for further analysis. In a similar test for compartmental modelling, we also investigated the effect of fixing 푉푏 to 0%. In this case, 푉푇 was not estimable and the 1TCM failed to fit in any tissue region, further supporting the notion that 푉푏 is a necessary and kinetically identifiable parameter for accurate fitting of a plasma input model with [18F]GE-180.

There were no group differences between HV and MS subjects using 퐵푃푁퐷, nor using SUVR as outcome measures. In order to account for possible suppression of HV/MS differences due to inter-subject variability, we also matched each MS subject by age, genotype and, where possible, gender, to one HV and directly compared their 푉푇, SUV_corr, SUVR and 퐵푃푁퐷. Using this approach, there were still no group differences in any standard region.

4.4.3.4 Lesion quantification

Total hemispheric lesion maps were quantified; 푉푇 was approximately bilateral in all subjects. Results for SUV_corr were also not significantly different between left and right hemispheres.

Lesions over the 100 mm3 volumetric cut-off were also quantified on an individual basis; a total of 108 lesions across all subjects. The 2TCM 푉푇 and SUV_corr results for these lesions, along with corresponding volumes, are presented in Table

4.5. The magnitudes of 푉푇 and SUV_corr showed no correlation with volume of lesions

(R = 0.02, R = 0.07 respectively). Notably, while 푉푇 was elevated in the Gd-enhancing lesion (subject 7, lesion 2), this was not the highest 푉푇 observed in this subject, nor 134 across all MS subjects. SUV_corr in the Gd-enhancing lesion, while the highest of quantified lesions in that subject, was also not the highest in the cohort. The majority of lesions did not have higher SUV_corrs than surrounding NAWM, with the notable exception of the Gd-enhancing lesion. Conversely, 푉푇 of NAWM was lower in several subjects than the individual quantified lesions (Figure 4.8). 퐵푃푁퐷 in NAWM was approximately in the middle of the range of individual lesion estimates, suggesting a similar level of specific binding of [18F]GE-180 in the NAWM of all MS subjects.

SRTM 퐵푃푁퐷 in individual lesions had large standard errors and the model again failed to converge in some regions. In those lesions where the model did converge, 퐵푃푁퐷 did not relate to 푉푇 from the 1TCM (R = 0.03). SUVR was generally in good agreement with SUV_corr (R = 0.74), but values were not > 1 in all regions, indicating that lesion uptake was, in those cases, lower than uptake in the striatum (Figure 4.8).

135

-3 3 Table 4.5 Summary of individual lesion SUV_corr and 푽푻 results (ml.cm ) for MS subjects from 1TCM, with corresponding lesion volumes (cm ). 1 2 3 4 5 6 7 8 9

VT SUV Vol VT SUV Vol VT SUV Vol VT SUV Vol VT SUV Vol VT SUV Vol VT SUV Vol VT SUV Vol VT SUV Vol corr corr corr corr corr corr corr corr corr 0.36 0.38 0.41 0.16 0.34 0.34 0.32 0.24 0.14 0.36 0.41 0.69 0.35 0.48 0.12 0.16 0.32 0.18 1.16 0.31 0.72 0.12 0.22 0.15 0.01 0.19 1.31 0.16 0.30 0.11 0.12 0.45 0.11 0.24 0.34 0.21 0.56 0.38 3.72 0.44 0.73 0.30 0.56 0.40 6.21 0.95 0.58 2.68 0.22 0.29 1.30 0.01 0.18 0.17 0.20 0.44 0.21 0.15 0.32 0.18 0.38 0.31 1.77 0.44 0.34 2.45 0.32 0.71 0.18 0.76 0.36 1.81 0.73 0.40 2.54 0.12 0.21 0.16 0.01 0.14 4.07 0.23 0.42 0.14 0.13 0.35 0.14 0.28 0.28 0.65 0.43 0.38 4.04 0.42 0.35 0.12 0.36 0.40 1.03 0.29 0.35 0.79 0.19 0.26 0.25 0.01 0.18 0.12 0.24 0.31 0.12 0.19 0.30 0.14 0.33 0.29 0.19 0.37 0.32 0.50 0.22 0.51 0.11 0.06 0.40 0.12 0.46 0.29 0.26 0.28 0.27 0.11 0.01 0.16 3.95 0.21 0.29 0.11 0.38 0.21 0.22 0.41 0.42 1.11 0.44 0.41 0.36 0.24 0.33 0.79 0.19 0.32 0.16 0.01 0.13 1.17 0.33 0.34 0.28 0.47 0.38 3.39 0.42 0.40 0.84 0.37 0.35 0.21 0.01 0.11 0.11 0.27 0.24 1.14 0.43 0.43 0.64 0.20 0.42 1.22 0.39 0.41 0.25 0.01 0.15 0.40 0.35 0.22 0.33 0.26 0.35 0.24 0.89 0.57 0.13 0.21 0.35 0.40 0.01 0.10 0.13 0.59 0.34 0.72 0.44 0.33 0.31 0.30 0.35 0.14 0.00 0.14 0.15 0.34 0.22 0.21 0.47 0.36 0.19 0.02 0.47 0.78 0.01 0.14 0.31 0.25 0.32 0.32 0.39 0.38 0.11 0.32 0.44 0.26 0.01 0.15 0.24 0.55 0.27 0.21 0.26 0.30 0.10 0.26 0.41 0.12 0.01 0.10 0.17 0.47 0.26 0.21 0.64 0.43 0.11 0.31 0.41 0.11 0.01 0.11 0.20 0.32 0.24 0.14 0.71 0.46 0.14 0.23 0.40 0.11 0.01 0.29 0.12 0.42 0.33 0.23 0.27 0.41 0.31 0.01 0.28 0.11 0.26 0.31 0.17 0.15 0.36 0.19 0.01 0.15 0.14 0.29 0.28 0.45 0.16 0.34 0.17 0.01 0.13 0.18 0.32 0.24 0.73 0.36 0.34 0.11 0.01 0.10 0.12 0.07 0.24 0.18 0.37 0.43 0.12 0.37 0.28 0.20 0.59 0.42 0.16 0.36 0.34 0.10

136

3 Figure 4.8 푽푻, corrected SUV and SUVR in lesions > 100 mm for individual MS subjects. Open symbols represent HABS. Red symbols represent NAWM value for each subject. Asterisks represent cases where lesions were significantly higher (or lower, where indicated) than NAWM.

In order to assess the uptake in dummy ‘lesion’ areas in HVs, we transformed the T1 MR from each MS subject and matched HVs to MNI (Montreal Neurological Institute) space and co-registered their respective PET images to them. We then warped the whole lesion masks from each MS subject into MNI space and applied them to the normalised PET images from both subjects. Performing a matched comparison in this way showed higher SUV_corr and SUVR in the lesion area of the MS subjects compared to the HVs (p = 0.03 and p = 0.05 respectively).

137

In order to assess for possible correlations between PET uptake and MR characteristics of lesions, we performed a lesion characterisation for each MS subject in FLAIR and T1 pre-contrast images. Loosely, lesions were grouped into: those along the edge of the ventricles/points of ventricular horns (thin hyperintensities in FLAIR); separate and more clearly defined hyperintensities in FLAIR and/or hypointensities in T1; and lesions at the GM/WM boundary, as well as the single Gd-enhancing lesion. Lesions were also sub-categorised as T1 black holes. Lesions were then evaluated for possible correlations between 푉푇 and/or SUV_corr in PET and location/category in MR.

Within each subject, there was a tendency for lesions located in the periventricular area to have slightly elevated 푉푇, SUV_corr and SUVR compared to other lesion groups, although the magnitude of this elevation was not significant. Lesions which were classified as black holes on T1 did not show significantly different 푉푇,

SUV_corr, SUVR or 퐵푃푁퐷 than other lesions within each patient. The Gd-enhancing lesion had the highest SUV_corr of lesions in that subject and tracer uptake was visibly higher than surrounding WM (Figure 4.4), with signal concentrated at the centre of the ROI and uptake at the rim being lower. Interestingly, some lesions also showed a focal absence of PET signal compared to background; these were not differentiable in FLAIR or T1 images from others which did show PET uptake, but were the same as those lesions with SUVR much lower than 1.

4.4.4 Subject demographics and observations from PET HVs were aged between 28 and 56 years and primarily male (n = 7). The majority drank alcohol moderately (~ 14 units per week) and were mild (~ 2 per day) or ex-smokers. In the patient cohort, the time since onset of MS varied between 2 and 42 years and EDSS scores had a range of 3.0 – 6.5. In terms of previous disease modifying therapies, subject 5 had taken interferon beta-1a therapy for 5 months (stopped approximately 1 year prior to PET) and fingolimod therapy for 5 months (stopped immediately prior to PET scan). Subject 7 took occasional steroids and subject 8 had recently stopped interferon beta-1a therapy (2 - 3 weeks prior to PET) and took amitriptyline for migraine relief. Subject 2 had experienced three or more relapses in the 12 months prior to PET, subject 7 had also experienced three and subjects 1, 3, 4 and 9 had had one. One subject was categorised as having definite progressive disease (with no definite relapses ever), and the others were either borderline (n = 5) or definitely relapsing-remitting. Most subjects did not drink

138

alcohol and those who did were mild drinkers (maximum of 10 units a week). Only one subject was a current smoker (10 cigarettes per day for 5 years).

There was no correlation between time since onset of disease and EDSS score (R = -0.24 and ρ = -0.17), nor between EDSS score and WM lesion load (R = -0.27 and ρ = -0.22). In subject 2, whose EDSS score was 6.5, the most recent relapse/s (12–24 months) were severe; the patient was admitted to hospital for two days and was unable to move her legs, and a previous relapse rendered her blind in the left eye. This subject had the lowest T2 lesion load (1.74 cm3) and only 6 lesions > 100 mm3, none of

푘3 which showed significantly higher 푉푇 and 퐵푃푁퐷 ( ) than other regions. SUVR in 푘4 NAWM for this subject, however, was elevated over all individual lesions and over the level of the middle frontal gyrus and thalamus, which, in other subjects, showed the highest SUVR. For other patients, severity of recent relapses was not correlated to overall EDSS score. Results from the 9HPT showed that those patients whose manual dexterity was compromised (i.e. who dropped pegs and took longer to complete the task) were not the same patients who had slower walking speeds (by the T25FW test) suggesting little correlation between manual and ambulatory impairment. Furthermore, the subjects with the highest and lowest WM lesion loads, respectively, completed the 9HPT in less time than the group average, although the subject with the lowest lesion load had the highest EDSS score and also took the longest time to walk 25 ft.

Subjects with the highest whole brain uptake were in general those who had highest total WM lesion loads, with the exception of subject 7, whose total lesion load was within 1 SD of the mean across subjects (mean ± SD = 9.44±7.60 cm3) , but whose overall brain uptake was high. Conversely, one MS subject (subject 9), whose lesion

3 load was 14.83 cm , showed universally low uptake in lesions (SUVR << 1, 푉푇 ~ 0.10 ml.cm-3 in all) in spite of having had a relapse within the previous 12 months.

139

4.5 Discussion

Activated microglia in the CNS are a hallmark of all inflammatory brain conditions. TSPO, the 18 kDa translocator protein, is over-expressed by activated microglia and is the target of several PET radioligands for neuroinflammation, including the prototypical [11C]-(R)-PK11195. A recently developed second generation TSPO-PET ligand, [18F]GE-180, has shown significantly improved signal-to-noise ratios and lower levels of non-specific binding compared to [11C]-(R)-PK11195 in animal models of focal and widespread diffuse neuroinflammation [115, 122, 294]. [18F]GE- 180 is a relatively new radioligand, with only two clinical papers currently published [278-279]. Both of these studies were in healthy volunteers only and focussed on investigating the appropriate compartmental model for kinetic analysis of [18F]GE-180 brain data. Fan et al. compared the performance of the unconstrained 1 and 2TCM and both models with a component of vascular trapping [154], as well as the irreversible

(3푘) 2TCM, concluding that the unconstrained (free 푉푏) 2TCM was preferred for fitting in healthy elderly brains. The Logan graphical method with parametric mapping was also found to produce 푉푇 estimates which were highly consistent with those from compartmental modelling. Feeney and colleagues performed a similar comparison of modelling approaches in a cohort of younger healthy subjects, also finding the best fitting model to the data to be the unconstrained 2TCM (푉푏 fixed to 5%). This group additionally used an SUVR approach with cortical grey matter as a reference region. A natural step is to investigate the performance of [18F]GE-180 in a diseased cohort; here, we compared semi-quantitative SUV and SUVR outcomes with plasma input compartmental models and reference tissue models for kinetic analysis of [18F]GE-180 brain PET in a cohort of relapsing-remitting multiple sclerosis patients and healthy control subjects. Simultaneously, we aimed to assess the utility and value of [18F]GE- 180 as a biomarker for disease activity in these patients compared to the healthy controls.

4.5.1 Blood activity Nine subjects with clinically definite RRMS and 10 neuro- and radiologically healthy control subjects underwent dynamic T1-weighted pre- and post-gadolinium contrast enhanced MRI and T2 FLAIR sequences, and 90 minute dynamic PET scans with arterial sampling. PET data was quantified on a regional basis with the one and

푘3 two tissue compartment models (1TCM/2TCM), with 푉푇 and 퐵푃푁퐷 ( ) as the 푘4 outcome measures and the SRTM with 퐵푃푁퐷 as outcome measure. In both the healthy and diseased cohorts, [18F]GE-180 activity concentration was lowest in the striatum

140

and highest in frontal, occipital and temporal cortical regions. The GM cerebellum, commonly used as a TSPO-PET reference region due to its low uptake and often- presumed negligible level of specific binding, also showed relatively high uptake of [18F]GE-180. This finding is in contrast to other TSPO-PET tracers. For instance, [11C]- (R)-PK11195 exhibits lower uptake in cortical areas and the GM cerebellum and higher uptake in the pituitary glands and midbrain [286], while [11C]PBR-28 shows high uptake in the thalamus and lower uptake in grey matter [295]. Additionally, [18F]GE-180 whole-brain uptake at 60-90 minutes was extremely low in both healthy and MS brains; 0.56±0.10% and 0.72±0.18% of injected dose of tracer respectively, compared to ~3% for [18F]FEDAA1106 [296] and ~2% for [11C]-(R)-PK11195 [297], [18F]PBR-06 [76] and [11C]PBR-28 [298] in the healthy human brain.

The peak of [18F]GE-180 blood activity was higher and earlier in MS subjects than in HVs and the corresponding whole brain uptake was also higher in MS subjects. Interestingly, in a binding-status-split analysis, while the former difference was only observed in the MAB group, the latter was only seen in HABs, suggesting that total brain uptake is not solely explained by the difference in the peak concentration of [18F]GE-180 in blood. In healthy humans, TSPO is present in larger quantities in the peripheral system than in the brain, in particular in the heart and liver [28, 299-300]. One might intuitively expect a higher expression of peripheral TSPO in inflammatory conditions such as MS, however a recent study by Harberts and colleagues [301], who used tritiated PBR-28 to assess levels of TSPO in peripheral blood mononuclear cells in vitro in MS subjects (n = 32) and healthy controls (n = 25), showed that the overall peripheral expression of TSPO was lower in the former than the latter. An earlier study performed using PK11195 showed similar findings [302]. The authors suggest that this reduction in peripheral blood TSPO may be a feature of neuroinflammatory disease and could in part be related to the infiltration of highly TSPO-expressing macrophages from blood into tissue. They also point out, however, that the same assay performed in traumatic brain injury (TBI) patients did not show the same reduction in TSPO [301]. While this finding could be related to the small sample size (n = 4), it could also suggest an effect more specific to central inflammatory conditions, or even MS exclusively, rather than simple neurological insult. Interestingly, in our cohort, POB ratios for MS patients were higher than those for HV subjects, suggesting that the concentration of [18F]GE-180 is higher in plasma than in red cells. With respect to the findings of Harberts et al., and assuming that the tracer binds specifically to TSPO in red cells, a lower peripheral expression of TSPO in these cells in MS subjects might be an underlying cause for [18F]GE-180 partitioning

141

preferentially into plasma. Differences in the plasma behaviour of TSPO tracers in vivo in different subject groups are not a new observation [152], however clearly our finding with [18F]GE-180 requires further investigation (see Chapter 6).

The slow metabolism of the tracer, usually a desirable feature of brain tracers as it minimises the signal from radiolabelled metabolites in blood over the duration of the scan, here negatively impacts quantification of the PET data. Furthermore, the

18 -3 -1 fraction of [ F]GE-180 which crosses the BBB is very low (퐾1 ~ 0.005 ml.cm .min

-3 -1 11 compared to 퐾1 ~ 0.05 ml.cm .min for [ C]-(R)-PK11195 [134]), suggesting a very small first pass extraction fraction. Indeed, assuming a cerebral blood flow in the average healthy human brain of approximately 50 ml.100ml-1.min-1 [260], [18F]GE-180 appears to have an extraction fraction of 1% or lower. This could be the result of interaction of the tracer with the endothelium, although Fan et al. have previously considered a compartmental model including this factor and found that it did not improve fitting of the data [278]. Another factor contributing to the low first pass extraction could be that a proportion of tracer binds non-displaceably to plasma proteins. This latter point is further supported by our findings of high POB ratios throughout the scan (average HAB/MAB POB = 1.69 ± 0.03 in HV and 1.65 ± 0.02 in MS compared to 1.55 ± 0.07 in the arterial blood of subjects with rheumatoid arthritis scanned with [11C]-(R)-PK11195 [303]). A finding of this study, therefore, is that the blood kinetic of [18F]GE-180 may be a surrogate marker for differentiation of patients and controls. Clearly, this requires further investigation in a larger cohort and ideally with drug intervention and/or blocking of TSPO [153] to assess the effect on the bioavailability of tracer.

4.5.2 Plasma input modelling In the human brain, the contribution of arterial blood (i.e. that which is measured to calculate the arterial plasma input function) to the overall cerebral blood volume is low (~30% [304]). Thus, the measured concentration of tracer in plasma as an ‘input function’ is not actually fully representative of the total blood volume in tissue, leading to a difference in activity concentration between arterial and venous blood. Lammertsma [305] clarified that the whole blood (uncorrected) curve should be used for estimation of the fractional blood volume, 푉푏, instead of the metabolite- corrected plasma input function, to avoid the presence of an apparent extra tissue compartment. This would be particularly pertinent in the case of a tracer with high vascular concentration compared to tissue, such as [18F]GE-180. Here, we used the uncorrected whole blood curve to estimate 푉푏 as a free parameter, as the low

142

extraction fraction and resultantly high vascular signal of [18F]GE-180 suggest that the tracer kinetics in tissue could be modelled well with 푉푏 representing the contribution from both arterial and venous blood.

In this data set, we found that the 2TCM was able to produce 푉푇 values with lower variability than the 1TCM in the majority of regions. In HVs, regional 푘3 and 푘4 estimates were in line or slightly higher than other tracers – up to 1.28 and 0.18 min-1 - in HABs and 0.39 and 0.30 min-1 in MABs, compared to [11C]-(R)-PK11195 of the order 0.06 and 0.04 min-1 [134] and [18F]DPA-714 up to 0.30 and 0.11 min-1 in HABs and

-1 0.53 and 2.00 min in MABs [240]. It should be noted that estimates of 푘3 and 푘4 are, however, relatively unreliable due to their high variability. The fact that the 1TCM was Akaike-preferred in small ROIs is not in itself surprising, as the noisier a TAC, the more accurate its fit to a simpler model. In addition, the 1TCM was more likely to miss the peak of the TACs than the 2TCM, leading to a possible underestimation of the vascular component of ROI signal. This finding is relevant to future clinical studies with [18F]GE-180, particularly in MS and when considering healthy control subjects.

Having thus selected the 2TCM for further kinetic analysis in this data set, we quantified on a regional level the 푉푇 in a large GM region, the thalamus, GM cerebellum and whole WM, as well as individual lesions and normal-appearing WM in

MS subjects. 푉푏, estimated as a free parameter in all standard regions was on average 5%, with no significant difference between HV and MS subjects. These estimates are consistent with the physiologically accepted values for fractional blood volume in the human brain [306], suggesting that the effects of dispersion on [18F]GE-180 in this setup were not substantial.

We compared HAB and MAB results in both HV and MS groups, finding a significantly higher 푉푇 in the NAWM of MAB MS subjects compared to MAB HVs. This finding is in line with that of previous studies in MS vs. control subjects with other TSPO tracers [20, 108, 112, 169, 307-308] and further indicates a diffuse inflammatory component in the white matter of RRMS subjects. Interestingly, while some of these other studies have found a correlation between the EDSS score of patients and elevated signal in NAWM [307, 309], we did not see one here. The EDSS in itself is not a metric measure of disability due to MS, and is correspondingly somewhat subjective; based more on the response of patients on the day of assessment than long-term/neurological functionality. Additionally, inflammation in the WM of MS subjects may precede the formation of new lesions [111]. Together these points may at least partially explain why the increased presence of activated

143

microglia in the NAWM of MS subjects does not correlate directly with their current EDSS scores. A longitudinal study (without interventional disease modifying therapy) might help to illuminate whether there is an association between this elevated signal in NAWM and the development of new WM lesions.

Previous MR studies have found significant changes in local blood volume [310-312] around acute lesions, although there is some discrepancy in whether this manifests as an increased blood volume due to vasodilation around Gd-enhancing lesions [312], or as reduced blood volume due to lowered capillary capacity [310- 311]. While our results were not significantly different to estimates in other regions, whole brain or HV subjects, there was a tendency towards lower 푉푏 in individual lesions than in any of these regions, for example the single Gd-enhancing lesion (푉푏 =

3%). 푉푇 in individual lesions was significantly higher than NAWM in the subject with this lesion, but in all other cases 푉푇 estimates were not elevated over the level of NAWM. Indeed, in subject 6, lesions were significantly lower than NAWM in terms

18 of 푉푇, suggesting a relative absence of [ F]GE-180 uptake. We hypothesise that this could relate to the level of activity of the lesions themselves; with lower uptake of tracer indicating a lower level of infiltrating microglia or astrocytes. Additionally, in

-3 the Gd-enhancing lesion, 푉푇 was 0.95 ml.cm , which, interestingly, was not the highest in the subject, nor in the cohort. Gd-enhancement is representative of localised BBB breakdown and presumably active infiltration of blood-born macrophages. A Gd- enhancing lesion should show increased tracer uptake and it is therefore likely that this unexpectedly low 푉푇 is again driven by the high plasma activity, such that even the signal in this lesion, which is visible the PET image, is not actually due to 퐶푡푖푠푠푢푒(푡), but primarily to 퐶푝푙푎푠푚푎(푡). Conversely, subjects with higher 푉푇 tended to be those whose peak activity of [18F]GE-180 in blood was lower, suggesting that the differences in 푉푇 may be overwhelmed by the differences in blood signal rather than reflective of true differences in disease-related pathology.

Of course, it is also possible to estimate 푉푇 using other modelling methods, including with Logan graphical analysis or spectral analysis. As mentioned, the former approach has been employed [278] with relative success and it would be of use to investigate such approaches in the RRMS dataset in the future.

4.5.3 Simplified reference region modelling The use of the SRTM was justified by the fact that the 1TCM provided a reasonable fit in whole brain, implying that tracer kinetics in tissue can be adequately described by a single compartment, a key assumption of the model [167]. In the

144

absence of a truly TSPO-free anatomical reference region, or of an alternative validated data-driven approach such as cluster analysis [135, 191, 197] for [18F]GE- 180, we elected to assess regional TACs and select as pseudo-reference that which was lowest.

Having identified the striatum as the region with the lowest TAC, we applied

11 the SRTM with this as input. Estimates of 퐵푃푁퐷 were somewhat lower than [ C]-(R)- PK11195-scanned healthy controls with a cerebellar reference input or a cluster- analysis derived input to the SRTM (퐵푃푁퐷 ~ 0.50 [135, 181]). Taken on its own, this would suggest that the component of specific binding in all regions, including lesions, is indeed very small. This is likely to again be a result of the low extraction fraction and slow washout of tracer. Theoretically, a tracer such as [18F]GE-180, whose kinetics in both target and reference region appear to be reasonably well described by a 1TCM,

푘3 should allow the SRTM to identify 퐵푃푁퐷 ( ) reliably [167]. The fact that the model 푘4 failed to converge in several regions, and that 퐵푃푁퐷 estimates were low with high standard errors, could indicate a component of specific binding in the striatum itself, introducing a negative bias. Another possibility is that the 퐾1 (and thus 푅1, a parameter in the SRTM) of the tracer is too low for the fit to converge. A further assumption of the SRTM is that the whole blood volume contribution to tissue is negligible, which, as we have shown, may not be the case in this cohort with [18F]GE- 180 (although see section 4.4.4). Subject motion and the partial volume effect could also have contributed to the spurious results. Recently, it has been shown that such violations of underlying assumptions of the SRTM can result in extreme over and underestimation of 퐵푃푁퐷 results, depending on the degree of violation [264]. Such effects have also been investigated with other PET tracers [249, 313]. Given the recent finding that accounting for the presence of TSPO binding sites in the endothelium did not significantly improve the fit of the 2TCM in healthy subjects [278] and also considering the above uncertainties regarding the striatum as a choice of reference region, we here elected not to investigate the equivalent SRTMV method proposed by Tomasi et al. [199].

Although the SRTM is the simplest version of a reference tissue approach to quantification of PET data, other methods exist, including MRTM and RS-ESA (see section 1.8.5), which may be investigated in the future. We also suggest an in vivo displacement experiment should be performed to establish the level of specifically bound tracer in this region before further analysis is performed using it as reference.

145

4.5.4 Standardised uptake values Given the unexpected findings with compartmental modelling and the variability of tracer behaviour in blood, as well as the underlying goal of eliminating the requirement for invasive arterial blood sampling, we also used a semi-quantitative approach to calculate SUV and SUVR outcomes. Since the SUV approach is unable to account for the ROI component of signal attributable to blood, and because of the high and persistent vascular signal of [18F]GE-180, we attempted to establish the fraction of apparent uptake which was truly tissue-related. In order to do this, we calculated at 60–90 minutes the SUV of the whole blood curve, incorporating an estimated average 5% fractional blood volume across regions, and subtracted this from the apparent regional SUVs taken from the PET image. We found that the maximum contribution of blood to the PET signal was approximately 35%. Neither original nor corrected SUVs showed significant differences across regions or subject groups. This is a relative reassurance as it implies that inter-subject variability is not driven by the large vascular component. When this correction was applied to lesions, those subjects which showed significantly higher uptake in lesions over NAWM in original SUV (data not shown) were also those whose SUV_corr were higher. This is again suggestive that the contribution of vascular signal to the total tissue signal, in spite of the slow washout of [18F]GE-180, does not drive the increases in signal seen in the lesion area.

In turn, this then implies that, where 푉푇 estimates in lesions and NAWM are elevated in MS subjects over HVs, the difference is not (solely) due to vascular signal.

Assuming a similar vascular contribution from target and reference regions, the SUVR approach advantageously does not require correction for the blood signal. SUVRs were not always > 1 in lesions, particularly in one subject (9). This absence of PET signal did not appear to be linked to peak of tracer in blood, nor to a lower POB ratio. Interestingly, the SUVRs in other brain regions of this subject were relatively high, thus lowering the possibility of mis-classification as a MAB (rather than a LAB).

In other subjects, neither the 푉푇 nor the SUV_corr/SUVR in lesions correlated strongly with their size, location or appearance on T1 or FLAIR MR images. Together, these findings suggest that, while the overall uptake of tracer is very low, the definite lesion- related signal from [18F]GE-180 may be identifying a component of inflammation in certain lesions which are not otherwise differentiable in conventional MRI. A follow- up study in these patients is currently ongoing and may help to address this tentative finding.

In order to assess the specificity of the signal to lesions, rather than anatomical location, we also matched each MS subject by age and TSPO genotype to a HV, 146

applying as an ROI the lesion mask from that subject. The fact that the SUV_corr and SUVR results for the whole lesion masks were significantly higher in MS than matched HV subjects provides support for our speculation that there is a lesion-specific focal increase in [18F]GE-180 uptake. This promising finding suggests that, in spite of its low extraction fraction and high blood retention, the tracer could have some use as a tool for identification of an inflammatory component in MS subjects.

4.5.5 HAB/MAB differences With all other second generation TSPO ligands, a single nucleotide polymorphism in the TSPO gene has been found to affect binding in the human brain to varying degrees, with [11C]PBR-28 exhibiting the highest HAB/LAB ratio of 50-fold [74-75, 240, 314-316] and other tracers affected to a lesser degree. Our observations in this cohort suggest that, for [18F]GE-180, HAB/MAB differences for healthy subjects with 푉푇, SUV_corr or SUVR, are not identifiable. In vitro work has suggested an approximate difference between affinity for HAB and LAB binding sites of 5 fold (personal communication, William Trigg) and the discrepancy in our findings indicates different behaviour of the tracer in vivo. It is possible that a contributing factor is the low overall binding of [18F]GE-180 and high vascular contribution across regions in all subjects, lowering distinguishability of any effects of TSPO genotype. This could be a particular factor in MS subjects, where the intensity of TSPO expression in a highly active (inflamed) MAB could be similar to a less active HAB. At least in HVs, the variability in uptake of tracer was low (26% and 7% in whole brain in HABs and MABs), suggesting relative consistency among binding groups. Until these discrepancies have been addressed, however, and under the assumption that differences in regional 푉푇 between HABs and MABs are reflective of binding status, we suggest ongoing stratification of subjects according to TSPO genotype.

4.5.6 Summary Our findings suggest that [18F]GE-180 brain PET data can be modelled reliably with a 2TCM in both our healthy volunteer and MS cohorts. We found an unexpected correlation between the presence of disease and the peak of tracer activity in whole blood, suggesting a possible reduction in peripherally-expressed TSPO in MS subjects (higher peaks). The overall brain uptake of [18F]GE-180 was very low, likely due to a small extraction fraction, itself possibly the result of a high level of avid non-specific binding to plasma proteins. Together, these effects lead to very small 푉푇 estimates in both subjects groups. Vascular contribution to SUV60-90 was estimated as a maximum of 35%, which, when accounted for, resulted in no observable group differences in

147

standard regions between HV and MS HABs or MABs. The SRTM showed no group differences in 퐵푃푁퐷 and in any case requires further validation of reference region selection to limit model violation related bias.

Overall, lesions defined in MS subjects showed higher SUV_corr and SUVR than the same areas in age and genotype-matched controls. A lesion-by-lesion analysis found decreased SUVR in several lesions, which were not apparently different in T1 or FLAIR MRI. Possible connotations of this finding are under investigation with longitudinal scans in this MS cohort.

Supplementary materials

Supplementary figure 4.1: HPLC chromatograms showing metabolite formation in plasma samples. Chromatograms are at a) 10 minutes, b) 30 minutes and c) 70 minutes post-injection of tracer. At all time points, there were four polar metabolites of [18F]GE-180 which were not expected to cross the blood brain barrier. [18F]GE-180 parent fraction decreased very slowly between 10 and 70 minutes from 0.78 to 0.70 and over the whole course of the scan.

148

Supplementary table 4.1 Summary of rate constant estimates from the 2TCM averaged over HAB and MAB MS subjects in standard anatomical regions.

-3 -1 -1 -1 -1 Region 퐊ퟏ (ml.cm .min ) 퐤ퟐ (min ) 퐤ퟑ (min ) 퐤ퟒ (min )

HAB MAB HAB MAB HAB MAB HAB MAB Middle frontal 0.004 (0.002) 0.011 (0.003) 0.114 (0.121) 0.201 (0.071) 0.280 (0.298) 0.076 (0.052) 0.181 (0.163) 0.013 (0.016) gyrus left Middle frontal 0.004 (0.001) 0.012 (0.003) 0.113 (0.078) 0.204 (0.062) 0.228 (0.168) 0.049 (0.026) 0.187 (0.119) 0.012 (0.012) gyrus right Thalamus left 0.004 (0.003) 0.012 (0.004) 0.086 (0.202) 0.276 (0.079) 0.292 (0.578) 0.162 (0.105) 0.198 (0.409) 1.076 (0.607)

Thalamus right 0.004 (0.002) 0.011 (0.003) 0.078 (0.050) 0.183 (0.128) 0.366 (0.231) 0.794 (0.611) 0.251 (0.184) 0.469 (0.580)

Cerebellum left 0.005 (0.002) 0.014 (0.004) 0.084 (0.064) 0.177 (0.103) 0.284 (0.235) 0.035 (0.046) 0.265 (0.183) 0.010 (0.029)

Cerebellum right 0.005 (0.002) 0.015 (0.007) 0.089 (0.066) 0.229 (0.116) 0.250 (0.182) 0.051 (0.045) 0.203 (0.158) 0.013 (0.022)

(NA)WM 0.004 (0.002) 0.010 (0.003) 0.115 (0.067) 0.213 (0.071) 0.223 (0.182) 0.064 (0.036) 0.169 (0.242) 0.015 (0.014)

149

Supplementary table 4.2 Summary of rate constant estimates from the 2TCM averaged over HAB and MAB HV subjects in standard anatomical regions.

-3 -1 -1 -1 -1 Region 퐊ퟏ (ml.cm .min ) 퐤ퟐ (min ) 퐤ퟑ (min ) 퐤ퟒ (min )

HAB MAB HAB MAB HAB MAB HAB MAB Middle frontal 0.008 (0.002) 0.007 (0.001) 0.243 (0.172) 0.216 (0.049) 0.862 (0.391) 0.163 (0.065) 0.057 (0.097) 0.213 (0.188) gyrus left Middle frontal 0.007 (0.002) 0.007 (0.001) 0.224 (0.086) 0.214 (0.071) 0.909 (0.483) 0.163 (0.108) 0.095 (0.072) 0.188 (0.231) gyrus right Thalamus left 0.008 (0.002) 0.006 (0.002) 0.236 (0.168) 0.107 (0.051) 0.552 (0.415) 0.268 (0.160) 0.057 (0.047) 0.259 (0.223)

Thalamus right 0.009 (0.004) 0.006 (0.002) 0.194 (0.244) 0.186 (0.072) 0.747 (0.472) 0.391 (0.287) 0.054 (0.101) 0.303 (0.216)

Cerebellum left 0.008 (0.003) 0.007 (0.001) 0.252 (0.276) 0.133 (0.046) 0.802 (0.609) 0.233 (0.099) 0.049 (0.092) 0.278 (0.243)

Cerebellum right 0.009 (0.003) 0.006 (0.002) 0.370 (0.297) 0.174 (0.119) 1.280 (1.314) 0.227 (0.162) 0.058 (0.095) 0.225 (0.281)

(NA)WM 0.001 (0.000) 0.001 (0.000) 0.078 (0.072) 0.074 (0.042) 0.230 (0.121) 0.059 (1.465) 0.177 (0.167) 0.015 (0.028)

150

5 Therapy response monitoring with [18F]GE-180 in a cohort of relapsing-remitting multiple sclerosis patients

Sujata Sridharan1, Rainer Hinz1, Alexander Gerhard1, Hervé Boutin1, William Trigg2, Christopher Buckley2, David Brooks3, 4, Richard Nicholas5, Joel Raffel5

1 Wolfson Molecular Imaging Centre, The University of Manchester, Manchester, U.K.

2 GE Healthcare, The Grove Centre, Amersham, Buckinghamshire, U.K.

3 Division of Brain Sciences, Imperial College London, Hammersmith Hospital Campus, U.K.

4 Institute of Clinical Medicine, Aarhus University, Aarhus, Denmark

5 Faculty of Medicine, Department of Medicine, Imperial College London, London, U.K.

5.1 Abstract

Introduction: Natalizumab is a second line disease modifying therapy which is effective in three quarters of MS patients with active disease. Current therapy response monitoring is performed using MRI/laboratory and clinical tests, but these are not always representative or predictive of long-term disease burden. The 18 kDa translocator protein (TSPO), expressed by activated microglia, can be used as a PET imaging target in inflammatory brain conditions such as multiple sclerosis (MS), with the potential to be a biomarker of therapy response. In this short pilot study, we investigate the possible utility of a new TSPO-PET tracer, [18F]GE-180, in this respect in a small cohort of relapsing-remitting MS (RRMS) subjects treated over 10 weeks with natalizumab. Methods: Five genotyped clinically-definite RRMS patients (three female) underwent MRI and 90 minute dynamic [18F]GE-180 PET scans at baseline and after 10 weeks of 300mg infusions of natalizumab. PET and T1 pre-contrast MR images were co-registered and regions of interest (ROIs) defined, including cortical grey matter, normal appearing white matter (NAWM), thalamus and cerebellum. Individual lesions were also identified as ROIs. Outcome parameters were the regional total volumes of distribution (푉푇) from the two tissue compartment model and standardised uptake values (SUV). Results: Three subjects were categorised as showing a clinical response to natalizumab therapy (i.e. improved or stable condition at follow up). Another patient, who experienced a relapse whilst on treatment, showed one Gd-enhancing lesion in T1 MR. In this subject, 푉푇 in lesion areas was elevated over NAWM at baseline (0.95 ml.cm-3 vs. 0.19 ml.cm-3), while at follow up,

151

uptake in lesions had decreased significantly (p = 0.009) so that this difference was no longer evident. Neither 푉푇 nor SUV showed any change at follow up in any standard region. While SUV in lesions was not decreased in the same subject as 푉푇 was, there was a significantly lower uptake in two of the subjects who showed clinical response to therapy at follow up. Conclusions: [18F]GE-180-PET uptake was significantly decreased in areas of MR pathology in some subjects after 10 weeks of natalizumab therapy. In two cases, this difference was correlated with clinical outcome measures of response to treatment. Further investigation over the full duration of treatment (13 months) and in a larger cohort (17 subjects) is ongoing.

152

5.2 Introduction

Multiple sclerosis (MS) is a progressive and debilitating central nervous system (CNS) disease and is the global leading cause of neurological disability in young adults. Particularly in its relapsing-remitting (RR) stage, MS is characterised by acute inflammatory processes which are driven by T and B cells, macrophages and microglia focally infiltrating tissue and fuelling demyelination and generation of sclerotic plaques [317]. While this inflammatory component has been observed both in vivo [20, 169, 307] and in histopathological studies [317-318] in RRMS and the precursor to MS, clinically isolated syndrome, and although some evidence exists for a diffuse component of inflammation in progressive disease [108], the nature and extent of inflammation during later stages of disease is generally contentious.

Nevertheless, several first-line disease modifying therapies have aimed at reducing central and peripheral inflammation in order to delay progression to clinically definite MS. First line treatments and early-intervention therapies such as interferon beta and glatiramer acetate have been successful at delaying progression, compared to placebo, by up to 45% over two or three years [319-323] and have shown a reduction of up to 34% in clinical relapse rate over 12 months [324]. One second line treatment, natalizumab, is a monoclonal antibody which blocks the adhesion of lymphocytes and monocytes to endothelial cells of the BBB and thus passage of inflammatory cells from the peripheral system into the CNS. It has been able to produce a 68% reduction in relapse rate and a 42% decrease in risk for disability progression when used as a monotherapy [325]. In spite of these promising results, it is well recognised that the optimal time window for effective treatment for long-term disability is the early phase of the disease [326-327] (with clinical measure, the Kurtzke expanded disability status scale, EDSS < 3.0) and not all patients respond to therapy (see [328] for review). Indeed, in a prospective study of response to the drug, while approximately 75% of subjects could be categorised as having a full or partial response (with improved or stable EDSS), the remaining 25% experienced ongoing relapses and/or worsening of EDSS [329].

Currently, brain MRI and laboratory measures are the recommended medical approach for diagnosis and therapy selection in MS patients [80]. While in the relapsing-remitting phase, total MR lesion loads have been reported to correlate well with clinical disability measures [102], it has become clear that a so-called clinico- radiological paradox [330] exists on a patient-by-patient basis, where this correlation and the long-term predictive value of MRI parameters to disease course are less

153

certain [331-333]. Furthermore, the non-functional nature of conventional MRI techniques does not allow investigation of the physiology of the MS brain. These discrepancies, combined with the lack of curative treatment for the disease, have somewhat fuelled the increasing use of PET imaging in multiple sclerosis research. In particular, PET may offer some elucidation on the role of inflammation in lesions in vivo, a component which has so far been primarily restricted to histopathological studies.

Activated microglia in the CNS are a hallmark of all inflammatory brain conditions. TSPO, the 18 kDa translocator protein, is over-expressed by activated microglia and is the target of several PET radioligands for neuroinflammation, including the prototypical [11C]-(R)-PK11195. Previous clinical PET studies imaging inflammation in MS patients have been mostly limited to baseline or non-therapeutic data and primarily using [11C]-(R)-PK11195, which is well known to exhibit high nonspecific binding among other unfavourable properties [20, 91, 106-108, 111-112, 169, 204, 276-277, 285, 307-308, 315, 334]. One study into the potential of PET for therapy response monitoring used [11C]-(R)-PK11195 to evaluate the changes in microglial activation of nine RRMS patients treated with glatiramer acetate [335]. Results showed a decrease in binding potential of the tracer in both cortical grey matter and white matter, but the small sample size limits the utility of this outcome.

A recently developed second generation TSPO-PET ligand, [18F]GE-180, has shown significantly improved signal-to-noise ratios and lower levels of non-specific binding compared to [11C]-(R)-PK11195 in animal models of focal and widespread diffuse neuroinflammation [115, 122, 294]. Thus, [18F]GE-180 may offer some insight into the level of microglial activation and corresponding inflammation in the MS brain; or more specifically, how the disease modifying therapy, natalizumab, affects this.

Therefore, the primary objective of this short pilot analysis is to assess the potential of [18F]GE-180 as a biomarker for response to therapy in MS. A small cohort of relapsing-remitting MS (RRMS) subjects were PET scanned at baseline and after three months of natalizumab treatment. A secondary objective is to establish which quantification methodology is able to best identify differences between pre and post- 10 week treatment groups and is able to categorise response to treatment.

154

5.3 Materials and methods

5.3.1 Subjects Five subjects with clinically definite RRMS (i.e. who met the revised McDonald criteria for MS [280]) were recruited from MS clinics within the Imperial College Healthcare NHS Trust, with ethics approved by the Riverside Research Ethics Committee. According to inclusion criteria, subjects were aged between 20 and 65 years old at baseline (mean±standard deviation, SD, 31.3±9.2 years), and high or mixed affinity binders (HAB/MAB) for the single nucleotide polymorphism in TSPO [75, 77] (4 high, 1 mixed). All subjects (3 female, 2 male) were considered eligible to commence natalizumab therapy with highly active disease; that is, they could be categorised as having ‘rapidly evolving severe MS’; with two or more disabling relapses within the previous year and one or more gadolinium (Gd) enhancing lesion in brain MRI or a significant increase in T2 lesion load compared with a previous MRI; or as a ‘suboptimal therapy group’; having failed to respond to a full and adequate course of interferon beta, having had at least one relapse in the previous year whilst on therapy and currently having at least nine T2 hyperintense lesions or at least one Gd-enhancing lesion. Subjects had discussed commencing natalizumab treatment with their clinician prior to recruitment. Pregnant or breastfeeding women were excluded from the study.

All subjects were required to show adequate visual, auditory and communication ability in order to provide informed consent for the study. Subjects were otherwise clinically healthy, with no evidence of other neurological, psychiatric or systemic disease than MS for at least one month before recruitment. They were also required to not be taking any medication which could interfere either with results from the study (i.e. other disease modifying therapies or anti-inflammatory medication), or which could pose a risk to participant safety.

5.3.2 Study timeline After recruitment, all patients attended a screening visit where consent was taken by a clinician and preliminary clinical assessment, including screening bloods and a genetic analysis for TSPO binding type, was performed. At this visit, patients who had not previously undergone clinical MRIs were also scanned. Approximately two weeks later, patients returned for a baseline visit, where blood samples (maximum 30 ml) were taken for blood biomarker assessment. Clinical and cognitive assessment was also performed, with MS-specific tests including the 9-hole peg test (9HPT), time to walk 25 feet (T25FW) and symbol digits modality test (SDMT), as well 155

as EDSS. Finally, patients underwent a number of MRI sequences, including T2- weighted FLAIR, and pre and post-Gd dynamic contrast enhanced imaging, as well as a 90 minute dynamic [18F]GE-180 PET scan. A minimum of 2 weeks after this visit, patients commenced natalizumab therapy (Tysabri®) with monthly infusions of 300 mg.

In order to assess the efficacy of the treatment, an early response biomarker such as [18F]GE-180 PET is sought. After 10 weeks (‘early time point’), patients returned for a follow-up visit consisting of the same battery of cognitive and clinical tests as the baseline visit, as well as blood biomarker assessment and MRI and [18F]GE-180 PET scans. After 13 months of therapy (‘late time point’), patients attended a final visit to assess their overall response to natalizumab treatment. Figure 5.1 summarises the timeline of the study. In this analysis, scans from n = 5 patients were available at baseline and after ten weeks of therapy.

Figure 5.1 Diagram summarising the timeline for patient visits in natalizumab study A target total of 25 patients were recruited from the Imperial College Healthcare Trust based on their prior discussion of commencement of natalizumab with their attending MS specialist. Consent was taken from patients and they underwent a genetic analysis with a QIAGEN QIAmp DNA Blood mini kit (see Chapter 6) to establish their TSPO genotype. Low affinity binders were excluded. Shortly after consenting, patients attended a baseline visit with a battery of clinical and cognitive tests and MR sequences, as well as a 90 minute dynamic [18F]GE-180 PET scan. Shortly after this baseline visit, at the ‘0 week’, patients commenced natalizumab therapy, with monthly infusions of 300 mg. Ten weeks later, patients visited again for a repeat of the battery of tests and scans. A thirteen month post-commencement of treatment visit was also part of the timeline; however, data was not available at the time of writing.

5.3.3 PET scanning [18F]GE-180 was synthesised on an automated FastLab™ platform as described previously by nucleophilic fluorination of an S-enantiomer mesylate precursor [223,

156

225]. The target dose of injected tracer was 185MBq (actual dose 180.05±4.99 MBq at baseline, 182.96±2.77 MBq at follow up, not significantly different, n.s.). Dynamic PET- CT scans were performed on a Siemens Biograph 6 with a field of view of 168 × 168 × 148 mm3. List-mode data were histogrammed into 24 time frames (6 × 90, 3 × 180, 5 × 600, 5 × 1500 and 5 × 3000 seconds) and reconstructed using filtered back projection with a ramp filter. Reconstructed voxel dimensions were 1.57 × 1.57 × 1.92 mm3.

Patients were cannulated in their radial artery, under local anaesthetic, with heparin-flushed tubing to avoid clotting, prior to the start of the PET scan. Blood was drawn continuously for the first 15 minutes of the PET scan at a target rate of 2.5 ml.min-1 (in total not exceeding 300 ml for each patient). Additionally, discrete whole blood samples were withdrawn at 4.5, 9.5, 14.5, 29.5, 49.5, 69.5 and 89.5 minutes for metabolite analysis (see Chapter 4 for details) and generation of a continuous plasma input function for kinetic modelling.

5.3.4 Kinetic analysis Reconstructed PET images were co-registered with T1 pre-Gd contrast MR images using the FLIRT tool in FMRIB’s (the Oxford Centre for Functional MRI of the Brain, Oxford, U.K., build 414) Software Library (FSL) by linear transformation with 6 degrees of freedom. MR images were segmented into grey and white matter maps and CSF using PMOD (PMOD Technologies Ltd., Switzerland, v3.7) and the 83-region probabilistic Hammers atlas [282] was used to define regions of interest (ROIs). Standard anatomical regions considered included unilateral (left and right) cortical grey matter (middle frontal gyrus), thalamus and grey matter cerebellum. A separate cortical grey matter ROI and normal appearing white matter (NAWM) were also quantified, the latter as whole white matter excluding lesions.

Lesions were defined using a semi-automated local thresholding technique (JIM, Imperial College London), considering T2 FLAIR, and T1 pre and post-Gd contrast images to improve lesion identification. Total lesions loads were calculated and lesions were also individually identified and quantified. In order to minimise the effects of partial volume, a cut off for the size of individual lesions was applied, taking into account the limit of spatial resolution of the PET scanner (~ 5 mm), of 100mm3. All standard anatomical regions excluded lesions.

All kinetic modelling was performed in PMOD. Calibrated continuous and discrete blood data were corrected for decay and delay. The parent fraction in plasma from metabolite analysis was fitted to a Hill function [336] using equation 5.1, where

푓푝 is the fraction of free parent tracer in plasma and 퐴, 퐵, 퐶 are scaling parameters.

157

1 푓 (푡) = 푓 . { 퐴푢퐵 } Eq. 5.1 푝푎푟푒푛푡 푝 1−( ) 푢퐵+퐶

The plasma-to-whole blood ratio (POB) was calculated at discrete time points for each subject and multiplied with the whole blood and parent fraction in plasma curves to obtain a continuous plasma input function. It has previously been shown (see Chapter 4 and [278-279]) that a two tissue compartment model with free fractional blood volume is suitable for describing the pharmacokinetic behaviour of [18F]GE-180 in the brain. Outcome parameters were therefore the total volume of distribution, 푉푇, and standardised uptake values at 60-90 minutes (SUV). Assuming that [18F]GE-180 PET signal is specific and representative of disease related inflammation, and that natalizumab therapy affects the level of inflammation in the MS brain, the hypothesis was that these outcome parameters should change, with a possible correlation to clinical outcomes, between baseline and follow up scans.

Statistical comparisons were performed between all pre and post-treatment scans using repeated measures ANOVA (with regions and subjects as factors). Significance was set at 5%.

158

5.4 Results

5.4.1 Blood data POB ratios did not vary over the course of the PET scan (average variability 푆퐷 calculated within subjects over time as was 2% both pre and post- 푎푣푒푟푎푔푒 푃푂퐵 treatment) and were not significantly different between pre and post-treatment scans (average 1.69±0.03 vs. 1.61±0.02 in HABs and MABs at pre and post-treatment respectively). Peak activity of tracer in whole blood was not significantly different between follow up and baseline and nor was parent fraction of tracer over time (see Figure 5.2).

Figure 5.2 Average parent fraction in plasma for baseline and post-treatment scans. Average parent fractions in plasma are shown for all patients at baseline (×) and post treatment (×).

5.4.2 Clinical observations At baseline, three subjects had suffered recent relapses (within 12-24 months prior to recruitment); one had experienced no major relapses in that time frame and one had never had any major relapses. Subjects had EDSS scores between 3 and 5.5 at baseline. After 10 weeks of treatment, EDSS was unchanged in two patients, decreased from 4 to 3.5 in one and 3.5 to 2.5 in another, and increased in another from 3 to 4. Three patients reported definite improvement of symptoms, with another indicating no major change to levels of fatigue (although no worsening of symptoms either) and the fifth (with worsened EDSS score at follow up) experiencing a relapse in the second month of treatment; in total, four of five patients were classed as stable or improved

159

after 10 weeks of treatment. The three patients reporting improvement were categorised as having a clinical response to natalizumab at this stage.

Other clinical and cognitive test scores (9HPT, T25FW and SDMT) remained unchanged between baseline and follow up assessments. These results are summarised in Table 5.1.

Table 5.1 Summary of results from baseline and follow up clinical assessment for all subjects.

Patient EDSS 9HPT T25FW SDMT Patient- Clinical (right R; left L) reported response to improvement natalizumab Baseline Follow Baseline Follow Baseline Follow Baseline Follow up up up up 1 3.0 3.0 20.8 R; 20.8 R; 4.2 4.2 52 52   21.2 L 21.1 L 2 4.0 3.5 22.0 R; 20.3 R; 3.0 3.2 43 49   20.3 L 27.9 L 3 5.5 5.5 94.0 R; 80.0 R; 9.4 7.4 42 48   58.0 L 33.0 L 4 3.0 4.0 18.3 R; 18.8 R; 4.2 5.2 51 54   27.9 L 25.7 L 5 3.5 2.5 19.0 R; 16.3 R; 5.4 4.1 76 79   17.0 L 16.5 L EDSS – Expanded disability status score, 9HPT – 9 hole peg test, T25FW – Time to walk 25 ft, SDMT – Symbol digits modality test

5.4.3 Imaging data One Gd-enhancing lesion was identified on post-contrast MR images (subject 4, Table 5.1). Figure 5.3 shows the baseline and follow up scans for this subject, with the Gd-enhancing lesion highlighted. Visually, the signal of [18F]GE-180 was lower in this lesion at follow up than baseline. T2 white matter lesion loads across all patients were between 2500 mm3 and 10,000 mm3 at baseline; these did not change significantly after 10 weeks of natalizumab treatment. Two new lesions were identified in patient 2 at follow up, while one lesion in patient 4 had shrunk below the 100mm3 cut-off for individual lesion analysis. Grey/white matter volume ratios (where white matter is NAWM) increased after 10 weeks of treatment in 3 patients (1, 4 and 5) by between 2 and 24%, but this change was not significant. In the remaining two patients, ratios were slightly decreased (3±2 %). Additionally, total grey and white matter volumes did not independently change over the treatment period (p = 0.81, p = 0.13 respectively).

160

Figure 5.3 Sum 60-90 minute [18F]GE-180 PET images with co-registered T1 pre-contrast images (top) and (below) corresponding post-contrast T1 images. Images are for subject 4 at baseline and after 10 weeks of treatment with natalizumab. A lesion which appeared as Gd-enhancing in dynamic contrast enhanced MR is highlighted.

In baseline scans, whole brain SUV was 0.56±0.10, which was not significantly higher than follow up scans (0.52±0.16). Figure 5.4 shows, in standard regions, 푉푇 for scans at baseline and 10 weeks after commencing treatment. Overall percentage changes in 푉푇 in standard regions varied between patients but were not significantly different at post-treatment to baseline. In normal appearing white matter (NAWM), assessed by subtracting whole lesion masks from whole white matter, baseline 푉푇 was

-3 -3 0.20±0.07 ml.cm , while follow up 푉푇 was 0.25±0.08 ml.cm (n.s.). Whole lesion mask

-3 -3 푉푇 estimates were 0.28±0.10 ml.cm at baseline and 0.25±0.05 ml.cm at follow up, with no significant differences between pre and post-treatment total lesion uptake. At baseline, all 푉푇 estimates in the total lesion area were elevated over NAWM by between 4% and 79% (p = 0.06). At follow up scans, 푉푇 had decreased in the total lesion area to below the level of NAWM in three of the five subjects.

푉푇 in individual lesions is shown for pre and post-treatment scans for each subject in Figure 5.4 (with median and interquartile range). In subject 4, there was a significant reduction in 푉푇 in individual lesions at the post-treatment stage, potentially

161

driven by a higher spread in values in the baseline scan than at follow up. At baseline, the Gd-enhancing lesion had a magnitude of 푉푇 which was over 500% greater than that of frontal grey matter respectively. A thin, periventricular lesion in this subject

-3 had a 푉푇 of 1.15 ml.cm , 22% greater than that of the Gd-enhancing lesion. After 10 weeks of treatment with natalizumab, both lesions had shrunk in volume and in 푉푇 was reduced by 77% and 46% respectively. While at baseline, 푉푇 in individual lesions was elevated over the level of NAWM in subject 4 (p<0.001), at follow up, this difference was no longer significant.

Figure 5.4 푽푻 estimates in standard regions at baseline (blue) and after 10 weeks of treatment with natalizumab (red). Open symbols represent HABs and closed symbols are MABs. Lines connect the same patient at baseline and follow up. There were no significant changes between time points in any region.

SUVs did not correlate well with 푉푇, in either standard regions or individual lesions, at baseline or follow up (Spearman’s correlation coefficient ρ < 0.40 in all cases). In standard regions, SUV did not change over time, but individual lesions showed significantly decreased SUV at follow up for patients 2 and 3 (p < 0.0001). In the single Gd-enhancing lesion, SUV decreased from 0.68 to 0.53 (22%). Other lesions in this patient showed a relative increase in SUV of between 3 and 12%, including the periventricular lesion (11% increase). In comparison, lesions in patients 2 and 3 had decreased at follow up by between 8 and 32%. In total, of 45 lesions identified across subjects, both at baseline and at follow up, 36 decreased in uptake of [18F]GE-180 after 10 weeks of natalizumab therapy, with one showing no change and the rest increasing. Splitting this into a patient-by-patient analysis showed that patients 1, 4 and 5 had some lesions which increased in SUV over time. 푉푇 decreased in 28 lesions,

162

with the magnitude of change being 37±24%. All patients had some lesions which increased in 푉푇 after treatment. In lesions which did decrease over time, however, this reduction was again significant only for patient 4 (p = 0.01) when considering 푉푇 and for patients 2 and 3 in terms of SUV, suggesting that the overall significant difference in outcome parameters seen across lesions in all patients is driven by these decreases.

SUV is a direct measure of tracer uptake in tissue, but also includes a component of vascular signal in the ROI. Given the high and persistent blood signal of [18F]GE-180, we previously implemented a correction by calculating the SUV of whole blood at 60-90 minutes and subtracting this from ROI SUVs, assuming an average 5% fractional blood volume in anatomical regions, Chapter 4. Repeating the analysis using this SUV_corr did not change the overall group differences between pre and post- treatment scans for patients 2 and 3. The proportion of blood signal potentially contributing to SUV in all regions (including lesions) was between 8 and 25% in baseline scans and between 7 and 27% at follow up.

Figure 5.5 푽푻 and SUV_corr in individual lesions for each patient at baseline (blue) and after 10 weeks of natalizumab treatment (red).

18 In subject 4, there was a significant decrease in [ F]GE-180 푉푇 (top), while overall SUV_corr decreased in subjects 2 and 3 (bottom). Open symbols represents the MAB subject. A single Gd- enhancing lesion in subject 4, which was still visible in T1 post-contrast MR at follow-up, is represented by half open symbols. 163

5.5 Discussion

Natalizumab is a highly effective second line treatment choice of disease modifying therapy for relapsing-remitting multiple sclerosis patients. The drug is an α4- integrin antagonist which blocks the passage of leukocytes across the BBB, thereby reducing central inflammation [337-338]. The effects of natalizumab have been reported to vary between patients, with full or partial response to therapy seen in 75% of patients [329]. The treatment is most effective during the early stage of the disease.

Therapy response is primarily monitored using MRI and clinical assessments. The latter includes measures such as EDSS and the multiple sclerosis functional composite, which have been recognised as suitable detectors of the effectiveness of therapy and monitoring of disease progression [339]. Nevertheless, there are notable caveats to the use of these scales; in particular, the EDSS is relatively subjective to the assessing clinician, insensitive to change, with a recognised ‘ceiling’ effect, and mostly weighted towards walking. Equally, while a high MRI lesion load at the early stage of the disease has been shown to be predictive of clinical disability development [82], this is not so at later stages, and the modality also does not allow a functional view of the brain, nor does it illuminate why a certain proportion of subjects do not respond to therapy. The inflammatory component of MS is known to drive disease progression in the relapsing- remitting phase by active infiltration of lymphocytes and other inflammatory cells, but it is also critical in the progressive phase, albeit through different processes [94, 340]. Indeed, inflammation in the CNS is of particular interest due to this discrepancy, which may be linked to the lack of response to anti-inflammatory disease modifying therapies in some patients.

Here, PET may offer some insight into the relationship between inflammation and response to therapy. We used a novel second generation TSPO-PET tracer, [18F]GE-180, which has good affinity [223] for the protein, to image microglial activation in a cohort of RRMS patients, at baseline and follow up after treatment with natalizumab. This sub- cohort of five RRMS patients are part of a larger ongoing study of 17 patients who are being treated with natalizumab for 13 months, with baseline, 10 week and 13 month follow up PET and MRI scans. In this short pilot analysis, we assess the potential efficacy of [18F]GE-180 as a biomarker of therapy response. Furthermore, we preliminarily evaluate different PET outcome measures in the context of detecting group differences in [18F]GE-180 binding.

At the early time point after therapeutic intervention (10 weeks), we were unable to detect significant differences in PET signal between baseline and follow up scans in

164

any standard anatomical regions. In one subject, who demonstrated T1 post-contrast enhancement in one lesion, indicating localised BBB breakdown, significantly lower 푉푇 was observed at follow up than at baseline. In spite of there being no significant differences in MRI lesion loads at follow up, PET findings suggest decreased tracer binding in the lesion area. All targeted anti-inflammatory disease modifying therapies are known to be most effective when taken during the acute phase of relapse; thus, one would expect this subject, with active Gd-enhancement, to respond well to therapy. Interestingly, however, this observation did not correlate with the EDSS score, which increased by one point between baseline and follow up, with the patient reporting a relapse in that time. This could be attributable to neuropathological and biological changes which were initiated before treatment. A more far-reaching possibility is that inflammation in the area of active lesions is not the driving force behind clinical disability. As mentioned previously, it has been observed that natalizumab is clinically ineffective in approximately 25% of patients; to our knowledge, an assessment of correlation between reduced inflammation as visualised by PET and changes in clinical symptomatology has not been published. Patients are expected to show a clinical response to natalizumab within 3 months [341]. Given the early time point of this analysis (< 3 months after treatment commencement), effects of treatment may not have manifested; evaluation at the 13 month time point may help to elucidate the relevance of our findings further.

In another subject, EDSS decreased from 4.0 to 3.5. At 10 weeks, this patient exhibited two new lesions in T2 and T1 pre-contrast MRI, which were not Gd-enhancing.

-3 The 푉푇 in these lesions was low (0.22 and 0.38 ml.cm respectively) compared to the

-3 highest lesion 푉푇 in this subject (0.64 ml.cm ). In this patient, NAWM had increased by 14% at follow up (n.s.). We hypothesise that these results tentatively suggest that an increase of microglial activation in NAWM might be linked to the development of these two new lesions, as opposed to an inflammatory component within the lesions themselves. Indeed, it has been shown that during the early stage of development of lesions, activated microglia, as well as astrocytes, which also express TSPO, are often present in NAWM, especially in perivascular areas (and bordering both active and inactive lesions) [342]. In this case, the increased binding of [18F]GE-180 is also consistent with previous reports of increased TSPO-PET signal in the NAWM of different cohorts of MS subject [20, 108].

An SUV analysis of uptake in standard regions, including NAWM, showed no significant differences between pre and post-treatment scans, suggesting no overall difference in binding of [18F]GE-180 at this early time point after commencing treatment.

165

Interestingly, in the two subjects (2 and 3) who exhibited significantly lower uptake in lesions at follow up, the distribution of lesion SUVs was similar (though lower, see Figure 5.4), and all decreased by a similar factor, suggesting that the difference is driven by a global decrease in uptake rather than an effect on particular lesions. A lesion characterisation based on FLAIR images showed little difference in the appearance of lesions between scans, however these two subjects reported improvement in symptoms after 10 weeks of natalizumab infusions and were also categorised as showing a clinical response to treatment. SUVs in patient 4, whose 푉푇 decreased in lesions after treatment, did not show a significant change, although uptake in the single Gd-enhancing lesion decreased by 21%. Although very preliminary, these findings might suggest a sensitivity of [18F]GE-180 PET to the expression of TSPO in the MS brain, which is different to information acquired through MRI.

The difference between results seen with quantitative kinetic methods (푉푇) and semi-quantitative SUV are likely to be in part attributable to the overall low penetration of [18F]GE-180 into the brain, see also Chapter 6. Simultaneously, the high blood signal of [18F]GE-180 may be responsible for a proportion of the SUV seen in ROI analysis. We thus performed a correction of SUV by subtracting the contribution of whole blood from online sampling in each region, assuming a global 5% vascular volume. While this correction did not change the global result of lower SUV_corr in two subjects, it should be noted that the correction performed was a simplification of the actual contribution of vascular signal. Equally, if [18F]GE-180 binds non-specifically to plasma proteins in the blood, the true 푓푝 may be much lower than this, leading to an underestimation of 푉푇 [343], see Chapter 6.

The NAWM of RRMS subjects has been shown to have significantly lower perfusion than that of healthy control subjects [344]. While estimates of blood volume from the 2TCM were between 3 and 6% across all brain regions, this might not be an insignificant effect and it cannot be ruled out that the overall blood volume contribution is lower in lesions than in other brain regions. Furthermore, in kinetic modelling with the 2TCM, these effects might not be detectable due to the relative sparing of major arteries to disease-related blood vessel changes [344]. Additionally, and again particularly in the case of lesions and NAWM, an association between vasculitis, or vascular inflammation, and the development of MS pathology has been observed [344-345]. TSPO is known to be highly expressed in the endothelium and could affect the binding of such high affinity tracers as [18F]GE-180 and [11C]PBR-28 [77, 154]. Where vasculitis exists in the MS brain, this might lead to a higher component of specific binding in the vessel walls, as well as altering the regional blood flow through narrower venous vessels. Using the arterial

166

plasma input function is performed under the assumption that the blood leaving the body is representative of the blood pool in the brain, i.e. that there are no biological or physiological processes which affect vessels within the brain, or more specifically that the input function generated from arterial sampling is the same as the input to tissue in the brain. If there are disease-specific effects on brain perfusion and vascular inflammation in the case of MS, as well as dispersion effects due to vascular narrowing/dilation, this might not be true. Ideally, a measure of cerebral blood flow in the MS brain using [15O]H2O might be performed. Although, in agreement with the work in Chapter 4, it has recently been shown that the 2TCM is the preferred fit to [18F]GE-180 human control data [278-279], without a measurement of flow rates, results should be interpreted cautiously. Furthermore, and in order to investigate the level of plasma- protein binding of [18F]GE-180 in human blood, we performed a study which is reported in greater detail in Chapter 6.

Clearly, the discrepancy between 푉푇 and SUV_corr measures requires further investigation. In this small cohort, each method appear to show some sensitivity to changes in the treated MS brain, but whether these measures are correlated to true reduction in inflammation due to natalizumab treatment or natural variation in microglial activation over the disease course, and concomitantly whether these changes are correlated to clinical indictors, remains to be clarified. Of course, it also cannot be ruled out that these results are due to a placebo effect in taking natalizumab treatment; a subject group which this study did not incorporate. Equally, since 푉푇, i.e. concentration ratio at equilibrium of tracer in target region to that in plasma, any post-treatment differences seen with this parameter could actually be representative of a change of tracer concentration or behaviour in blood. Namely, as described in Chapter 4, the baseline peak and time-to-peak for tracer in the blood of RRMS subjects were higher and earlier than those in healthy volunteers (HV). Assessing the same parameters at 10 weeks post-treatment in this subset of patients showed that, although there were no significant differences between pre and post-treatment, both the time to peak and the height decreased by up to 12 seconds and 35% respectively after 10 weeks of natalizumab therapy, outside of one SD. These lower results were more in line with those from healthy controls. Analysis of results from the larger longitudinal study may be able to shed further light on the findings presented here. It has previously been noted, both here and in other studies that the total brain uptake and extraction fraction of this tracer

-1 -1 are quite low (퐾1 ~0.005 ml.min .ml ). However, if the tracer is able to show such differences in binding, and more specifically if these have a pathological relevance to disease progression, and correlate with clinical parameters (both in conjunction or

167

complementary to MRI), its use as a biomarker of therapeutic response could be very valuable.

5.6 Conclusion

[18F]GE-180 TSPO-PET shows differences in uptake in the brains of RRMS subjects treated with natalizumab over the course of 10 weeks. Of the methods assessed

(푉푇 and SUV_corr), kinetic modelling was able to identify a decrease in signal in lesions in a single subject, whose clinical presentation was worse at follow up than at baseline. Conversely, decreased SUV_corr in two subjects correlated with clinical and patient- reported positive response to natalizumab. In all cases, there were no changes in MRI parameter outcomes. Results from this short pilot study are therefore promising, suggesting that [18F]GE-180 PET could be a valuable and sensitive tool for therapy monitoring. This deduction warrants further investigation in the full cohort of 17 patients and over the course of 13 months of natalizumab therapy.

168

6 Plasma-protein binding of the 18 kDa TSPO-PET tracer [18F]GE-180

Sujata Sridharan1, Elizabeth Barnett1*, Carrie-Anne Mellor1*, Rainer Hinz1, William Trigg2, Alexander Gerhard1, Christopher Clark3, Hervé Boutin1

1 Wolfson Molecular Imaging Centre, The University of Manchester, Manchester U.K.

2 GE Healthcare, The Grove Centre, Amersham, Buckinghamshire U.K.

3 Cancer Research UK Manchester Institute, The University of Manchester, Manchester U.K.

* Joint first authors

6.1 Abstract

Introduction: The 18 kDa translocator protein (TSPO) is upregulated in inflammatory diseases of the central nervous system. Several positron emission tomography (PET) radioligands for TSPO have been developed. One such ligand, [18F]GE- 180, has shown good signal to noise ratio and selectivity for TSPO in animal models, especially compared to the original ligand [11C]-(R)-PK11195, but clinical subjects have shown low brain penetration. Plasma-protein binding of PET tracers and the corresponding plasma free fraction, 푓푝, can have a significant effect on the availability of the tracer in tissue. Here, we experimentally determined the plasma-protein binding of [18F]GE-180. Methods: Blood samples were taken from two sub-cohorts of healthy volunteers (n = 5 and n = 3) and genotyped for the TSPO single nucleotide polymorphism. Plasma-protein binding assays were performed on a Centrifree® device and activity of free tracer measured in a well counter. Samples were incubated as hot tracer only, hot tracer with cold GE-180 and hot tracer with cold PK11195 to determine the displaceable fraction of tracer binding. Results: The percentage of free [18F]GE-180 in hot tracer only assays (i.e. 푓푝) was approximately 0.1%. In displacement assays, the percentage of free tracer increased by up to 60 fold with GE-180 and up to 35 fold with PK11195, indicating a difference in binding sites of the two tracers on TSPO, as well as specific binding in the blood. Importantly, the low 푓푝 indicates large proportion of non-displaceable binding of [18F]GE-180 to plasma proteins. Conclusions: Experimental techniques require further development, with an extension proposed to measure plasma-protein binding of other TSPO tracers and in other species. Preliminary results indicate a high non-specific binding of the tracer to plasma proteins in human blood.

169

6.2 Introduction

All positron emission tomography (PET) ligands are susceptible to a certain level of binding to the proteins found in human blood. Typically, this is via pharmacodynamically rapid and loose lipophilic or hydrogen-bonding interactions between the ligand and plasma proteins or the membranes of blood cells (which are rich in lipids). In general, the higher the lipophilicity of the tracer, the greater the level of this binding [346]. While the protein-bound tracer cannot easily cross the BBB through capillary membranes [347], for tracers exhibiting low levels bound in plasma, as well as a rapid equilibrium between free and bound states in plasma, this does not significantly decrease the amount of tracer available to tissue, especially considering the time course of a typical PET scan [348]. Conversely, in cases where equilibrium is not reached over the residence time of the tracer in blood and/or where the level of bound tracer in plasma is high, the transport rate constant from plasma to tissue, 퐾1, is reduced by a factor depending on the level of binding to plasma components [348].

A large proportion of plasma protein binding of PET tracers is to human serum albumin (HSA), which comprises approximately 60% of total serum proteins. The molecular structure of HSA allows it to bind non-saturably and irreversibly to many ligands with high affinity [349]. Unfortunately, the affinity of a ligand or drug for HSA has no correlation with its affinity for serum albumin from other species [350], perhaps in part explaining the difference in pharmacokinetic behaviour of some PET tracers between animals and humans. Measurement of binding to plasma proteins in vivo is impossible, but reliable in vitro techniques have been developed to give a good indication of the behaviour of a tracer within a region of interest [351-352].

The 18 kDa translocator protein (TSPO), which is expressed at low levels in the healthy brain, but is found in greater concentrations in the heart, liver and testes among other organs [299-300, 353-354], increases in neuroinflammatory disease. Since neuroinflammation is concurrent with the release of an array of pro-inflammatory cytokines by microglia and other immune defence cells (see [236, 355] for review), this may affect the availability of TSPO-PET tracers for passage into tissue, especially in cases where peripheral expression of inflammatory markers is high [301]. Indeed, the corresponding plasma free fraction (푓푝) of many TSPO ligands is very low (for example ~1% for the prototypical ligand, [11C]-(R)-PK11195, and 9% for the second generation

11 tracer [ C]DPA-713 [356]). Table 6.1 summarises the 푓푝 values for selected TSPO tracers which can be found in the literature.

170

Among other characteristics, the ideal TSPO-PET tracer should have: a measurable 푓푝; a lipophilicity in the appropriate range (logP between 1.5 and 4) for brain penetration by passive diffusion, but not so high as to amplify non-specific binding; reasonable affinity for the target; and no brain-penetrating radiolabelled metabolites [343]. Unfortunately, several of these features are paradoxical; for example, while affinity for the target must be high, usually in the nanomolar range, it must also not be too high (femtomolar) so that binding equilibrium is reached within the duration of the scan. Additionally, the higher the lipophilicity, the more likely a tracer is to cross the blood brain barrier, but this also increases the component of non-specific binding of a tracer to plasma proteins and lipids, dampening the signal to noise ratio in in vivo imaging. A further possible confounding factor specific to TSPO tracers is the recent discovery of a single nucleotide polymorphism [77] which affects their affinity for TSPO binding sites in the brain and could be correlated with the behaviour of the tracer in blood [314].

[18F]GE-180 is a novel TSPO tracer which has shown good signal to noise ratio and specific brain uptake in preclinical studies of neuroinflammatory conditions [115, 122, 294]. Personal observations (chapters 4 and 5), as well as other early clinical data

[278-279] have indicated that the brain penetration in humans, in particular 퐾1, is very low, both in healthy and disease states. Additionally, the blood signal of [18F]GE-180 is high and persistent, suggesting a low extraction fraction through the capillary membrane into tissue. One possible explanation for this retention in the blood is the presence of binding of the tracer to plasma proteins.

In order to investigate this observation further, an in vitro plasma-protein displacement assay was performed in a group of volunteers. Radiolabelled [18F]GE-180 was displaced with unlabelled (cold) GE-180 and PK11195 in order to account for possible differences in binding sites of TSPO between the two tracers [75]. Volunteers were also genotyped for the TSPO single nucleotide polymorphism.

171

Table 6.1 Summary of free fraction of tracer in plasma (풇풑) in healthy subjects for some common TSPO-PET tracers. * This study was performed in rhesus monkey † Personal communication, William Trigg

Tracer 풇풑(%) Lipophilicity In vitro affinity Reference 풇풑, cLogP, 푲 (nM) ([c]LogP) 풊 푲풊

[11C]-(R)-PK11195 1 3.4 9.3 [356], [357], [358]

[18F]-PBR111 HAB 5 3.2 3.7 [314], [359], [359] MAB 7 LAB 6 [11C]DPA-713 9 2.4 5 [356], [357], [357]

[11C]PBR-28 6* 3.0 4 (high), 200 [360], [361], [75] (low)

[18F]PBR-06 4* 4.4 0.997 [360], [362], [362]

18 † [ F]GE-180 - 3.8 0.87 [223] (퐾푖)

172

6.3 Materials and methods

Five volunteers were recruited from the Wolfson Molecular Imaging Centre (WMIC) at the University of Manchester and blood analysis was performed on site. 40ml of blood was withdrawn from each subject by an experienced radiographer into ethylenediaminetetraacetic acid (EDTA)-lined Vacutainer® tubes to minimise blood coagulation. Samples were anonymised but all subjects were Caucasian. Approximately 10ml per subject was used for genotyping, with the rest devoted to plasma-protein binding assessment.

6.3.1 Genotyping assays

6.3.1.1 Blood preparation and DNA extraction

Each blood sample was prepared for genotyping by performing DNA extraction using a QIAGEN QIAamp® Blood and Cell Culture DNA Mini kit. Genotyping was performed in-triplicate to allow for potential errors in processing. In short, blood samples were split into 3 × 1 ml aliquots in 10ml EDTA falcons. 1 ml of buffer C1 (see Supplementary Materials for full description of buffer compositions) and 3 ml of distilled water at 4°C were added to each 1ml sample and the tubes were inverted approximately 10 times to disperse buffer through the sample. Buffer C1 lyses cells (specifically erythrocytes first) while stabilising the nuclei, releasing haemoglobin and making the resulting suspension translucent. Lysed samples were then centrifuged in a swing out rotor at 1300×g (4°C) for 15 minutes. Supernatant was pipetted out and discarded, leaving a small nuclear pellet. To remove residual haemoglobin, a further 0.25ml of buffer C1 and 0.75ml distilled water were added and the pellet re-suspended by vortexing briefly. Samples were then centrifuged again for 15 minutes at 1300×g and the supernatant again removed and discarded. Pellets were then frozen in the falcons at - 20°C overnight.

The following day, 1 ml of another buffer (G2) was added and each sample was vortexed thoroughly to achieve full re-suspension. Buffer G2 lyses nuclei and denatures proteins, including nucleases and viral particles. 25 µl of QIAGEN Proteinase K solution, which digests denatured proteins, thus facilitating their full removal in the following steps, was added to each sample. Samples were then incubated in a water bath at 50°C for 60 minutes.

QIAGEN genomic-tips, provided in the Blood and Cell Culture Mini kit, were first equilibrated with 1 ml of buffer QBT, which contains 0.15% Triton® X-100, a detergent

173

which reduces surface tension in the tip. The samples for each subject were taken from the water bath and vortexed for 10s at maximum speed (to improve flow rates and reduce risk of clogging) before being loaded promptly onto the genomic-tips (1 tip per sample). Each tip was then washed 3 times with 1 ml each of buffer QC, emptying through the tip by gravity flow. DNA was then eluted by placing the tip over clean microcentifuge tubes (2 per sample) and applying 2 x 1 ml of buffer QF at 50°C. DNA was precipitated by adding 1.4ml room temperature isopropanol to each sample, mixing and centrifuging at 5500×g for 15-20 minutes at 4°C. A washing step to purify was then performed by adding 1 ml ice cold 70% ethanol, vortexing and centrifuging again at 5500×g (4°C) for 10 minutes. Ethanol removes the precipitated salts, allowing for improved re-dissolution. This washing step was repeated to maximise DNA yields.

After removing the supernatant from the second wash, the resulting pellet was then air-dried for at least 10 minutes and then re-suspended in 1ml of Tris-EDTA buffer (pH 8.0). Samples were left to dissolve overnight on a shaker at room temperature.

6.3.1.2 Genotyping

Genomic DNA (gDNA) samples were genotyped using a TaqMan® single nucleotide polymorphism (SNP) genotyping assay (Applied Biosystems, Thermo Fisher Scientific Inc., USA). As per the associated protocol, stock solutions were diluted to 20x working stock with TE buffer. The probes were custom made by the supplier to identify the alanine (ACG) for threonine (GCG) substitution associated with the TSPO SNP rs6971. In this case, A is the mutant (allele 1, VIC® probe) and G is the wild-type (allele 2, FAM® probe).

Each sample was made up to 2.75 µl in total. The 384-well reaction plate was prepared with 2× TaqMan Universal polymerase chain reaction (PCR) Master Mix - 2.5 µl/well; 20× primer and probe dye mix (VIC/FAM)- 0.25 µl/well, vortexed before use, and the gDNA samples. The plate was sealed to prevent evaporation of the samples and PCR amplification was performed on an Applied Biosystems 7900HT Fast Real-Time PCR System with 384-Well Block Module, with pre and post-amplification background runs to normalise readings.

6.3.2 Plasma-protein binding displacement assay Whole blood samples were received from volunteers at staggered intervals across one day and immediately placed on ice. [18F]GE-180 was synthesised on a FASTLabTM platform as described previously [223]. The specific activity of radiotracer on

174

the day was 219.15 GBq.µmol-1, with a cold concentration of 0.51 µg/ml (typical for synthesis of [18F]GE-180 on the FASTLabTM). Cold standards (excess) of PK11195 and GE- 180 were made up as per the concentrations in Table 6.2.

Table 6.2 Cold spike concentrations for displacement of hot [18F]GE-180 in plasma samples. Cold standard Stable spiked concentration per 1ml plasma sample (µg)

PK11195 1400

GE-180 2557.5

Whole blood samples from each patient were centrifuged at room temperature for 2 minutes at 2800×푔 to separate plasma. Plasma from each subject was processed in triplicate in EDTA tubes, with samples for hot tracer only, hot tracer spiked with excess cold GE-180 and hot tracer spiked with excess cold PK11195. In total, 3.25 ml plasma was used for each subject (3×1 ml samples and 0.25 ml aliquots for counting only).

With cold tracers added to the samples first, [18F]GE-180 was made up to approximately 14 kBq.µl-1 and spiked into each as shown in Table 6.3. Samples were incubated for 15 minutes at 37°C and counted in a well counter to confirm the amount of activity in each pot. Plasma samples were then pipetted into a Centrifree® Ultrafiltration device (EMD Millipore, ref 4104, lot: R5BA49658), which allows separation of free and unbound solutes through a cellulose membrane. The device was weighed in this full state. Samples were then centrifuged again at 1500×g for 15 minutes at room temperature. Ultrafiltrate (unbound tracer) was pipetted off into a pre-weighed tube and counted in the well counter.

Table 6.3 Volumes of hot [18F]GE-180 spikes for each conditions (hot tracer only, hot tracer with excess cold GE-180 and hot tracer with excess cold PK11195). Subject Condition 3. Concentration of hot tracer, cold tracer, cold PK11195 (µM, mM) 1 Hot, cold GE-180 or cold PK11195 2 Hot, cold GE-180 or cold PK11195 3 Hot, cold GE-180 or cold PK11195 4 Hot only 64, 3.96, 7.38 4 Cold (GE180 or cold PK11195) 5 Hot only 5 Cold (GE180 cold PK11195)

175

6.4 Results

6.4.1 Genotyping For optimum PCR amplification, gDNA samples should be at a concentration of at least 0.7 ng/µl in each well. Of 6 samples per subject (30 in total), 8 were of too low a yield to meet this requirement. There was, however, enough gDNA per subject to determine their genotype using an allelic discrimination plot (see Figure 6.1). The cluster of points in the lower left of the plot are those of non-template controls. The allelic frequency of G (allele 2, y-axis), the wild-type, was 5-6 times higher than that of A (mutant) for all samples, indicating that all 5 volunteers were high affinity binders (HABs).

6.4.2 Plasma-protein binding A large concentration of excess unlabelled GE-180 (mmol vs. nmol labelled tracer) was used in displacement assays; almost double of that of cold PK11195. This was because, during optimisation of the experiment, it was noted that lower concentrations of GE-180 seemed to be unable to produce an excess; i.e. to displace all labelled [18F]GE- 180. This was a result of the concentration of hot tracer used also being extremely high; for displacement experiments, radioligand concentrations should be of the order of 퐾푑, the equilibrium dissociation constant (i.e. approximately 2 nM for [3H]-PK11195 [363]). Additionally, preparation for assays, including the addition of hot tracer, was performed in plastic EDTA tubes. It is known that the ability of plastics to bind lipophilic drugs can interfere with experimental results of ligand-binding studies [364-365]. The lipophilicity of [18F]GE-180 has been measured as moderate (cLogP = 3.76, see Table 6.1) and it is thus possible that the level of non-specific binding of the hot tracer to the plastic tubes is enough to alter the results. Indeed, it is likely that this effect also interfered with the steps to optimise the concentrations of cold ligands needed to fully displace [18F]GE-180.

We therefore chose to repeat the study in a further subset of volunteers, this time using a recalculated concentration of unlabelled ligands and plastic-free equipment (including low-binding pipette tips and glass tubes). We also incubated hot and cold tracer with pre-extracted plasma and, separately, incubated whole blood before centrifuging to obtain plasma. The latter approach was to simulate in vivo conditions; accounting for that fact that tracer enters whole blood before partitioning between cells and plasma. Inconclusive results from the first stage of the experiment are presented in the Supplementary Materials.

176

Figure 6.1 Allelic discrimination plot for genotyping of the TSPO single nucleotide polymorphism rs6971 in five volunteer subjects. All subjects were high affinity binders, as indicated by the 6-7 times greater allelic frequency of allele Y.

177

6.4.3 Repeat experiment Three different volunteers were recruited from the WMIC and informed consent was given prior to blood samples being taken. Blood samples of 60 ml were withdrawn from each subject by an experienced radiographer (more blood was required to perform the incubation in whole blood as well as extracted plasma). DNA extraction and genotyping were performed as previously, with three new subjects also being categorised as HABs. Specific activity of [18F]GE-180 was 165.22 GBq.µmol-1.

6.4.3.1 Method adjustments Table 6.4 is a summary of the quantities of hot and cold tracer used in whole blood and plasma. As control, buffer was also co-incubated with hot tracer instead of blood. Pre-extracted plasma samples were processed as previously but in duplicate (2.85 ml per condition). After incubation, plasma samples were held at 4°C to minimise further biological metabolism of the tracer while other samples were processed. Whole blood samples (total 20ml/subject – 5.70ml for each condition) were incubated for 1 hour at 37°C with either hot [18F]GE-180 only, [18F]GE-180 and cold GE-180, or [18F]GE-180 and cold PK11195, before being centrifuged for 2 minutes at room temperature (2800×푔) to obtain plasma. 250 µl of plasma from each condition was then weighed and counted in a well counter. Of the remaining plasma, 1 ml was added to the Centrifree® device, which was then weighed full. Samples were centrifuged at 1500×g for 15 minutes at room temperature. Ultra-filtrate was pipetted off into a pre-weighed tube and counted in the well counter. The empty Centrifree® device was also counted in the well counter, as were the used (low-binding) pipette tips.

Table 6.4 Volumes and spike concentrations of cold and hot tracer used in blood samples. Condition Volume of tracer Concentration of (µl) hot/cold tracer stock solution (nM)

Plasma [18F]GE-180 150 20

Plasma [18F]GE-180 + GE-180 150/150 20/20000

Plasma [18F]GE-180 + PK11195 150/150 20/20000

Whole blood [18F]GE-180 300/300 20

Whole blood [18F]GE-180 + GE-180 300/300 20/20000

Whole blood [18F]GE-180 + PK11195 300/300 20/20000

178

An additional experiment was designed to measure the protein-bound concentration of [18F]GE-180 (as opposed to the amount of free tracer washed out through the Centrifree®) in whole blood, plasma and the red blood cell pellet by filtration through a vacuum filter rig. This was performed by collecting the whole blood, plasma and red blood cell pellets from the whole blood Centrifree® part of the experiment (stored on ice until use, after the Centrifree® experiment). A buffer of 50 mM Tris, 120 mM NaCl and 1 ml 0.05% Tween was mixed 1:1 with a polyethlyeneimine (PEI) membrane solution. PEI was also used to pre-soak the filters used to collect the plasma proteins, for 1 hour at room temperature, prior to use. Each sample of whole blood, plasma and red blood cell pellet was applied through these pre-soaked filters using a vacuum filter rig. The filters were then washed with 4×5 ml of buffer before being removed and allowed to diffuse into 1 ml of distilled water. The filter paper was then counted in the well counter to measure the remaining activity in the trapper proteins from plasma.

6.4.3.2 Results Results from the Centrifree® experiments are summarised in Table 6.5. The total remaining percentage of free [18F]GE-180 after ultrafiltration was very low in samples incubated with hot tracer only. As seen previously, the fraction of free tracer increased when hot tracer was co-incubated with cold PK11195, and more so when co-incubated with cold GE-180. 푓푝 measured in this data as the fraction of free tracer in plasma measured with hot tracer only was 0.05±0.01%.

Table 6.5 Average percentage of unbound [18F]GE-180 from whole blood incubated with tracer and cold spikes before plasma extraction. Subject Average % Average % unbound tracer Average % unbound tracer unbound tracer after cold GE180 addition after cold PK11195 addition

1 0.06 1.86 1.03 2 0.04 0.56 0.44 3 0.06 2.40 0.33

Furthermore, the discrepancy in absolute percentages of free tracer in all conditions between the repeat and first Centrifree® experiments suggests that there may indeed have been a fraction of displaceable binding to the plastic equipment used in the former. Tracer delivery occurred in the morning, and due to limitations with the number of samples which could be well-counted simultaneously, the filter counting could not be performed until after the Centrifree® experiment. The resultant loss of activity and the

179

fact that the first half of the experiment was time-consuming meant that results for this extension were not accurately quantifiable by the time of counting (i.e. at the level of background counts). They are therefore not presented here.

180

6.5 Discussion

TSPO, at a low basal level in the human central nervous system, is upregulated in a wide variety of inflammatory brain conditions [20, 166, 183, 214]. The original and still most widely used TSPO radioligand, [11C]-(R)-PK11195, suffers from a high level of non- specific binding. The push to develop improved PET radioligands for TSPO has resulted in several new tracers, including [18F]GE-180. Compared to [11C]-(R)-PK11195, the tracer has shown improved specificity and a higher signal to noise ratio in preclinical models of neuroinflammation [115, 122]. Unfortunately, this has not translated to clinical models using [18F]GE-180. One possible cause for the low apparent brain uptake in humans is a high level of non-specific binding to plasma proteins. Thus, the aim of this (short pilot) study was to quantify the plasma-protein binding behaviour of the second generation TSPO-PET tracer [18F]GE-180.

As mentioned, [18F]GE-180 human brain scans have revealed a low overall signal in the brains of both healthy [278-279] and diseased human subjects (Chapters 4 and 5). Although low lipophilicity is a desirable characteristic of a TSPO tracer in order to enable its passage across the BBB by passive diffusion, this is not the only mechanism by which it can enter the brain. Here, we investigated a possible contributing factor to the low apparent brain uptake of [18F]GE-180; the pharmacological binding of the tracer in blood. Another possible explanation, however, could be that the tracer is a substrate of an ABC (ATP-binding cassette) transporter such as P-glycoprotein (P-gp) or breast cancer resistance protein (BCRP). In this case, the overall retention of the tracer in brain tissue over the duration of the scan would be similarly low due to the active efflux mechanism of these proteins. It should be noted that P-gp and BCRP both also function in rats, where [18F]GE-180 shows a good brain uptake. Nevertheless, it is possible that the mechanism of transport for the tracer is species-dependent, as the behaviour of ABC-transporter substrates has been reported to be variable [366-367]. It is thus worth investigating pharmacological blockade of one or multiple efflux pumps (for example for P-gp, with cyclosporine) prior to PET scanning of rats and/or humans. This might allow an understanding of whether a given efflux pump is preventing [18F]GE-180 from binding to TSPO in brain tissue (see [368] for review).

Results suggest that the amount of specific binding of the tracer in blood is low – as indicated by a small increase in free hot tracer when specific binding sites were blocked with either cold GE-180 or cold PK11195. Furthermore, the difference between the level of displacement with the two cold ligands suggests a likely difference in binding sites on TSPO. The fact that free tracer was extremely low, even after blocking with cold

181

ligands, implies that a large proportion of [18F]GE-180 (~99%) binds non-specifically in plasma. Addressing the fact that a pre-extracted plasma sample is not representative of the in vivo behaviour of the tracer by also incubating whole blood before centrifuging for plasma increased the amount of free tracer only very slightly (>95% of tracer still bound).

Although the results are highly suggestive of a large component of strong non- specific binding to plasma proteins, there were a few limitations to this study. Most notably, there were difficulties with scheduling of this long set of measurements which meant that quantification of the protein-bound tracer was not possible in the repeat experiment as counts were at the level of background activity. We were also unable to assess possible differences in the plasma-protein binding between high and mixed or low affinity binders as the results of genotyping indicated that all eight volunteers were HABs. Especially given that the reported prevalence of HABs in the Caucasian population is between 65 and 70%, the probability of finding all HAB subjects is extremely small [77]. However, given our small sample size, it is possible that the smaller MAB fraction of the population was missed.

We therefore suggest, given the potential relevance of this tentative finding to future clinical studies planned with [18F]GE-180, an extension of this pilot study. This would be designed as a comparative investigation of the plasma-protein binding behaviour of [18F]GE-180 compared to [3H]-PK11195 (to avoid the short half-life of 11C) and another TSPO tracer such as [18F]DPA-714, which is easily synthesised in the WMIC. A parallel arm to this extended study could be to assess the protein binding of these tracers in rodent blood, since the behaviour of [18F]GE-180 in particular is noticeably different between species. The genotyping of blood samples could be performed again at a different site to account for possible bias in results, and ideally with a sample of known genotype as control. Finally, the experiment should be split over the course of two days to allow measurement of unbound tracer (ultrafiltration with the Centrifree®) and of the protein-bound tracer using the vacuum filter rig without significant loss of activity by the end of the day.

182

Supplementary materials Supplementary table 6.1 summarises binding results. The percentage of unbound [18F]GE-180 in ultrafiltrate was low for all subjects. After addition of cold GE-180, the amount of unbound tracer increased by between 30 and 50 fold, indicating a large proportion of [18F]GE-180 binding to components of plasma. Adding an excess of cold PK11195 similarly increased the percentage of unbound tracer by up to 30-fold.

Supplementary table 6.1 Individual subject results for hot tracer only and with excess cold GE-180 and PK11195. Subject Average % Average % unbound tracer Average % unbound tracer unbound tracer after cold GE180 addition after cold PK11195 addition

1 0.12 5.86 3.29

2 0.12 5.85 3.88

3 0.20 7.62 4.75

4 0.17 6.11 3.50

5 0.23 6.81 3.85

Supplementary table 6.2 Composition of buffer solutions used in blood preparation and gDNA extraction from Blood and Cell Culture DNA Mini Kit, QIAGEN. Buffer (function) Composition

C1 (cell lysis) 1.28 M sucrose; 40 mM Tris·Cl, pH 7.5; 20 mM MgCl2; 4% Triton X-100

G2 (digestion) 800 mM guanidine HCl; 30 mM Tris·Cl, pH 8.0; 30 mM EDTA, pH 8.0; 5% Tween- 20; 0.5% Triton X-100

QBT (equilibration) 750 mM NaCl; 50 mM MOPS, pH 7.0; 15% isopropanol, 0.15% Triton X-100

QC (washing) 1.0 M NaCl; 50 mM MOPS, pH 7.0; 15% isopropanol

QF (elution) 1.25 M NaCl; 50 mM Tris·Cl, pH 8.5; 15% isopropanol

TE (storage/suspension of DNA) 10 mM Tris·Cl, pH 8.0; 1 mM EDTA, pH 8.0

183

7 Summary and conclusion

7.1 Recapitulation of research aims

This project was designed in conjunction with General Electric (GE) Healthcare in order to evaluate a newly developed PET radiotracer for TSPO, [18F]GE-180. The main impetus behind the research is the fact that inflammation plays a role in almost all neurological diseases. Its exact contribution to the aetiology and progression of these diseases is a broad and ongoing field of research, to which I have endeavoured to contribute a small fraction. Although this thesis is presented in alternative format and each chapter necessarily already contains a discussion and conclusion, I have here summarised the overall results and explained their relevance to the field, including an outlook on possible future avenues of work.

The overall aim of the project was to quantitatively characterise the behaviour of the tracer in and ex vivo in different disease models, as well as to compare and contrast its efficacy with other TSPO-PET radiotracers, including the prototypical ligand [11C]-(R)- PK11195. While each study presented in this thesis was designed to address different gaps in our understanding of inflammation in brain diseases, in the interests of succinctness, let us condense these various research aims into two main questions and tackle them in the context of each chapter:

1. What is the most appropriate methodology for kinetic quantification of [18F]GE-180 animal and human PET data? 2. Is [18F]GE-180 a potentially useful PET imaging tracer? In other words, can the tracer provide a specific and selective signal compared to other TSPO-PET tracers in models of neuroinflammation?

7.2 Preclinical studies

7.2.1 The LPS model of inflammation – Chapter 2 The aims of the preclinical study presented in Chapter 2 were two-fold: 1) is the current kinetic modelling methodology of small animal TSPO-PET data commonly used in stroke and excito-toxic lesion models still valid in cases where inflammation is of a lower level? And 2) how does [18F]GE-180 compare with [11C]-(R)-PK11195 and another second generation TSPO tracer, [18F]DPA-714?

In order to explore the first question, we adopted a model of low-dose inflammation in the rodent brain; administering 1 µg of LPS, a potent endotoxin

184

inflammagen found on gram-negative bacteria, stereotactically to the right striatum of adult rats. The dose is lower than that previously and recently used in other preclinical TSPO-PET studies [122-124] and produced a lower lesion-to-reference ratio which is clinically more comparable to the level of inflammation seen in AD or MS. We then performed a comparison of the traditionally used modelling methodology (contralateral reference tissue input to the SRTM) with a cerebellar reference tissue and a modified supervised clustering approach. Results showed that both the cerebellar and the supervised clustering derived references consistently underestimated 퐵푃푁퐷, suggesting a component of specific binding in these tissues which has not been observed in acute excito-toxic models [124]. Cluster analysis has shown some success in clinical studies, where reference voxels are identifiable based on their kinetic similarity to a reference class of grey matter [20, 135, 196-197]. The interest in attempting to use a supervised clustering approach stemmed from the fact that the currently used method of quantification in preclinical PET data at the WMIC, which is based on ROI definition via automated local means analysis in the BrainVisa/Anatomist analysis package, may not be appropriate in models where activated microglia are dispersed throughout brain tissue. Rodent brain data from a stroke model was interrogated to define the number of classes into which voxels from the LPS brains would be partitioned during supervised clustering. The shapes of TACs from the automatic segmentation performed in the BrainVisa/Anatomist framework were reviewed and grouped based on their TAC shapes, also taking into account the percentage contribution of each class to the whole brain. Ultimately, this number was reduced from 4, which is typically used in clinical studies, to 3. Given that the number of voxels in the rodent brain is considerably lower than in humans, and the disease model used for class definition was relatively simple (unilateral infarct) this modification was not unexpected.

The fact that the resulting 퐵푃푁퐷 estimates in LPS animals using the clustering method were significantly lower than those from the contralateral reference input to the SRTM is highly suggestive that the ‘normal’ binding class of voxels identified in the LPS brain by the supervised clustering algorithm contained some element of specific binding. It is currently unclear as to whether this is due to the choice of the stroke model for identification of classes; if the ‘normal’ tissue defined from this model by automatic segmentation in BrainVisa is actually contaminated by some specific signal, then the tissue class identified as reference in the LPS brain is also likely to be. Additionally, the group difference identified with a contralateral reference input ([18F]GE-180 exhibited

11 18 higher 퐵푃푁퐷 than [ C]-(R)-PK11195 in dual scanned animals, while [ F]DPA-714 did not) was not replicated when using the supervised clustering input, further suggesting a

185

mis-identification of reference tissue in the latter. Interestingly, significantly lower 퐵푃푁퐷 was observed with [18F]DPA-714 than [11C]-(R)-PK11195 in dual scanned animals with the supervised clustering method. It is possible that this was representative of a higher level of specificity for TSPO of [18F]DPA-714 than [11C]-(R)-PK11195; in this case, any activated microglia (and thus TSPO) in the reference tissue voxels selected by the supervised clustering method would affect SRTM estimates for [18F]DPA-714 proportionately more. Furthermore, however, this would imply that the specificity of binding of [18F]DPA-714 is greater than that of [18F]GE-180, while displacement studies have shown a similar level of specific binding in (for example) stroke and excito-toxic models [115, 208]. In itself, this explanation thus seems unlikely to drive the large discrepancy seen in results with the cluster-derived input and the contralateral reference input, since we also demonstrated that the contralateral uptake in stroke (and LPS) animals was not significantly different to that from naïve control animals. Additionally, it should be considered that the basal level of TSPO expression in the healthy brain (in microglia, astrocytes and endothelial cells) is non-pathological. Given that we aim to identify a pathological component of disease, the basal presence of TSPO in the reference region might not spell the end for reference region modelling, but merely limit the signal to background ratio of a tracer. Some clinical studies have also concluded that a cerebellar reference region is preferred over supervised clustering [182, 369] for [11C]- (R)-PK11195. The same phenomenon here is thus not entirely unexpected, though clearly more work is needed to establish whether this systematic discrepancy can be reduced by optimising the definition of supervised clustering classes. This might be investigated further by redefining classes in a different model of neuroinflammation, potentially including a blood class using the left ventricle of the rodent heart as blood pool rather than relying on the LMA segmentation algorithm. Additionally, the method should be validated against an arterial plasma input function (see Chapter 3 and section 7.2.2) and ideally in a more representative model of subtle inflammation, such as a transgenic AD model. Equally, an in vitro autoradiography study of the other two tracers in this study, [11C]-(R)-PK11195 and [18F]GE-180, would help to clarify the level of specific binding of each tracer compared to [18F]DPA-714. Of course, the overall result may be that the current methodology for quantification using a contralateral reference region input is the least biased, and that quantification in preclinical AD models can best be performed semi-quantitatively rather than with full kinetic modelling.

In summary, to answer the questions posed in section 7.1, the findings of this study are: 1) the contralateral reference region remains the best modelling approach to identify group differences in low-dose LPS-injected vs. naïve animals. This is consistent

186

with previous findings [124], however, a novel finding is that the cerebellum is not an appropriate choice of reference input due to the likely presence of specific binding in the region, which seems to be particular to the low-dose model compared to more acute models [124]. The second key finding is that, using this methodology 2) while [18F]GE-

11 180 is able to identify a higher signal to noise ratio and 퐵푃푁퐷 than [ C]-(R)-PK11195 in animals dual scanned with both tracers, there was no significant difference between [18F]GE-180 and [18F]DPA-714. In other words, the tracer is more sensitive than [11C]- (R)-PK11195 at identifying inflammation in this low-dose preclinical model, while [18F]DPA-714 is not, but in the wider context, [18F]GE-180 does not demonstrate a significant improvement over [18F]DPA-714 in terms of lesion to background ratio.

7.2.2 Quantification of [18F]DPA-714 brain data – Chapter 3 Since arterial sampling in rodents is technically challenging, it has not previously been performed on-site at the WMIC. Its use in preclinical studies could be invaluable, however, especially in the validation of new PET tracers and disease models. The motivation behind this experimental study was thus two-fold: 1) to set up a detector and sampling system for rodent studies using a well-described tracer and 2) to characterise its use in the LPS model from Chapter 2. Due to practical constraints with availability of precursor and FastLab® cassettes, and also considering the broader applicability of certain preclinical TSPO tracers within the WMIC, we elected to perform this characterisation of an arterial sampling system using [18F]DPA-714 rather than [18F]GE- 180. As a parallel aim, we also investigated the potential use of a novel external population-based reference input for kinetic modelling as an alternative to the contralateral region.

Naturally, since it was designed as a pilot study, the statistical power of this analysis is relatively limited. Nevertheless, the key findings were that 1) the 2TCM fits data better than the 1TCM in this model and 2) tentatively, the external reference input may have some advantage over the contralateral region, however, 3) arterial sampling for [18F]DPA-714 in the current setup is subject to large dispersion effects. Clearly these findings are not mutually exclusive; dispersion effects are tracer specific and require more thorough investigation and the system requires optimisation; until this is done, it is difficult to draw conclusions about the validity of an external reference tissue input for modelling. The results from plasma input functions are dependent on the accurate measurement of whole blood; specifically, large dispersion effects lead to an underestimation of the true plasma input function, and consequently an overestimation of tracer influx, fractional blood volume contribution to tissue signal and volumes of

187

퐾1 푘3 distribution. The definition of 푉푇 in the 2TCM as (1 + ) necessarily of course 푘2 푘4 includes the calculation of rate constants. Assuming that estimates of these are affected by dispersion to the same degree may not be appropriate, given that they are by definition tissue-dependent. A method described by Munk et al. [246] was used to correct the whole blood sampling curves for dispersion and a comparison of parameter estimates with and without this correction revealed a substantial reduction in 퐾1 and 푉푏 to within a physiological range. Clearly, the effects of dispersion must be accounted for and while it appears that this was successfully performed here, a large delay was also present (~80s), thus optimisation of the system to minimise the effects is still ongoing.

While this study did not directly address the research questions posed in section 7.1, there is value to the results, as demonstrated. Ultimately, the value of arterial sampling in preclinical studies at the WMIC must be assessed, as well as its use as a validation tool for other kinetic quantification methods. Shortening of the catheter sampling tubing and material should help to reduce dispersion of the input functions. Additionally, kinetic validation in this larger cohort may elucidate whether the external reference tissue approach is more suitable for analysis of rodent data from [18F]DPA-714 LPS-injected animals.

7.3 Clinical studies

7.3.1 Baseline RRMS and healthy data – Chapter 4 The work presented in Chapter 4 pertains to a larger cohort of 17 RRMS subjects and 11 healthy volunteers who are part of a study into natalizumab therapy response monitoring with [18F]GE-180 (Chapter 5). With any novel tracer, it is important to establish baseline behaviour and a kinetic quantification methodology before assessing its potential use longitudinally. Thus, Chapter 4 was designed as a preliminary investigation into quantification of [18F]GE-180 PET brain data in healthy volunteers and patients with active RRMS at the pre-treatment time point.

A sub-cohort of 9 RRMS and 10 HV subjects was chosen from the baseline data available at the time of analysis and writing. The analysis focussed mainly on two points: 1) what is the appropriate compartmental model to describe [18F]GE-180 brain data? And 2) is [18F]GE-180 able to identify group differences in binding between RRMS and HV cohorts? A tertiary aim was to assess the in vivo difference in binding of HABs and MABs.

In order to address the first of these questions, blood data from the subjects was used to generate arterial plasma input functions for use with the one and two tissue compartment models. Tailored kinetic models have been suggested for use with other

188

TSPO-PET tracers, most notably compartmental modelling with correction for a component of endothelial binding with [11C]PBR-28 [154], or, for example, Logan graphical analysis with plasma input parametric mapping of [11C]-(R)-PK11195 data [158]. Typically, when a plasma input function is available, the binding of TSPO-PET tracers is frequently characterised using the 2TCM with various fixed or free parameters [134, 155]. Recent clinical studies using [18F]GE-180 have concluded that the 4푘 2TCM is the appropriate choice for modelling healthy brain data [278-279]; our aim was to assess the validity of this approach in RRMS subjects. We compared the performance of the unconstrained free 푉푏 1 and 2TCM.

Results showed that the 2TCM was indeed the preferred fit in the majority of brain regions in the real data set, a preference which was retained even under the addition of simulated noise. Overall, however, the total volumes of distribution in target regions of RRMS subjects and HVs were very low, indicating a low brain penetration of [18F]GE-180. Compounding this was the large amount of tracer retained in the blood, particularly in the diseased cohort. Indeed, an overriding finding of this clinical work is that the blood binding of [18F]GE-180 appears to be extremely high, dominating the brain signal, while the metabolism of the tracer is much slower than has been observed in preclinical studies (~ 70% parent vs.~20% parent at 60 minutes in the Wistar rat brain [115]). The latter is usually a desirable attribute of a PET tracer, since the slower the metabolism, the slower the formation of radiolabelled metabolites, which, over time, may enter the brain tissue via passive diffusion. In conjunction with this, a low lipophilicity (such as that exhibited by [18F]GE-180) can inhibit the crossing of the tracer through the BBB, although advantageously this feature should also limit the amount of nonspecific binding to plasma proteins [343, 346]. We investigated and quantified the plasma protein binding of [18F]GE-180 in a short experiment described in Chapter 6.

In response to the second research question, and in spite of the low 푉푇 estimates in all anatomical regions, we were able to detect a significantly higher [18F]GE-180 signal in the NAWM of RRMS subjects compared to that of HVs. This finding, consistent with those of previous studies with other TSPO-PET tracers ([108, 112], see [309] for review), is suggestive of a diffuse component of microglial activation in apparently healthy white matter in RRMS subjects. In an individual lesion analysis, only one subject, who was also the only patient to exhibit a Gd-enhancing lesion in T1 post-contrast MR, showed significantly higher overall lesion uptake than NAWM; suggestive of there being an increased presence of TSPO and inflammatory signal in the lesions of patients with active RRMS compared both to the NAWM and to the lesion/NAWM uptake ratio for patients with less active disease. This finding is consistent with previously reported results with

189

other TSPO-tracers and ex vivo work [112, 370]. In terms of clinical observations, and further supporting this hypothesis, is the fact that the subject with Gd-enhancement had also recently (within the previous 12 months) experienced three relapses, while another patient, whose lesion/NAWM uptake was less than 1 (significantly lower uptake in the lesions than NAWM), had shown no obvious relapses in the same time frame. However, given that Gd-enhancement is reflective of local BBB breakdown, and given the observed high blood signal of [18F]GE-180, it is likely that a large proportion of the signal seen in this lesion is not related to specific binding of the tracer to TSPO in tissue, but rather to the blood. To semi-quantitatively assess whether [18F]GE-180 in blood was dominating the signal from this lesion, we drew an ROI around the venous sinus to represent whole blood and compared the shape of the TACs from both regions, observing that they were not related; the lesion showed a slowly increasing uptake of tracer, while washout of the blood was very rapid (see Figure 7.1), suggesting that the uptake in the lesion was not fully driven by influx of tracer in blood through a damaged BBB.

Figure 7.1 TACs showing uptake in whole blood (venous sinus) and Gd-enhancing lesion. TAC shapes (red is the venous sinus) bear little resemblance to one another, suggesting that uptake in the Gd-enhancing lesion is not primarily driven by the high concentration of [18F]GE-180 in blood.

While plasma input functions are seen as preferredfor quantification of PET images, due to the highly invasive nature of arterial sampling, there is an underlying desire to eventually render them obsolete. To this end, we also investigated the possible use of a non-invasive semi-quantitative outcome measure – SUV, as well as employing a

190

reference region approach with the SRTM. The ubiquity of TSPO in even the healthy human brain is well reported [23, 44, 67, 286, 371]. Nevertheless, due to the relatively low expression at basal level, its relative sparing from much disease-related pathology and the group differences which have been observed when using it, cerebellar grey matter has often been used as input to reference tissue models, even in MS [166, 172, 335, 369, 372]. After assessment of the regional TACs from all Hammers atlas brain regions in both HV and MS subjects, we found the grey matter cerebellum did not exhibit the lowest uptake of [18F]GE-180, but rather a rank position of uptake which was approximately in the middle of the analysed regions, suggesting that it may contain some level of specific binding (assuming all regions show a similar level of non-specific binding). Furthermore, while supervised (and unsupervised) clustering methods have been employed with some success in other MS studies [20, 108] specifically with [11C]- (R)-PK11195, the former has not currently been validated for use with other TSPO tracers. Thus in the absence of a data-driven methodology for extraction of a [18F]GE-180 reference tissue class, we elected to use the anatomical region which consistently showed the lowest tracer uptake in both cohorts; the striatum. It should be observed at this point that the ‘ventral striatum’ (caudate nucleus, putamen and nucleus accumbens) is known to be affected by MS pathology, in particular the caudate nucleus [373-375] has been observed to exhibit demyelination in deep grey matter. For the purpose of assessing the usefulness of the striatum as a reference region input for kinetic modelling, and having subtracted all lesions from analysed standard ROIS, however, we assume that remaining uptake in the region is due to non-specific binding of [18F]GE-180, as is implicit in using any anatomical reference region. Of course, there may be specific binding in the striatum; thus violating a key assumption of the SRTM [167]. In our analysis, the SRTM failed to converge in several regions. This could be due to possible issues with partial volume effects and/or subject motion, or another feature of the overall low extraction into tissue of [18F]GE-180.

Having corrected SUV results for the estimated contribution of vascular signal to each region, we did not find a significant difference in any region between RRMS and HV groups, contrary to results observed with 푉푇. This could be due to several reasons.

Firstly, our correction for blood contribution was relatively primitive; fixing 푉푏 to 5% in all standard ROIs is not truly representative of the small variations in fractional blood volume which were actually observed in modelling with the 2TCM. Furthermore, a direct comparison of lesion areas in RRMS patients vs. the same ROIs in age and genotype matched HV subjects revealed a significantly higher SUVR and 푉푇 in patients. This further

191

supports the suggestion that the uptake of tracer in the lesions of RRMS subjects is related to pathology rather than anatomical location.

In terms of binding affinity differences based on the TSPO single nucleotide polymorphism, in vitro work has suggested that HABs exhibit a minimum of 5 times higher binding than MABs (personal communication, William Trigg). An unexpected finding of this study is that there were no significant differences in any outcome measure observed between HABs and MABs. Furthermore, the MAB cohort frequently exhibited somewhat higher 푉푇 values than the HAB group (see Figure 4.7). While this is somewhat understandable in the RRMS cohort, where variable levels of TSPO expression could be related to disease activity and state, the same finding was observed in healthy volunteers, where the level of TSPO expression in tissue should be approximately consistent within genotypes. In contrast, for [11C]PBR-28, SUV has been shown to be 30% lower [376] and

푉푇 up to 40% lower in MABs than HABs [153]. For other tracers, differences in binding affinity have been reported to be very similar in vitro and in vivo [315]. It is possible, however, that this discrepancy is driven by [18F]GE-180 binding in vivo to a different site, or with a different mechanism to that seen in in vitro binding assays. In order to further investigate this possibility, a study extension is underway. In this experiment, 6 MS subjects (different to those participating in the main study) will be recruited. Each patient will attend two imaging visits within a one month period, with no clinical/cognitive assessments but otherwise identical MR and TSPO-PET protocols to those in the main study. The first visit will establish baseline data, while 2 hours before the follow-up PET scans, patients will receive an oral dose of XBD173. This drug is a highly selective and specific TSPO agonist with nanomolar affinity which, aside from being an extremely effect anxiolytic [377-378], also acts as a blocking agent against TSPO-PET radioligands. It has recently been shown that the drug is susceptible to the same polymorphism in the TSPO gene as second generation radioligands [316] and, as such, doses must be carefully calculated according to genotype in order to produce equivalent occupancy of TSPO binding sites in all subjects. Once this is done, given that [18F]GE-180 binds with high selectivity and specificity to TSPO, the second scan will allow quantification of the non-specific signal of [18F]GE-180 related to TSPO. Arterial blood sampling will be performed at both visits to allow the generation of plasma input functions, as will blood biomarker assessment. Given the unusual PET findings to date, we hypothesise that the XBD-173 follow up [18F]GE-180 scan may show a substantial amount of non-specific binding, particularly in blood. Until this study has been completed, stratification for TSPO genotype remains the recommended approach to analysis of [18F]GE-180 brain PET data.

192

While we believe we have demonstrated a good fit of this healthy human and RRMS data to the 2TCM, there are other methodological approaches which may help to reduce the noisiness and corresponding large errors seen in smaller regions, such as lesions. In a very recent paper [278] investigating the optimum modelling approach in elderly healthy control subjects scanned with [18F]GE-180, the authors also concluded that the unconstrained 2TCM provided the best fit to 90 minute data, but that a Logan graphical analysis produced estimates which were consistent with those from compartmental modelling. The authors also noted a high plasma retention of the tracer over the course of the scans and HAB/MAB differences were only observed in the thalamus and cerebellum of these subjects, regions which have previously also been shown to exhibit age-related increases in TSPO expression [200, 286]. Another very recent study in [18F]GE-180-scanned healthy control subjects (average age 41±9 years; younger than those of Fan et al.) showed similar results, with low 퐾1 values of the order 0.005 ml.cm-3.min-1 and the 2TCM proving the recommended model for fitting of data [279]. Importantly, however, and in agreement with our analysis, there were no apparent differences between HAB and MAB subjects. It is promising that the results of our study into analysis of disease vs. control subject data are consistent with the findings of Feeney and Fan et al. in healthy subjects only, particularly in the case of the former, whose healthy control data was the same as that used in the analysis in Chapter 4. Given also the relatively promising result of an identification of main effect group differences between RRMS and HV subjects in our study, [18F]GE-180 warrants a more detailed examination in order to establish the optimum modelling approach; in the full data-set of 17 baseline scans, an assessment of comparability of other techniques, including parametric mapping and spectral analysis will be investigated, as well as the feasibility of using reference region/tissue methods.

In summary, and in response to the research questions posed at the start of this section, [18F]GE-180 PET brain data is well modelled by the 2TCM with free fractional blood volume. Also, the tracer does appear to exhibit a relative increase in binding of the disease state compared to the healthy brain, both in terms of overall brain uptake and in terms of regional binding in NAWM and the lesions of active patients. Further analysis is ongoing in the full cohort; clearly, some aspects of the current methodology require more investigation; namely, is the unconstrained 2TCM able to accurately and robustly account for the unusually high and persistent blood activity of [18F]GE-180, or would a model incorporating this slow component, such as that suggested by Rizzo et al. [154] for [11C]PBR-28, be more appropriate? Furthermore, is there a more suitable anatomical reference region than the striatum for use as input to the SRTM, or in an SUVR approach?

193

Results from the current preliminary analysis are promising and we believe we have demonstrated that [18F]GE-180 is able to identify a component of increased binding in these RRMS subjects compared to healthy controls. The true significance of this observation in terms of clinical study of the disease, and whether the tracer will allow further illumination on the role of inflammation in the progression of MS remains to be clarified.

7.3.2 Monitoring treatment in RRMS patients – Chapter 5 As mentioned, the initial aim of the study from which subjects in section 7.3.1 and Chapter 4 were analysed was to assess the potential utility of [18F]GE-180 as a marker of response to natalizumab therapy. The larger study is still ongoing, but a subset of the data was analysed (Chapter 5) in a preliminary assessment. In short, the research question posed by this analysis was: can [18F]GE-180 TSPO-PET visualise changes in TSPO expression in the treated RRMS brain?

A subset of five patients was selected from those described in Chapter 4 based on the available scans. The baseline (Chapter 4) and 10 week follow-up scans after monthly infusions of natalizumab were analysed for each patient. As a result of the findings described in Chapter 4, the 2TCM was used to derive estimates of 푉푇 in standard regions and individual lesions. We also evaluated SUV in these regions, however, given the findings in Chapter 4 regarding the likely component of specific binding in the ‘reference’ tissue (striatum), an SUVR approach was not considered reliable. The results of this analysis showed a reduced ratio of 푉푇 in lesions compared to NAWM in the subject with Gd-enhancing lesion after 10 weeks of treatment with natalizumab.

Notably, this was the only patient to exhibit a significant difference in binding in the lesion area, but they were also the only patient to show a clinical worsening of EDSS score (by one point). Naturally, it is possible and indeed an increasingly supported notion that the component of inflammation in lesions does not drive long-term clinical disability or progression (see [379] for review). Equally, the reduction in [18F]GE-180 signal may not be reflective of a true reduction in TSPO (see section 7.3.1), explicitly if the tracer binding observed in the region is non-specific. Another consideration, however, is that at this relatively early time point after commencing therapy, it is possible that clinical or pathological worsening was instigated prior to the first infusion of natalizumab and manifested over the 10 week period. In this case, any beneficial reductions in the level of inflammation in the brain as a result of treatment might not correlate with what might otherwise be classed as clinical improvement. Additionally, post-relapse assessment of EDSS can vary depending on how close to the date of relapse measurement is made

194

[380]. Ultimately, however, it cannot currently be ruled out that the subject is part of the 25% of RRMS patients who do not respond to natalizumab anti-inflammatory therapy [329]. Indeed, it was pointed out in this prospective study by Prosperini et al. that the drug appeared to perform best in patients with less aggressive, early stage disease [329], while this patient was in an active stage of disease in terms of Gd-enhancement in T1 MR and in terms of number of recent relapses (three in the 12 months prior to starting natalizumab and one in the second month after starting treatment).

We found significantly lower SUV_corr in two of the patients who, prior to starting natalizumab, had experienced either one or no recent (within 12 months) relapses and who reported improvement to symptoms after 10 weeks of therapy. These two patients were not necessarily classed as having ‘early’ stage RRMS in terms of length of time since onset of disease; however, the lack of recent relapses suggests that they were not in the aggressively advancing phase where disability accumulates rapidly [380]. Tentatively, they may then be in the group which Prosperini et al. refer to as ‘full’ or ‘partial responders’, i.e. those subjects which do exhibit some clinical improvement in relation to natalizumab. Importantly, in terms of imaging findings, there was no significant difference between T2 lesion load pre and post-treatment, while [18F]GE-180 PET appeared to be able to identify a decrease in signal over the time frame.

There is a discrepancy between estimates of 푉푇 and SUV_corr. It is not yet clear what drives these differences, although it is likely that the level of binding of [18F]GE-180 in blood plays a role. High blood binding and a low extraction fraction lead to low 푉푇 estimates, since the parameter is directly related to 퐾1, which itself is defined as flow multiplied by the net extraction fraction. Moreover, the variability in 푉푇 in individual lesions was higher than the variability of SUV; this could be driven by poor model fitting in these noisy regions, or more particularly by a sensitivity of compartmental modelling to 퐾1 and 푉푏. Evidently, without results from a test-retest study in healthy control subjects, it is not possible to quantify the variability between pre and post-treatment observations, and, more importantly, whether the differences seen in 푉푇 and SUV_corr are still significant when accounting for this variability compared to natural variation in the general population. Even so, when teamed with the findings described in section 7.3.1 and Chapter 4, namely the overall main effect of disease on uptake in NAWM, the results of this preliminary longitudinal evaluation demonstrate an ability of the current methodology to identify changes both statically and over time of [18F]GE-180 PET signal in the MS compared to the healthy brain.

195

These observational correlations illustrate a potential sensitivity of [18F]GE-180 PET to changes in the treated RRMS brain, associated with clinical improvements, which conventional MRI techniques do not appear to have. Since the study does not involve a placebo-treated group, it is difficult to establish whether these changes are truly reflective of an effect of natalizumab. Nevertheless, if the tracer is able to identify these differences longitudinally, its potential use in therapy, or at least in disease status monitoring, could be considerable.

7.3.3 Plasma-protein binding of [18F]GE-180 – Chapter 6 Having commenced an analysis of the human imaging data from [18F]GE-180, it rapidly became apparent that the behaviour of the tracer in blood was somewhat different to that observed in the preclinical study reported in Chapter 2, and in the literature in different animal models of disease [115, 119, 122, 213, 294]. A hypothesis to explain the high level of retention in blood is that [18F]GE-180 binds non-specifically with a slowly dissociating component to plasma proteins in human blood. Further supporting this notion is the fact the plasma-to-blood ratios observed in the analyses in Chapters 4 and 5 were very high compared to, for example, [11C]-(R)-PK11195 [135-136]. In order to investigate this possibility further, a study was designed to determine the fraction of non- specific tracer binding in the blood of healthy volunteers from the WMIC. Having not performed such quantification previously, the study also contained a developmental aspect relating to the methodology of measurement.

In vitro measurement of plasma-protein binding is not strictly accurate, given that the fraction of tracer available in tissues of interest during a dynamic PET scan is also dependent on pH and temperature [381-382], which cannot be determined with accuracy in vivo. Nevertheless, as an estimation of the relative fraction of bound/unbound tracer in blood, we performed a displacement experiment with labelled [18F]GE-180. In the first instance, displacement was performed by spiking unlabelled GE- 180 or unlabelled PK11195 into plasma samples and adding radiolabelled [18F]GE-180. Measurement via a Centrifree® ultrafiltration device of the difference in radioactivity present in tracer-only and cold-spiked samples revealed an increase in free tracer of up to 50 fold after displacement with GE-180 (~30 fold with PK11195), suggesting the presence of specific binding of [18]GE-180 in the blood. Importantly, the fraction of unbound (whether specifically or non-specifically) tracer was up to 10 times lower than that of [11C]-(R)-PK11195 (reported as ~1% [356]), which itself is notably low. Given that more lipophilic tracers generally suffer from higher levels of plasma-protein binding [343], one might expect that [18F]GE-180 should have a lower plasma-bound fraction and

196

higher corresponding free fraction than, for example, [11C]-(R)-PK11195, since its lipophilicity is somewhat lower (see Table 6.1). Equally, the difference in the magnitude of displacement when using unlabelled GE-180 compared to PK11195 might suggest a difference in the binding sites of the two tracers to TSPO, but also a more far-reaching indication is that GE-180 binds to a different target with some ‘specificity’ too.

Clearly, measurement of such low levels of activity comes with its own unreliability, and additionally there were practical issues with the use of lipophilic plastic pipette tips, the concentration of unlabelled tracer used to displace [18F]GE-180, and the fact that measurement was made in incubated plasma, rather than whole blood, all leading to a potential underestimation of free tracer after displacement. Repetition of the experiment in another group of healthy volunteers revealed similar results; namely, the fraction of unbound tracer was approximately 0.1%, increasing by up to 60 and 35 fold when displaced with GE-180 and PK11195 respectively.

There was a large variability in the measurement of unbound tracer fraction across all subjects. Furthermore, genotyping of blood samples for the TSPO single nucleotide polymorphism revealed a HAB-only cohort, which is highly improbable given the reported prevalence of each genetic group in the healthy Caucasian population [383]. Comparing results from [18F]GE-180, however, recent measurement of the plasma free fraction of [11C]PBR-28 has also shown large variability within genetic subgroups, as well as no HAB/MAB differences in 푓푝 [314]. Indeed, variability in the level of plasma-protein binding within cohorts depending on age has also been reported, presumably due to changing concentrations of certain plasma proteins, such as human serum albumin [384]. It has also recently been suggested that the fraction of unbound tracer available for passage into tissue could vary depending on disease state [152], in part because expression of TPSO on platelets can vary, particularly in psychiatric disorders (specific binding) [385-388], but also because levels of certain plasma proteins may increase peripherally in inflammatory brain diseases (non-specific binding) [389]. In the study by Lockhart et al. [389], the authors point out that the primary plasma-protein to which PK11195 binds is AGP (α1-acid glycoprotein), which, they suggest, could be locally synthesised by glial cells at the site of tissue injuries such as MS lesions. In addition, it has been observed that AGP exhibits some anti-inflammatory properties [390] and can bind to the endothelium capillaries [391-393]. Thus, if [18F]GE-180 also binds to AGP, or any other blood protein which is dysregulated in disease, 푓푝 could be significantly altered compared to a healthy cohort. Moreover, local expression of such proteins would increase the level of non-specific binding at the site of lesions.

197

Furthermore, and in the case of all TSPO tracers, but particularly those with low plasma free fractions, the higher the level of binding to plasma-proteins, the greater the effect on the derivation of the arterial plasma input function, where 푓푝 is often taken to be 1 due to a combination of the afore-mentioned inter-subject variability and the difficulty of accurately and reliably measuring 푓푝 (either in vivo or in vitro). A subsequent overestimation of the fraction of tracer actually available to cross into tissue would lead to falsely high estimates of 퐾1 and associated macroparameters, including 푉푇.

Justification for using 푓푝 = 1 often arises both from the fact that correction of 푉푇 for 푓푝 has been shown to have high variability (presumably due to the inter-subject variations in 푓푝)

[172], and because, as mentioned previously, the measurement of 푓푝 is somewhat error prone when the values are so low. Thus, fixing the fraction across subject groups allows for more stable estimates of parameters such as 푉푇, with the rather indiscriminate assumption that 푓푝 is not significantly different between cohorts, which, as discussed previously, is not always true. Indeed, not correcting for individual 푓푝 measurements could lead to exaggeration of possible signal differences related to pathology, particularly if the quantity of plasma proteins is higher in disease vs. healthy cohorts (leading to a lower actual free fraction and overestimation of the amount of tracer bound).

In summary, the findings of the investigation presented in Chapter 6 are 1) [18F]GE-180 is subject to a high level of non-specific binding to plasma-proteins. Given the low fraction of free tracer observed in our human blood samples from healthy volunteers, this measurement is subject to some error, but does appear to be lower than the free fractions of other tracers reported in the literature [314, 356, 360, 394]. A second finding is that 2) [18F]GE-180 may bind either with higher affinity (as observed in vitro, see Supplementary Table 6.1 and Table 6.5) or in part to a different site on TSPO than [11C]-(R)-PK11195, as suggested by the different levels of displacement observed with unlabelled tracers. Given that GE-180 displaced almost double the amount of radiolabelled tracer as PK11195, this finding warrants further investigation. Simultaneously, this could be representative of a proportion of displaceable binding to a different target altogether. In this case, a blocking study with a high affinity drug such as XBD-173 as described in section 7.3.1 would allow quantification of the signal from [18F]GE-180 not related to the occupancy of TSPO sites.

7.4 The polymerisation of TSPO and other considerations

Here, we will revisit TSPO as described in Chapter 1, with more specific reference to the findings of this project and how they might be relevant to our understanding of the

198

protein. It is a well-established fact that TSPO expression is altered in several neurological brain disorders and correlates with the presence of activated microglia, astrocytes and other cells of macrophage-lineage (see [24, 395] for reviews). It is therefore a natural step to target the protein as a PET imaging biomarker for inflammation in the CNS and to draw sweeping conclusions regarding the true significance of its presence/absence in disease. While the structure and function of the protein have been studied to some extent ([34, 56], see [52, 396] for review), its full characterisation and how it interacts with (radio)ligands is still ongoing.

Figure 7.2 Representation of the binding of a TSPO ligand to the protein. TSPO is part of a larger complex, including three channels: the 18 kDa TSPO, the 32 kDa VDAC and the 30 kDa ANT.

In the last decade, it has come to light that TSPO exhibits multiple binding sites; Korkhov et al. [34] suggest that two TSPO monomers, each consisting of 5 transmembrane domains, are tightly associated in a dimer with two binding sites. While TPSO ligands are generally understood to bind primarily to the 18 kDa TSPO channel of the protein complex described in Figure 7.2, the VDAC and ANT channels are recognised as key components to its functionality [397-398] and thus their presence could affect the binding of some ligands to the protein as a whole [399]. The polymerisation of TSPO is not well-documented, but general consensus is that the protein polymerises at least during steroidogenesis [25, 400], though it is possible that TSPO polymers also develop via other mechanisms. If this is the case, and if the process of steroidogenesis (cholesterol metabolism) is modified in the course of inflammatory diseases under study with TSPO-

199

PET ligands, including both MS and AD ([401-402], see [403] for review), then polymerised TSPO, which plays a key role in these steroid-related processes, could be present. Specifically, the polymerisation of TSPO, possibly due to the presence of reactive oxygen species [400] found in many inflammatory diseases (see [404] for review), may lead to a structural change in TSPO binding sites, introducing a potentially significant prejudice for ligands which target only the mono or dimeric forms of the protein. Furthermore, the polymerisation process may actually be dynamic; modified during particular stages of cholesterol transport and not others, which would also affect the binding of a TSPO tracer differently within subject cohorts depending on their disease stage. It is also possible that the binding of TSPO ligands to one of these sites affects the polymerisation process itself, which could alter the subsequent binding of ligands [25].

The TSPO single nucleotide polymorphism was originally thought not to affect PK11195 binding, however recent evidence suggests that in the presence of polymeric TSPO, a phenomenon which itself may be related to increased steroid-synthesis [405], tritiated PK11195 binds to two different sites with different affinities in (murine) Ehrlich tumour cells [406]. While this has still not been observed in vivo in the human brain, [11C]-(R)-PK11195 appears to bind with different affinities in organs such as the lungs and heart [25, 73]. The reason for this differential binding profile between the brain, a so- called ‘immune-privileged’ organ, and those organs in the periphery, is unknown. If this is again related to the polymerisation of TSPO, however, it may be necessary to characterise the ligand-binding site interaction for each tracer in each disease state before results from imaging studies can be fully understood.

Detailed discussion of the literature relevant to these findings is outside the scope of this thesis, but the key points to note are 1) the binding of ligands to the TSPO complex is not well-characterised and 2) particularly in studies involving comparison of more than one tracer, it would be wise to interpret results cautiously since, without knowledge of the binding sites of TSPO available in given disease states, it is not easy to determine the true significance of a tracer having ‘higher’ apparent signal. Furthermore, 3) the future of PET imaging in inflammation may not lie with targeting TSPO, but if we wish to determine the usefulness of the protein as a biomarker, deeper understanding of its functionality and configuration, specifically in different disease states, is pivotal.

7.5 Final comments

To return to the research questions posed in Chapter 1 and in section 7.1, the work presented in this thesis demonstrates for the first time, in preclinical models of inflammation, that the new second generation TSPO-PET tracer [18F]GE-180 produces a

200

good target/reference contrast ratio compared to [11C]-(R)-PK11195, but not significantly higher than that exhibited by another second generation tracer, [18F]DPA- 714. In our model of low-dose inflammation in the rodent brain, all three tracers were best modelled with a contralateral reference region input to the SRTM. Full characterisation of an arterial sampling system for preclinical studies, which will allow, in future, full kinetic validation of models such as that used here, is continuing. Tentatively, however, the use of an external reference tissue may produce less biased results than using a contralateral input.

In clinical RRMS subjects, [18F]GE-180 is well modelled kinetically by the 4푘 2TCM with free fractional blood volume, but does show a high plasma retention which is not observed in rodents; we have measured the high level of non-displaceable binding and corresponding very low free fraction of [18F]GE-180 and suggest that these factors could be linked to this feature. Nevertheless, it appears that the tracer has some sensitivity to inflammation in the NAWM of MS subjects compared to healthy subjects, an observation which has previously been noted. Also, we observed that uptake in the lesion area of two subjects reporting clinical improvement after trialling natalizumab for 10 weeks was significantly reduced. Interestingly,SUVs in the healthy rodent brain is comparable to healthy grey matter in the human brain, suggesting a similar over all brain penetration. There thus remain some currently unresolved points raised in these studies: 1) there appears to be no significant in vivo difference in uptake of [18F]GE-180 between high and mixed affinity binders in the young healthy subjects in our cohort (and that of Feeney et al. [279]), contrary to findings in elderly healthy controls and in vitro work; 2) is the plasma-protein binding of [18F]GE-180 different between rodents (for example) and humans?; 3) other plasma input modelling methodologies warrant investigation, particularly considering the high blood signal of the tracer. Further work to develop these analyses in the full clinical cohort (3 scans from 17 MS subjects, as well as a supplementary blocking study in 6 MS subjects) is necessary and ongoing. Some other questions, such as whether the apparent reduction of inflammation in the treated MS brain is truly a result of natalizumab therapy, or a reflection of the placebo effect, cannot be answered by this study but are important research questions in themselves and should be addressed separately, particularly if it emerges that [18F]GE-180 could be a sensitive marker of these changes.

The true role of inflammation in the progression of diseases such as MS is still not clear. There is, however, a clear indication of the presence of neuroinflammation in such neurological diseases and studying the functionality of the brain under this inflammatory burden in vivo is our greatest opportunity to illuminate the significance of the process.

201

Key to understanding the relationship between inflammation and the development of clinical disease is being able to visualise the process, which can be achieved with the molecular sensitivity of PET. Selecting the correct target to best represent the extent of inflammation in the diseased brain is challenging; as presented above, there are several issues with our current understanding of TSPO as a biomarker of neuroinflammation, but, outside the scope of this thesis, there also exist several other inflammatory targets which are under investigation. The potential utility of TSPO PET ligands such as [18F]GE- 180 should not be de-valued as a result; however, our interpretation of the results of studies such as those presented in this thesis may evolve with our developing understanding of TSPO as an imaging target.

Ultimately, on a planet with a population which is not only increasing, but ageing rapidly, and with an ever-growing demand for effective therapy, if not curative treatment, we seek the ability to understand disease via tools such as PET and manipulate factors such as inflammation accordingly. It is my hope that, with time and ongoing efforts in both preclinical and clinical studies, the continually expanding field of PET imaging in inflammation will enable its characterisation in diseases such as MS; eventually allowing quantitative clinical interpretation of its effects on disease course and the subsequent development of tailored therapy.

202

Appendix 1 Definitions of macroparameters The in vitro equilibrium interaction of a free radioligand, F, with a receptor, R, can be characterised by equation 1.1A.

푘표푛 푅 + 퐹 ⇄ 퐵 Eq. 1.1A 푘표푓푓

퐵 is the bound complex of the receptor and the ligand and 푘표푛, 푘표푓푓 are reaction rate constants. This basic relationship led to the definition of a ‘binding potential’. The binding potential of a radioligand receptor such as TSPO can be measured in different ways. The term was first coined to describe the ‘capacity of a given tissue…for ligand- binding site interaction’ and is the product of 퐵푚푎푥, or the total number of binding sites, −1 and 퐾퐷 , the radioligand binding affinity [407], i.e.:

퐵푃 = 퐵푚푎푥⁄퐾퐷 Eq. 1.2A

퐵푚푎푥 and 퐾퐷 cannot be accessed individually, only the binding potential can be derived from the data (see Table 1.1A). Essentially, the term ‘binding potential’ refers to the ratio of ‘the equilibrium concentration of specific binding’ to the concentration of binding in a predefined reference region [408]. Since the term was first described, however, there has been some convolution of the meaning. Innis et al. [408] designed a consensus nomenclature to define the different forms of binding potential as follows:

Table 1.1A Summary of different definitions for binding potential (rate constant equations are for the 2TCM).

Abbreviation Reference Mathematical formula Units

-3 푩푷푭 Free concentration in plasma 퐾1푘3 ml.cm 퐵푎푣푎푖푙⁄퐾퐷 = 푓푃푘2푘4

-3 푩푷푷 Total parent concentration in plasma 퐾1푘3 ml.cm 푓푃퐵푎푣푎푖푙 ⁄퐾퐷 = 푘2푘4

푩푷푵푫 Non-displaceable binding 푘3 None 푓푁퐷퐵푎푣푎푖푙⁄퐾퐷 = 푘4

n.b. 푓푃 is the free fraction of radioligand in plasma, 푓푁퐷 is the free fraction of radioligand in non-displaceable tissue; i.e. the fraction not bound to plasma proteins at equilibrium; both are unitless. 퐵푎푣푎푖푙 refers to the density of receptors available to bind

203

the radioligand (in vivo) and often has units of nmol.L-1; that is, nmol receptor concentration per litre (1000 cm3) of tissue.

퐵푃퐹 (ratio at equilibrium of concentration of specifically bound radioligand to free radioligand in given tissue) and 퐵푃푃 (ratio at equilibrium of concentration of specifically bound radioligand in tissue to concentration of parent ligand in plasma) require arterial plasma measurements to be acquired, but 퐵푃푁퐷 does not; although this seems an advantage in terms of simplicity, the authors suggest that 퐵푃푃 is more useful when the plasma free fraction cannot be easily measured. 퐵푃푁퐷 is often calculated in studies where a reference tissue rather than arterial plasma sampling is used. The in vitro binding potential, 퐵푃, is unitless. In vivo, in healthy subjects, the value of 퐵푃푁퐷 has been reported as between 0.23 and 0.45 [135]. The rate constant 퐾1, which describes transfer from arterial plasma to tissue, has units of ml.cm-3.min-1, while the other rate constants

-1 푘2,3,4 have units of min . The authors explain this discrepancy by the fact that 퐾1 describes volume of blood (or plasma) per volume of tissue per minute, where the other constants describe the rate of transfer of radioligand between compartments. In vivo, equation 1.3A defining 푘3 applies, although equation 1.4A defining 푘4 shows some discrepancy in that the in vitro constant 푘표푓푓 has been shown to be much faster than the in vivo 푘4. It has been suggested that this may be due to the slow ‘rate measured in typical baseline conditions of negligible receptor occupancy’ [408].

푘3 = 푓푁퐷푘표푛퐵푎푣푎푖푙 Eq. 1.3A

푘4 = 푘표푓푓 Eq. 1.4A

The volume of distribution, 푉, is generally defined in in vivo imaging as the ‘ratio of the concentration of radioligand in a region of tissue to that in plasma (or blood)’. Innis et

-3 al. [408-410] define the units as ml.cm to avoid confusion. In the same way that 퐵푃푁퐷 shows some inconsistency in its definition in the literature, the volume of distribution sometimes refers to 푉푇, defined as the ‘distribution volume of total ligand uptake in tissue relative to total concentration of ligand in plasma’. It is a combination of the volumes of distribution in the free, specific and non-specifically bound compartments. 푉푁퐷, on the other hand, is ‘the distribution volume of [the] non-displaceable compartments relative to [the] total concentration of ligand in plasma’.

푉푇 is only calculable for reversible models and is related to 퐾1 and 푘2,3,4 as shown in the following equations.

퐾1 푉푇 = for the 1TCM Eq. 1.5A 푘2

204

퐾1 푘3 푉푇 = (1 + ) for the 2TCM Eq. 1.6A 푘2 푘4

The volume of distribution is related to the binding potential as shown in equation 1.7A.

푉푇 퐵푃푁퐷 = − 1 Eq. 1.7A 푉푁퐷

푉 푇 is known as the distribution volume ratio (DVR) and is a parameter derived 푉푁퐷 from Logan analysis with a reference tissue input. While binding potential is the ratio of specific binding at equilibrium to the concentration in other compartments, the volume of distribution can describe specific, non-displaceable or total binding as a ratio to the total ligand concentration in plasma. All forms of 푉 also have units of ml.cm-3.

The blood volume, 푉푏, often refers to the whole blood concentration; that is, the blood concentration of tracer not corrected for metabolites.

퐾퐼 is the parameter of interest in irreversible models. It is the ‘steady-state rate of solute flux across the BBB complex from plasma (constant concentration) into brain extracellular fluid divided by the plasma concentration of the solute’ [142].

205

Appendix 2 Posters and conference presentations During my time as a PhD student at the University of Manchester, and with the financial support of the EPSRC and GE Healthcare, I have been lucky enough to attend several national and international conferences and showcases. At each of these, I have had the opportunity to share my work with numerous colleagues in the field of PET research. Unquestionably, the discussions which have taken place at these conferences have allowed me to improve and challenge my own scientific thoughts and ideas; affording me a perspective which I would not otherwise have been gained. Conferences which I have attended are listed below in chronological order.

Working at the WMIC has also allowed me to take part in regular seminars and academic discussions such as ‘Journal Club’ meetings. At these, I have been in equal part encouraged and challenged, allowing me to hone my skills in the performance of scientific oral and poster presentations, as well as critical analysis of research articles, and for which I owe my colleagues a great deal for their attendance and engagement.

Poster presentations

 Neuroscience Research Institute (NRI) Showcase, July 2013, University of Manchester, UK – ‘Kinetic analysis of [11C]-(R)-PK11195 human brain Positron Emission Tomography (PET) data’  Biomedical Imaging Institute (BII) Showcase, November 2013, University of Manchester, UK – ‘Parametric Mapping of [11C]-(R)-PK11195 human brain PET data’  All-centre Institute of Population Health (IPH) Postgraduate Research Showcase, March 2014, University of Manchester, UK - ‘Comparison of three reference input functions for quantifying [11C]-(R)-PK11195 human brain PET data’  Imaging of Neuroinflammation in Neurodegenerative Diseases (INMiND) TSPO Symposium – Microglia Imaging and Biology, April 2014, University of Manchester, UK – ‘A data-driven method for automatic extraction of a reference tissue kinetic from [11C]-(R)-PK11195 rodent brain scans’  10th International Symposium on Functional NeuroReceptor Mapping of the Living Brain (NRM), May 2014, Amsterdam, the Netherlands – ‘Comparison of three reference input functions for quantifying [11C]-(R)-PK11195 human brain PET data’  Biotechnology and Biological Sciences Research Council (BBSRC) and Biochemical Society “Sparking Impact” meeting – Mitochondrial Stress Response Pathways: Functions and Applications of the 18 kDa Protein TSPO, December

206

2014, The Royal Veterinary College, London, UK – ‘Comparative evaluation of three TSPO PET radiotracers in a rodent model of diffuse brain inflammation’  European Society for Molecular Imaging (ESMI) – Hot Topics in Molecular Imaging (TOPIM) – Imaging Inflammation, February 2015, Les Houches, France – ‘Comparative evaluation of three TSPO PET radiotracers in a rodent model of diffuse brain inflammation’  Faculty Postgraduate Summer Research Showcase (PSRS), June 2015, University of Manchester, UK – ‘Initial evaluation of [18F]GE-180 PET imaging in relapsing- remitting multiple sclerosis patients’

Oral presentations

 Centre for Imaging Sciences (CIS) Postgraduate Student Symposium 2014, 2015, 2016, University of Manchester, UK – ‘Quantitative analysis of positron emission tomography (PET) with novel TSPO agents’ (2014, 2015), ‘Initial Evaluation of [18F]GE-180 PET Imaging in Relapsing-Remitting Multiple Sclerosis Patients’ (2016)  13th Turku PET Symposium, May 2014, Turku, Finland – winner, Young Investigator Award - ‘A data-driven method for automatic extraction of a reference tissue kinetic from [11C]-(R)-PK11195 rodent brain scans’  All-centre Institute of Population Health (IPH) Postgraduate Research Showcase, March 2015, University of Manchester, UK – centre representative - ‘Comparative evaluation of three TSPO PET radiotracers in a rodent model of diffuse brain inflammation’  Annual Congress of the European Association of Nuclear Medicine (EANM), October 2015, Hamburg, Germany – ‘Initial Evaluation of [18F]GE-180 PET Imaging in Relapsing-Remitting Multiple Sclerosis Patients’  Kinetic Modelling and Analysis workshop, July 2016, Columbia University Medical Centre, New York, NY, USA – ‘Methodological considerations in clinical [18F]GE- 180 TSPO-PET data analysis’  11th International Symposium on Function NeuroReceptor Mapping of the Living Brain (NRM), July 2016, Boston, MA, USA – winner, Young Investigator Award – ‘Initial evaluation of [18F]GE-180 as an imaging biomarker in natalizumab therapy of multiple sclerosis’

207

References 1. Jacobs AH, Tavitian B (2012) Noninvasive molecular imaging of neuroinflammation. J Cereb Blood Flow Metab 32:1393-1415. 2. Murray KN, Buggey HF, Denes A, Allan SM (2013) Systemic immune activation shapes stroke outcome. Mol Cell Neurosci 53:14-25. 3. Ginhoux F, Greter M, Leboeuf M, et al. (2010) Fate mapping analysis reveals that adult microglia derive from primitive macrophages. Science 330:841-845. 4. Ginhoux F, Lim S, Hoeffel G, Low D, Huber T (2013) Origin and differentiation of microglia. Front Cell Neurosci 7:45. 5. Saijo K, Glass CK (2011) Microglial cell origin and phenotypes in health and disease. Nat Rev Immunol 11:775-787. 6. Kierdorf K, Erny D, Goldmann T, et al. (2013) Microglia emerge from erythromyeloid precursors via Pu.1- and Irf8-dependent pathways. Nat Neurosci 16:273- 280. 7. Schulz C, Gomez Perdiguero E, Chorro L, et al. (2012) A lineage of myeloid cells independent of Myb and hematopoietic stem cells. Science 336:86-90. 8. Prinz M, Mildner A (2011) Microglia in the CNS: immigrants from another world. Glia 59:177-187. 9. Prinz M, Priller J (2014) Microglia and brain macrophages in the molecular age: from origin to neuropsychiatric disease. Nat Rev Neurosci 15:300-312. 10. Lawson LJ, Perry VH, Dri P, Gordon S (1990) Heterogeneity in the distribution and morphology of microglia in the normal adult-mouse brain. Neuroscience 39:151-170. 11. Politis M, Su P, Piccini P (2012) Imaging of microglia in patients with neurodegenerative disorders. Front Pharmacol 3:96. 12. Davalos D, Grutzendler J, Yang G, et al. (2005) ATP mediates rapid microglial response to local brain injury in vivo. Nat Neurosci 8:752-758. 13. Raivich G (2005) Like cops on the beat: the active role of resting microglia. Trends Neurosci 28:571-573. 14. Nimmerjahn A, Kirchhoff F, Helmchen F (2005) Resting microglial cells are highly dynamic surveillants of brain parenchyma in vivo. Science 308:1314-1318. 15. Czeh M, Gressens P, Kaindl AM (2011) The yin and yang of microglia. Dev Neurosci 33:199-209. 16. Tang Y, Le W (2016) Differential roles of M1 and M2 microglia in neurodegenerative diseases. Mol Neurobiol 53:1181-1194. 17. Ransohoff RM (2016) A polarizing question: do M1 and M2 microglia exist? Nat Neurosci 19:987-991. 18. Kim CC, Nakamura MC, Hsieh CL (2016) Brain trauma elicits non-canonical macrophage activation states. J Neuroinflammation 13:117. 19. McGeer PL, McGeer EG (1995) The inflammatory response system of brain: Implications for therapy of Alzheimer and other neurodegenerative diseases. Brain Research Reviews 21:195-218. 20. Banati RB, Newcombe J, Gunn RN, et al. (2000) The peripheral benzodiazepine binding site in the brain in multiple sclerosis: quantitative in vivo imaging of microglia as a measure of disease activity. Brain 123 ( Pt 11):2321-2337. 21. Benavides J, Fage D, Carter C, Scatton B (1987) Peripheral type benzodiazepine binding-sites are a sensitive indirect index of neuronal damage. Brain Research 421:167- 172. 22. Da Pozzo E, Giacomelli C, Barresi E, et al. (2015) Targeting the 18-kDa translocator protein: recent perspectives for neuroprotection. Biochemical Society Transactions 43:559-565. 23. Papadopoulos V, Baraldi M, Guilarte TR, et al. (2006) Translocator protein (18kDa): new nomenclature for the peripheral-type benzodiazepine receptor based on its structure and molecular function. Trends Pharmacol Sci 27:402-409.

208

24. Scarf AM, Ittner LM, Kassiou M (2009) The translocator protein (18 kDa): central nervous system disease and drug design. J Med Chem 52:581-592. 25. Scarf AM, Kassiou M (2011) The translocator protein. Journal of Nuclear Medicine 52:677-680. 26. Chang YJ, McCabe RT, Rennert H, et al. (1992) The human "peripheral-type" benzodiazepine receptor: regional mapping of the gene and characterization of the receptor expressed from cDNA. DNA Cell Biol 11:471-480. 27. Riond J, Mattei MG, Kaghad M, et al. (1991) Molecular cloning and chromosomal localization of a human peripheral-type benzodiazepine receptor. Eur J Biochem 195:305-311. 28. Woods MJ, Williams DC (1996) Multiple forms and locations for the peripheral- type benzodiazepine receptor. Biochemical Pharmacology 52:1805-1814. 29. Beurdeley-Thomas A, Miccoli L, Oudard S, Dutrillaux B, Poupon MF (2000) The peripheral benzodiazepine receptors: a review. Journal of Neuro-Oncology 46:45-56. 30. Dolle F, Luus C, Reynolds A, Kassiou M (2009) Radiolabelled molecules for imaging the translocator protein (18 kDa) using positron emission tomography. Current Medicinal Chemistry 16:2899-2923. 31. Olson JM, Ciliax BJ, Mancini WR, Young AB (1988) Presence of peripheral-type benzodiazepine binding sites on human erythrocyte membranes. Eur J Pharmacol 152:47-53. 32. Kuhlmann AC, Guilarte TR (2000) Cellular and subcellular localization of peripheral benzodiazepine receptors after trimethyltin neurotoxicity. J Neurochem 74:1694-1704. 33. Gavish M, Bachman I, Shoukrun R, et al. (1999) Enigma of the peripheral benzodiazepine receptor. Pharmacol Rev 51:629-650. 34. Korkhov VM, Sachse C, Short JM, Tate CG (2010) Three-dimensional structure of TspO by electron cryomicroscopy of helical crystals. Structure 18:677-687. 35. Jaremko L, Jaremko M, Giller K, Becker S, Zweckstetter M (2014) Structure of the mitochondrial translocator protein in complex with a diagnostic ligand. Science 343:1363-1366. 36. Jaremko M, Jaremko L, Jaipuria G, Becker S, Zweckstetter M (2015) Structure of the mammalian TSPO/PBR protein. Biochem Soc Trans 43:566-571. 37. Midzak A, Zirkin B, Papadopoulos V (2015) Translocator protein: pharmacology and steroidogenesis. Biochem Soc Trans 43:572-578. 38. Jaremko L, Jaremko M, Giller K, Becker S, Zweckstetter M (2015) Conformational flexibility in the transmembrane protein TSPO. Chemistry 21:16555-16563. 39. Jaremko M, Jaremko L, Giller K, Becker S, Zweckstetter M (2015) Structural integrity of the A147T polymorph of mammalian TSPO. Chembiochem 16:1483-1489. 40. Li F, Liu J, Valls L, Hiser C, Ferguson-Miller S (2015) Identification of a key cholesterol binding enhancement motif in translocator protein 18 kDa. Biochemistry 54:1441-1443. 41. Krueger KE, Papadopoulos V (1990) Peripheral-type benzodiazepine receptors mediate translocation of cholesterol from outer to inner mitochondrial membranes in adrenocortical cells. J Biol Chem 265:15015-15022. 42. Papadopoulos V, Mukhin AG, Costa E, Krueger KE (1990) The peripheral-type benzodiazepine receptor is functionally linked to Leydig cell steroidogenesis. J Biol Chem 265:3772-3779. 43. Choi J, Ifuku M, Noda M, Guilarte TR (2011) Translocator protein (18 kDa)/peripheral benzodiazepine receptor specific ligands induce microglia functions consistent with an activated state. Glia 59:219-230. 44. Chen MK, Guilarte TR (2008) Translocator protein 18 kDa (TSPO): molecular sensor of brain injury and repair. Pharmacol Ther 118:1-17. 45. Zhao AH, Tu LN, Mukai C, et al. (2016) Mitochondrial translocator protein (TSPO) function is not essential for heme biosynthesis. J Biol Chem 291:1591-1603.

209

46. Guidotti A, Forchetti CM, Corda MG, Konkel D, Bennett CD, Costa E (1983) Isolation, characterization, and purification to homogeneity of an endogenous polypeptide with agonistic action on benzodiazepine receptors. Proc Natl Acad Sci U S A 80:3531-3535. 47. Besman MJ, Yanagibashi K, Lee TD, Kawamura M, Hall PF, Shively JE (1989) Identification of des-(Gly-Ile)-endozepine as an effector of corticotropin-dependent adrenal steroidogenesis: stimulation of cholesterol delivery is mediated by the peripheral benzodiazepine receptor. Proc Natl Acad Sci U S A 86:4897-4901. 48. Snyder SH, Verma A, Trifiletti RR (1987) The peripheral-type benzodiazepine receptor: a protein of mitochondrial outer membranes utilizing porphyrins as endogenous ligands. FASEB J 1:282-288. 49. Verma A, Nye JS, Snyder SH (1987) Porphyrins are endogenous ligands for the mitochondrial (peripheral-type) benzodiazepine receptor. Proc Natl Acad Sci U S A 84:2256-2260. 50. Costa B, Da Pozzo E, Martini C (2012) Translocator protein as a promising target for novel anxiolytics. Current Topics in Medicinal Chemistry 12:270-285. 51. Papadopoulos V, Aghazadeh Y, Fan J, Campioli E, Zirkin B, Midzak A (2015) Translocator protein-mediated pharmacology of cholesterol transport and steroidogenesis. Mol Cell Endocrinol 408:90-98. 52. Li F, Liu J, Liu N, Kuhn LA, Garavito RM, Ferguson-Miller S (2016) Translocator protein 18 kDa (TSPO): an old protein with new functions? Biochemistry 55:2821-2831. 53. Veenman L, Gavish M (2012) The role of 18 kDa mitochondrial translocator protein (TSPO) in programmed cell death, and effects of steroids on TSPO expression. Curr Mol Med 12:398-412. 54. Campanella M, Turkheimer FE (2015) TSPO: functions and applications of a mitochondrial stress response pathway. Biochem Soc Trans 43:593-594. 55. Gatliff J, Campanella M (2015) TSPO is a REDOX regulator of cell mitophagy. Biochem Soc Trans 43:543-552. 56. Li F, Liu J, Zheng Y, Garavito MR, Ferguson-Miller S (2015) Crystal structures of translocator protein (TSPO) and mutant mimic of a human polymorphism. Science 347:555-558. 57. Guo Y, Kalathur RC, Liu Q, et al. (2015) Protein structure. Structure and activity of tryptophan-rich TSPO proteins. Science 347:551-555. 58. Farde L (2015) TSPO binding may also represent 'resting' microglia. Clin Trans Imaging 3:491-492. 59. Meikle SR, Badawi, R. D. (2003) Quantitative techniques in PET. In Positron Emission Tomography: Basic Science and Clinical Practice, Eds. Valk PE, Bailey DL, Townsend DW, Maisey MN. London, United Kingdom: Spring-Verlag, pp 115-147. 60. Bailey DL, Karp, J. S., Surti, S. (2003) Physics and instrumentation in PET. In Positron Emission Tomography: Basic Science and Clinical Practice, Eds. Valk PE, Bailey DL, Townsend DW, Maisey MN. London, United Kingdom: Springer-Verlag, pp 41-69. 61. Defrise M, Kinahan, P. E., Michel, C. (2003) Image reconstruction algorithms in PET. In Positron Emission Tomography: Basic Science and Clinical Practice, Eds. Valk PE, Bailey DL, Townsend DW, Maisey MN. London, United Kingdom: Springer-Verlag, pp 91- 115. 62. Soret M, Bacharach SL, Buvat I (2007) Partial-volume effect in PET tumor imaging. J Nucl Med 48:932-945. 63. Hoffman EJ, Phelps ME (1986) Positron emission tomography and autoradiography: principles and applications for the brain and heart. New York: Raven Press. 64. Erlandsson K, Buvat I, Pretorius PH, Thomas BA, Hutton BF (2012) A review of partial volume correction techniques for emission tomography and their applications in neurology, cardiology and oncology. Phys Med Biol 57:R119-159.

210

65. Hutton BF, Thomas BA, Erlandsson K, et al. (2013) What approach to brain partial volume correction is best for PET/MRI? Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 702:29-33. 66. Lavisse S, Guillermier M, Herard AS, et al. (2012) Reactive astrocytes overexpress TSPO and are detected by TSPO positron emission tomography imaging. J Neurosci 32:10809-10818. 67. Cosenza-Nashat M, Zhao ML, Suh HS, et al. (2009) Expression of the translocator protein of 18 kDa by microglia, macrophages and astrocytes based on immunohistochemical localization in abnormal human brain. Neuropathol Appl Neurobiol 35:306-328. 68. Shah F, Hume SP, Pike VW, Ashworth S, McDermott J (1994) Synthesis of the enantiomers of [N-methyl-11C]PK 11195 and comparison of their behaviours as radioligands for PK binding sites in rats. Nucl Med Biol 21:573-581. 69. Banati RB (2002) Visualising microglial activation in vivo. Glia 40:206-217. 70. Imaizumi M, Kim HJ, Zoghbi SS, et al. (2007a) PET imaging with [11C]PBR28 can localize and quantify upregulated peripheral benzodiazepine receptors associated with cerebral ischemia in rat. Neurosci Lett 411:200-205. 71. Imaizumi M, Briard E, Zoghbi SS, et al. (2008) Brain and whole-body imaging in nonhuman primates of [11C]PBR28, a promising PET radioligand for peripheral benzodiazepine receptors. Neuroimage 39:1289-1298. 72. Petit-Taboue MC, Baron JC, Barre L, et al. (1991) Brain kinetics and specific binding of [11C]PK 11195 to omega 3 sites in baboons: positron emission tomography study. Eur J Pharmacol 200:347-351. 73. Kreisl WC, Fujita M, Fujimura Y, et al. (2010) Comparison of [(11)C]-(R)-PK 11195 and [(11)C]PBR28, two radioligands for translocator protein (18 kDa) in human and monkey: Implications for positron emission tomographic imaging of this inflammation biomarker. Neuroimage 49:2924-2932. 74. Owen DR, Howell OW, Tang SP, et al. (2010) Two binding sites for [3H]PBR28 in human brain: implications for TSPO PET imaging of neuroinflammation. J Cereb Blood Flow Metab 30:1608-1618. 75. Owen DRJ, Gunn RN, Rabiner EA, et al. (2011) Mixed-affinity binding in humans with 18-kDa translocator protein ligands. Journal of Nuclear Medicine 52:24-32. 76. Fujimura Y, Kimura Y, Simeon FG, et al. (2010) Biodistribution and radiation dosimetry in humans of a new PET ligand, F-18-PBR06, to image translocator protein (18 kDa). Journal of Nuclear Medicine 51:145-149. 77. Owen DR, Yeo AJ, Gunn RN, et al. (2012) An 18-kDa translocator protein (TSPO) polymorphism explains differences in binding affinity of the PET radioligand PBR28. J Cereb Blood Flow Metab 32:1-5. 78. Gerhard A, Banati RB, Cagnin A, Brooks DJ (2001) In vivo imaging of activated microglia with [C-11]PK11195 positron emission tomography (PET) in idiopathic and atypical Parkinson's disease. Neurology 56:A270-A270. 79. Gerhard A, Pavese N, Hotton G, et al. (2006) In vivo imaging of microglial activation with [C-11](R)-PK11195 PET in idiopathic Parkinson's disease. Neurobiology of Disease 21:404-412. 80. Lublin FD, Reingold SC (1996) Defining the clinical course of multiple sclerosis: results of an international survey. National Multiple Sclerosis Society (USA) Advisory Committee on Clinical Trials of New Agents in Multiple Sclerosis. Neurology 46:907-911. 81. Compston A, Coles A (2008) Multiple sclerosis. Lancet 372:1502-1517. 82. Fisniku LK, Brex PA, Altmann DR, et al. (2008) Disability and T2 MRI lesions: a 20-year follow-up of patients with relapse onset of multiple sclerosis. Brain 131:808-817. 83. Langrish CL, Chen Y, Blumenschein WM, et al. (2005) IL-23 drives a pathogenic T cell population that induces autoimmune inflammation. J Exp Med 201:233-240.

211

84. Grigoriadis N, van Pesch V, Paradig MSG (2015) A basic overview of multiple sclerosis immunopathology. Eur J Neurol 22 Suppl 2:3-13. 85. Stys PK, Zamponi GW, van Minnen J, Geurts JJ (2012) Will the real multiple sclerosis please stand up? Nat Rev Neurosci 13:507-514. 86. Bruck W (2005) Inflammatory demyelination is not central to the pathogenesis of multiple sclerosis. J Neurol 252 Suppl 5:v10-15. 87. Inglese M, Mancardi GL, Pagani E, et al. (2004) Brain tissue loss occurs after suppression of enhancement in patients with multiple sclerosis treated with autologous haematopoietic stem cell transplantation. J Neurol Neurosurg Psychiatry 75:643-644. 88. DeLuca GC, Williams K, Evangelou N, Ebers GC, Esiri MM (2006) The contribution of demyelination to axonal loss in multiple sclerosis. Brain 129:1507-1516. 89. Evangelou N, DeLuca GC, Owens T, Esiri MM (2005) Pathological study of spinal cord atrophy in multiple sclerosis suggests limited role of local lesions. Brain 128:29-34. 90. Kornek B, Storch MK, Weissert R, et al. (2000) Multiple sclerosis and chronic autoimmune encephalomyelitis: a comparative quantitative study of axonal injury in active, inactive, and remyelinated lesions. Am J Pathol 157:267-276. 91. Faria DD, Copray S, Buchpiguel C, Dierckx R, de Vries E (2014) PET imaging in multiple sclerosis. Journal of Neuroimmune Pharmacology 9:468-482. 92. Kutzelnigg A, Lucchinetti CF, Stadelmann C, et al. (2005) Cortical demyelination and diffuse white matter injury in multiple sclerosis. Brain 128:2705-2712. 93. Magliozzi R, Howell O, Vora A, et al. (2007) Meningeal B-cell follicles in secondary progressive multiple sclerosis associate with early onset of disease and severe cortical pathology. Brain 130:1089-1104. 94. Frischer JM, Bramow S, Dal-Bianco A, et al. (2009) The relation between inflammation and neurodegeneration in multiple sclerosis brains. Brain 132:1175-1189. 95. Krumbholz M, Theil D, Derfuss T, et al. (2005) BAFF is produced by astrocytes and up-regulated in multiple sclerosis lesions and primary central nervous system lymphoma. J Exp Med 201:195-200. 96. Meinl E, Krumbholz M, Derfuss T, Junker A, Hohlfeld R (2008) Compartmentalization of inflammation in the CNS: a major mechanism driving progressive multiple sclerosis. J Neurol Sci 274:42-44. 97. Pérez-Cerdá F, Sánchez-Gómez MV, Matute C (2016) The link of inflammation and neurodegeneration in progressive multiple sclerosis. Mult Scler Demyel Disord 1. 98. Venneti S, Lopresti BJ, Wiley CA (2006) The peripheral benzodiazepine receptor (Translocator protein 18kDa) in microglia: from pathology to imaging. Prog Neurobiol 80:308-322. 99. Cutter GR, Baier ML, Rudick RA, et al. (1999) Development of a multiple sclerosis functional composite as a clinical trial outcome measure. Brain 122 ( Pt 5):871-882. 100. Rudick R, Antel J, Confavreux C, et al. (1997) Recommendations from the National Multiple Sclerosis Society clinical outcomes assessment task force. Ann Neurol 42:379- 382. 101. Filippi M, Horsfield MA, Morrissey SP, et al. (1994) Quantitative brain MRI lesion load predicts the course of clinically isolated syndromes suggestive of multiple sclerosis. Neurology 44:635-641. 102. Brex PA, Ciccarelli O, O'Riordan JI, Sailer M, Thompson AJ, Miller DH (2002) A longitudinal study of abnormalities on MRI and disability from multiple sclerosis. N Engl J Med 346:158-164. 103. Masek M, Vaneckova M, Krasensky J, et al. (2008) Secondary-progressive form of multiple sclerosis: MRI changes versus clinical status. Neuro Endocrinol Lett 29:461-466. 104. Filippi M, Paty DW, Kappos L, et al. (1995) Correlations between changes in disability and T2-weighted brain MRI activity in multiple sclerosis: a follow-up study. Neurology 45:255-260. 105. Kappos L, Moeri D, Radue EW, et al. (1999) Predictive value of gadolinium- enhanced magnetic resonance imaging for relapse rate and changes in disability or

212

impairment in multiple sclerosis: a meta-analysis. Gadolinium MRI Meta-analysis Group. Lancet 353:964-969. 106. Vas A, Shchukin Y, Karrenbauer VD, et al. (2008) Functional neuroimaging in multiple sclerosis with radiolabelled glia markers: preliminary comparative PET studies with [11C]vinpocetine and [11C]PK11195 in patients. J Neurol Sci 264:9-17. 107. Versijpt J, Debruyne JC, Van Laere KJ, et al. (2005) Microglial imaging with positron emission tomography and atrophy measurements with magnetic resonance imaging in multiple sclerosis: a correlative study. Mult Scler 11:127-134. 108. Rissanen E, Tuisku J, Rokka J, et al. (2014) In vivo detection of diffuse inflammation in secondary progressive multiple sclerosis using PET imaging and the radioligand 11C-PK11195. J Nucl Med 55:939-944. 109. Banati RB, Gehrmann J, Schubert P, Kreutzberg GW (1993) Cytotoxicity of microglia. Glia 7:111-118. 110. Li H, Cuzner ML, Newcombe J (1996) Microglia-derived macrophages in early multiple sclerosis plaques. Neuropathol Appl Neurobiol 22:207-215. 111. Oh U, Fujita M, Ikonomidou VN, et al. (2011) Translocator protein PET imaging for glial activation in multiple sclerosis. J Neuroimmune Pharmacol 6:354-361. 112. Colasanti A, Guo Q, Muhlert N, et al. (2014) In vivo assessment of brain white matter inflammation in multiple sclerosis with 18F-PBR111 PET. J Nucl Med 55:1112- 1118. 113. Boutin H, Prenant C, Maroy R, et al. (2013) [(18) F]DPA-714: direct comparison with [C-11]PK11195 in a model of cerebral ischemia in rats. Plos One 8. 114. Harhausen D, Sudmann V, Khojasteh U, et al. (2013) Specific imaging of inflammation with the 18kDa translocator protein ligand DPA-714 in animal models of epilepsy and stroke. Plos One 8. 115. Boutin H, Murray K, Pradillo J, et al. (2014) F-GE-180: a novel TSPO radiotracer compared to C-R-PK11195 in a preclinical model of stroke. Eur J Nucl Med Mol Imaging. 116. Martin A, Boisgard R, Theze B, et al. (2010) Evaluation of the PBR/TSPO radioligand [F-18]DPA-714 in a rat model of focal cerebral ischemia. Journal of Cerebral Blood Flow and Metabolism 30:230-241. 117. Elder GA, Gama Sosa MA, De Gasperi R (2010) Transgenic mouse models of Alzheimer's disease. Mt Sinai J Med 77:69-81. 118. Mattner F, Staykova M, Berghofer P, et al. (2013) Central nervous system expression and PET imaging of the translocator protein in relapsing-remitting experimental autoimmune encephalomyelitis. J Nucl Med 54:291-298. 119. Airas L, Dickens AM, Elo P, et al. (2015) In vivo PET imaging demonstrates diminished microglial activation after fingolimod treatment in an animal model of multiple sclerosis. J Nucl Med 56:305-310. 120. Liu M, Bing GY (2011) Lipopolysaccharide animal models for Parkinson's disease. Parkinsons Disease 2011. 121. Deng XH, Li M, Ai W, et al. (2014) Lipolysaccharide-induced neuroinflammation is associated with Alzheimer-like amyloidogenic axonal pathology and dendritic degeneration in rats. Advances in Alzheimer's Disease 3:78-93. 122. Dickens AM, Vainio S, Marjamaki P, et al. (2014) Detection of microglial activation in an acute model of neuroinflammation using PET and radiotracers C-11-(R)-PK11195 and F-18-GE-180. Journal of Nuclear Medicine 55:466-472. 123. Ory D, Planas A, Dresselaers T, et al. (2015) PET imaging of TSPO in a rat model of local neuroinflammation induced by intracerebral injection of lipopolysaccharide. Nucl Med Biol 42:753-761. 124. Ory D, Postnov A, Koole M, et al. (2015) Quantification of TSPO overexpression in a rat model of local neuroinflammation induced by intracerebral injection of LPS by the use of [F]DPA-714 PET. Eur J Nucl Med Mol Imaging. 125. Streit WJ (2002) Microglia as neuroprotective, immunocompetent cells of the CNS. Glia 40:133-139.

213

126. Phelps ME, Huang SC, Hoffman EJ, Kuhl DE (1979a) Validation of tomographic measurement of cerebral blood volume with C-11-labeled carboxyhemoglobin. Journal of Nuclear Medicine 20:328-334. 127. Phelps ME, Huang SC, Hoffman EJ, Selin C, Sokoloff L, Kuhl DE (1979b) Tomographic measurement of local cerebral glucose metabolic rate in humans with (F- 18)2-fluoro-2-deoxy-D-glucose: validation of method. Ann Neurol 6:371-388. 128. Pike VW, Halldin C, Crouzel C, et al. (1993) Radioligands for PET studies of central benzodiazepine receptors and PK (peripheral benzodiazepine) binding sites--current status. Nucl Med Biol 20:503-525. 129. De Vos F, Dumont F, Santens P, Slegers G, Dierckx R, De Reuck J (1999) High- performance liquid chromatographic determination of [11C]1-(2-chlorophenyl)-N- methyl-N-(1-methylpropyl)-3-isoquinoline carboxamide in mouse plasma and tissue and in human plasma. J Chromatogr B Biomed Sci Appl 736:61-66. 130. Roivainen A, Nagren K, Hirvonen J, et al. (2009) Whole-body distribution and metabolism of [N-methyl-11C](R)-1-(2-chlorophenyl)-N-(1-methylpropyl)-3- isoquinolinecarboxamide in humans; an imaging agent for in vivo assessment of peripheral benzodiazepine receptor activity with positron emission tomography. Eur J Nucl Med Mol Imaging 36:671-682. 131. Greuter HN, van Ophemert PL, Luurtsema G, et al. (2005) Optimizing an online SPE-HPLC method for analysis of (R)-[11C]1-(2-chlorophenyl)-N-methyl-N-(1- methylpropyl)-3-isoquinolinecarboxamide [(R)-[11C]PK11195] and its metabolites in humans. Nucl Med Biol 32:307-312. 132. Jucaite A, Cselenyi Z, Arvidsson A, et al. (2012) Kinetic analysis and test-retest variability of the radioligand [11C](R)-PK11195 binding to TSPO in the human brain - a PET study in control subjects. EJNMMI Res 2:15. 133. Charbonneau P, Syrota A, Crouzel C, Valois JM, Prenant C, Crouzel M (1986) Peripheral-type benzodiazepine receptors in the living heart characterized by positron emission tomography. Circulation 73:476-483. 134. Kropholler MA, Boellaard R, Schuitemaker A, et al. (2005) Development of a tracer kinetic plasma input model for (R)-[11C]PK11195 brain studies. J Cereb Blood Flow Metab 25:842-851. 135. Turkheimer FE, Edison P, Pavese N, et al. (2007) Reference and target region modeling of [C-11]-(R)-PK11195 brain studies. Journal of Nuclear Medicine 48:158-167. 136. Lamare F, Hinz R, Gaemperli O, et al. (2011) Detection and quantification of large- vessel inflammation with 11C-(R)-PK11195 PET/CT. Journal of Nuclear Medicine 52:33- 39. 137. Koeppe RA, Holthoff VA, Frey KA, Kilbourn MR, Kuhl DE (1991) Compartmental analysis of [11C]flumazenil kinetics for the estimation of ligand transport rate and receptor distribution using positron emission tomography. J Cereb Blood Flow Metab 11:735-744. 138. Gunn RN, Gunn SR, Cunningham VJ (2001) Positron emission tomography compartmental models. J Cereb Blood Flow Metab 21:635-652. 139. Carson RE (1986) Parameter estimation in positron emission tomography. In Positron emission tomography and autoradiography: principles and applications for the brain and heart, Ed. Phelps M, Mazziotta, J., Shelbert H. New York: Raven Press, pp 347- 390. 140. Gunn RN, Lammertsma AA, Hume SP, Cunningham VJ (1997) Parametric imaging of ligand-receptor binding in PET using a simplified reference region model. Neuroimage 6:279-287. 141. Logan J, Fowler JS, Volkow ND, et al. (1990) Graphical analysis of reversible radioligand binding from time-activity measurements applied to [N-11C-methyl]-(-)- cocaine PET studies in human subjects. J Cereb Blood Flow Metab 10:740-747.

214

142. Patlak CS, Blasberg RG, Fenstermacher JD (1983) Graphical evaluation of blood- to-brain transfer constants from multiple-time uptake data. J Cereb Blood Flow Metab 3:1-7. 143. Ko JH, Koshimori Y, Mizrahi R, et al. (2013) Voxel-based imaging of translocator protein 18 kDa (TSPO) in high-resolution PET. J Cereb Blood Flow Metab 33:348-350. 144. Slifstein M, Laruelle M (2000) Effects of statistical noise on graphic analysis of PET neuroreceptor studies. J Nucl Med 41:2083-2088. 145. Cunningham VJ, Jones T (1993) Spectral analysis of dynamic PET studies. J Cereb Blood Flow Metab 13:15-23. 146. Press WH, Flannery, B. P., Teukololsky, S. A., Vetterling, W. T. (1988) Numerical recipes in C. New York: Cambridge University Press. 147. Lawson CL, Hanson, R. J. (1974) Solving least squares problems. Englewood Cliffs, New Jersey: Prentice-Hall. 148. Edison P, Archer HA, Gerhard A, et al. (2008) Microglia, amyloid, and cognition in Alzheimer's disease: an [11C](R)PK11195-PET and [11C]PIB-PET study. Neurobiology of Disease 32:412-419. 149. Akaike H (1974) A new look at the statistical model identification. IEEE Transactions on Automatic Control AC-19:716-723. 150. Turkheimer FE, Hinz R, Cunningham VJ (2003) On the undecidability among kinetic models: from model selection to model averaging. J Cereb Blood Flow Metab 23:490-498. 151. Turkheimer FE, Rizzo G, Bloomfield PS, et al. (2015) The methodology of TSPO imaging with positron emission tomography. Biochem Soc Trans 43:586-592. 152. Bloomfield PS, Selvaraj S, Veronese M, et al. (2016) Microglial activity in people at ultra high risk of psychosis and in schizophrenia: an [(11)C]PBR28 PET brain imaging study. Am J Psychiatry 173:44-52. 153. Owen DR, Guo Q, Kalk NJ, et al. (2014) Determination of [C]PBR28 binding potential in vivo: a first human TSPO blocking study. J Cereb Blood Flow Metab. 154. Rizzo G, Veronese M, Tonietto M, Zanotti-Fregonara P, Turkheimer FE, Bertoldo A (2014) Kinetic modeling without accounting for the vascular component impairs the quantification of [(11)C]PBR28 brain PET data. J Cereb Blood Flow Metab 34:1060-1069. 155. Golla SS, Boellaard R, Oikonen V, et al. (2015) Quantification of [F]DPA-714 binding in the human brain: initial studies in healthy controls and Alzheimer's disease patients. J Cereb Blood Flow Metab. 156. Schuitemaker A, van Berckel BN, Kropholler MA, et al. (2007a) SPM analysis of parametric (R)-[11C]PK11195 binding images: plasma input versus reference tissue parametric methods. Neuroimage 35:1473-1479. 157. Herholz K, Patlak CS (1987) The influence of tissue heterogeneity on results of fitting nonlinear model equations to regional tracer uptake curves: with an application to compartmental models used in positron emission tomography. J Cereb Blood Flow Metab 7:214-229. 158. Schuitemaker A, van Berckel BN, Kropholler MA, et al. (2007b) Evaluation of methods for generating parametric (R-[11C]PK11195 binding images. J Cereb Blood Flow Metab 27:1603-1615. 159. Chen K, Bandy D, Reiman E, et al. (1998) Noninvasive quantification of the cerebral metabolic rate for glucose using positron emission tomography, 18F-fluoro-2- deoxyglucose, the Patlak method, and an image-derived input function. J Cereb Blood Flow Metab 18:716-723. 160. Weinberg IN, Huang SC, Hoffman EJ, et al. (1988) Validation of PET-acquired input functions for cardiac studies. Journal of Nuclear Medicine 29:241-247. 161. Krejza J, Arkuszewski M, Kasner SE, et al. (2006) Carotid artery diameter in men and women and the relation to body and neck size. Stroke 37:1103-1105.

215

162. Mourik JE, Lubberink M, Klumpers UM, Comans EF, Lammertsma AA, Boellaard R (2008) Partial volume corrected image derived input functions for dynamic PET brain studies: methodology and validation for [11C]flumazenil. Neuroimage 39:1041-1050. 163. Litton JE (1997) Input function in PET brain studies using MR-defined arteries. J Comput Assist Tomogr 21:907-909. 164. Mourik JE, Lubberink M, Schuitemaker A, et al. (2009) Image-derived input functions for PET brain studies. Eur J Nucl Med Mol Imaging 36:463-471. 165. Groom GN, Junck L, Foster NL, Frey KA, Kuhl DE (1995) PET of peripheral benzodiazepine binding sites in the microgliosis of Alzheimer's disease. Journal of Nuclear Medicine 36:2207-2210. 166. Gerhard A, Schwarz J, Myers R, Wise R, Banati RB (2005) Evolution of microglial activation in patients after ischemic stroke: a [11C](R)-PK11195 PET study. Neuroimage 24:591-595. 167. Lammertsma AA, Hume SP (1996) Simplified reference tissue model for PET receptor studies. Neuroimage 4:153-158. 168. Hume SP, Myers R, Bloomfield PM, et al. (1992) Quantitation of carbon-11-labeled raclopride in rat striatum using positron emission tomography. Synapse 12:47-54. 169. Debruyne JC, Versijpt J, Van Laere KJ, et al. (2003) PET visualization of microglia in multiple sclerosis patients using [C-11]PK11195. European Journal of Neurology 10:257-264. 170. Versijpt JJ, Dumont F, Van Laere KJ, et al. (2003) Assessment of neuroinflammation and microglial activation in Alzheimer's disease with radiolabelled PK11195 and single photon emission computed tomography. A pilot study. Eur Neurol 50:39-47. 171. Albrecht D, Shcherbinin S, Wooten D, et al. (2016) Occipital lobe as a pseudo- reference region for [11C]PBR28 PET imaging: validation in chronic pain and amyotrophic lateral sclerosis cohorts. Journal of Nuclear Medicine 57:1814. 172. Lyoo CH, Ikawa M, Liow JS, et al. (2015) Cerebellum can serve as a pseudo- reference region in Alzheimer disease to detect neuroinflammation measured with PET radioligand binding to translocator protein. Journal of Nuclear Medicine 56:701-706. 173. Cunningham VJ, Hume SP, Price GR, Ahier RG, Cremer JE, Jones AK (1991) Compartmental analysis of diprenorphine binding to opiate receptors in the rat in vivo and its comparison with equilibrium data in vitro. J Cereb Blood Flow Metab 11:1-9. 174. Logan J, Fowler JS, Volkow ND, Wang GJ, Ding YS, Alexoff DL (1996) Distribution volume ratios without blood sampling from graphical analysis of PET data. J Cereb Blood Flow Metab 16:834-840. 175. Holden JE, Sossi, V., Chan, G., Doudet, D. J., Stoessl, A. J. and Ruth, T. J. (2001) Effect of population k2 values in graphical estimation of DV ratios of reversible ligands. In Physiological Imaging of the Brain with PET, Ed. Gjedde A. San Diego, CA: Academic Press, pp 127-129. 176. Turkheimer FE, Hinz R, Gunn RN, Aston JA, Gunn SR, Cunningham VJ (2003) Rank-shaping regularization of exponential spectral analysis for application to functional parametric mapping. Phys Med Biol 48:3819-3841. 177. Turkheimer FE, Aston JA, Asselin MC, Hinz R (2006) Multi-resolution Bayesian regression in PET dynamic studies using wavelets. Neuroimage 32:111-121. 178. Gunn RN, Gunn SR, Turkheimer FE, Aston JA, Cunningham VJ (2002) Positron emission tomography compartmental models: a basis pursuit strategy for kinetic modeling. J Cereb Blood Flow Metab 22:1425-1439. 179. Chen SS (1995) Basis Pursuit. Stanford University. 180. Koeppe RA, Holden JE, Ip WR (1985) Performance comparison of parameter estimation techniques for the quantitation of local cerebral blood flow by dynamic positron computed tomography. J Cereb Blood Flow Metab 5:224-234.

216

181. Kropholler MA, Boellaard R, Schuitemaker A, Folkersma H, van Berckel BN, Lammertsma AA (2006) Evaluation of reference tissue models for the analysis of [11C](R)-PK11195 studies. J Cereb Blood Flow Metab 26:1431-1441. 182. Kropholler MA, Boellaard R, van Berckel BN, et al. (2007) Evaluation of reference regions for (R)-[(11)C]PK11195 studies in Alzheimer's disease and mild cognitive impairment. J Cereb Blood Flow Metab 27:1965-1974. 183. Cagnin A, Brooks DJ, Kennedy AM, et al. (2001b) In-vivo measurement of activated microglia in dementia. Lancet 358:461-467. 184. Buchert R, Wilke F, Chakrabarti B, et al. (2005) Adjusted scaling of FDG positron emission tomography images for statistical evaluation in patients with suspected Alzheimer's disease. J Neuroimaging 15:348-355. 185. Ichise M, Ballinger JR, Golan H, et al. (1996) Noninvasive quantification of dopamine D2 receptors with iodine-123-IBF SPECT. Journal of Nuclear Medicine 37:513- 520. 186. Toth M, Doorduin J, Haggkvist J, et al. (2015) Positron emission tomography studies with [11C]PBR28 in the healthy rodent brain: validating SUV as an outcome measure of neuroinflammation. PLoS One 10:e0125917. 187. Yoder KK, Territo PR, Hutchins GD, et al. (2015) Comparison of standardized uptake values with volume of distribution for quantitation of [C-11]PBR28 brain uptake. Nuclear Medicine and Biology 42:305-308. 188. Hartigan JA, Wong MA (1979) Algorithm AS 136: a k-means clustering algorithm. Journal of the Royal Statistical Society Series C (Applied Statistics) 28:100-108. 189. MacKay DJC (2003) Information theory, inference and learning algorithms. Cambridge University Press. 190. Chapple M K-means clustering. http://databases.about.com/od/datamining/a/kmeans.htm (accessed 22/03/2013). 191. Ashburner J, Haslam J, Taylor C, Cunningham VJ, Jones T (1996) A cluster analysis approach for the characterization of dynamic PET data. In Quantification of Brain Function Using PET, Eds. Myers R, Cunningham VJ, Bailey D, Jones T. San Diego: Academic Press, pp 301-306. 192. Hartigan JA (1975) Clustering algorithms. New York: Wiley. 193. Myers R, Gunn RN, Cunningham VJ, Banati RB, Jones T (1999) Cluster analysis and the reference tissue model in the analysis of clinical [11C]PK11195 PET. Journal of Cerebral Blood Flow and Metabolism 19:S789. 194. Kurtzke JF (1983) Rating neurologic impairment in multiple sclerosis: an expanded disability status scale (EDSS). Neurology 33:1444-1452. 195. Grira N, Crucianu, M., Nozha, B. Unsupervised and semi-supervised clustering: a brief survey. http://cedric.cnam.fr/~crucianm/src/BriefSurveyClustering.pdf (accessed 09/01/2016). 196. Boellaard R, Turkheimer F, Hinz R, et al. (2008) Performace of a modified supervised cluster algorithm for extracting reference region input functions from (R)- [11C]PK11195 brain PET studies [abstract]. 5400-5402P. 197. Yaqub M, van Berckel BN, Schuitemaker A, et al. (2012) Optimization of supervised cluster analysis for extracting reference tissue input curves in (R)- [(11)C]PK11195 brain PET studies. J Cereb Blood Flow Metab 32:1600-1608. 198. Schuitemaker A, van der Doef TF, Boellaard R, et al. (2012) Microglial activation in healthy aging. Neurobiol Aging 33:1067-1072. 199. Tomasi G, Edison P, Bertoldo A, et al. (2008) Novel reference region model reveals increased microglial and reduced vascular binding of 11C-(R)-PK11195 in patients with Alzheimer's disease. Journal of Nuclear Medicine 49:1249-1256. 200. Cagnin A, Myers R, Gunn RN, et al. (2001a) Imaging activated microglia in the aging human brain. San Diego, CA: Academic Press.

217

201. Anderson AN, Pavese N, Edison P, et al. (2007) A systematic comparison of kinetic modelling methods generating parametric maps for [(11)C]-(R)-PK11195. Neuroimage 36:28-37. 202. Pavese N, Gerhard A, Tai YF, et al. (2006) Microglial activation correlates with severity in Huntington disease: a clinical and PET study. Neurology 66:1638-1643. 203. Kukreja SL, Gunn RN (2004) Bootstrapped DEPICT for error estimation in PET functional imaging. Neuroimage 21:1096-1104. 204. Debruyne JC, Van Laere KJ, Versijpt J, et al. (2002) Semiquantification of the peripheral-type benzodiazepine ligand [11C]PK11195 in normal human brain and application in multiple sclerosis patients. Acta Neurol Belg 102:127-135. 205. Turkheimer FE, Aston JA, Banati RB, Riddell C, Cunningham VJ (2003) A linear wavelet filter for parametric imaging with dynamic PET. IEEE Trans Med Imaging 22:289-301. 206. Ching AS, Kuhnast B, Damont A, Roeda D, Tavitian B, Dolle F (2012) Current paradigm of the 18-kDa translocator protein (TSPO) as a molecular target for PET imaging in neuroinflammation and neurodegenerative diseases. Insights Imaging 3:111- 119. 207. James ML, Fulton RR, Vercoullie J, et al. (2008) DPA-714, a new translocator protein-specific ligand: synthesis, radiofluorination, and pharmacologic characterization. Journal of Nuclear Medicine 49:814-822. 208. Chauveau F, Van Camp N, Dolle F, et al. (2009) Comparative evaluation of the translocator protein radioligands C-11-DPA-713, F-18-DPA-714, and C-11-PK11195 in a rat model of acute neuroinflammation. Journal of Nuclear Medicine 50:468-476. 209. Doorduin J, Klein HC, Dierckx RA, James M, Kassiou M, de Vries EF (2009) [11C]- DPA-713 and [18F]-DPA-714 as new PET tracers for TSPO: a comparison with [11C]-(R)- PK11195 in a rat model of herpes encephalitis. Mol Imaging Biol 11:386-398. 210. Lee DE, Yue X, Ibrahim WG, et al. (2015) Lack of neuroinflammation in the HIV-1 transgenic rat: an [(18)F]-DPA714 PET imaging study. J Neuroinflammation 12:171. 211. Arlicot N, Vercouillie J, Ribeiro MJ, et al. (2012) Initial evaluation in healthy humans of [18F]DPA-714, a potential PET biomarker for neuroinflammation. Nucl Med Biol 39:570-578. 212. Hamelin L, Lagarde J, Dorothee G, et al. (2016) Early and protective microglial activation in Alzheimer's disease: a prospective study using 18F-DPA-714 PET imaging. Brain 139:1252-1264. 213. Liu B, Le KX, Park MA, et al. (2015) In vivo detection of age- and disease-related increases in neuroinflammation by 18F-GE180 TSPO microPET imaging in wild-type and Alzheimer's transgenic mice. J Neurosci 35:15716-15730. 214. Gerhard A, Banati RB, Goerres GB, et al. (2003) [11C](R)-PK11195 PET imaging of microglial activation in multiple system atrophy. Neurology 61:686-689. 215. Serriere S, Tauber C, Vercouillie J, et al. (2015) Amyloid load and translocator protein 18 kDa in APPswePS1-dE9 mice: a longitudinal study. Neurobiol Aging 36:1639- 1652. 216. Lartey FM, Ahn GO, Shen B, et al. (2014) PET imaging of stroke-induced neuroinflammation in mice using [18F]PBR06. Mol Imaging Biol 16:109-117. 217. Folkersma H, Boellaard R, Vandertop WP, et al. (2009) Reference tissue models and blood-brain barrier disruption: lessons from (R)-[11C]PK11195 in traumatic brain injury. J Nucl Med 50:1975-1979. 218. Shao X, Wang X, English SJ, et al. (2013) Imaging of carrageenan-induced local inflammation and adjuvant-induced systemic arthritis with [(11)C]PBR28 PET. Nucl Med Biol 40:906-911. 219. Winkeler A, Boisgard R, Awde AR, et al. (2012) The translocator protein ligand [(1)(8)F]DPA-714 images glioma and activated microglia in vivo. Eur J Nucl Med Mol Imaging 39:811-823.

218

220. Hinz R, Boellaard, R. (2015) Challenges of quantification of TSPO in the human brain. Clin Transl Imaging 3:403-416. 221. Camsonne R, Crouzel C, Comar D, et al. (1984) Synthesis of N-(C-11) methyl, N- (methyl-1 propyl), (chloro-2 phenyl)-1 isoquinoleine carboxamide-3 (PK-11195) - a new ligand for peripheral benzodiazepine receptors. Journal of Labelled Compounds & Radiopharmaceuticals 21:985-991. 222. Cremer JE, Hume SP, Cullen BM, et al. (1992) The distribution of radioactivity in brains of rats given [N-methyl-11C]PK 11195 in vivo after induction of a cortical ischaemic lesion. Int J Rad Appl Instrum B 19:159-166. 223. Wadsworth H, Jones PA, Chau WF, et al. (2012) [(1)(8)F]GE-180: a novel fluorine- 18 labelled PET tracer for imaging Translocator protein 18 kDa (TSPO). Bioorg Med Chem Lett 22:1308-1313. 224. Landini A (1976) [Traumatic lesions of the urinary tract. Radiographic aspects]. Radiol Med 62:855-872. 225. Wickstrom T, Clarke A, Gausemel I, et al. (2014) The development of an automated and GMP compliant FASTlab Synthesis of [(18) F]GE-180; a radiotracer for imaging translocator protein (TSPO). J Labelled Comp Radiopharm 57:42-48. 226. Maroy R, Boisgard R, Comtat C, et al. (2008) Segmentation of rodent whole-body dynamic PET images: an unsupervised method based on voxel dynamics. IEEE Trans Med Imaging 27:342-354. 227. Maroy R, Boisgard R, Comtat C, et al. (2010) Quantitative organ time activity curve extraction from rodent PET images without anatomical prior. Med Phys 37:1507- 1517. 228. Maroy R, Boisgard R, Comtat C, Trebossen R, Tavitian B (2007) Local means analysis: an unsupervised method for the segmentation of rodent whole-body dynamic PET images. 2007 4th Ieee International Symposium on Biomedical Imaging : Macro to Nano, Vols 1-3:324-327. 229. Chauveau F, Boutin H, Van Camp N, et al. (2011) In vivo imaging of neuroinflammation in the rodent brain with [C-11]SSR180575, a novel indoleacetamide radioligand of the translocator protein (18 kDa). European Journal of Nuclear Medicine and Molecular Imaging 38:509-514. 230. Chauveau F, Boutin H, Van Camp N, Dolle F, Tavitian B (2008) Nuclear imaging of neuroinflammation: a comprehensive review of [11C]PK11195 challengers. Eur J Nucl Med Mol Imaging 35:2304-2319. 231. Boutin H, Pinborg LH (2015) TSPO imaging in stroke: from animal models to human subjects. Clinical and Translational Imaging 3:423-435. 232. Varrone A, Nordberg A (2015) Molecular imaging of neuroinflammation in Alzheimer's disease. Clinical and Translational Imaging 3:437-447. 233. Cagnin A, Kassiou M, Meikle SR, Banati RB (2007) Positron emission tomography imaging of neuroinflammation. Neurotherapeutics 4:443-452. 234. Alliot F, Godin I, Pessac B (1999) Microglia derive from progenitors, originating from the yolk sac, and which proliferate in the brain. Brain Res Dev Brain Res 117:145- 152. 235. Gehrmann J, Mies G, Bonnekoh P, et al. (1993) Microglial reaction in the rat cerebral cortex induced by cortical spreading depression. Brain Pathol 3:11-17. 236. Dheen ST, Kaur C, Ling EA (2007) Microglial activation and its implications in the brain diseases. Curr Med Chem 14:1189-1197. 237. Kreutzberg GW (1996) Microglia: a sensor for pathological events in the CNS. Trends Neurosci 19:312-318. 238. Starosta-Rubinstein S, Ciliax BJ, Penney JB, McKeever P, Young AB (1987) Imaging of a glioma using peripheral benzodiazepine receptor ligands. Proc Natl Acad Sci U S A 84:891-895.

219

239. Wang Y, Yue X, Kiesewetter DO, Niu G, Teng G, Chen X (2014) PET imaging of neuroinflammation in a rat traumatic brain injury model with radiolabeled TSPO ligand DPA-714. Eur J Nucl Med Mol Imaging. 240. Lavisse S, Garcia-Lorenzo D, Peyronneau MA, et al. (2015) Optimized quantification of translocator protein radioligand (1)(8)F-DPA-714 uptake in the brain of genotyped healthy volunteers. J Nucl Med 56:1048-1054. 241. Wyss-Coray T, Rogers J (2012) Inflammation in Alzheimer disease-a brief review of the basic science and clinical literature. Cold Spring Harb Perspect Med 2:a006346. 242. Constantinescu CS, Farooqi N, O'Brien K, Gran B (2011) Experimental autoimmune encephalomyelitis (EAE) as a model for multiple sclerosis (MS). Br J Pharmacol 164:1079-1106. 243. Sridharan S, Lepelletier FX, Trigg W, et al. (2016) Comparative evaluation of three TSPO PET radiotracers in a LPS-Induced Model of mild neuroinflammation in rats. Mol Imaging Biol. 244. Dhawan V, Jarden JO, Strother S, Rottenberg DA (1988) Effect of blood curve smearing on the accuracy of parameter estimates obtained for 82Rb/PET studies of blood-brain barrier permeability. Phys Med Biol 33:61-74. 245. Yeung WT, Lee TY, Del Maestro RF, Kozak R, Bennett JD, Brown T (1992) An absorptiometry method for the determination of arterial blood concentration of injected iodinated contrast agent. Phys Med Biol 37:1741-1758. 246. Munk OL, Keiding S, Bass L (2008) A method to estimate dispersion in sampling catheters and to calculate dispersion-free blood time-activity curves. Medical Physics 35:3471-3481. 247. Kleiber M (1947) Body size and metabolic rate. Physiol Rev 27:511-541. 248. Kleiber M (1932) Body size and metabolism. Hilgardia 6:315-353. 249. Turkheimer FE, Selvaraj S, Hinz R, et al. (2012) Quantification of ligand PET studies using a reference region with a displaceable fraction: application to occupancy studies with [(11)C]-DASB as an example. J Cereb Blood Flow Metab 32:70-80. 250. Watabe H, Channing MA, Der MG, et al. (2000) Kinetic analysis of the 5-HT2A ligand [11C]MDL 100,907. J Cereb Blood Flow Metab 20:899-909. 251. Watabe H, Carson, R.E., Iida, H. (2000) The reference tissue model: three compartments for the reference region. Neuroimage 11. 252. Repalli J (2014) Translocator protein (TSPO) role in aging and Alzheimer's disease. Curr Aging Sci 7:168-175. 253. Peyronneau MA, Saba W, Goutal S, et al. (2013) Metabolism and quantification of [(18)F]DPA-714, a new TSPO positron emission tomography radioligand. Drug Metab Dispos 41:122-131. 254. Mauvais-Jarvis F (2015) Sex differences in metabolic homeostasis, diabetes, and obesity. Biol Sex Differ 6:14. 255. Klein SL, Schiebinger L, Stefanick ML, et al. (2015) Opinion: sex inclusion in basic research drives discovery. Proc Natl Acad Sci U S A 112:5257-5258. 256. Meyer E (1989) Simultaneous correction for tracer arrival delay and dispersion in CBF measurements by the H215O autoradiographic method and dynamic PET. J Nucl Med 30:1069-1078. 257. Iida H, Kanno I, Miura S, Murakami M, Takahashi K, Uemura K (1986) Error analysis of a quantitative cerebral blood flow measurement using H2(15)O autoradiography and positron emission tomography, with respect to the dispersion of the input function. J Cereb Blood Flow Metab 6:536-545. 258. Iida H, Higano S, Tomura N, et al. (1988) Evaluation of regional differences of tracer appearance time in cerebral tissues using [O-15]water and dynamic positron emission tomography. Journal of Cerebral Blood Flow and Metabolism 8:285-288. 259. Cremer JE, Seville MP (1983) Regional brain blood flow, blood volume, and haematocrit values in the adult rat. J Cereb Blood Flow Metab 3:254-256.

220

260. Ito H, Kanno I, Kato C, et al. (2004) Database of normal human cerebral blood flow, cerebral blood volume, cerebral oxygen extraction fraction and cerebral metabolic rate of oxygen measured by positron emission tomography with 15O-labelled carbon dioxide or water, carbon monoxide and oxygen: a multicentre study in Japan. Eur J Nucl Med Mol Imaging 31:635-643. 261. Lavisse S, Inoue K, Jan C, et al. (2015) [18F]DPA-714 PET imaging of translocator protein TSPO (18 kDa) in the normal and excitotoxically-lesioned nonhuman primate brain. Eur J Nucl Med Mol Imaging 42:478-494. 262. Sandor P, Cox-van Put J, de Jong W, de Wied D (1986) Continuous measurement of cerebral blood volume in rats with the photoelectric technique: effect of morphine and naloxone. Life Sci 39:1657-1665. 263. Yaqub M, Boellaard R, van Berckel BN, et al. (2009) Evaluation of tracer kinetic models for analysis of [18F]FDDNP studies. Mol Imaging Biol 11:322-333. 264. Salinas CA, Searle GE, Gunn RN (2015) The simplified reference tissue model: model assumption violations and their impact on binding potential. J Cereb Blood Flow Metab 35:304-311. 265. Hemmer B, Kerschensteiner M, Korn T (2015) Role of the innate and adaptive immune responses in the course of multiple sclerosis. Lancet Neurol 14:406-419. 266. Noseworthy JH, Lucchinetti C, Rodriguez M, Weinshenker BG (2000) Multiple sclerosis. N Engl J Med 343:938-952. 267. Bo L, Geurts JJ, Mork SJ, van der Valk P (2006) Grey matter pathology in multiple sclerosis. Acta Neurol Scand Suppl 183:48-50. 268. Lassmann H (2014) Mechanisms of white matter damage in multiple sclerosis. Glia 62:1816-1830. 269. Reynolds R, Roncaroli F, Nicholas R, Radotra B, Gveric D, Howell O (2011) The neuropathological basis of clinical progression in multiple sclerosis. Acta Neuropathol 122:155-170. 270. Hochmeister S, Grundtner R, Bauer J, et al. (2006) Dysferlin is a new marker for leaky brain blood vessels in multiple sclerosis. J Neuropathol Exp Neurol 65:855-865. 271. Lassmann H, Raine CS, Antel J, Prineas JW (1998) Immunopathology of multiple sclerosis: report on an international meeting held at the Institute of Neurology of the University of Vienna. J Neuroimmunol 86:213-217. 272. Newcombe J, Hawkins CP, Henderson CL, et al. (1991) Histopathology of multiple sclerosis lesions detected by magnetic resonance imaging in unfixed postmortem central nervous system tissue. Brain 114 ( Pt 2):1013-1023. 273. Kutzelnigg A, Lassmann H (2014) Pathology of multiple sclerosis and related inflammatory demyelinating diseases. Handb Clin Neurol 122:15-58. 274. Magliozzi R, Howell OW, Reeves C, et al. (2010) A gradient of neuronal loss and meningeal inflammation in multiple sclerosis. Ann Neurol 68:477-493. 275. Moll NM, Rietsch AM, Thomas S, et al. (2011) Multiple sclerosis normal-appearing white matter: pathology-imaging correlations. Ann Neurol 70:764-773. 276. Park E, Gallezot JD, Delgadillo A, et al. (2015) (11)C-PBR28 imaging in multiple sclerosis patients and healthy controls: test-retest reproducibility and focal visualization of active white matter areas. Eur J Nucl Med Mol Imaging 42:1081-1092. 277. Takano A, Piehl F, Hillert J, et al. (2013) In vivo TSPO imaging in patients with multiple sclerosis: a brain PET study with [18F]FEDAA1106. EJNMMI Res 3:30. 278. Fan Z, Calsolaro V, Atkinson RA, et al. (2016) Flutriciclamide (18F-GE180) PET: first in human PET study of novel 3rd generation in vivo marker of human translator protein. J Nucl Med. 279. Feeney C, Scott G, Raffel J, et al. (2016) Kinetic analysis of the translocator protein positron emission tomography ligand [18F]GE-180 in the human brain. Eur J Nucl Med Mol Imaging. 280. Polman CH, Reingold SC, Banwell B, et al. (2011) Diagnostic criteria for multiple sclerosis: 2010 revisions to the McDonald criteria. Ann Neurol 69:292-302.

221

281. Parmenter BA, Weinstock-Guttman B, Garg N, Munschauer F, Benedict RH (2007) Screening for cognitive impairment in multiple sclerosis using the Symbol digit Modalities Test. Mult Scler 13:52-57. 282. Hammers A, Allom R, Free SL, et al. (2002) Three-dimensional probabilistic atlas of the human brain. Neuroimage 16:S86-S86. 283. Nelson F, Poonawalla AH, Hou P, Huang F, Wolinsky JS, Narayana PA (2007) Improved identification of intracortical lesions in multiple sclerosis with phase-sensitive inversion recovery in combination with fast double inversion recovery MR imaging. AJNR Am J Neuroradiol 28:1645-1649. 284. Lucchinetti CF, Popescu BF, Bunyan RF, et al. (2011) Inflammatory cortical demyelination in early multiple sclerosis. N Engl J Med 365:2188-2197. 285. Kiferle L, Politis M, Muraro PA, Piccini P (2011) Positron emission tomography imaging in multiple sclerosis-current status and future applications. Eur J Neurol 18:226- 231. 286. Kumar A, Muzik O, Shandal V, Chugani D, Chakraborty P, Chugani HT (2012) Evaluation of age-related changes in translocator protein (TSPO) in human brain using (11)C-[R]-PK11195 PET. J Neuroinflammation 9:232. 287. Anderson VM, Wheeler-Kingshott CA, Abdel-Aziz K, et al. (2011) A comprehensive assessment of cerebellar damage in multiple sclerosis using diffusion tractography and volumetric analysis. Mult Scler 17:1079-1087. 288. Nakashima I, Fujihara K, Okita N, Takase S, Itoyama Y (1999) Clinical and MRI study of brain stem and cerebellar involvement in Japanese patients with multiple sclerosis. J Neurol Neurosurg Psychiatry 67:153-157. 289. Poser S, Wikstrom J, Bauer HJ (1979) Clinical data and the identification of special forms of multiple sclerosis in 1271 cases studied with a standardized documentation system. J Neurol Sci 40:159-168. 290. Weier K, Banwell B, Cerasa A, et al. (2015) The role of the cerebellum in multiple sclerosis. Cerebellum 14:364-374. 291. Love S (2006) Demyelinating diseases. J Clin Pathol 59:1151-1159. 292. Savio S, Hakulinen U, Ryymin P, et al. (2015) Hemispheric asymmetry measured by texture analysis and diffusion tensor imaging in two multiple sclerosis subtypes. Acta Radiol 56:844-851. 293. Burnham KP, Anderson DR (1998) Model selection and inference: a practical information-theoretical approach. 1st ed., New York: Springer-Verlag. 294. Lemere CA, Liu B, Le KX, et al. (2014) 18F-GE180 PET imaging of neuroinflammation in aging and Alzheimer's disease in mice. Neurobiology of Aging 35:S14-S14. 295. Collste K, Forsberg A, Varrone A, et al. (2016) Test-retest reproducibility of [C- 11]PBR28 binding to TSPO in healthy control subjects. European Journal of Nuclear Medicine and Molecular Imaging 43:173-183. 296. Takano A, Gulyas B, Varrone A, et al. (2011) Biodistribution and radiation dosimetry of the 18 kDa translocator protein (TSPO) radioligand [18F]FEDAA1106: a human whole-body PET study. Eur J Nucl Med Mol Imaging 38:2058-2065. 297. Hirvonen J, Roivainen A, Virta J, Helin S, Nagren K, Rinne JO (2010) Human biodistribution and radiation dosimetry of 11C-(R)-PK11195, the prototypic PET ligand to image inflammation. Eur J Nucl Med Mol Imaging 37:606-612. 298. Brown AK, Fujita M, Fujimura Y, et al. (2007a) Radiation dosimetry and biodistribution in monkey and man of 11C-PBR28: a PET radioligand to image inflammation. Journal of Nuclear Medicine 48:2072-2079. 299. Doble A, Benavides J, Ferris O, et al. (1985) Dihydropyridine and peripheral type benzodiazepine binding sites: subcellular distribution and molecular size determination. Eur J Pharmacol 119:153-167. 300. O'Beirne GB, Woods MJ, Williams DC (1990) Two subcellular locations for peripheral-type benzodiazepine acceptors in rat liver. Eur J Biochem 188:131-138.

222

301. Harberts E, Datta D, Chen S, Wohler JE, Oh U, Jacobson S (2012) Translocator protein 18 kDa (TSPO) expression in multiple sclerosis patients. J Neuroimmune Pharmacol. 302. Ferrero P, Rocca P, Benna P, et al. (1992) An analysis of peripheral type benzodiazepine receptors on blood mononuclear cells during high dose steroid treatment of multiple sclerosis. Ital J Neurol Sci 13:685-691. 303. Kropholler MA, Boellaard R, Elzinga EH, et al. (2009) Quantification of (R)- [11C]PK11195 binding in rheumatoid arthritis. Eur J Nucl Med Mol Imaging 36:624-631. 304. Ito H, Kanno I, Iida H, et al. (2001) Arterial fraction of cerebral blood volume in humans measured by positron emission tomography. Ann Nucl Med 15:111-116. 305. Lammertsma AA (2002) Radioligand studies: imaging and quantitative analysis. Eur Neuropsychopharmacol 12:513-516. 306. Leenders KL, Perani D, Lammertsma AA, et al. (1990) Cerebral blood flow, blood volume and oxygen utilization. Normal values and effect of age. Brain 113 ( Pt 1):27-47. 307. Giannetti P, Politis M, Su P, et al. (2015) Increased PK11195-PET binding in normal-appearing white matter in clinically isolated syndrome. Brain 138:110-119. 308. Politis M, Giannetti P, Su P, et al. (2012) Increased PK11195 PET binding in the cortex of patients with MS correlates with disability. Neurology 79:523-530. 309. Airas L, Rissanen E, Rinne J (2015) Imaging neuroinflammation in multiple sclerosis using TSPO-PET. Clin Transl Imaging 3:461-473. 310. Jensen CV, Rostrup E, Blinkenberg M, Larsson HBW, Henrikson O (1995) Relative regional cerebral blood-volume in multiple sclerosis [abstract]. 1995: 1297P. 311. Petrella JR, Yang Y, Richert N, et al. (1997) Dynamic Contrast MR Measurements of Relative Cerebral Blood Volume in Enhancing Multiple Sclerosis Lesions: Comparison to Nonenhancing lesions and Normal Appearing White Matter [abstract]. 1997: 647P. 312. Haselhorst R, Kappos L, Bilecen D, et al. (2000) Dynamic susceptibility contrast MR imaging of plaque development in multiple sclerosis: application of an extended blood-brain barrier leakage correction. J Magn Reson Imaging 11:495-505. 313. Gunn RN, Murthy V, Catafau AM, et al. (2011) Translational characterization of [11C]GSK931145, a PET ligand for the glycine transporter type 1. Synapse 65:1319-1332. 314. Guo Q, Colasanti A, Owen DR, et al. (2013) Quantification of the specific translocator protein signal of 18F-PBR111 in healthy humans: a genetic polymorphism effect on in vivo binding. J Nucl Med 54:1915-1923. 315. Kreisl WC, Jenko KJ, Hines CS, et al. (2013) A genetic polymorphism for translocator protein 18 kDa affects both in vitro and in vivo radioligand binding in human brain to this putative biomarker of neuroinflammation. J Cereb Blood Flow Metab 33:53-58. 316. Owen DR, Lewis AJ, Reynolds R, et al. (2011) Variation in binding affinity of the novel anxiolytic XBD173 for the 18 kDa translocator protein in human brain. Synapse 65:257-259. 317. Bruck W (2005) The pathology of multiple sclerosis is the result of focal inflammatory demyelination with axonal damage. Journal of Neurology 252:V3-V9. 318. Stadelmann C, Wegner C, Bruck W (2011) Inflammation, demyelination, and degeneration - recent insights from MS pathology. Biochim Biophys Acta 1812:275-282. 319. Comi G, Filippi M, Barkhof F, et al. (2001) Effect of early interferon treatment on conversion to definite multiple sclerosis: a randomised study. Lancet 357:1576-1582. 320. Comi G, Martinelli V, Rodegher M, et al. (2009) Effect of glatiramer acetate on conversion to clinically definite multiple sclerosis in patients with clinically isolated syndrome (PreCISe study): a randomised, double-blind, placebo-controlled trial. Lancet 374:1503-1511. 321. Miller AE, Wolinsky JS, Kappos L, et al. (2014) Oral teriflunomide for patients with a first clinical episode suggestive of multiple sclerosis (TOPIC): a randomised, double-blind, placebo-controlled, phase 3 trial. Lancet Neurol 13:977-986.

223

322. Kappos L, Polman CH, Freedman MS, et al. (2006) Treatment with interferon beta-1b delays conversion to clinically definite and McDonald MS in patients with clinically isolated syndromes. Neurology 67:1242-1249. 323. Jacobs LD, Beck RW, Simon JH, et al. (2000) Intramuscular interferon beta-1a therapy initiated during a first demyelinating event in multiple sclerosis. CHAMPS Study Group. N Engl J Med 343:898-904. 324. Khan O, Rieckmann P, Boyko A, Selmaj K, Zivadinov R, Group GS (2013) Three times weekly glatiramer acetate in relapsing-remitting multiple sclerosis. Ann Neurol 73:705-713. 325. Polman CH, O'Connor PW, Havrdova E, et al. (2006) A randomized, placebo- controlled trial of natalizumab for relapsing multiple sclerosis. N Engl J Med 354:899- 910. 326. Andersen O (2010) Predicting a window of therapeutic opportunity in multiple sclerosis. Brain 133:1863-1865. 327. Leray E, Yaouanq J, Le Page E, et al. (2010) Evidence for a two-stage disability progression in multiple sclerosis. Brain 133:1900-1913. 328. Bitsch A, Bruck W (2002) Differentiation of multiple sclerosis subtypes: implications for treatment. CNS Drugs 16:405-418. 329. Prosperini L, Gianni C, Barletta V, et al. (2012) Predictors of freedom from disease activity in natalizumab treated-patients with multiple sclerosis. J Neurol Sci 323:104-112. 330. Barkhof F (1999) MRI in multiple sclerosis: correlation with expanded disability status scale (EDSS). Mult Scler 5:283-286. 331. Bakshi R, Thompson AJ, Rocca MA, et al. (2008) MRI in multiple sclerosis: current status and future prospects. Lancet Neurol 7:615-625. 332. Miki Y, Grossman RI, Udupa JK, et al. (1999) Relapsing-remitting multiple sclerosis: longitudinal analysis of MR images--lack of correlation between changes in T2 lesion volume and clinical findings. Radiology 213:395-399. 333. Nijeholt GJ, van Walderveen MA, Castelijns JA, et al. (1998) Brain and spinal cord abnormalities in multiple sclerosis. correlation between MRI parameters, clinical subtypes and symptoms. Brain 121 ( Pt 4):687-697. 334. Vowinckel E, Reutens D, Becher B, et al. (1997) PK11195 binding to the peripheral benzodiazepine receptor as a marker of microglia activation in multiple sclerosis and experimental autoimmune encephalomyelitis. J Neurosci Res 50:345-353. 335. Ratchford JN, Endres CJ, Hammoud DA, et al. (2012) Decreased microglial activation in MS patients treated with glatiramer acetate. J Neurol 259:1199-1205. 336. Gunn RN, Sargent PA, Bench CJ, et al. (1998) Tracer kinetic modeling of the 5- HT1A receptor ligand [carbonyl-11C]WAY-100635 for PET. Neuroimage 8:426-440. 337. Cobo-Calvo A, Bau L, Matas E, et al. (2015) Effectiveness of natalizumab in patients with highly active relapsing remitting multiple sclerosis. Eur Neurol 73:220-229. 338. Yednock TA, Cannon C, Fritz LC, Sanchez-Madrid F, Steinman L, Karin N (1992) Prevention of experimental autoimmune encephalomyelitis by antibodies against alpha 4 beta 1 integrin. Nature 356:63-66. 339. Meyer-Moock S, Feng YS, Maeurer M, Dippel FW, Kohlmann T (2014) Systematic literature review and validity evaluation of the Expanded Disability Status Scale (EDSS) and the Multiple Sclerosis Functional Composite (MSFC) in patients with multiple sclerosis. BMC Neurol 14:58. 340. Prineas JW, Kwon EE, Cho ES, et al. (2001) Immunopathology of secondary- progressive multiple sclerosis. Ann Neurol 50:646-657. 341. Kappos L, O'Connor PW, Polman CH, et al. (2013) Clinical effects of natalizumab on multiple sclerosis appear early in treatment course. J Neurol 260:1388-1395. 342. Henderson AP, Barnett MH, Parratt JD, Prineas JW (2009) Multiple sclerosis: distribution of inflammatory cells in newly forming lesions. Ann Neurol 66:739-753.

224

343. Laruelle M, Slifstein M, Huang Y (2003) Relationships between radiotracer properties and image quality in molecular imaging of the brain with positron emission tomography. Mol Imaging Biol 5:363-375. 344. Law M, Saindane AM, Ge YL, et al. (2004) Microvascular abnormality in relapsing- remitting multiple sclerosis: Perfusion MR imaging findings in normal-appearing white matter. Radiology 231:645-652. 345. Adams CWM, Poston RN, Buk SJ, Sidhu YS, Vipond H (1985) Inflammatory vasculitis in multiple-sclerosis. Journal of the Neurological Sciences 69:269-283. 346. Zoghbi SS, Anderson KB, Jenko KJ, Luckenbaugh DA, Innis RB, Pike VW (2012) On quantitative relationships between drug-like compound lipophilicity and plasma free fraction in monkey and human. J Pharm Sci 101:1028-1039. 347. Tillement JP, Houin G, Zini R, et al. (1983) Binding of drugs to plasma-proteins. Semaine Des Hopitaux 59:1413-1422. 348. Berridge MS (2009) The importance of kinetic enhancement. J Nucl Med 50:1203- 1204. 349. Carter DC, Ho JX (1994) Structure of serum albumin. Adv Protein Chem 45:153- 203. 350. Basken NE, Mathias CJ, Lipka AE, Green MA (2008) Species dependence of [64Cu]Cu-Bis(thiosemicarbazone) radiopharmaceutical binding to serum albumins. Nucl Med Biol 35:281-286. 351. Rowland M (1980) Plasma protein binding and therapeutic drug monitoring. Ther Drug Monit 2:29-37. 352. Wright JD, Boudinot FD, Ujhelyi MR (1996) Measurement and analysis of unbound drug concentrations. Clin Pharmacokinet 30:445-462. 353. Oke BO, Suarez-Quian CA, Riond J, Ferrara P, Papadopoulos V (1992) Cell surface localization of the peripheral-type benzodiazepine receptor (PBR) in adrenal cortex. Mol Cell Endocrinol 87:R1-6. 354. Garnier M, Boujrad N, Oke BO, et al. (1993) Diazepam binding inhibitor is a paracrine/autocrine regulator of Leydig cell proliferation and steroidogenesis: action via peripheral-type benzodiazepine receptor and independent mechanisms. Endocrinology 132:444-458. 355. Smith JA, Das A, Ray SK, Banik NL (2012) Role of pro-inflammatory cytokines released from microglia in neurodegenerative diseases. Brain Res Bull 87:10-20. 356. Endres CJ, Pomper MG, James M, et al. (2009) Initial evaluation of 11C-DPA-713, a novel TSPO PET ligand, in humans. J Nucl Med 50:1276-1282. 357. James ML, Fulton RR, Henderson DJ, et al. (2005) Synthesis and in vivo evaluation of a novel peripheral benzodiazepine receptor PET radioligand. Bioorg Med Chem 13:6188-6194. 358. Selleri S, Bruni F, Costagli C, et al. (2001) 2-Arylpyrazolo[1,5-a]pyrimidin-3-yl acetamides. New potent and selective peripheral benzodiazepine receptor ligands. Bioorg Med Chem 9:2661-2671. 359. Fookes CJR, Pham TQ, Mattner F, et al. (2008) Synthesis and biological evaluation of substituted [(18)F]imidazo[1,2-a]pyridines and [(18)F]pyrazolo[1,5-a]pyrimidines for the study of the peripheral benzodiazepine receptor using positron emission tomography. Journal of Medicinal Chemistry 51:3700-3712. 360. Imaizumi M, Briard E, Zoghbi SS, et al. (2007b) Kinetic evaluation in nonhuman primates of two new PET ligands for peripheral benzodiazepine receptors in brain. Synapse 61:595-605. 361. Briard E, Zoghbi SS, Imaizumi M, et al. (2008) Synthesis and evaluation in monkey of two sensitive 11C-labeled aryloxyanilide ligands for imaging brain peripheral benzodiazepine receptors in vivo. J Med Chem 51:17-30. 362. Briard E, Zoghbi SS, Simeon FG, et al. (2009) Single-step high-yield radiosynthesis and evaluation of a sensitive 18F-labeled ligand for imaging brain peripheral benzodiazepine receptors with PET. J Med Chem 52:688-699.

225

363. Butlen D (1984) Benzodiazepine receptors along the nephron: [3H]PK 11195 binding in rat tubules. FEBS Lett 169:138-142. 364. Minder R, Weder HJ, Bickel MH (1970) Experimental errors resulting from uptake of lipophilic drugs by soft plastic materials. Biochem Pharmacol 19:2179-2180. 365. Forrest IS, Bolt AG, Serra MT (1966) "Applejack" concentration of urinary chlorpromazine metabolites: a widely applicable and mild procedure. Life Sci 5:473-479. 366. Bauer M, Zeitlinger M, Karch R, et al. (2012) Pgp-Mediated Interaction Between (R)-[C-11]Verapamil and Tariquidar at the Human Blood-Brain Barrier: A Comparison With Rat Data. Clinical Pharmacology & Therapeutics 91:227-233. 367. Liu L, Collier AC, Link JM, et al. (2015) Modulation of P-glycoprotein at the Human Blood-Brain Barrier by Quinidine or Rifampin Treatment: A Positron Emission Tomography Imaging Study. Drug Metab Dispos 43:1795-1804. 368. Toyohara J (2016) Importance of P-gp PET imaging in pharmacology. Curr Pharm Des. 369. Su Z, Herholz K, Gerhard A, et al. (2013) [(11)C]-(R)PK11195 tracer kinetics in the brain of glioma patients and a comparison of two referencing approaches. Eur J Nucl Med Mol Imaging 40:1406-1419. 370. Trapp BD, Bo L, Mork S, Chang A (1999) Pathogenesis of tissue injury in MS lesions. J Neuroimmunol 98:49-56. 371. Veenman L, Gavish M (2000) Peripheral-type benzodiazepine receptors: Their implication in brain disease. Drug Development Research 50:355-370. 372. Gerhard A, Watts J, Trender-Gerhard I, et al. (2004) In vivo imaging of microglial activation with [C-11](R)-PK 11195 PET in corticobasal degeneration. Movement Disorders 19:1221-1226. 373. Hashiguchi S, Ogasawara N, Mine H, Kawachi Y (2001) Multiple sclerosis with caudate lesions on MRI. Intern Med 40:358-362. 374. Vercellino M, Masera S, Lorenzatti M, et al. (2009) Demyelination, inflammation, and neurodegeneration in multiple sclerosis deep gray matter. J Neuropathol Exp Neurol 68:489-502. 375. Haider L, Simeonidou C, Steinberger G, et al. (2014) Multiple sclerosis deep grey matter: the relation between demyelination, neurodegeneration, inflammation and iron. J Neurol Neurosurg Psychiatry 85:1386-1395. 376. Yoder KK, Nho K, Risacher SL, Kim S, Shen L, Saykin AJ (2013) Influence of TSPO genotype on 11C-PBR28 standardized uptake values. J Nucl Med 54:1320-1322. 377. Rupprecht R, Rammes G, Eser D, et al. (2009) Translocator protein (18 kD) as target for anxiolytics without benzodiazepine-like side effects. Science 325:490-493. 378. Costa B, Da Pozzo E, Cavallini C, Taliani S, Da Settimo F, Martini C (2016) Long Residence Time at the Neurosteroidogenic 18 kDa Translocator Protein Characterizes the Anxiolytic Ligand XBD173. ACS Chem Neurosci. 379. Confavreux C, Vukusic S (2006) Accumulation of irreversible disability in multiple sclerosis: from epidemiology to treatment. Clin Neurol Neurosurg 108:327-332. 380. Hirst C, Ingram G, Pearson O, Pickersgill T, Scolding N, Robertson N (2008) Contribution of relapses to disability in multiple sclerosis. J Neurol 255:280-287. 381. Hinderling PH, Hartmann D (2005) The pH dependency of the binding of drugs to plasma proteins in man. Ther Drug Monit 27:71-85. 382. Kochansky CJ, McMasters DR, Lu P, et al. (2008) Impact of pH on plasma protein binding in equilibrium dialysis. Molecular Pharmaceutics 5:438-448. 383. Fan Z, Harold D, Pasqualetti G, Williams J, Brooks DJ, Edison P (2015) Can studies of neuroinflammation in a TSPO genetic subgroup (HAB or MAB) be applied to the entire AD cohort? Journal of Nuclear Medicine 56:707-713. 384. Viani A, Rizzo G, Carrai M, Pacifici GM (1992) The effect of ageing on plasma albumin and plasma protein binding of diazepam, salicylic acid and digitoxin in healthy subjects and patients with renal impairment. Br J Clin Pharmacol 33:299-304.

226

385. Abelli M, Chelli B, Costa B, et al. (2010) Reductions in platelet 18-kDa translocator protein density are associated with adult separation anxiety in patients with bipolar disorder. Neuropsychobiology 62:98-103. 386. Chelli B, Pini S, Abelli M, et al. (2008) Platelet 18 kDa translocator protein density is reduced in depressed patients with adult separation anxiety. Eur Neuropsychopharmacol 18:249-254. 387. Nakamura K, Fukunishi I, Nakamoto Y, Iwahashi K, Yoshii M (2002) Peripheral- type benzodiazepine receptors on platelets are correlated with the degrees of anxiety in normal human subjects. Psychopharmacology (Berl) 162:301-303. 388. Gavish M, Laor N, Bidder M, et al. (1996) Altered platelet peripheral-type benzodiazepine receptor in posttraumatic stress disorder. Neuropsychopharmacology 14:181-186. 389. Lockhart A, Davis B, Matthews JC, et al. (2003) The peripheral benzodiazepine receptor ligand PK11195 binds with high affinity to the acute phase reactant alpha1-acid glycoprotein: implications for the use of the ligand as a CNS inflammatory marker. Nucl Med Biol 30:199-206. 390. Libert C, Brouckaert P, Fiers W (1994) Protection by alpha 1-acid glycoprotein against tumor necrosis factor-induced lethality. J Exp Med 180:1571-1575. 391. Huxley VH, Curry FE (1991) Differential actions of albumin and plasma on capillary solute permeability. Am J Physiol 260:H1645-1654. 392. Schnitzer JE, Pinney E (1992) Quantitation of specific binding of orosomucoid to cultured microvascular endothelium: role in capillary permeability. Am J Physiol 263:H48-55. 393. Johnsson E, Haraldsson B (1993) Addition of purified orosomucoid preserves the glomerular permeability for albumin in isolated perfused rat kidneys. Acta Physiol Scand 147:1-8. 394. Pike VW (2009) PET radiotracers: crossing the blood-brain barrier and surviving metabolism. Trends Pharmacol Sci 30:431-440. 395. Rupprecht R, Papadopoulos V, Rammes G, et al. (2010) Translocator protein (18 kDa) (TSPO) as a therapeutic target for neurological and psychiatric disorders. Nat Rev Drug Discov 9:971-988. 396. Selvaraj V, Stocco DM (2015) The changing landscape in translocator protein (TSPO) function. Trends Endocrinol Metab 26:341-348. 397. McEnery MW, Snowman AM, Trifiletti RR, Snyder SH (1992) Isolation of the mitochondrial benzodiazepine receptor: association with the voltage-dependent anion channel and the adenine nucleotide carrier. Proc Natl Acad Sci U S A 89:3170-3174. 398. Joseph-Liauzun E, Farges R, Delmas P, Ferrara P, Loison G (1997) The Mr 18,000 subunit of the peripheral-type benzodiazepine receptor exhibits both benzodiazepine and isoquinoline carboxamide binding sites in the absence of the voltage-dependent anion channel or of the adenine nucleotide carrier. J Biol Chem 272:28102-28106. 399. Scarf AM, Auman KM, Kassiou M (2012) Is there any correlation between binding and functional effects at the translocator protein (TSPO) (18 kDa)? Current Molecular Medicine 12:387-397. 400. Delavoie F, Li H, Hardwick M, et al. (2003) In vivo and in vitro peripheral-type benzodiazepine receptor polymerization: functional significance in drug ligand and cholesterol binding. Biochemistry 42:4506-4519. 401. Noorbakhsh F, Ellestad KK, Maingat F, et al. (2011) Impaired neurosteroid synthesis in multiple sclerosis. Brain 134:2703-2721. 402. Di Paolo G, Kim TW (2011) Linking lipids to Alzheimer's disease: cholesterol and beyond. Nat Rev Neurosci 12:284-296. 403. Orth M, Bellosta S (2012) Cholesterol: its regulation and role in central nervous system disorders. Cholesterol 2012:292598. 404. Mittal M, Siddiqui MR, Tran K, Reddy SP, Malik AB (2014) Reactive oxygen species in inflammation and tissue injury. Antioxid Redox Signal 20:1126-1167.

227

405. Papadopoulos V, Liu J, Culty M (2007) Is there a mitochondrial signaling complex facilitating cholesterol import? Mol Cell Endocrinol 265:59-64. 406. Sakai M, Ferraz-de-Paula V, Pinheiro ML, et al. (2010) Translocator protein (18 kDa) mediates the pro-growth effects of diazepam on Ehrlich tumor cells in vivo. Eur J Pharmacol 626:131-138. 407. Mintun MA, Raichle ME, Kilbourn MR, Wooten GF, Welch MJ (1984) A quantitative model for the in vivo assessment of drug binding sites with positron emission tomography. Ann Neurol 15:217-227. 408. Innis RB, Cunningham VJ, Delforge J, et al. (2007) Consensus nomenclature for in vivo imaging of reversibly binding radioligands. J Cereb Blood Flow Metab 27:1533-1539. 409. Laruelle M, Baldwin RM, Rattner Z, et al. (1994) SPECT quantification of [123I]iomazenil binding to benzodiazepine receptors in nonhuman primates: I. Kinetic modeling of single bolus experiments. J Cereb Blood Flow Metab 14:439-452. 410. Robertson MW, Leslie CA, Bennett JP, Jr. (1991) Apparent synaptic dopamine deficiency induced by withdrawal from chronic cocaine treatment. Brain Res 538:337- 339.

228