A Metabonomics Study of the Modulation of Lipid and Bile Acid Metabolism by the Gut Microbiota and Consequences on Obesity and Fatty Liver Disease

A metabonomics study of the modulation of lipid and bile acid metabolism by the gut microbiota and consequences on obesity and fatty liver disease

Thesis submitted by Magali H. SARAFIAN

Supervisors: Dr. Marc Emmanuel Dumas & Prof. Elaine Holmes

For thesis submitted for the degree of Doctor of Philosophy and Diploma of Imperial College London

Section of Biomolecular Medicine Division of Computational Systems Medicine Department of Surgery and Cancer Imperial College London 2016 ABSTRACT

Obesity and fatty liver disease are characterised by an imbalance between energy intake and energy expenditure. Genetic and environmental factors strongly contribute to impaired energy homoeostasis and can trigger metabolic disorders such as hypertension, hyperglycaemia and hypercholesterolemia which are associated with type 2 diabetes, cardiovascular diseases and metabolic syndrome. The mechanisms underpinning obesity and its co-morbidity are poorly understood but dysregulation of lipid metabolism is likely to contribute and can be studied by metabolic profiling using UPLC-MS. Moreover, the host-gut axis has a strong influence on obesity and fatty liver disease development by modulation of metabolic pathways. The analytical quality of the metabolic profiling methods is critical to understand the aetiopathogenesis of obesity. Hence the aim of this PhD project is to investigate lipid metabolism in obesity and fatty liver disease. I developed and optimised a strategy for characterising the global lipid profile of plasma based on isopropanol precipitation and supplemented this with the development of a targeted assay for obtaining in depth profile of bile acids (BAs) (n=145) since BAs are known to play a specific role in obesity and fatty liver disease. Furthermore, BAs can provide insight into the role of the gut microbiota in obesity as they are metabolised by the gut microbiota. The analytical pipeline for lipid and BA analysis are subsequently applied for two human clinical studies. i) The metabolic signatures associated with subcutaneous and visceral obesity were investigated. A significant increase of LPC (16:0), unconjugated BAs and sulphated BAs with a decrease of PC (16:0/20:3), taurine conjugates was observed in visceral obesity compared to subcutaneous obesity. ii) Subtle disease

1 progression in fatty liver disease, NAFLD and NASH was evaluated. A signiﬁcant increase of triacylglycerols, hyocholic acid and tauro-conjugated BAs with a decrease of phosphocholines were observed in NASH compared to NAFLD. This phD project illustrated that metabonomic instrumentations (UPLC-MS and UPLC-MS/MS) are non-invasive and powerful analytical techniques to diagnose obesity and fatty liver disease. In addition, metabonomic analysis of urine and blood samples may oﬀer the possibility to optimise individual diagnosis and management of patients with obesity and fatty liver disease.

2 STATEMENT OF ORIGINALITY

I certify that this thesis, and the research to which it refers, are the product of my own work, and that any ideas or quotations from the work of other people, published or otherwise, are fully acknowledged in accordance with the standard referencing prac- tices of the discipline.

I would like to mention the contributions of Mathieu Gaudin and Matthew Lewis for the BA targeted assay presented in chapter 3. The design of the BA sulfation reaction and optimisation was developed with the help of Dr Mathieu Gaudin and BA UPLC method was optimised with the help of Dr Matthew Lewis.

Magali H. SARAFIAN, January 2016

3 COPYRIGHT DECLARATION

The copyright of this thesis rests with the author and is made available under a Creative Commons Attribution Non-Commercial No Derivatives licence. Researchers are free to copy, distribute or transmit the thesis on the condition that they attribute it, that they do not use it for commercial purposes and that they do not alter, transform or build upon it. For any reuse or redistribution, researchers must make clear to others the licence terms of this work.

4 ACKNOWLEDGEMENTS

I would like to thank my supervisors Dr Marc E. Dumas and Prof Elaine Holmes for your guidance, patience and trust. You were constant support throughout my ups and downs during my PhD thesis and have been inspirational mentors and teachers. I would like also to thank the head of Surgery and Cancer; Prof. Jeremy Nicholson who contributed to the success of my PhD projects.

I acknowledge the support of Nestl´e for funding this PhD project. Thank you to Dr Francois-Pierre Martin, Dr Serge Rezzi, Dr Sebastiano Collino and Dr Sunil Kochhar for their advice on projects and interpretation of data analysis.

I am especially grateful to Dr Matthew Lewis, a research collaborator and MS mentor who has become a great friend and a constant source of positivity. I would like to thanks also your team at the NPC, you are amazing source of knowledge and fun.

I would like thank the people that I worked with in Computational and System Medicine at Imperial College, I had great time working with you, Mathieu, Julien, Laura, Ana, Andrea, Badrin, and others that I cannot mention as it would be too long. I would like also to thank my friends from Norwich, place where my interest for metabonomics started, Dr Gwenaelle Le Gall, Mark Philo, Ian Colquhoun, Rob and Ruth.

Special mention to the ”London Family” and especially to Lea, a fantastic friend who has also helped me through many moments of panic! Including Anita, Panos, Claire, Abby, Gaby, Kathy, Alex F., Evie, Alex P., Dina, Harris, Vassilli, Michael, Maria, Nandish, Anthony, Ebony, Paul and Florian. I would like to thank my old friends from Casteljaloux, Bordeaux and Toulouse.

I would like to dedicate my thesis to my family, Mum, Dad, my twin Romain, Ana¨ıs, Jean-Maurice, Marie-Aude, Adela¨ıde, Cyril, Tha¨ıs, Stephane, Nadine, Maxime, Hélène, Frédéric, Sophie and Bastien. I will not forget my nieces and nephews, Elisa, Colin, Olivia, Maud and Alexis, what a great little team taking on the world and making it better! And my favourite little friends, Toulouse, Violette, Winnie and my dear Undavril. Especially thank you to my parents, Catherine and Jean who believed in me and gave me the opportunity to have this incredible life in England. Thank you Dad you always have been a source of inspiration.

Dear Julien thank you for being there through it all you are a constant source of support. I really couldn’t have done it without you!

5 TABLE OF CONTENTS

Abstract ...... 1 Statement of originality ...... 3 Copyright declaration ...... 4 Acknowledgements ...... 5 Table of contents ...... 6 List of ﬁgures ...... 11 List of tables ...... 13

1 Introduction 18 1.1 Obesity, non-alcoholic fatty liver disease and metabolic syndrome . . 18 1.1.1 Epidemiology ...... 18 1.1.2 Obesity to NAFLD disease and metabolic syndrome . . . . . 20 (a) Aetiology ...... 20 (b) Pathogenesis ...... 20 (c) Clinical diagnosis ...... 22 1.2 Classiﬁcation and biochemistry of lipids based on Lipidmaps . . . . . 24 1.2.1 Fatty acyls ...... 25 1.2.2 Glycerolipids ...... 26 1.2.3 Glycerophospholipids ...... 26 1.2.4 Sphingolipids ...... 27 1.2.5 Sterol lipids ...... 28 1.2.6 Other lipid classes: prenols, sacccharolipids and polyketides . 31 1.3 Lipid metabolism ...... 32 1.3.1 Lipids and lipoprotein metabolism ...... 32 1.3.2 Role of the gut microbiota and lipid metabolism ...... 33 1.3.3 BA metabolism and the gut microbiota ...... 34 1.4 Metabonomics ...... 41 1.5 Thesis outlines and aims ...... 42

2 The metabonomic approach 44 2.1 Sample preparation ...... 45 2.1.1 Sample type ...... 45 2.1.2 Storage ...... 46 2.1.3 Sample preparation: Lipid extractions ...... 46 2.2 UPLC-MS analytical techniques ...... 48 2.2.1 LC ...... 48 (a) Theory of chromatography ...... 48 (b) Column chemistry in LC ...... 49 (c) Mobile phase in LC ...... 50 2.2.2 Mass spectrometry ...... 50

6 (a) Ionisation ...... 50 (b) Analysers ...... 51 (c) Detector ...... 53 2.3 The metabonomic run strategy ...... 54 2.4 Data pre-processing ...... 55 2.4.1 Peak detection ...... 55 2.4.2 Peak alignment ...... 56 2.4.3 Peak ﬁltering ...... 56 2.4.4 Normalisation ...... 56 2.5 Data analysis ...... 57 2.5.1 Chemometrics ...... 57 2.5.2 Structural identiﬁcation ...... 58

3 An Objective Set of Criteria for Optimisation of Sample Prepara- tion Procedures for Ultra-High Throughput Untargeted Blood Plasma Lipid Profiling by UPLC-MS 60 3.1 Introduction ...... 60 3.2 Material and methods ...... 63 3.2.1 Materials ...... 63 3.2.2 Sample preparation for UPLC-MS: Precipitation ...... 63 3.2.3 Sample preparation for UPLC-MS: Extraction ...... 63 3.2.4 Protein quantification ...... 65 3.2.5 Ultra Performance Liquid Chromatography ...... 66 3.2.6 Lipid profiling by UPLC-Q-ToF Mass Spectrometry ...... 66 3.2.7 Structural identification ...... 67 3.2.8 MS data pre-processing ...... 67 3.2.9 Multivariate statistical analysis ...... 67 3.2.10 Univariate statistical analysis ...... 68 3.2.11 Lipid recovery ...... 68 3.3 Results ...... 69 3.3.1 Simplicity of protocols ...... 69 3.3.2 Protein removal efficiency ...... 71 3.3.3 Lipid coverage ...... 72 3.3.4 Repeatability ...... 77 3.3.5 Lipid recovery ...... 78 3.4 Discussion ...... 82 3.5 Conclusion ...... 84

4 UPLC-MS/MS targeted method for bile acids 85 4.1 Introduction ...... 85 4.2 Materials and methods ...... 89 4.2.1 Materials ...... 89 4.2.2 Collection of human plasma for targeted analysis ...... 89 4.2.3 Sample preparation for targeted analysis of human pre-/ post- prandial study ...... 89 4.2.4 UPLC-MS proﬁling and MS/MS conditions ...... 90 4.2.5 Optimisation of BA sulfation ...... 95 4.2.6 Puriﬁcation of sulfated BAs ...... 95 4.2.7 Optimisation of MRM transitions and SIR ...... 95

7 4.2.8 Method validation ...... 96 4.2.9 MS data pre-processing ...... 97 4.2.10 Multivariate Statistical Analysis ...... 97 4.3 Results ...... 98 4.3.1 Bile acid sulfation and purification ...... 98 4.3.2 Targeted MS/MS analysis of 145 BAs ...... 100 4.3.3 Method validation of the BA targeted UPLC-MS/MS method 101 (a) Intra- and Inter-day precision and accuracy . . . . . 101 (b) BA recoveries in human serum, plasma and urine . . 104 4.3.4 Application of the targeted UPLC/MS-MS methodology to a human fed/fasted study showing post-prandial differences in BA quantification and detection ...... 107 4.4 Discussion ...... 110 4.4.1 Limits of targeted UPLC-MS/MS method regarding the BA diversity ...... 110 4.4.2 BA sulfation ...... 110 4.4.3 Variations in BA solubility ...... 111 4.4.4 Robustness of the targeted UPLC-MS/MS BA method . . . . 111 4.4.5 BA quantification in human blood ...... 112 4.5 Conclusion ...... 113

5 The post-prandial response is influenced by increased of visceral fat and disturbed lipid-bile acid metabolism 114 5.1 Introduction ...... 114 5.2 Materials and methods ...... 117 5.2.1 Materials ...... 117 5.2.2 Participants and experimental design ...... 117 5.2.3 Sample preparation ...... 118 5.2.4 Lipid profiling. Ultra-Performance Liquid Chromatography. . . 118 5.2.5 Lipid profiling. Quadrupole-Time-of-Flight Mass Spectrometry 118 5.2.6 BA targeted method. Ultra-Performance Liquid Chromatog- raphy...... 118 5.2.7 BA targeted method. Triple Quadrupole Mass Spectrometry. 118 5.2.8 MS data preprocessing...... 118 5.2.9 Structural identification...... 119 5.2.10 Univariate and multivariate statistical analysis...... 119 5.3 Results ...... 120 5.3.1 Clinical biochemistry analysis highlighted changes in visceral fat compared to subcutaneous fat ...... 120 5.3.2 Lipid metabolism modifications correlate with increased visceral obesity ...... 122 5.3.3 BA metabolism modifications correlate with increased visceral obesity ...... 124 5.4 Discussion ...... 129 5.4.1 Visceral obesity is associated with impaired insulin levels and insulin resistance ...... 129 5.4.2 Visceral obesity is associated with variations of phosphocholine levels ...... 129

8 5.4.3 Circulating BA are modiﬁed in visceral obesity ...... 130 5.4.4 Key outcomes and drawbacks ...... 131 5.5 Conclusion ...... 133

6 Metabonomic analysis of lipids and bile acids circulating in NAFLD and NASH 134 6.1 Introduction ...... 134 6.2 Materials and methods ...... 136 6.2.1 Materials ...... 136 6.2.2 Human plasma samples ...... 136 6.2.3 Sample preparation ...... 136 6.2.4 Lipid profiling. Ultra-Performance Liquid Chromatography . . 137 6.2.5 Lipid profiling. Quadrupole-Time-of-Flight Mass Spectrometry 137 6.2.6 BA targeted assay. Ultra-Performance Liquid Chromatography 137 6.2.7 BA targeted assay. Triple Quadrupole Mass Spectrometry . . 137 6.2.8 MS Data Preprocessing ...... 137 6.2.9 Drift correction ...... 138 6.2.10 Univariate and multivariate statistical analysis ...... 138 6.2.11 Lipid structural identification ...... 138 6.3 Results ...... 139 6.3.1 Changes observed between plasma of NAFLD vs. control participants ...... 139 6.3.2 Lipid metabolism modifications that correlates with NAFLD and NASH ...... 142 (a) UPLC-MS lipid profiling ...... 142 (b) OPLS and OPLS-DA models demonstrated systemic differences in lipid profiles between controls, NAFLD and NASH profiles ...... 143 (c) Identification of potential lipid biomarkers of NAFLD and NASH ...... 147 6.3.3 BA metabolism impaired in NAFLD patients ...... 152 (a) Targeted BA assay ...... 152 (b) Increase in circulating BA observed in NAFLD and NASH ...... 152 (c) Identification of potential BA biomarkers ...... 156 6.4 Discussion ...... 159 6.4.1 NAFLD and NASH diagnosis by metabonomics ...... 159 6.4.2 Blood clinical data associated with NAFLD and NASH . . . . 160 6.4.3 Limitations of the predictive lipidomic model for NAFLD . . . 161 6.4.4 Fatty Liver Disease-associated with lipidomic variations . . . 161 6.4.5 Fatty Liver Disease-associated with variations in circulating BApool ...... 162 (a) Impaired BA metabolism related to the host function 163 (b) Impaired BA metabolism that are related to the gut microbiota activity ...... 164 6.5 Conclusion ...... 166

9 7 General discussion 167 7.1 Key results criteria established for the analytical pipelines for lipid and BA measurements ...... 168 7.1.1 Isopropanol precipitation and its suitability for high - throughput UPLC/MS lipid profiling analysis ...... 168 7.1.2 Targeted UPLC-MS/MS methodology for the detection and quantification of 145 BAs ...... 169 7.2 Lipid and BA application of profiles to obesity and liver disease in human population ...... 171 7.2.1 Key results ...... 171 (a) Contribution of lipid and BA metabolism dysfunction to ectopic fat deposition in obesity ...... 171 7.2.2 Contribution of lipid and BA metabolism dysfunction to NAFLD and NASH ...... 172 7.2.3 Overall coherence of the markers between application studies 172 7.3 New findings in obesity and fatty liver disease ...... 173 7.4 Strength and weakness of the experiments ...... 174 7.5 Preliminary assessment of FXR activation by the 57 BAs quantified by the targeted assay ...... 176 7.6 Future works and perspectives ...... 178

I Appendix:Copyrights 209

II Appendix:Published articles 212

III Appendix:Bile acid standards 216

10 List of Figures

1.1 Cellular events leading to obesity and NAFLD-NASH ...... 22 1.2 Lipidmaps database ...... 24 1.3 Lipid classes ...... 28 1.4 Cholesterol metabolism ...... 29 1.5 Bile acid enterohepatic circulation ...... 30 1.6 BA signalling and regulation of cardiometabolic risk factors . . . . . 40 1.7 Thesis workﬂow ...... 43

2.1 Metabonomic approach and experimental design ...... 44 2.2 Example of mass spectrometry principle with Q-ToF ...... 52 2.3 Run strategy in metabonomics for proﬁling ...... 54 2.4 Lipid structural identiﬁcation ...... 59

3.1 Sample preparation steps ...... 65 3.2 TG stability for the four sample preparations by precipitation . . . . . 70 3.3 Lipid proﬁling chromatograms of the eight sample preparations . . . 73 3.4 PCA analysis and loading plots of the eight sample preparations . . . 75 3.5 Extraction eﬃciency of the eight sample preparations ...... 76 3.6 Repeatability of the eight methods ...... 78 3.7 Lipid recoveries for the eight sample preparations ...... 81

4.1 Sulfation scheme ...... 98 4.2 Optimisation of BA sulfation ...... 99 4.3 Chromatograms of 145 BAs ...... 102 4.4 Accuracy distribution of QCs ...... 103 4.5 Recoveries of the 16 deuterated standards ...... 106 4.6 PCA score plots of pre-/post-prandial human plasma, serum and urine 108

5.1 Design of experiment for PPLR ...... 117 5.2 Lipid markers subcutaneous vs. visceral obesity ...... 123 5.3 OPLS-DA scores plot of BAs on PPLR ...... 126 5.4 OPLS-DA scores plot of BAs on PPLR by time points ...... 126 5.5 PPLR curve of signiﬁcant BAs ...... 128

6.1 Correlation structure between physiological variables ...... 141 6.2 Lipid profiling chromatograms of controls , NAFLD and NASH . . . . 142 6.3 Drift correction of run order effect occurring in UPLC-MS lipid profiling ...... 143 6.4 One OPLS and four OPLS-DA score plot of control, NAFLD and NASH positive mode lipid profiles ...... 145

11 6.5 Predictive OPLS-DA score plot of control participants vs. NASH . . 146 6.6 Correlations between control participants, NAFLD and NASH . . . . 148 6.7 Correlations control vs. NAFLD and vs. NASH ...... 149 6.8 Correlations between NAFLD and NASH ...... 151 6.9 Quantities of BAs species circulating in control participants, NAFLD and NASH patients ...... 154 6.10 Correlations for BA targeted method ...... 158 6.11 Synopsis of the results ...... 165

7.1 Top ten BAs able to activate FXR ...... 177 7.2 Ten BAs out of the 57 activated FXR overexpressed in CHO cells . . 178

12 List of Tables

1.1 Deconjugation performed on primary bile acids, cholic acid and chenodeoxycholic acid by gut microbiota ...... 36 1.2 7α-Dehydroxylation performed on primary bile acids, cholic acid and chenodeoxycholic acid by gut microbiota ...... 37 1.3 Epimerization performed on primary bile acids, cholic acid and chenodeoxycholic acid by gut microbiota ...... 37 1.4 Oxidation performed on primary bile acids, cholic acid and chenodeoxycholic acid by gut microbiota ...... 38

2.1 Scan experiments using MS and MS/MS ...... 53 2.2 Settings setup for pre-processing steps ...... 57

3.1 Chromatographic gradient used for UPLC-MS lipid proﬁling . . . . . 66 3.2 Protein precipitations eﬃciency for the eight sample preparation . . . 72 3.3 Distributions of the eight sample preparations ...... 77 3.4 Repeatability of the eight methods ...... 78 3.5 List of lipid standards used for the recovery study ...... 79 3.6 Lipid recoveries for the eight sample preparations ...... 80 3.7 Synopsis of the criteria evaluated for the eight sample preparation methods ...... 83 3.8 Evaluation criteria for the eight sample preparations ...... 83

4.1 Chromatographic gradient of BA profiling and targeted analysis . . . 91 4.2 UPLC-MS/MS settings for the 57 BA standards ...... 92 4.3 UPLC-MS/MS settings for sulfated BA standards ...... 94 4.4 Yield of the 16 sulfation reactions ...... 99 4.5 Intra- and inter-day validation of accuracy and precision ...... 104 4.6 Matrix effect in BA UPLC-MS/MS method...... 105 4.7 Carry-over in BA UPLC-MS/MS method ...... 105 4.8 Chromatographic gradient used for UPLC-MS lipid profiling . . . . . 109

5.1 Clinical biochemistry data from 40 participants with subcutaneous and visceral obesity ...... 121 5.2 OPLS-DA analysis of subcutaneous obesity (Q1) and visceral obesity (Q4) lipid profiles ...... 122 5.3 Quantified BA in subcutaneous and visceral obesity ...... 125 5.4 OPLS-DA analysis of quantified BA for subcutaneous obesity (Q1) to visceral obesity (Q4) ...... 126 5.5 Student’s t-test of quantified BA to compare Q1 and Q4 ...... 127

13 6.1 Summary of blood clinical biochemistry data ...... 140 6.2 Evaluation of one OPLS model and four OPLS-DA models ...... 144 6.3 Lipids identiﬁcation of correlations between NAFLD vs. NASH patients ...... 151 6.4 Means of 57 BA species quantiﬁed...... 155

14 LIST OF ABBREVIATIONS

ACN Acetonitrile ALT Alanine amino transaminase AST Aspartate amino transaminase BA Bile acid BEH Ethylene bridge hybrid BMI Body mass index CA Cholic acid CDCA Chenodeoxycholic acid CE Cholesteryl esters Cer Ceramides CSH Charged surface hybrid CT Computed tomography CV Coeﬃcient of variation CYP3A4 Cytochrome P450 3A4 DDA Data-dependent acquisition DCA Deoxycholic acid DG Diacylglycerols EGFR Epidermal growth factor receptor ESI Electrospray ionisation FA Fatty acids FFAs Free FAs FDR False discovery rate FGF Fibroblast growth factor FXR Farnesoid X receptor GC Gas chromatography GCA Glycocholic acid GCDCA Glycochenodeoxycholic acid GDCA Glycodeoxycholic acid GGT γ-glutamyl transferase GHCA Glycohyocholic acid GHDCA Glycohyodeoxycholic acid GLCA Glycolithocholic acid GLP-1 Glucagon-like-peptide-1 GR Glucocorticoid receptor GS Glucagon synthase GUDCA Glycoursodeoxycholic acid HbA1c Glycated hemoglobin HCA Hyocholic acid HDL High density lipoproteins HNF4α Hepatocyte nuclear factor 4 α HOMA-IR Homeostatic model assessment of insulin resistance HsCRP High-sensitivity C-reactive protein IDL Intermediate density lipoprotein IPA Isopropanol LCA Lithocholic acid

15 LDL Low density lipoprotein LLOQ Lower limit of quantiﬁcation LOD Limit of detection LPC Lysophosphocholines LPE Lysophosphoethanolamines MAP Mean arterial pressure MG Monoacylglycerol MRM Multiple reaction monitoring MS/MS Tandem mass spectrometry MTBE Methyl-tert butyl ether MuroCA Murocholic acid NAFLD Non-alcoholic fatty liver diseases NAS NAFLD activity score NASH Non-alcoholic steatohepatitis NMR Nuclear resonance magnetic OPLS-DA Orthogonal partial least square discriminant analysis PA Phosphatidic acid PC Phosphatidylcholines PCA Principal components analysis PE Phosphatidylethanolamines PG Phosphatidylglycerols PI Phosphatidylinositols PPARγ Peroxisome proliferator-activated receptor γ PPGR Postprandial glycemic response PPLR Postprandial lipemic response PS Phosphatidylserines PXR Pregnane X receptor Q Quadrupole QC Quality control ROS Reactive oxygen species RT Retention time S Sulfur SCFA Short-chain fatty acids SIM Selected ion monitoring SM Sphingomyelins SRM Selected reaction monitoring SREBP-1c Sterol regulatory element-binding protein 1 TCA cycle Tricarboxylic pathways TCA Taurocholic acid TCDCA Taurochenodeoxycholic acid TDCA Taurodeoxycholic acid TG Triacylglycerols THCA Taurohyocholic acid THDCA Taurohyodeoxycholic acid TαMCA Tauro-α-muricholic acid TβMCA Tauro-β-muricholic acid TωMCA Tauro-ω-muricholic acid TNFα Tumor necrosis factor

16 TOF Time-of-ﬂight TUDCA Tauroursodeoxycholic acid UDCA Ursodeoxycholic acid ULOQ Upper limit of quantiﬁcation UPLC-MS Ultra-performance liquid chromatography-mass spectrometry VLDL Very low density lipoprotein

17 Chapter 1

Introduction

In this thesis, scientiﬁc knowledge of metabolic disorders resulting from obesity and liver disease was expanded upon by using newly improved and developed analytical techniques: ultra-liquid performance chromatography MS analysis (UPLC-MS) lipid proﬁling and UPLC-MS/MS targeted bile acid (BA) assays.

1.1 Obesity, non-alcoholic fatty liver disease and

metabolic syndrome

Overweight and obesity are defined as excessive fat deposition in adipose tissue and non-alcoholic fatty liver disease (NAFLD) is defined as excessive fat deposition in liver tissue. Both obesity and NAFLD coexist with chronic inflammation. Diagnosis of obesity and NAFLD is recognised as an important indicator of chronic metabolic disorders such as hyperglycemia, hypertriglyceridemia, dylipidemia and hypertension. The presence of at least three of these complications are considered as metabolic syndrome, which ultimately can result in type 2 diabetes and cardiovascular diseases (Alberti et al. 2005).

1.1.1 Epidemiology

Obesity is a complex disease, which is multi-factorial and results mostly from interactions among genetic, behavioural and modern environmental factors. The consequences of obesity are a lack of balance in energy homoeostasis between energy

18 Chapter 1-Introduction storage and expenditure which causes metabolic disorders. Obesity is currently a significant public health problem and is one of the most common chronic conditions in Europe, with approximately 30-70% of the population overweight and approximately 10-30% obese (WHO 2015). According to the World Health Organisation (WHO) there has been an alarming increase in the number of overweight adults: from 1.6 billion in 2005 to approximately 2.3 billion in 2015. Of those adults who are overweight, between 400 million (2005) and approximately 700 million (2015) are obese (WHO 2009; Ng et al. 2014). In addition, overweight children represent 20 million infants under the age of 5 (WHO 2009). Obesity can co-occur with cardiovascular diseases, type 2 diabetes and NAFLD, which contribute to increase overall morbidity and mortality with high rates of disease and death seen in older adult patients and the elderly (Dulloo et al. 2010). Obesity and related complications such as liver disorders (i.e. NAFLD) have become especially problematic in developed countries and represent significant healthcare expenditure of the total national health costs (2-7% of budget) (Kopelman 2000). The progression of NAFLD is not only related to obesity, and was estimated to be present in 60% of a lean Asian population, highlighting the association with other risks factors such as type 2 diabetes (Perry et al. 2014). NAFLD is present in 30-90% of obese and 40-80% of type 2 diabetic populations in the US and Europe (Rinella 2015). Both obesity and NAFLD increase the incidence of metabolic syndrome, with 50% of obese and 22% of NAFLD people in the US affected by metabolic syndrome (Adams et al. 2005; Eckel et al. 2005; Pollex et al. 2006). Molecular mechanisms underpinning interactions among obesity, NAFLD and metabolic syndrome remain to be fully explored. The scope of this thesis aims to take advantage of recent develop- ments in high-throughput analytical techniques to offer a comprehensive overview of these mechanisms by using a metabonomic approach.

19 Chapter 1-Introduction

1.1.2 Obesity to NAFLD disease and metabolic syndrome

(a) Aetiology

Development of obesity is associated with genetic and environmental factors. Twin studies clearly highlighted and confirmed the importance of heritability of obesity in humans (Bouchard et al. 1990). The last few decades have seen a rapid increase in the incidence of obesity, mainly explained by environmental rather than genetic factors (Barsh et al. 2000). Prevalence of obesity has been drastically increased by a deleterious western lifestyle (mainly high caloric diet and physical inactivity) (WHO 2009). Altogether, weight gain and metabolic disorders are associated with ”obesogenic” environmental factors such as a high carbohydrate, high fat diet and a sedentary lifestyle. These factors can promote an imbalance in energy homeostasis and in the presence of other metabolic diseases such as NAFLD can lead to metabolic syndrome. Similarly, NAFLD is of increasing concern to those in the healthcare sector, as it is now the most common liver disorder in developed countries. Prevalence of NAFLD varies between developed and non-developed countries, again reflecting the importance of lifestyle in human health (Browning et al. 2004b). However, NAFLD has also been shown to comprise a substantial heritable component (Anstee et al. 2013b). Multiple genes are known to contribute to the development of metabolic syndrome and are associated with risk factors such as dyslipidemia (FABP2, APOC3, LMNA, AGT and IL6 genes), insulin resistance (IL6 and LTA genes) and hypertension (NOS3, GNB3 and AGT genes) (Alberti et al. 2005; Pollex et al. 2006). Despite these findings, metabolic syndrome is a concept that remains difficult to define due to multiplicity of phenotypes and is not diagnosed based on genetic markers.

(b) Pathogenesis

Obesity is accompanied by adverse physiological functions and by accumulation of stored body fat, which leads to alterations in adipose tissue structure and deposi-

20 Chapter 1-Introduction tion. Healthy adipose tissue is composed of adipocytes and various cells contributing to its growth and function including: pre-adipocytes; fibroblasts that provide an extracellular matrix component; vascular cells; macrophages that coordinate the inflammatory response to eliminate apoptotic cells, restore tissue function and repair processes. Cellular events considered to be essential for obesity development include adipocyte hypertrophy, overproduction of extracellular matrix, increased angiogenesis and macrophage infiltration (Figure 1.1.A) (Cinti et al. 2005; Corvera et al. 2014; Weisberg et al. 2003). Upper body obesity can be characterised by an accumulation of subcutaneous fat and/or visceral fat. The subcutaneous fat is located directly under the skin (e.g. abdominal). The visceral fat is located around internal organs such as the stomach, intestines, liver and kidneys. Both are erroneous fat depositions contribute severely to lipotoxicity, organ dysfunctions, hypertension, elevated insulin and insulin resistance (Kopelman 2000; Després et al. 2006; Unger et al. 2010). In obesity, adipocytes dysfunction triggers insulin resistance, by increasing the release of various critical modulators such as non-esterified fatty acids (FAs), glycerols, hormones (e.g. leptin and adiponectin) and pro-inflammatory cytokines. By contrast, deposition of fat in the lower body has been demonstrated to maintain a good balance between energy intake and expenditure, and to protect against ectopic fat deposition (Tran et al. 2008). Insulin resistance has been demonstrated to be a link between obesity and NAFLD (Petersen et al. 2006). In NAFLD, insulin resistance is triggered through hepatocyte dysfunction, which increases hepatic diacyglycerols (DGs) and activates pathways to reduce insulin sensitivity with impaired glucose tolerance (Petersen et al. 2006). Some NAFLD patients can develop important histological changes such as steatosis, lobular inflammation, hepatocellular ballooning and fibrosis that can lead to irreversible non-alcoholic steatohepatitis (NASH) (Figure 1.1.B). NASH may lead ultimately to advanced fibrosis, cirrhosis and hepatocellular carcinoma (Anstee et al. 2013a; Yu et al. 2013).

21 Chapter 1-Introduction

Obese populations have an increased likelihood of developing dyslipidemia, arterial hypertension, impaired glucose tolerance and insulin resistance (Phillips et al. 2008; Unger et al. 2010). These risk factors are also interrelated to diseases associated with obesity such as cardiovascular disease and type 2 diabetes, and increased incidence of cancer, respiratory problems and osteoarthritis. Diagnosis of at least three of these risk factors can predict metabolic syndrome (Alberti et al. 2005). Taken together, obesity and NAFLD are part of a constellation of serious complications of metabolic syndrome including insulin resistance and cardiometabolic diseases (Hertle et al. 2014; Grundy 2008; Dietrich et al. 2014). Nowadays, it is critical and challenging to understand the molecular mechanisms behind these major pathophysiological processes.

Figure 1.1: Cellular events leading to obesity (A) and to NAFLD and NASH linked by insulin resistance (B).

Obesity is defined as an excess of FAs deposited in adipose tissue and is often characterised in clinical practice by the body mass index (BMI). BMI is calculated by dividing the weight (in kilograms) of an individual by their height (in metres) squared, and patients are classified as being overweight (BMI ≥ 25), obese (BMI ≥ 30) or morbidly obese (BMI ≥ 40) (Seidell et al. 1997). BMI measurements have some limitations and cannot discriminate fat mass from lean mass (e.g. a very muscular person will be misclassified). Additional anthropometric measurements (e.g. waist circumference, waist-to-hip ratio and computed tomography CT scan to measure upper body fat deposition) are required to better describes the obesity and distribution of fat. Altogether, they offer a good first indicator of health risk and

22 Chapter 1-Introduction likelihood of developing diseases associated with obesity (Kopelman 2000). Identification of disease linked to metabolic syndrome can be complex and more evaluation tests are required. For example, NAFLD is asymptomatic in most patients and more tests are required for its diagnosis. Clinical biochemistry tests on blood and imaging studies are the first approaches to evaluate the severity of tissue damage in liver disease. NAFLD patients are diagnosed with abnormal levels (normal <30-40 U/L) of aspartate amino transaminase (ASP) and alanine amino transaminase (ALT). In addition, elevation of the ASP/AST ratio (>1) indicates progression of liver disease to cirrhosis (Angulo et al. 1999). However, these tests might be insensitive for the diagnosis of NAFLD in patients as measures can be normal for NAFLD (Fracanzani et al. 2008; Mofrad et al. 2003). The enzyme gamma-glutamyltransferase (GGT) is another liver marker used to diagnose NAFLD (normal <50 U/L). Although liver biopsy is an invasive technique, it is the only method currently available for accurate diagnosis of NAFLD according to NAFLD score (NAS). This score is calculated from histopathological evaluation of the degree of fibrosis (0,1a,1b,1c-3), lobular inflammation (0 to 3), ballooning (0 to 2) and steatosis (1 to 3), and allowing differentiation of NAFLD/simple steatosis to NASH (1 to 3) (Torres et al. 2008). Additional tests can be implemented to exclude other risk factors, such as diabetes (glucose, normal <100 mg/dL, insulin, normal <5-20 mU/L and glycated haemoglobin HbA1c, normal <4- 5.9 %), dyslipidemia (cholesterol, normal <120-180 mg/dL, high-density lipoprotein HDL normal <40-60 mg/dL, low-density lipoprotein LDL normal <100-129 mg/dL and triglycerides, normal <150 mg/dL) and hypertension (high-sensitivity c-reactive protein Hs.CRP, normal <3 mg/L) (Yun et al. 2009). There is an urgent need for new straightforward and less invasive approaches to diagnose and predict metabolic syndrome-associated diseases such as obesity and NAFLD. Studies of lipid diversity and lipid function have identified essential features of metabolic syndrome development, and lipids represent a good target for diagnostics as they directly impact on energy homeostasis.

23 Chapter 1-Introduction

1.2 Classiﬁcation and biochemistry of lipids based

on Lipidmaps

Lipids are a broad group of naturally occurring organic compounds composed of FAs or derivatives. Lipids are hydrophobic and amphiphilic biomolecules, insoluble in water and soluble in organic solvents, regulating energy storage (Beller et al. 2010), structure of cell membranes and physiological responses through cell signalling: all these processes have an important impact in disease development (Wymann et al. 2008; Fletcher 2013). Altogether, identification and quantification of lipid pathways and profiles are studied Discoveries of new lipid species implicated in disease progression strongly contribute to the extension of the lipidomics field, best exemplified by the efforts made by the American Lipidmaps gateway project funded by the NIH (Fahy et al. 2009). Lipidomic research aim to identify and quantify lipids in tissue and biofluids profiles to understand metabolic disorders such as obesity and NAFLD. A comprehensive classification of lipids was achieved due to the wide range of structural diversity found within these molecules. Lipids were divided into eight main categories (then class and subclass) according to the biosynthetic pathways fatty acyls, glycerolipids, glycerophospholipids, sphingolipids, sterol lipids, prenol lipids, saccharolipids and polyketides. Currently, a total of 40,300 known lipid species are referenced in the Lipidmaps database (Figure 1.2).

Figure 1.2: Lipidmaps database has 40,360 lipid species referenced which belong to eight distinct categories.

24 Chapter 1-Introduction

1.2.1 Fatty acyls

Fatty acyls are synthesised from chain elongation of their precursor acetyl CoA. Fatty acyls are composed of aliphatic tails of carbons which are saturated (e.g. palmitic acid) or unsaturated (i.e. double bonds, e.g. oleic acid) and engaged in ester or amide bonds to form complex lipids (Figure 1.3A). Nomenclature of fatty acyls is based on the length of the aliphatic tail, and the number and position of double bonds (unsaturated). According to the chain length we can differentiate four types of fatty acyls; short chain fatty acyls (>C6), medium chain fatty acyls (>C6-12), long chain fatty acyls (>C13-21) and very long chain fatty acyls (>C22). An abbreviated nomenclature was adopted to simplify the description of the fatty acyl structure and has been used in this thesis. Fatty acyls are written as follows; Cx:y where x represents the number of carbons and y represents the number of double bonds. The symbol ”∆-” or ”n-” or ”ω-” can be used to give details on the unsaturation position and (Z)(zusammen=”together”)/cis- or (E)(entgegen=”apart”)/trans- to give information on the fatty acyl conformation. FAs are fatty acyls containing a carboxylic group which are metabolised by the host and the gut microbiota. FAs can derived from breakdown of dietary triacylglycerols (TGs) by lingual, pancreatic and gastric lipases in adipose tissue. Furthermore, the fermentation of dietary fibres (carbohydrates) by intestinal bacteria lead particularly to three main short-chain fatty acids (SCFAs) acetate, propionate and butyrate, which are known to have beneficial effects on energy balance and obesity (Russell 2003; Hoyles et al. 2010; Hoyles et al. 2010). Some FAs, called essential FAs (ω-3 α linoleic acid and ω-6 linoleic acid), cannot be synthesised in the body and are directly available through dietary lipids. FAs are required for physiological processes including cell signalling, transport and energy metabolism, and contribute to the formation of TGs, which are stored in tissues (i.e. adipose and hepatic) during the postprandial phase. In the context of reduced calories, mobilisation of stored energy in the form of FAs along with glucose and amino acids aims to supply the tricarboxylic pathway (TCA cycle) with substrates

25 Chapter 1-Introduction from which ATP can be produced to maintain normal cell and organ function. This happens in situations of stress when energy available from carbohydrates and amino acids plummets (i.e. lipolysis), which is induced by fasting or by β-agonists (i.e. stress signalling) through β-adrenergic response. Excess FAs have a major impact on cellular pathways through exacerbated activation of nuclear receptor (i.e. peroxisome proliferator-activated receptor PPARγ), production of reactive oxygen species (ROS) and pro-inﬂammatory cytokines (i.e. tu- mour necrosis factor-α, TNFα). These reactions culminate in chronic inﬂammation and in insulin resistance which are main contributors to obesity (i.e. adipocytes), to NAFLD (i.e. hepatocytes) and to metabolic syndrome (Wymann et al. 2008).

1.2.2 Glycerolipids

Glycerol is the backbone structure of glycerolipids and can be esteriﬁed by one (Monoacylglycerols, MGs), two (DGs) or three (TGs) fatty acyls (Figure 1.3B). Dietary energy supplies TGs as well as adipose tissue, where TGs are stored to maintain the vital functions of the body. TGs are the main form of energy intake and excess of energy storage results in activation of TG catabolism by lipases to form glycerolipid intermediates, DGs and MGs, with the release of fatty acyls (i.e. lipolysis). Cellular damage induced by an excess of fatty acyls involves an increase in the synthesis of glycerophospholipids. DGs have a key role in mediating pro- inﬂammatory regulator responses and the inhibition of insulin sensitivity in obesity, NAFLD and metabolic syndrome (Wymann et al. 2008).

1.2.3 Glycerophospholipids

Glycerophospholipids are composed of glycerolipids with a phosphate group at the sn-3 position (Figure 1.3C). Diversity of glycerophospholipids resides in various combinations with head groups such as cholines, phosphatidylcholine (PC); ethanolamines, phosphatidylethanolamine (PE); serines, phosphatidylserine (PS); in- ositols, phosphatidylinositol (PI); glycerols and phosphatidylglycerol (PG) or without polar compounds binding the phosphate group, called phosphatidic acid (PA).

26 Chapter 1-Introduction

Glycerophospholipids are major components of the cell membrane. This can be explained by their amphipathic properties and ability to form bilayers. PCs are abundant in the outer membranes of cells. The reverse is observed for the amino glycerophospholipids (PE and PS) and PI. Head groups of glycerophospholipids have an important impact on cell membrane rigidity. This is caused by different fusion temperatures between each class of glycerophospholipids. For instance, PEs have higher fusion temperature than PCs (Wan et al. 2008). Fatty acyls can bind through three different ways to the glycerol at sn-1: withO- alkyl (-O-CH2-) or O-alkenyl ether (-O-CH=CH-) or O (-O-CO-CH-). Other glycerophospholipids are known such as lysophospholipids (LPs), characterised by having no fatty acyl on one of the position sn and cardiolipids corresponding to a dimer of PG. Glycerophospholipids aid the emulsification of dietary fats by formation of mixed micelles (including other lipids such as TGs, cholesterol and BAs). In addition, glycerophospholipids are a source of different lipid classes. For example, before absorption in small intestine glycerophospholipids can be processed by pancreatic phospholipase to provide fatty acyls and LPs (Phan et al. 2001). Similarly, bacterial phospholipases produce DGs from glycerophospholipids (Morotomi et al. 1990).

1.2.4 Sphingolipids

Sphingolipids comprise an aliphatic amino alcohol and a sphingosine backbone (sphingoid base) (Figure 1.3D). Sphingolipids can bind various hydrophilic molecules. Ceramides (Cer) have a fatty acyl attached to the sphingoid base by an amide bond. Sphingomyelins are formed by ester linkage between Cer and phosphocholine. Gly- cosphingolipids are formed by linkage between a hydroxyl group of Cer and sugar (e.g. glucose, galactose) and derivates with sulfate on the sugar are called sulfatides. Glycosphingolipids with one sugar are cerebrosides and with more than three sugars are gangliosides. Sphingolipids are important components of cell membranes, especially nerve cell membranes. Sphingolipids are synthesised from palmitoyl CoA and serine at the cytoplasmic face of the endoplasmic reticulum (Gault et al. 2010). Subsequently,

27 Chapter 1-Introduction cascades of transferase allow synthesis of the different sphingolipids. The class of sphingolipids is regulated by lysosomal enzymes. For example, sphyngomyelinase provides Cers by cleaving sphingomyelins, and these have deleterious effects on cells by promoting cell inflammation and death receptors (Wymann et al. 2008).

Figure 1.3: Lipid classes: Fatty acyls (A), Glycerolipids (B), Glycerophospholipids (C) and Sphingolipids (D).

1.2.5 Sterol lipids

Sterols are tetracyclic triterpenes with a hydroxyl in position C3 derived from cholesterol. A cascade of reactions that involves acetyl CoA and the squalene precursor lanosterol is at the origin of cholesterol synthesis. Cholesterol plays a pivotal role as a precursor of steroid synthesis pathways, such as those associated with steroid hormones, vitamin D3, oxysterols and primary BA biosynthesis. The scope of this thesis intends to investigate more closely BA pathways to characterise obesity and liver disease. Oxysterols and BAs are products of cholesterol metabolism in the liver. Oxys- terols are synthesised via two major pathways, the classic and alternative, and two minor pathways, the 24- and 25-hydroxylase. The four pathways are distinct in their initial step, which hydroxylates cholesterol at the C7, C27, C24 or C25 position (by

28 Chapter 1-Introduction decreasing order of importance). Oxysterols diﬀer by the introduction of hydroxyl, ketone and/or carboxyl groups in the sterol nucleus (C3, C6, C7 and/or C12) or in the side chain (C24, C26 and/or C27). Primary BAs are the end-products of these pathways and are cholic acid (CA) and chenodeoxycholic acid (CDCA) in humans (Figure 1.4).

Figure 1.4: Cholesterol metabolism. Classical, alternative, 24-hydroxylase and 25- hydroxylase pathways.

29 Chapter 1-Introduction

Through the enterohepatic circulation of BAs (Figure 1.5), a pool of primary BAs and derivatives such as, secondary and tertiary BA is maintained. Primary BAs are synthesised in the liver and undergo conjugation with glycine or taurine before excretion into the intestine. Gut microbial enzymes are responsible for secondary BA synthesis by deconjugation and structural modiﬁcation of primary BAs, such as epimerisation and dehydroxylation (e.g. deoxycholic acid;DCA and lithocholic acid;LCA), permitting passive absorption. The BA pool is then reabsorbed in the terminal ileum through the portal vein to the liver. Around 95% are reabsorbed (≈20g/day) and around 5% escaped absorption (≈0.2-0.5g/day). Secondary and primary BAs are then again conjugated in the liver and excreted into the intestine, resulting in the in formation of tertiary BAs. Enterohepatic circulation of BAs is regulated through sulfation and glucuronidation, which increase hydrophobicity, toxicity and elimination of BAs in the urine and faeces (Figure 1.5) (Mackie et al. 1997).

Figure 1.5: Bile acid enterohepatic circulation.

30 Chapter 1-Introduction

1.2.6 Other lipid classes: prenols, sacccharolipids and polyke-

tides

Prenols are connected units of isoprene (CH2=C(CH3)CH=CH2) synthesised from acetyl CoA. Polyisoprenic lipids are largely represented by liposolubles vitamins (A, D, E, K1 and K2). Vitamin E, for example, protects the mitochondrial membrane from oxidative stress to promote production of free radicals (Traber et al. 1999). Saccharolipids are composed of polysaccharides linked to acyl chains. Saccha- rolipids are mainly found in plants, bacteria and fungi where they play an important role in membranes, best exempliﬁed by lipopolysaccharides in Gram-negative bacteria (Raetz et al. 2002). Polyketides are a wide class of polycyclic lipids and found in plants, bacteria and fungi and animals. Polyketides often have anti-microbial (e.g erythromycin), anti- parasitic, and anti-cancer properties (Cortes et al. 1990).

31 Chapter 1-Introduction

1.3 Lipid metabolism

1.3.1 Lipids and lipoprotein metabolism

To carry out physiological processes, lipids and other compounds (carbohydrates and proteins) are crucial to maintain energy homeostasis. Fat components (such as TGs) of ingested meals are hydrolysed by pancreatic lipase into free FAs (FFAs) and MGs in the intestinal lumen. Lipids emulsified by BAs are transported into the cytoplasm of enterocytes to be processed and to form chylomicrons composed of TGs, cholesterol esters (cholesterol derivatives), phospholipids and apolipoproteins before entering the lymphatic then systemic circulation. In response to increased levels of circulating TGs and FAs, very low density lipoproteins (VLDLs) are produced by the liver. Eventually, chylomicrons and VLDLs provide small circulating lipoproteins with low TG level called chylomicron remnants, intermediate-density lipoproteins (IDLs) and low density lipoproteins (LDLs, ”bad” cholesterol). These products return to the liver by endocytosis and are degraded by lysosomal actions. Lipolysis of transported TGs by VLDLs and chylomicrons contributes to deliver a stock of FAs to different organs and tissues (e.g. adipocytes and myocytes). Lipid uptake is, for example, mediated by an increase in insulin, which inhibits lipases during the postprandial response. Conversely, during the fasting state, energy expenditure is insufficient and adipocytes have to mobilise their stock of FAs to assure vital functions. Besides FA metabolism, cholesterol uptake from cells is regulated by high density lipoproteins (HDLs, ”good” cholesterol) synthesised in the liver (Bronk 1999). Substantial evidence suggests that the metabolic syndrome is closely associated with disruption of lipid metabolism (Wymann et al. 2008). Obesity is characterised by increased levels of TGs at fasting and postprandial state compared to healthy subjects. Impaired lipolysis of TGs due to lipase deficiency results in compromised lipoprotein transport and clearance (Couillard et al. 1998; Taira et al. 1999). Furthermore, obese adipose tissue can lead to insulin resistance via excess FAs (Karpe et al. 2011). NAFLD signature was also shown to be characterised by multiple lipid classes (Dumas

32 Chapter 1-Introduction et al. 2014).

1.3.2 Role of the gut microbiota and lipid metabolism

The gut microbiota refers to the trillions of commensal bacteria and archaea living in the gastrointestinal tract in symbiosis with their host. Although many members of this community remain incompletely characterized, molecular-based (16S rRNA gene) studies have shown that the human faecal microbiota comprises seven phyla of bacteria (Firmicutes, Bacteroidetes, Actinobacteria, Fusobacteria, Proteobacteria, Verrucomicrobia, Cyanobacteria), with (in order of abundance) the Bacteroidetes, Fir- micutes, Actinobacteria and Proteobacteria predominating (Turnbaugh et al. 2006). The main role of the gut microbiota in the large intestine is to transform indigestible nutrients such as dietary fibres (e.g. polysaccharides) into digestible compounds. The gut microbiota is also able to protect the gastrointestinal tract against opportunistic pathogens (Bäckhed et al. 2005). The close relationship between the gut microbiota and its host can directly affect energy harvesting and can have an impact on host energy balance and health (Nicholson et al. 2005). Exploration of the gut microbiota and its relationship with the host in health and disease is done using diversity studies based on 16S rRNA gene sequencing of bacteria and archaea in faecal samples and, to a lesser extent, gut microbiota biopsy samples. Metagenomics and metabonomics are also being used to investigate the gut microbiota’s influence on host health. Using germ-free mice, studies have demonstrated the link between gut microbiota activity and host weight gain. In addition, a decrease in dietary fiber intake resulted in an increase in body weight and the development of diabetes. Therefore, changes in gut microbiota composition/activity induced alterations in the nutrient supply and digestion. As a consequence one study has suggested that signaling molecules such as BAs are altered and affect the regulation of lipid metabolism (Sayin et al. 2013). Furthermore, it is known that the switch to a high fat diet also results in rapid alterations of the gut microbiota. Similarly, modification of both the gut microbiota and the metabolic phenotype has been demonstrated in bariatric surgery (Aron-Wisnewsky

33 Chapter 1-Introduction et al. 2012). To counteract the obesity epidemic, the gut microbiota is being considered as a therapeutic target and diﬀerent approaches (e.g. synbiotics, prebiotics, probiotics, functional foods) are being used to manipulate the activity and/or the growth of bacterial populations, to maintain an optimal microbial balance in the intestine (Nicholson et al. 2005). In summary, lipids are at the boundary of host and gut microbiota entities. Modu- lation of some lipid classes (e.g. FAs, BAs) by the gut microbiota has been suggested to be responsible for functional disturbances related to obesity, NAFLD and metabolic syndrome.

1.3.3 BA metabolism and the gut microbiota

It is now widely accepted that BAs and BA precursors (i.e. oxysterols) are versatile molecules. The emulsification and absorption of lipids from the diet by BAs is not their only role. BAs and oxysterols contribute to the activation of nuclear hormone receptors and G-coupled receptor proteins. Extensive investigations have showen the implication of BAs in down-regulation of metabolic syndrome factors such as insulin resistance, hyperglycaemia and atherogenic dyslipidaemia. The gut microbiota contributes to the structural diversity of circulating BAs - primary BAs (CA and CDCA) synthesised by the liver are conjugated with taurine or glycine and during enterohepatic circulation, primary BAs are deconjugated, dehydroxylated, oxidized and epimerized by gut microbiota enzymes to form secondary BAs (Ridlon et al. 2006). The enterohepatic circulation of these secondary BAs generates tertiary BAs. The bacterial enzymes require a free carboxyl group (C-24) to allow the modifications of hydroxyls groups in the steroid nucleus of BAs. In this respect, bile salt hydrolases were shown to deconjugate BAs from taurine and glycine before any further conversion. For example, taurocholic acid (TCA) and glycocholic acid (GCA) are deconjugated to CA. Furthermore, these amino acids were observed to be metabolised by some gut bacteria (e.g. Clostridium and Bifidobacterium) for energy supply (Hui-

34 Chapter 1-Introduction jghebaert et al. 1982; Tanaka et al. 2000; Van Eldere et al. 1996) (Table 1.1). 7α- dehydroxylation can then be completed and aims to remove the hydroxyl group on position C7 of the BA. This 7α-dehydroxylation activity can have a signiﬁcant impact on health as the product are more toxic than the substrate. For example, 3α,7α,12α- trihydroxy-5β-cholanoic acid (i.e. CA) lead to 3α,12α-trihydroxy-5β-cholanoic acid (DCA) which was shown to be hepatotoxic (Delzenne et al. 1992). Bacterial genera such as Bacteroides, Clostridium, Eubacterium and Escherichia were shown to pos- sess the activity of 7α-dehydroxylation (Table 1.2). Epimerization and oxidation is performed by hydroxysteroid dehydrogenases on hydroxyl groups (C3, C7 and C12). Epimerization is a stereochemical conversion of an hydroxyl group from α to β conformation. For example, epimerization of 3α,7α,12α-trihydroxy-5β-cholanoic acid can lead to 3β,7α,12α-trihydroxy-5β-cholanoic acid. Oxidation is the loss of an hy- drogen from an hydroxyl group to form a carbonyl group. For example, oxidation of 3α,7α,12α-trihydroxy-5β-cholanoic acid can lead to 3-keto,7α,12α-trihydroxy-5β- cholanoic acid. Epimerization and oxidation of BAs is mediated by various bacterial genera such as Clostridium, Bacteroides, Bacillus, Eubacterium, Escherichia, Arthrobacter, Pseudomonas, Alcaligenes (Table 1.3 and 1.4).

35 Chapter 1-Introduction 1988) 1965, et al. et al. 1978) 2005) 1988) 1988) 1988) 1988) 1988) 1992) 1978, et al. et al. et al. et al. et al. et al. et al. et al. et al. 1982) 1975; Masuda 1981; Shindo 1989; Masuda 1981) 1999; Kobashi 1975; Masuda 1981; Nair 1977) 1977) 1977) 1977) 1977) 1977) 1977; Shindo 1977; McAuliffe 1978) 1978; Shindo 1978) 1978) 1978) 1978) 1978) 1978; Shindo 1978; Shindo 1978; Shindo 1988) 1988) 1988) 1988) 1988) 2001) 1988) 1972; Kobashi 1972; Lundeen et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. (Tenneson (Kobashi (Kobashi (Shindo (Tenneson (Kobashi Shindo (Shindo (Masuda 1981) (Kobashi Macdonald (Masuda 1981) (Floch (Kobashi (Kobashi (Gilliland (Elkins (Gilliland (De Boever (Gilliland (Floch (Gilliland (Kobashi (Gilliland (Masuda 1981) (Kawamoto (Macdonald (Gilliland (Kobashi (Kobashi (Shindo (Shindo (Kobashi (Shindo (Macdonald Pseudomononas spp. Proteus mirabilis Streptococcus lactis Escherichia coli Streptococcus faecalis Peptostreptococcus intermedius Clostridium paraputrificum Clostridium sordellii Clostridum sphenoides Clostridium bifermentans Clostridium perfringens Sporolactobacillus inulinus Lactobacillus xylosu Lactobacillus johnsonii Lactobacillus leichmanni Lactobacillus plantarum Lactobacillus salivarius Lactobacillus sp. Lactobacillus fermenti Lactobacillus casei Lactobacillus brevis Lactobacillus buchneri Lactobacillus acidophilus Bacteroides vulgatus Eubacterium aerofaciens Bacteroides ovatus Bifidobacterium bifidum Bifidobacterium breve Bifidobacterium liberorum Bifidobacterium longum Bifidobacterium parvulorum Bacteroides fragilis Bifidobacterium adolescentis Abbreviations: taurochenodeoxycholic acid;TCDCA, glycochenodeoxycholic acid;GCDCA Deconjugation performed on primary bile acids, cholic acid and chenodeoxycholic acid by gut microbiota Table 1.1: TCA, TCDCA, GCA, GCDCA CA, CDCA GCA CA TCA, TCDCA, GCA, GCDCATCA, TCDCA, GCA, CA, GCDCA CDCA CA, CDCA Proteobacteria Gamma Proteobacteria TCA, TCDCA, GCA, GCDCA CA, CDCA Enterococcaceae TCA, TCDCA CA, CDCA TCA, TCDCA, GCA, GCDCATCA, TCDCA CA, CDCA CA, CDCA GCA CA TCA, TCDCATCA, TCDCA, GCA, GCDCA CA, CDCA CA, CDCA Clostridia GCA CA TCA, TCDCA CA, CDCA TCA, TCDCA, GCA, GCDCATCA, GCA CA, CDCA GCATCA, GCATCA, TCDCA, GCA, GCDCA CA, CDCA CA CA CA TCA, GCA CA GCA CA TCA, TCDCA, GCA, GCDCA CA, CDCA TCA, TCDCA, GCA, GCDCA CA, CDCA TCA, TCDCA, GCA, GCDCA CA, CDCA Firmicutes Bacilli TCA, TCDCATCA, TCDCA, GCA, GCDCA CA, CDCA CA, CDCA TCA, TCDCA, GCA, GCDCA CA, CDCA TCA, TCDCA, GCA, GCDCATCA, TCDCA, GCA, CA, GCDCA CDCA TCA, TCDCA, GCA, CA, GCDCA CDCA TCA, TCDCA, GCA, CA, GCDCA CDCA TCA, TCDCA, GCA, CA, GCDCA CDCA TCA, TCDCA, GCA, CA, GCDCA CDCA CA, CDCA Bacteroidetes Bacteroidetes BA substrateTCA, TCDCA, GCA, GCDCA CA, CDCA Actinobacteria Actinobacteria BA product Phylum Class Species References

36 Chapter 1-Introduction 1985) 1984) 1984) 1989) 1989) 1984) 1982) 1982) 1982) 1985) 1985) 1983) 1985) 2004) 1982a) 1990) et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. 1985) et al. et al. et al. et al. 1970) (Edenharder (Edenharder (Macdonald (Macdonald (Hirano (Edenharder (Sutherland (Macdonald (Kole (Lepercq (Sutherland (Huang (Edenharder (Aragozzini (Edenharder (Edenharder 1997; White 1979) 1997) et al. et al. et al. et al. 1975) 1975) 1987; Doerner 1979) 1997) 1997; Hayakawa 1997) 1997; Stellwag 1982b) 1982b) 1982b) 1982b) 1977; Doerner et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. Peptostreptococcus productus Clostridum absonum Eubacterium aerofaciens Clostridum paraputriﬁcum Bacteroides fragilis Bacteroides uniformis Clostridum limosum Eubacterium aerofaciens Clostridum absonum Clostridum baratii Clostridum perfringens Clostridum limosum Clostridum perfringens Clostridum innocuum Eubacterium lentum Clostridum innocuum (Macdonald (Hirano (Hirano (Hirano (Stellwag (Hirano (Doerner (Doerner (Coleman (Doerner (Doerner (Macdonald (Ferrari Firmicutes Clostridia Escherichia coli Bacteroides fragilis Bacteroides ovatus Bacteroides thetaiotaomicron Clostridium leptum Escherichia coli Clostridium sordellii Clostridium sp. Eubacterium sp. Clostridium scindens Clostridium leptum Bacteroides fragilis Clostridium bifermentans -trihydroxy 5-cholenoic acid Firmicutes Clostridia -trihydroxy 5-cholenoic acid Firmicutes Clostridia β α ,12 -dihydroxy 5-cholenoic acid Firmicutes Clostridia ,12 α α α Abbreviations: ursocholic acid;UCA and ursodeoxycholic acid;UDCA ,7 ,7 ,7 α β β Proteobacteria Gamma Proteobacteria Proteobacteria Gamma Proteobacteria Firmicutes Bacilli Firmicutes Clostridia Epimerization performed on primary bile acids, cholic acid and chenodeoxycholic acid by gut microbiota -Dehydroxylation performed on primary bile acids, cholic acid and chenodeoxycholic acid by gut microbiota CDCA 3 CDCA UDCA Firmicutes Clostridia α 7 Table 1.3: epimerization CA 3 epimerization CA UCA Bacteroidetes Bacteroidetes epimerization CA 3 β β β Table 1.2: 7 activity3 BA substrate BA product Phylum Class Species References 12 CDCA LCA Bacteroidetes Bacteroidetes BA substrate BACA product Phylum DCA Class Bacteroidetes Bacteroidetes Species References

37 Chapter 1-Introduction 1985) 1982) 1979) 1979) 1979) 1982) 1979) 1979) 1982) 1987) 1987) 1987) 1987) 1993) 1993) 1985) 1993) 1993) 1982) 1995) 1995) 1994) 1994) 1994) 2009) 2009) 1983) et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. 1985, et al. et al. et al. et al. et al. et al. et al. et al. et al. et al. (Mukherjee (Kimura (Macdonald (Mazumder (Macdonald (Owen (Macdonald (Aragozzini (Macdonald (Mazumder (Macdonald (Fukiya (Kimura (Macdonald (Sutherland (Sutherland (Takamine (Fukiya (Kole Macdonald (Sutherland (Sutherland (Takamine (Macdonald (Mukherjee (Kimura (Edenharder (Macdonald Alcaligenes recti Pseudomonas sp. Escherichia coli Bacillus sp. Eubacterium lentum Alcaligenes recti Bacteroides fragilis Clostridum perfringens Eubacterium lentum Bacteroides fragilis Bacteroides intestinalis Bacillus sp. Clostridum absonum Clostridium bifermentans Clostridum limosum Clostridium sordellii Bacteroides intestinalis Clostridum absonum Clostridum absonum Clostridium bifermentans Clostridum limosum Clostridium sordellii Eubacterium lentum Arthrobacter simplex Bacillus sp. Clostridum paraputriﬁcum Eubacterium lentum Arthrobacter simplex Gammaproteobacteria Clostridia Proteobacteria Betaproteobacteria Clostridia Firmicutes Bacilli FirmicutesProteobacteria Bacilli Betaproteobacteria Firmicutes Clostridia Firmicutes Clostridia Firmicutes Bacilli Firmicutes Clostridia hydroxy 5-cholenoic acid Bacteroidetes Bacteroidetes hydroxy 5-cholenoic acid Clostridia dihydroxy 7-oxo - 5-cholenoic acid Bacteroidetes Bacteroidetes dihydroxy 3-oxo 5-cholenoic acid Actinobacteria Actinobacteria α α dihydroxy 12-oxo 5-cholenoic acid Actinobacteria Actinobacteria α α α hydroxy,7-oxo 5-cholenoic acid, ketoLCA Bacteroidetes Bacteroidetes ,12 ,7 ,12 Oxidation performed on primary bile acids, cholic acid and chenodeoxycholic acid by gut microbiota α α α α Table 1.4: CDCA 3-oxo, 7 CDCA 3 CDCA 3-oxo,7 C7 oxidation CA 3 C12 oxidation CA 3 activityC3 oxidation CA BA substrate BA product 7 Phylum Class Species References

38 Chapter 1-Introduction

The BAs and their precursor oxysterols are versatile molecules with i) bacteriostatic (Begley et al. 2005), ii) emulsifying and iii) signalling properties. BAs have a beneficial effect in cardiometabolic disease by improving insulin sensitivity, hyperglycaemia and dyslipidaemia. Bariatric surgery studies also demonstrated that microbiome interventions increased circulating BAs, which appeared to be critical for body weight loss and glucose homeostasis (Ryan et al. 2014; Penney et al. 2015). BAs regulate glucose homeostasis via several pathways, i.e. activation of Farnesoid X Receptor (FXR or NR1H4), Gαi protein-dependent receptor and TGR5 receptor (or GP-BAR1, or M-BAR) (Figure 1.6) (Watanabe et al. 2006; Ma et al. 2006). Activation of FXR by BAs downregulates fatty acid and TG synthesis in the liver and decreases circulating TGs and VLDL production (Figure 1.6) (Watanabe et al. 2004; Zhang et al. 2006). BAs also increase energy expenditure through cAMP-mediated activation of thyroid hormone (Watanabe et al. 2006). In vitro experiments show that DCA is a stronger TGR5 agonist and it is a greater antimicrobial agent compared to CA (Kawamata et al. 2003; Begley et al. 2005). However, the role of DCA is contrasted by the fact that is hepatotoxic and can trigger hepatocellular carcinoma through senescence secretome (Bernstein et al. 2005; Yoshimoto et al. 2013). BAs such as DCA and LCA are modified by the gut microbiota and are able to activate alternative pathways through interactions with the epidermal growth factor receptor (EGFR)/FAS and the PXR/Vitamin D receptor, respectively (Staudinger et al. 2001; Qiao et al. 2001). For instance, a mouse study highlighted the gut’s ability to modulate the BA pool circulating, particularly of the primary BA tauro-β- muricholic acid (TβMCA) and to modulate Fibroblast Growth Factor 15 (FGF15, in human FGF19) implicated in the inhibition of BA synthesis. However, key outcomes of gut-mediated modification of the BA pool are still unclear. In addition, findings obtained with mice are contentiously comparable to human because, for example, the main primary BAs are different.

39 Chapter 1-Introduction

Figure 1.6: BA signalling and regulation of cardiometabolic risk factors. BA activation of the FXR regulates lipid metabolism via activation of apolipoproteins A1, C II and C III (apo A1, apoC II and apo C III) and inhibition of Sterol Regulatory Element Binding Protein 1c (SREBP-1c). In addition, glucose homeostasis is regulated by various pathways including; FOX01, Glucocorticoid Receptor (GR) and Hepatocyte nuclear factor 4 α (HFN4α) activated through FXR, Glucagon-Like Peptide-1 (GLP-1) by TGR5 receptor (or GP-BAR1, or M-BAR) and Glucagon Synthase (GS) by Gαi protein (Neves et al. 2015).

40 Chapter 1-Introduction

1.4 Metabonomics

Given the complexity of lipid metabolism and the multitude of genetic and environmental factors that can affect metabolic pathways, it is necessary to investigate obesity and metabolic syndrome from a holistic point of view. Metabonomics is defined as ”the quantitative measurement of the dynamic mul- tiparametric metabolic response of living systems to pathophysiological stimuli or genetic modification” (Nicholson et al. 1999). This approach is considered as the evaluation of a global metabolic status which is the end result of perturbations occurring at the proteome, transcriptome, genome and external environmental levels. Information emerging from each of these approaches offers a wide overview and in- depth understanding of the variations associated with biological system phenotypes. The metabolome activity of a biological system depends on the flux of metabolites, described in analytical research as ”low molecular weight molecules” typically less than 2kDa in mass. The scope of the emerging metabonomic field is dependent on advances in modern analytical techniques enabling the detection and quantification of metabolites to capture an individual metabolic profile. Analysis of diverse metabolites relies on two powerful analytical techniques; nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS). MS is used to implement a metabonomic approach to obesity and NAFLD in this thesis.

41 Chapter 1-Introduction

1.5 Thesis outlines and aims

Having reviewed the literature there is a critical lack of suitable analytical methods for studying subtle lipid variations involved in disease progression in subcutaneous vs. visceral obesity and the transition from NAFLD to NASH. These unaddressed challenges in the ﬁeld are identiﬁed as follows:

• lack of suitable extraction method for lipid proﬁling

• lack of comprehensive BA quantiﬁcation method

• gap in knowledge about the inﬂuence of central obesity on postprandial lipemic response (PPLR)

• gap in knowledge about lipids involved in the transition from NAFLD to NASH

Accordingly, this thesis intends to investigate these unaddressed challenges in the field by a metabonomic approach with four main objectives: 1) development of a suitable extraction for lipid profiling (Chapter 3) 2) development of a suitable BA quantification method (Chapter 2) 3) application to PPLR (Chapter 4) 4) lipid signature of NAFLD/NASH transition (Chapter 5)

Materials and methods used in this thesis are explained in Chapter 2. In this thesis, preliminary method development of sample preparation for lipid proﬁling (Chapter 3) and method development of the BA targeted assay (Chapter 4) were implemented before applying this methods to the human cohorts (Chapter 5 and 6). As such, the two cohorts were evaluated by lipid proﬁling (Chapter 5) and the targeted BA method with UPLC-MS instrumentation (Chapter 6). A general discussion is presented at the end of this thesis (Chapter 7) (Figure 1.7).

42 Chapter 1-Introduction

Figure 1.7: Thesis workﬂow presenting methodology in Chapter 2, results Chapters 3 to 6 and discussion in Chapter 7.

43 Chapter 2

The metabonomic approach

As mentioned previously (cf. Chapter 1), metabonomics is influenced by genetic and environmental variations and is a systemic approach that identifies metabolic signatures. In this thesis, molecular profiling and targeted analysis were applied to the study of the lipid and bile acid metabolism using human samples. As NMR is not adapted to decipher the complexity of lipidomic patterns, UPLC-MS was conducted in this work (Roberts et al. 2008). The experimental design adopted here can be divided in five major steps: sample preparation, data acquisition, data treatment, statistical analysis and identification of biomarkers (Figure 2.1).

Figure 2.1: Metabonomic approach and experimental design.

44 Chapter 2-Methodologies

2.1 Sample preparation

Sample preparation is a critical step which sets the foundation for the quality of the analysis. The ﬁrst step in sample preparation is to break the cells and to separate metabolites according to their aﬃnity for aqueous or organic solvent environments (structural properties). High-quality sample preparation is designed to obtain conditions to optimise the extraction of the lipids and limit their alteration or degradation. In this thesis, sample preparation was applied to blood samples (EDTA-plasma) for analytical method development (cf Chapters 3 and 4) and biological applications (cf. Chapters 5 and 6).

2.1.1 Sample type

Selection of the blood sample matrix can impact on the outcome of sample preparation and results obtained with lipidomic analysis. There are two types of processed blood: serum and plasma. The latter can be collected into tubes with EDTA, citrate, or heparin anticoagulants. Serum has been reported to contain all lipid classes including phosphocholines (PCs) and triglycerides (TGs) (Breier et al. 2014; Ishikawa et al. 2014). However, the profile of the lipid extract can vary according to the serum clotting temperature, centrifugation speed and centrifugation temperature (Cequier-Sánchez et al. 2008). Citrate plasma was described as a poor matrix for lipid extraction resulting in significant loss of PCs, phosphoethanolamines (PEs), sphingomyelins (SMs) and TGs (Jørgenrud et al. 2015). Similarly to serum, EDTA plasma appeared to offer high concentration of diverse lipid classes, mainly SMs (Hammad et al. 2010). However, processing temperature of the EDTA plasma was observed to have no impact on the final lipid extraction (Jørgenrud et al. 2015). EDTA plasma was selected to conduct experiments presented in this thesis.

45 Chapter 2-Methodologies

2.1.2 Storage

Sample storage is associated with lipid stability. Freezer storage at -80°C is recommended to avoid lipid degradation by oxidation. However, multiple freeze- thaw cycles is not recommended as losses of LCs, PCs and TGs can occur and aliquots can be prepared to prevent these losses (Breier et al. 2014; Ishikawa et al. 2014). Studies carried out at 4°C after 72h shown oxidation and hydrolysis of lipids, especially lysophospholipids (Jørgenrud et al. 2015). With regard to these previous ﬁndings, in the current study samples were stored at -80°C, aliquots were prepared for each analysis to avoid multiple freeze-thaw cycles and samples were left at 4°C for a maximum of 24h during UPLC-MS analysis.

2.1.3 Sample preparation: Lipid extractions

Biological membranes are composed of various lipid classes with different polarities and thereby lipid extraction methods demonstrate varying selectivity when using various organic solvents. In addition, lipids are both hydrophobic (carbon chain) and hydrophilic (head group) and optimal lipid extraction relies on combination of organic solvents (apolar) and water (polar). Removal of polar compounds (e.g. proteins, sugars or salts) is decisive as they can deteriorate the column and impact on the chromatographic process in UPLC-MS analysis by critical increase of the pressure (cf. Chapter 2.3). Lipid extraction methods can be divided in two groups (cf. Chapter 3), biphasic (liquid-liquid extraction) and monophasic (simple protein precipitation). The classic and most ubiquitous biphasic lipid extraction method is the Folch method using chloroform and methanol solvents (CHCl3:MeOH, 2:1) (Folch et al. 1957). Similarly, the Bligh and Dyer method is a well-known adaptation of the Folch method biphasic lipid extraction with some variations in terms of solvents ratio and chloroform (Bligh et al. 1959). Efforts have been made to replace the use of chloroform in these methods with a less toxic solvent, dichloromethane (CH2Cl2) (Cequier-Sánchez et al. 2008). Non-chlorinated solvents (e.g. methyl-tert-butyl ether, MTBE and hexane)

46 Chapter 2-Methodologies

with lower densities than CHCl3 and CH2Cl2, have been evaluated to facilitate the lipid extraction procedure (Matyash et al. 2008). However, these chlorinated solvents are not compatible with the UPLC-MS method as it can result in poor chromatographic separation and can damage parts of the instrument. Therefore, these biphasic sample preparation methods require a drying step where loss of lipids can be expected. Conversely, monophasic lipid extraction is based on protein precipitation followed by centrifugation and oﬀers straightforward advantages compared to biphasic method (Saraﬁan et al. 2014). In this thesis, eight methods were evaluated (cf. Chapter 3) and monophasic protein precipitation was selected as a robust method to prepare EDTA-plasma samples (cf. Chapters 4, 5 and 6).

47 Chapter 2-Methodologies

2.2 UPLC-MS analytical techniques

Advances in liquid chromatography (LC) combined with mass spectrometry (MS) allows high detection sensitivity and selectivity of lipid classes. Ultra-performance liquid chromatography mass spectrometry (UPLC-MS) is a proven analytical technique and was implemented in this work (Niessen 2006).

2.2.1 LC

A LC system is composed of stationary phase (column particles) and mobile phase (solvents). UPLC allows separation of analytes according to their diﬀerential aﬃnity for the stationary or mobile phases.

(a) Theory of chromatography

UPLC performance oﬀers advantages such as high resolution, high analysis speed, and increased sensitivity. Resolution (Rs) is deﬁned as the separation between two eluting peaks (Gaussian shape) measured as the retention time (tR) divided by the average width (wav) of the respective peaks.

tR Rs = wav

High resolution is achieved by decreasing the peak width (increased eﬃciency) and increasing peak separation which may be achieved by changing the chemical selectivity of the chromatographic system. Optimisation of these parameters maximises the peak capacity (k’ which is the measure of the hypothetical separation capability of a gradient elution on a particular column. Moreover, plate theory was developed to explain the association between these parameters (Martin et al. 1941; Van Deemter et al. 1956).The plate theory illustrates a column as divided into adjacent segments called theoretical plates. In each theoretical plate, each analyte equilibrates between the stationary phase and mobile phase. Resolution can be determined with plate number (N), selectivity factor (α) corresponding to separation between two analytes

48 Chapter 2-Methodologies and capacity factor as follows;

√ N k’ ! α-1 Rs = 4 1+k’ α

The increase in number of theoretical plates (N) is proportional to the decrease of plate heights (H) for a corresponding column length (L) and results in improved eﬃciency of the column. L = N×H

The number of theoretical plates generated by a chromatographic system can be calculated using the chromatographic peak width at its base (W ) and the retention time (t) which is the volume of mobile phase needed to elute the analyte from the column.

V !2 N = 16 R Wb

Reduced particle size of the stationary phase (column packing material) can be used to generate a higher number of theoretical plates as can optimising the flow rate of mobile phase flow rate for minimised peak dispersion and increasing the column length. Improved sensitivity is afforded by combined high resolution and speed in UPLC column which result in narrower peaks with greater height (Plumb et al. 2004).

(b) Column chemistry in LC

Compared to gas chromatography (GC), LC can separate non-volatile analytes without derivatisation oﬀering a better direct compatibility with bioﬂuids and liquid tissue. Classic examples of LC separation include the use of a hydrophilic (silica or alumina particles, i.e. normal phase) or hydrophobic (alkyl chains, i.e. reversed-phase) stationary phase (column), although many additional modes exist (e.g. ion exchange, hydrophilic interaction liquid chromatography HILIC). In this thesis, reversed-phase chromatography was selected as the most compatible method for lipid separation according to hydrophobicity (cf. Chapter 4, C8 column for bile acid targeted assay

49 Chapter 2-Methodologies and Chapter 3, C18 column for lipid proﬁling).

Solvents are eluted through the column to enable separation of analytes. There are two types of elution, isocratic elution where the mobile phase is unchanged during the run and gradient elution where mobile phase’s quantities are modified during the run. In the chromatographic systems utilised within this thesis, gradient elution was used within reversed phase columns. The first solvent is a primarily aqueous solvent, used to load analytes onto the column and for the elution of hydrophilic analytes (mobile phase A). The second solvent is a primarily organic solvent which is used to elute hydrophobic analytes (mobile phase B). In addition, volatile acids such as formic acid and acetic acid were added to mobile phases in low concentrations (0.1 %) to help control the pH of the mobile phase, improve chromatographic peak shape, and promote ionisation. Volatile buffer salts such as ammonium formate and ammonium acetate were similarly used to control mobile phase pH, stabilising analyte retention time. Specifically, the lipid profiling method used herein required the use of C18 charged surface hybrid (CSH) column (cf. Chapter 3), and the bile acid method required use of C8 ethylene bridged hybrid column (BEH) (cf. Chapter 4).

2.2.2 Mass spectrometry

MS aims to separate ions by their mass-to-charge ratios (m/z) and quantitatively detect their abundance. MS instruments contain three distinct components: the ionisation source, the mass analyser(s) and a detector.

(a) Ionisation

In this thesis, an electrospray (ESI) source was used for analyte ionisation and conversion of LC eluent to the gas phase. ESI aims to disperse droplets containing charged molecular species as ﬁne aerosol producing gas-phase ions (Fenn et al. 1989) (2002 Nobel Prize in Chemistry). Charged droplets (solvent and analytes) are ejected

50 Chapter 2-Methodologies from a Taylor cone formed by the application of high voltage to a capillary, forming a plume or spray which is sheathed in nebulisation gas (nitrogen). Solvent removal is enhanced by the application of a desolvation gas (heated nitrogen) to the spray. The droplets produced progressively decrease in size until the charge density is increased to a point where ions repel each other in a Coulomb ﬁssion event. This process repeats until ions are virtually free of solvent and able to enter the MS in the gas phase. The capillary voltage, gas ﬂow rates, and applied temperatures required in this process are typically adjustable ensuring optimal compatibility with the LC eluent.

(b) Analysers

The MS instrumentation utilised in this thesis contain one or more of two types of mass analysers; time-of-flight (TOF) and quadrupole (Q). In a TOF mass analyser, an electric field accelerates ions to a velocity dependent on their mass and charge. The precise time ions take to reach the detector after acceleration is dependent on their velocity, and therefore the ”time-of-flight” can be used to calculate the ion’s m/z. Electrostatic mirrors (reflectrons) present in some TOF analyser correct for minor variations in kinetic energy among the ions of a given molecular species, as well as multiply the effective length of the flight path, increasing the achievable mass spectral resolution. Quadrupole mass analysers are able to stabilise the trajectory of ion species within a narrow m/z window (equivalent to approximately one Dalton) and destabilise the trajectories of both higher mass and lower mass species using two pairs of parallel metal rods. Ion trajectory is controlled by application of both direct current (DC) and alternating current (AC) voltages, former with a polarity offset between the two pairs of rods and the latter being applied at radio frequency (RF). The resulting effect is oscillating charge on each pair of poles which is capable of selectively transmitting ions with a controllable range of m/z values through the quadrupole system and toward the detector. MS instrumentation is often composed of multiple analysers which enhance the

51 Chapter 2-Methodologies functionality of the system by enabling tandem mass spectrometry (MS/MS) (Figure 2.2). First, a Q analyser is used to filter intended parent ions. Into the collision cell (CID) precursor/parent ions can collide (with argon gas) to create under controlled conditions fragments, called product /daughter ions. Second, Tof (QqToF) or Q (QqQ, TripleQ TQ) can be used after the collision cell to analyse product ions. On Q-ToF, data are acquired in full scan mode with low collision energy in the collision cell to detect all ionisable analytes and this allows a profiling mode (Figure 2.2). Simultaneous acquisition in full scan and using the fragmentation methods such as MSE (Plumb et al. 2006) or data-dependent acquisition (DDA) give MS/MS spectra with structural information. In MSE, a collision energy ramp is applied (10-40 V) to fragment gradually parent ions and in DDA the most intense parent ions in the full-scan spectra are fragmented. On TQ, ions collision is controlled to detect specific precursor or product by various ways (Table 2.1). TQ instruments generally provide better sensitivity and linearity than Q-ToF instruments but achieve lower resolution and typically lower mass accuracy.

Figure 2.2: Example of mass spectrometry principle with Q-ToF.

In this thesis, the instruments used were as follows; Q-ToF for lipid proﬁling and TQ for targeted bile acid analysis. On the Q-Tof, MSE and DDA acquisition was applied. On the TQ, selected reaction monitoring (SRM) and multiple reactions monitoring (MRM) were used to acquire data in the targeted bile acid method presented in this thesis.

52 Chapter 2-Methodologies

Table 2.1: Scan experiments using MS and MS/MS.

Type of scan Type of data Ions detected Full scan qualitative All ionisable analytes Collision energy ramp 10-40 V produce MSE qualitative gradually fragments

PROFILING DDA qualitative Fragments of intense ions in the full-scan Precursor/parent Precursor are scanned and product ion is quantitative ion selected Product Precursors ion is selected and product ions are quantitative ion/daughter scanned Both precursor and product ions is scanned; Neutral loss quantitative neutral losses are selected Selected reaction

TARGETED quantitative Precursor ions is selected Monitoring (SRM) Multiple Reactions Multiple precursor and product ions are quantitative Monitoring (MRM) selected

Ions are detected by an electron multiplier detection system and signal is ampliﬁed then digitised to obtain a spectrum of m/z values and their intensities.

53 Chapter 2-Methodologies

2.3 The metabonomic run strategy

Metabonomic studies highlight metabolic variations occurring in samples. In this thesis, there are two human metabonomic studies which were divided as follows: four groups of overweight to obese volunteers (cf. Chapter 5) and healthy vs liver disease patients (cf. Chapter 6). Additional samples were prepared to monitor the robustness of the analytical run. Blank samples are analysed first to identify contaminants in mobile phases and last to assess carry over and contaminants in samples during data analysis for subsequent removal. Quality controls (QCs) are pooled aliquots of every sample in the study and are essential for column conditioning as well as monitoring the quality of the sample run and for use in future structural identification. QCs are injected at least 10 times for the conditioning of the column which aims to equilibrate the system and reduce peak drift before the sample run. Diluted QCs (1/2, 1/3 and 1/4) are included before and after the sample run to assess the linearity of ion response (Eliasson et al. 2012). QC replicates are analysed through the analytical run every 10 samples to determine the instrumental stability and analyte reproducibility. Furthermore, to avoid the injection order to influence the results, samples are randomised. Finally, QCs are analysed after the sample run by MSE and DDA acquisition to obtain supplementary information about metabolite structure (cf. Chapter 2.3) (Figure 2.3).

Figure 2.3: Run strategy in metabonomics for proﬁling.

54 Chapter 2-Methodologies

2.4 Data pre-processing

Data pre-processing is a fundamental and complex step that extracts LC-MS raw data and describes each feature with a unique retention time (RT), mass to charge (m/z) and intensity. Lipid profiling raw data files were acquired in centroid mode (discrete m/z with zero line width) as it reduces the dataset size compared to continuum mode (record all signal) and were converted to NetCDF format using Databridge (MassLynx Version 4.1) prior to pre-processing. Many valuable tools for LC-MS data are available (Katajamaa et al. 2007) such as open-source XCMS software which was applied on UPLC-MS data presented in this thesis (Smith et al. 2006). The final data sheet is produced with XCMS package and a cascade of data treatments was implemented; peak detection, peak alignment, filtering and normalisation. Several settings (e.g peak width, intensity and CV threshold) for data pre-processing were tested to determine the best extraction of relevant peaks observed in chromatograms. The final functions and settings used in this thesis are presented in Table 2.2.

2.4.1 Peak detection

Peak detection aims to ﬁnd and extract peaks (i.e. features) from background noise with minimum loss of information. Each feature is characterised by RT, m/z and intensities obtained for each sample. For this crucial ﬁrst pre-processing step, the XCMS centWave peak detection algorithm was the selected method as it was recommended for complex matrices (e.g plasma) (Tautenhahn et al. 2008). Settings of this algorithm considered are as follows; chromatographic peak width deviation, m/z deviation due to instrument’s response (ppm, ratio of the mass error and the expected mass), signal to noise threshold and intensity threshold (Table 2.2).

55 Chapter 2-Methodologies

2.4.2 Peak alignment

Peak alignment and peak grouping aim to align and combine features across samples, accounting for RT shifts. A smoother bandwidth on peak alignment facilitates to diﬀerentiate one peak from another. Furthermore, a second phase in peak alignment uses an algorithm to correct non-linear retention time shift with outliers’ removal and missing features identiﬁcation.

2.4.3 Peak ﬁltering

Peak filtering was implemented by using the minfrac function which validates a peak when it is identified in a minimum fraction of samples within a user-specified group. In addition, reproducibility between samples is evaluated by the coefficient of variation (CV), defined as the standard deviation divided by the mean of features intensities across QC samples. Randomisation of the samples was applied to avoid the impact of run order and to obtain reliable CV values.

2.4.4 Normalisation

Data normalisation was implemented to remove variations not related to biological variation. The most appropriate normalisation for UPLC-MS data was shown to be the median fold change method (Veselkov et al. 2011). This method adjusts for the median of log fold changes of peak intensities between samples to zero. Changes in peak intensity are attributed to the dilution eﬀect between samples which is minimised by this method.

56 Chapter 2-Methodologies

Table 2.2: Settings setup for pre-processing steps.

Function Settings Method centWave Peak width 1 to 30 sec ppm Peak detection 30 Signal to noise threshold 3 Intensity threshold At least 3 peaks with intensity ≥ 40 Method group m/z 0.05 RT error 0.6 bandwith Peak alignment 1 Method retcor Method peakgroups Missing features 0 Outlier removal family=’s’ Minfrac Peak ﬁltering 0.2 and 0.3 CV 30% Normalisation Method Median fold change

2.5 Data analysis

2.5.1 Chemometrics

Chemometrics is a key tool in metabonomics that facilitate the visualisation of the data and highlight signals that are statistically significant, therefore delivering a better understanding of biological profiles. In this thesis both, supervised and unsupervised statistical analyses were applied. Unsupervised statistical analyses typically include principal components analysis (PCA) and models are built without information about class membership. PCA cap- tures the maximum of data variation in X matrix. The highest variation is deciphered by the first principal component, the second highest variation is decipher by the second principal component and so on for the following components. By construction, successive components are orthogonal to each other (Wold et al. 1987). Graphically, PCA can be interpreted via two projection plots, the scores plot which correspond to the projection of individual observation and the loading plots which correspond to the projection of the variables. Therefore, when relatively low amplitude of effect between samples is subtle, supervised method is applied to discern occurring variations (Trygg et al. 2007). Supervised statistical methods include partial least squares discriminant analysis (PLS-DA) and/or orthogonal partial least squares discriminant

57 Chapter 2-Methodologies analysis (OPLS-DA)(Eriksson et al. 2004; Trygg et al. 2007). In PLS approaches, class membership (e.g. Chapter 6; class 1 for control subjects, class 2 for NAFLD steatosis and class 3 for steatosis) is introduced in the model (Y matrix) making the selection of discriminating features easier (X matrix). These analyses provide visualisation of the variation of interest and minimises other variation uncorrelated to the Y matrix. R2Xhat and R2Yhat indicate the proportion of explained variance from the X matrix and Y matrix. Q2Yhat indicate the proportion of the predicted variance from the Y matrix in the model. For model validation, a permutation test can be applied to estimate the predicted power of Q2Yhat Statistically signiﬁcant compounds (biomarkers) with high correlation and high covariance were identiﬁed from these analyses (Fonville et al. 2010).

2.5.2 Structural identiﬁcation

Structural assignment of biomarkers was achieved by extensive literature review, and interrogation of both in-house and online metabolite databases (Lipidmaps, HMDB, METLIN) and by complementary experiments such as MS/MS acquisition (DDA and MSE) and spiking of authentic standards (Figure 5). In this thesis, UPLC- MS lipidomic analysis positive mode ESI (+) induce formation of molecular ions with protons ([M+H]+) and/or various adducts such as [M+NH4]+. Negative mode ESI (-) induce ion formation of molecular ions such as [M+H]-, adducts [M+HCOO]- and [M+Cl]- major for glucose.

58 Chapter 2-Methodologies

Figure 2.4: Steps followed for lipid structural identiﬁcation.

59 Chapter 3

An Objective Set of Criteria for Optimisation of Sample Preparation Procedures for Ultra-High Throughput Untargeted Blood Plasma Lipid Proﬁling by UPLC-MS

3.1 Introduction

Metabolic profiling and phenotyping of biological samples provide a global understanding of complex metabolism using two major analytical techniques: NMR and MS (Nicholson et al. 1999; Nicholson et al. 2002; Nicholson et al. 2008). Due to the complex composition of blood, the characterisation of this biofluid greatly benefits from the multidimensional separation and sensitive detection provided by UPLC-MS. UPLC columns can operate with higher flows and pressures, which directly results in shorter acquisition times of analysis for similar resolution to HPLC (Lenz et al. 2007). However, due to the large diversity in physico-chemical properties of the metabolites found in plasma, polar metabolites and lipids are often analysed separately by two different LC-MS methods (Want et al. 2006; Bruce et al. 2008; Tulipani et al. 2012; Bruce et al. 2009; Sommer et al. 2006; Bird et al. 2011). For untargeted UPLC-MS profiling, non-selectivity during sample preparation is critical for successful analysis of these two molecular classes since sample preparation issues may impact the final quality of the data. Sample preparation is a key, yet often overlooked step for suc-

60 Chapter 3-Results cessful UPLC-MS based lipid profiling and can have a strong impact on the quality of subsequent spectral data. Given the importance of lipidomics in systems biology and in various disease areas such as obesity and the metabolic syndrome (Pietiläinen et al. 2007; Pietiläinen et al. 2011; Oreˇsiˇc et al. 2008; Quehenberger et al. 2010), there is a strong analytical need for developing robust and efficient methods fit-for-purpose in large-scale epidemiological and clinical studies, requiring optimised profiling of thousands of samples (Want et al. 2013). Typically, two types of sample preparation strategies can be considered for lipid sample preparation: protein precipitation (monophasic) and lipid extraction (biphasic). In this chapter, eight conventional sample preparation methods for plasma lipid analysis by UPLC-MS were evaluated. The four precipitation methods evaluated here were based on the use of organic solvents such as methanol (MeOH), acetonitrile (ACN), isopropanol (IPA) and IPA-ACN (Figure 3.1). The role of these solvents is to precipitate proteins and solubilise lipids. The four liquid-liquid extraction methods evaluated were based on MeOH combined with various amounts of chloroform (CHCl3), dichloromethane (CH2Cl2), methyl-tert-butyl ether/H2O (MTBE), and hexane/IPA (Hexane) mixtures. These extraction methods were evaluated as they partition lipids from polar metabolites. The

Folch and Bligh-Dyer methods involving MeOH/CHCl3 are widely used for lipid extraction (Bligh et al. 1959; Folch et al. 1957; Nygren et al. 2011), but since the lipid extract is the lower phase due to the high density of halogenated solvents, this can lead to contamination of the organic extract by proteins at the interface between the two phases. MeOH/hexane/IPA and MeOH/MTBE/H2O are alternative methods making the organic phase collection easier due to the lower density of the organic solvents and allow higher-throughput protocols (Matyash et al. 2008; Reis et al. 2013; Hara et al. 1978). Yet, despite the availability of well-documented sample preparation protocols, there is still a need to identify one of these preparation methods, as efficient, robust, repeatable, cost-efficient and reflecting the original sample composition for global

61 Chapter 3-Results lipid profiling analysis (Duportet et al. 2012). In this chapter, the benefits and limitations for each protocol were evaluated with respect to the subsequent results generated by UPLC-MS analysis (Theodoridis et al. 2012; Vuckovic 2012). The eight protocols (Figure 3.1) were benchmarked using the following assessment criteria: simplicity (pipetting, drying, storage, cost and safety), protein removal efficiency, lipid coverage, repeatability of lipids measurements and recovery efficiency, which are necessary considerations in high-throughput workflows. The aim of the current chapter was to develop and apply the ideal sample preparation method for UPLC-MS lipid profiling for biological studies presented in chapter 5 and chapter 6.

62 Chapter 3-Results

3.2 Material and methods

3.2.1 Materials

Pooled human plasma was purchased from Sigma Aldrich (Dorset, UK). Or- ganic solvents used for the extractions and precipitations were HPLC grade and obtained from Sigma Aldrich (Dorset, UK). All mobile phases were prepared with LC-MS grade solvents, formic acid and ammonium formate from Sigma Aldrich (Dorset, UK). Bradford reagents were obtained for protein quantiﬁcation from Bio- Rad (Hertfordshire, UK). The ten internal and non-endogenous lipid standards; fatty acid FA(17:0); lysophosphocholine, LPC(15:0/0:0); phosphoglycerol, PG(15:0/15:0); phosphocholine, PC(15:0/15:0); phosphoethanolamine, PE(15:0/15:0); sphingomyelin SM(d18:1/17:0); phosphoserine, PS(17:0/17:0); Cer(d18:1/17:0); DG(17:0/17:0/0:0) D5; triacylglyceride, TG(15:0/15:0/15:0), were purchased from Avanti Polar Lipids (Alabaster, Alabama, USA).

3.2.2 Sample preparation for UPLC-MS: Precipitation

For each precipitation condition (Figure 3.1.A), ten replicate plasma samples (200 µL each) were precipitated by the addition of three volumes of organic solvent pre-cooled to -20°C. Selected solvents were: MeOH, ACN, IPA and ACN/IPA (1:2 v:v). Samples were vortex mixed for 1 min. After 10 min of incubation at room temperature, samples were stored overnight at -20°C to improve protein precipitation and then centrifuged at 14,000 g for 20 min. The supernatant was collected (600 µL) and stored at -80°C awaiting MS analysis. The sample was diluted to adjust the water content at 50% and analysed by UPLC-MS. The analytical workﬂow is summarised in Figure 3.1.A.

3.2.3 Sample preparation for UPLC-MS: Extraction

For each extraction condition (Figure 3.1.B), ten replicate plasma samples (200 µL each) were extracted in glass tubes by addition of the following organic sol-

63 Chapter 3-Results

vents: 600 µL MeOH/CHCl3 (4:4 v:v), 600 µL MeOH/CH2Cl2 (4:4 v:v), 1.5 mL of MeOH/MTBE-H2O (2:10:3 v:v:v) and 1.1 mL MeOH/hexane/IPA (3:7:1 v:v:v). Samples were vortex mixed for 1 min. After 10 min of incubation at room temperature, samples were stored overnight at -20°C to improve protein precipitation and then centrifuged at 14,000g for 20 min. Organic and aqueous phases were collected and 200 µL was dried in a vacuum centrifuge. The samples were stored at -80°C awaiting

MS analysis. Organic phases were reconstituted in 200 µL of IPA/ACN/H2O (2:1:1 v:v:v) and aqueous phases were reconstituted in ACN-H2O (1:1 v:v). The analytical workﬂow is summarised in Figure 3.1.B.

64 Chapter 3-Results

Figure 3.1: Sample preparation steps for (A) Precipitations: MeOH, ACN, IPA and IPA combined with ACN (B) Extractions: MeOH combined with (CH2Cl2), (CHCl3), hexane/IPA (Hexane) and MTBE/H2O (MTBE).

3.2.4 Protein quantiﬁcation

Proteins were quantiﬁed using the Bradford method (Bradford 1976). Protein concentration was estimated for each extraction and precipitation method. A standard curve of bovine serum albumin was prepared as a quantitative reference. Precipitated or extracted samples were ﬁrst dried in the vacuum centrifuge and solubilised with water in order to avoid solvent incompatibility with the Bradford reagent. Samples were incubated at room temperature for at least 5 min. Absorbance was measured at 595 nm.

65 Chapter 3-Results

3.2.5 Ultra Performance Liquid Chromatography

Chromatographic analysis was performed using an Acquity UPLC system (Waters Ltd, Elstree, UK). Precipitated and extracted samples (organic phase) were injected onto a C18 CSH column (100 x 2.1 mm, 1.7 µm; Waters) at 55°C. Flow rate was 400

µL/min. The mobile phase A consisted of: ACN/H2O(60:40, v:v) mixed with 10 mM ammonium formate and 0.1% formic acid and mobile phase B: IPA/ACN (90:10, v:v) mixed with 10 mM ammonium formate and 0.1% formic acid. The injection volume was 5 µL. The gradient use for this chromatographic approach (Table 3.1) allowed an eﬀective separation of the diﬀerent lipid species (Isaac et al. 2011).

Table 3.1: Chromatographic gradient used for UPLC-MS lipid proﬁling. Time(min) %A %B Curve Initial 60 40 Initial 2 57 43 6 2.1 50 50 1 12 46 54 6 12.1 30 70 1 18 1 99 6 18.1 60 40 6 20 60 40 1

3.2.6 Lipid proﬁling by UPLC-Q-ToF Mass Spectrometry

After separation by UPLC, mass spectrometry was performed using a Q-Tof Pre- mier (Waters, Manchester, UK) for global lipid profiling and a Xevo G2-S Q-ToF for the recovery study with an ESI (Waters, Manchester, UK). Dynamic range enhancement was applied throughout the MS run to improve isotopic distribution and mass accuracy and reduce high ion intensities. In positive ion-mode, MS parameters were as follows: capillary voltage was set at 2.5 kV, cone voltage at 30 V, source temperature 120°C, desolvation temperature at 400°C, desolvation gas flow at 800 L/h, cone gas flow at 20 L/h. Centroided data were acquired for each sample from m/z 100 to 1500. In negative ion mode, MS parameters were as follows: capillary voltage was set at 2.5 kV, cone voltage at 25 V, source temperature 120°C, desol-

66 Chapter 3-Results vation temperature at 500°C, desolvation gas ﬂow at 800 L/h, cone gas ﬂow at 25 L/h. Acquisition was performed from m/z 100 to 1500. For both ionisation modes, leucine enkephalin (m/z 556.2771 in ESI+, m/z 554.2615 in ESI-) was continuously infused at 30 µL/min and used as a lock mass correction.

3.2.7 Structural identiﬁcation

Metabolite annotation was made by searching m/z ratios against online databases such as METLIN (http://metlin.scripps.edu), Lipidmaps (http://www.lipidmaps.org) and HMDB (http://www.hmdb.ca) (Smith et al. 2005; Fahy et al. 2005; Fahy et al. 2009; Wishart et al. 2007). The mass error used was 5 ppm. Further structural elucidation was performed using collision-induced dissociation experiments, with (DDA) and without (MSE) selection of the precursor ion by the quadrupole of the Q-Tof mass spectrometer (Bateman et al. 2007; Rainville et al. 2007). Diagnostic fragments of the polar head group or the fatty acyl chains were investigated to conﬁrm the annotation proposed by the databases and discriminate isomers.

3.2.8 MS data pre-processing

MassLynxTM software version 4.1 was used for data acquisition and analysis. Waters raw data ﬁles were converted to NetCDF format and the metabolite features reported hereafter were entirely generated using the freely available data analysis software package R (v2.11)/XCMS (v1.24.1) to preprocess the raw data. MassLynxTM software version 4.1 was used for data acquisition and analysis. Target Lynx 4.1 was used to process the data from the lipid standards for the recovery experiment (Smith et al. 2006).

3.2.9 Multivariate statistical analysis

PCA was carried out on the XCMS extracted intensities using SIMCA P+ v13 (Umetrics, Ume˚a, Sweden) and Matlab (The Math-Works, Natick, MA). Spectra variables were scaled to unit-variance. Discriminating features between metabolic

67 Chapter 3-Results profiles were identified for each model by displaying loadings plots (model coefficients vs. covariance).

3.2.10 Univariate statistical analysis

Variability of sample preparation was also assessed by univariate statistics. The CV deﬁned as the standard deviation divided by the mean was calculated from the mean intensities of each feature detected in the ten replicates. A histogram of the CV values was plotted for each sample preparation protocol. Instrumental variability was estimated by computation of the CV in quality control (QC) samples, constituted by pooling 10 µL of each of the 80 extracts. Standard deviations were also calculated for each lipid standard in the recovery study.

3.2.11 Lipid recovery

Recovery for each of the eight methods was evaluated by spiking a 20 ng/µL mixture of the ten internal and non endogenous lipid standards into the sample before (pre-spiked) and after (post-spiked) the sample preparation method. The repeatability of the method was conﬁrmed with six pre-spiked samples and six post-spiked samples. Prior analysis of the ten standards by MSE and DDA modes allowed the characterisation of m/z and retention time of the molecular ion adducts and fragments. For each standard the most intense ion peak was selected for the recovery calculations.

68 Chapter 3-Results

3.3 Results

To identify the most suitable procedure for plasma sample preparation for UPLC- MS lipid proﬁling, an exhaustive assessment of quantitative criteria for benchmarking the simplicity, protein removal eﬃciency, lipid coverage, repeatability and recovery, of each sample preparation methods was developed. Together, these criteria were used to select the optimal method from the eight sample preparations studied herein.

3.3.1 Simplicity of protocols

Careful inspection of protocol steps showed that biphasic extractions are more time- and resource-consuming than precipitation methods (Figure 3.1). - Pipetting. For extraction methods based on halogenated solvents, collection of the lipid extract fraction required careful pipetting of the lower organic phase while avoiding any contamination by proteins at the interface, which could cause instrumental failure and decrease throughput. MeOH/MTBE-H2O and MeOH/hexane/IPA protocols make this step slightly easier, since the lipid extract is on top of the aqueous phase and the protein pellet at the bottom of the tube after centrifugation. These methods still require the interface of the aqueous and lipid layer to remain undis- turbed while collecting the lipid extract. Protein precipitation protocols resulted in a monophasic mixture of solvents with a protein pellet at the bottom of the tube, thus avoiding all the issues associated with the presence of an interface and decreasing risks of contamination of the lipid extract with proteins. In the current study, pipetting was manual, but these pipetting steps are critical when robotic sample preparation is considered. For instance, handling low viscosity solvents used in biphasic extraction protocols is not straightforward, as is the collection of organic phases located at the bottom of the container. - Drying. Extraction protocols also involve a drying step followed by a reconstitution by a mixture of solvents close to the initial mobile phase of the chromatographic system. This is due to the fact that direct injection of chlorinated solvents can result in poor chromatographic separation of lipid classes and can be detrimental to some

69 Chapter 3-Results parts of the autosampler and cause leakage of contaminants from polyetheretherke- tone (PEEK) tubing. This solvent exchange extends the time of sample preparation and increases the risk of introducing procedural error or altering the metabolic proﬁle by loss of volatile metabolites or degradation of labile molecules. - Storage. In case immediate analysis is not possible, one must also consider the stability of the extract when stored at low temperature (Zivkovic et al. 2009). Lipid losses occurs with MeOH and ACN precipitation; chromatogram inspection showed no signiﬁcant loss of lipids for IPA, IPA-ACN, CH2Cl2, CHCl3, MTBE and hexane sample preparations. This has also been reported in the literature. One of the direct advantages is that the IPA precipitation supernatant is stable until injection, therefore bypassing unnecessary sample preparation before re-injection in the case of instrumental failure (Figure 3.2) (Vuckovic 2012; Wolf et al. 2008; Yang et al. 2013).

Figure 3.2: Chromatograms of plasma samples precipitated with MeOH, ACN, IPA and IPA ACN obtained in positive mode respectively before A, B, C and D, and after storage at -80°C E, F, G and H.

- Cost. Aiming at deploying these methods in an ultra-high throughput workﬂow, particular attention needs to be paid to solvent cost and the impact of the solvents on

70 Chapter 3-Results the environment (i.e. preference for ”green” solvents). First, extraction solvents are, at least twice as expensive as solvents for precipitation with ACN being three-times more expensive than IPA and MeOH, which are used for precipitation. Second, compared to precipitation methods, biphasic extraction methods require greater solvent volumes, longer labour time, use of glass vials and dedicated instrumentation for the drying step (centrifugal vacuum concentrator or under nitrogen to evaporate organic phases). For instance, we estimated that consumables for extraction methods are roughly 5 times more expensive than precipitation methods, whilst the total time for manual interventions can be 2 to 3 times longer than precipitation methods. - Safety. Finally, the solvents used in extraction and in precipitation protocols (MeOH and ACN) are particularly harmful to human health: chloroform is known to be carcinogenic and n-hexane is neurotoxic (Frenia et al. 1993; Winslow et al. 1978; Takeuchi et al. 1980). In contrast, an alcohol such as isopropanol is reasonably safe making its handling easier. Altogether, the addition of a protein precipitant solvent and the removal of the total liquid contents from a solid pellet is a simple and straightforward procedure, which is suitable for high-throughput workﬂows. However, a systematic quantitative assessment is needed to make a clear recommendation.

3.3.2 Protein removal eﬃciency

Plasma composition is complex and the predominant critical step for sample preparation is to remove proteins. The presence of proteins in a liquid sample poses a challenge to the analytical instrumentation, decreasing column lifetime, causing ion suppression, and thus impacting on data quality (Polson et al. 2003). After each procedure the percentage of proteins remaining in the supernatant was evaluated with the Bradford method (Table 3.2) (Bradford 1976; Want et al. 2006). The resid- ual amount of proteins in samples was measured at 1% for precipitations and for

MeOH/MTBE-H2O and MeOH/hexane-IPA extractions in the organic phase. How- ever, MeOH /CH2Cl2 and MeOH /CHCl3 extractions were less eﬃcient as 5% of

71 Chapter 3-Results proteins still remain in the organic phase. These results are in general agreement with previous studies on sample preparation for plasma (Polson et al. 2003; Lai et al. 2009; Want et al. 2006). MeOH precipitation is the standard method of plasma sample preparation because of the eﬃcient protein precipitation (98%). To improve protein precipitation the solvents were kept at 4°C and added slowly to the sample (Bruce et al. 2009; Bird et al. 2011; Gika et al. 2011). It was also important that the mixture samples-solvent were stored overnight at -20°C before centrifugation in order to maximise the precipitation of the large amount of proteins.

Table 3.2: Percentage protein precipitation eﬃciency for the eight sample preparation methods after 2 h and 24 h. Precipitations: MeOH, ACN, IPA and IPA combined with ACN. Extractions: MeOH combined with CH2Cl2, CHCl3, hexane/IPA (Hexane) and MTBE/H2O (MTBE). Mean +/- SEM were calculated for three batches of sample preparations.

MeOH ACN IPA IPA ACN CH2Cl2 CHCl3 MTBE Hexane Protein removal eﬃciency % 2 hours 99.74 99.51 99.78 99.81 96.65 96.82 98.19 95.39 Standard deviation 0.86 1.35 0.02 0.09 0.08 0.09 0.02 0.16 Protein removal eﬃciency % 24 hours 99.77 99.57 99.82 99.82 97.05 97.14 98.22 98.55 Standard deviation 0.73 1.45 0.08 0.02 0.20 0.09 0.02 0.02

3.3.3 Lipid coverage

Inspection of chromatograms indicated that plasma samples precipitated with ACN did not achieve a suitable detection level of sphingomyelins (SMs), DGs and TGs (Figure 3.3.B). Likewise, we observed that MeOH/hexane/IPA extraction did not allow the detection of lysophospholipids (Figure 3.3.H). However, samples precipitated with MeOH (Figure 3.3.A) and IPA (Figure 3.3.C) or samples extracted with MeOH combined with CH2Cl2 (Figure 3.3.E), CHCl3 (Figure 3.3.F) and MTBE/H2O (Figure 3.3.H) oﬀered broader lipid coverage.

72 Chapter 3-Results

Figure 3.3: Representative chromatograms of plasma lipid proﬁling in positive mode Pre- cipitations: methanol (MeOH) A, acetonitrile (ACN) B, isopropanol (IPA) C and IPA combined with ACN D, Extractions: MeOH combined with dichloromethane (CH2Cl2) E, chloroform (CHCl3) F, hexane/IPA (Hexane) G and methyl-tert-butyl ether/H2O (MTBE) H.

PCA was used to evaluate the differences between the sample preparation methods in ESI+ and ESI- modes (Figure 3.4). Each dataset was comprised of spectra acquired on forty aliquots (five sample preparation replicates for each sample preparation method). Firstly, the repeatability of the UPLC-MS run was verified by multiple injections of the pooled quality control (QC) sample constituted by pooling a 10 µL aliquot from all the samples: QCs were tightly clustered in the middle of the PCA scores plots, which confirmed the reliability of the analytical platform. As anticipated after initial inspection of the chromatograms, the unsupervised PCA model clearly indicated that the eight plasma sample preparation methods generated different LC-MS profiles. Moreover, in negative mode PC1 showed a separation between extractions and precipitations. Confounding effect of the drying step in the extraction protocol, which could potentially result in loss of analytes cannot be ruled

73 Chapter 3-Results

out as accounting for this separation. The MeOH/MTBE/H2O and MeOH/hexane-

IPA extracted samples were better clustered than MeOH /CH2Cl2 and MeOH /CHCl3 extracted samples in the scores plot of PC1 vs. PC2. This difference may be due to the physical location of the lipids during the biphasic extraction: withdrawing the organic phase directly at the top may confer a better clustering (for MeOH/MTBE-H2O and MeOH/hexane-IPA). Precipitation resulted in tighter clustering in the PCA plot than extraction. This broad-sense repeatability can be interpreted as being related to the simplicity of the protocols (which only involves the addition of a single solvent). The main sources of variation in the data were investigated, by inspecting PCA scores plots (Figure 3.4.A-B) and loadings plots (Figure 3.4.C-D). To interpret the loadings, putative identification from the different databases of the features detected was carried out. Detailed inspection of features resulted in filtering out noisy and unknown features (Bird et al. 2011; Fahy et al. 2005). Data-Dependent Analysis (DDA) and MSE result in the characterisation of lipid species in both positive and negative ion modes. Numerous lipid classes were identified in lipid profiling such as lysophosphatidylcholi- nes (LPCs), lysophosphatidyl- ethanolamines (LPEs), phosphatidylinositol (PIs), phosphatidylcholines (PAs), phosphatidylethanolamines (PEs), phosphatidylserines (PSs), phosphatidylglycerols (PGs), sphingomyelins (SMs), Cers, DGs, TGs and cholesteryl esters (CEs). Altogether, PCA loadings plots (Figure 3.4.C-D) showed that the main difference between sample preparation methods is due to the intrinsic lipid selectivity of each solvent.

74 Chapter 3-Results

Figure 3.4: Scores plots (A and B) and loading plots (C and D) obtained from precipitation and extraction of samples analysed by lipid proﬁling in positive mode and negative mode. In PC1 and PC2 precipitation protocols lead to tight clustering of the samples while extraction protocols exhibit larger variability.

75 Chapter 3-Results

The Venn diagrams in Figure 3.5 summarise the lipid selectivity of each sample preparation method. Analysis for both polarities lead to the detection of 724 features including 185 identiﬁed lipids. The peak intensities of the 185 lipids were compared between each method of sample preparation. As demonstrated in Figure 3.5.A, IPA precipitation enables the inclusion of all classes of lipids and in particular improved TG detection compared to the other precipitations. These results also demonstrate that extraction with a combination of MeOH and CHCl3 resulted in an increase of lysophospholipids and phospholipids compared to the other extraction methods

(Figure 3.5.B). Moreover, it appears that MeOH combined with CHCl3 only extracted 6 more PCs than IPA, whereas the IPA precipitation protocol considerably increased our ability to study TGs (Figure 3.5.C). Clearly each lipid has a speciﬁc interaction with the solvent according to their solubility. This may explain the diﬀerences of extraction observed between phospholipids (amphipathic) and TG (neutral). However, PCs are more easily observed than TGs as they ionised well in both modes and were more broadly represented than TGs in plasma. Altogether, these results imply that

IPA has a similar lipid selectivity and coverage as MeOH combined with CHCl3 and is clearly superior in terms of standard recovery when comprehensive analysis of all lipid classes is required.

Figure 3.5: Venn diagrams show the overlapping selectivity between extractions (A) precipitations (B) and IPA vs. MeOH combined with CHCl3 (C) based on identiﬁed lipids. Lipid species detected after using the IPA protocol are highly similar to those detected after using the well-established CHCl3/MeOH-based protocols.

76 Chapter 3-Results

3.3.4 Repeatability

Untargeted lipid profiling is based on the relative comparison of spectral profiles and therefore a systematic QC strategy is necessary to assess repeatability of analytical workflows. All sample preparation methods were tested for their ability to provide similar profiles between replicates under the same conditions: each sample was split into 5 aliquots, which were extracted or precipitated independently. Stable retention times were obtained after conditioning the column with multiple (n = 15) injections of QC samples as advised in previous work (Want et al. 2013). Typical variation of the retention times with an Acquity UPLC under these lipid profiling conditions did not exceed 0.3 seconds (CV around 0.16%) which is critical for correct data preprocessing and feature identification. The CV distribution for the detected features (RT and m/z) in the pooled QC samples indicated that the instrument and samples were remarkably consistent over time. In fact, 66.6% of the features from the pooled QC had a CV lower than 20%, which defines the instrumental repeatability estimate. To compare the CV distributions, we implemented a t-test on CV distribution frequencies (Table 3.3) and demonstrated that all sample preparation methods were significantly different from the sample-preparation of QC (p<0.05), apart from IPA (p=0.892) and IPA ACN (p=0.360).

Table 3.3: Student’s t-test of the mean of the log-transformed distributions of each sample preparation method.

MeOH ACN IPA IPA ACN CH2Cl2 CHCl3 MTBE Hexane p-value 9×10-171 8.9×10-134 0.892 0.360 0.049 1.04 ×10-12 3.58×10-67 5.7×10-216 Mean of the data 2.11 2.22 2.77 2.75 2.74 2.92 2.36 2.00

To evaluate the repeatability of each sample preparation method, the CVs for sample preparation were calculated for features detected in the ﬁve replicates and independently prepared in diﬀerent batches for each method (Figure 3.6 and Table 3.4). The results provide compelling evidence that IPA enabled a robust lipid extraction since 61.1% of the features had a CV lower than 20%. The IPA repeatability at 20% threshold (61.1% of features with a CV<20%) was similar to the intrinsic instrumentation repeatability determined with the pooled QC (66%), whereas the

77 Chapter 3-Results other methods are less repeatable (< 55.4%). Interestingly, the IPA protocol was not only the best precipitation method for lipid proﬁling but also presented a better data reproducibility (CV distribution) than the well-established MeOH/MTBE/H2O and

MeOH/CHCl3-based extraction protocols.

Figure 3.6: Histograms of the CVs to assess repeatability of the eight methods. Quality control QC, Precipitations; MeOH, ACN, IPA, IPA ACN, and Extractions MeOH combined with CH2Cl2 or CHCl3, MTBE/H2O (MTBE) and, hexane/IPA (Hexane).

Table 3.4: Percentage of features under and below 20 per-cent of variation coeﬃcients for the eight sample preparation methods. Precipitations: MeOH, ACN, IPA and IPA combined with ACN. Extractions: MeOH combined with CH2Cl2, CHCl3, hexane-IPA (Hexane) and MTBE-H2O (MTBE) MeOH ACN IPA IPA ACN Precipitations <20% 49.9 55.4 61.1 50.4 >20% 50.1 44.6 38.9 49.6 CH2Cl2 CHCl3 Hexane MTBE-H2O Extractions <20% 27.0 38.2 40.7 36.2 >20% 73.0 61.8 59.3 63.8

3.3.5 Lipid recovery

The eﬃciency of each one of the eight sample preparation methods was then tested for the capacity to quantitatively recover ten lipid standards spiked into the samples before and after the treatment (Table 3.5). Lipid standards selected for this

78 Chapter 3-Results purpose were non-endogenous lipids (i.e. odd-carbon chains) and their structures were similar to endogenous lipids found in human plasma (Wolk et al. 2001). The recovery was calculated as the ratio of peak areas of the internal standards in sample spiked before sample-preparation (pre-spiked samples, deﬁning the amount of exogenous standard detected) to those added after sample-preparation (post-spiked samples used to deﬁne the amount of detectable exogenous standard, see Figure 3.7 and Table 3.6).

Table 3.5: List of lipid standards used for the recovery study.

Name Formula Positive mode m/z Negative mode m/z - FA(17:0) C17H34O2 N/A N/A [M-H] 269.2474 + - LPC(15:0/0:0) C23H48NO7P [M+H] 482.3246 [M-H] 480.3094 + - PG(15:0/15:0) C36H70O10PNa [M+H] 717.4677 [M-H] 693.4710 + - PC(15:0/15:0) C38H76NO8P [M+H] 706.5386 [M-H] 704.5234 + - PE(15:0/15:0) C35H70NO8P [M+H] 664.4916 [M-H] 662.4764 + - SM(d18:1/17:0) C40H81N2O6P [M+H] 717.5906 [M-H] 715.5754 + - PS(17:0/17:0) C40H77NO10P [M+H] 764.5440 [M-H] 762.5290 + - Cer(d18:1/17:0) C35H69NO3 [M+H] 552.5356 [M-H] 550.5204 + DG(17:0/0:0/17:0) D5 C37H67D5O5 [M+NH4] 619.6026 N/A N/A + TG(15:0/15:0/15:0) C48H92O6 [M+Na] 787.6782 N/A N/A

As initially expected, the area under the peak for internal standards in pre-spiked samples was lower than in post-spiked samples (≈12% difference), which allowed the derivation of a sample preparation yield. However, some of the standards were only observed in one mode. This is typically the case for the FA(17:0) standard, which was only observed in ESI- mode. Likewise, the standards for TG(15:0/15:0/15:0) and DG(17:0/0:0/17:0)D5 were only observed in positive mode. The other lipid fam- ilies (lysophosphocholines, phosphoglycerols, phosphatidylcholines, phosphatidylethanolamines, sphingomyelins, phosphatidylserines and Cers), can ionise in both positive and negative modes and in this case, although, there were small variations in recovery and were not significant between positive and negative modes. Finally, as for variations between sample preparation methods, these results provide key under- reported data and fills a huge gap in terms of sample preparation yield. As previously observed in the chromatograms (Figure 3.3.G), MeOH/hexane/IPA extraction only allowed a selective extraction, which resulted in losing lysophospholipids, phospholipids and SMs, as demonstrated by their poor recovery (<20%). Similar conclusions were drawn for ACN sample preparation methods regarding the SM, DG and TG

79 Chapter 3-Results lipid species (Figure 3.3.B). Traditional MeOH precipitation presented comparable recoveries to the ACN precipitation. However, the PC recovery appeared to be better with ACN (>80%) compared to MeOH method (<60%). Lipid extraction by

MeOH/(CH2Cl2), MeOH/CHCl3 and MeOH/MTBE-H2O yielded a recovery higher than 50% for most of the lipid species but some exceptions such as TG recovery decreased the universality of these methods. Conversely, IPA precipitation oﬀered a highly consistent and reliable recovery (>60-80%) for the ten lipids standards tested. These results indicated that IPA precipitation had signiﬁcantly higher recovery than the other methods.

Table 3.6: Summary of recoveries obtained for 10 standards for each sample preparation method

PRECIPITATIONS EXTRACTIONS MeOH ACN IPA IPA ACN CH2Cl2 CHCl3 Hexane MTBE-H2O NEG POS NEG POS NEG POS NEG POS NEG POS NEG POS NEG POS NEG POS FA(17:0) Standard deviation 22.5 18.1 1.5 3.3 2.7 12.1 18.2 1.0 CV 27.2 21.2 11.1 15.7 26.3 34.0 21.6 23.6 Recovery 76.4 54.2 85.3 57.5 21.6 52.1 69.4 3.9 LPC(15:0/15:0) Standard deviation 20.7 8.6 6.1 21.9 11.0 16.8 6.9 18.4 13.4 8.4 4.5 8.5 5.6 15.2 0.3 1.2 CV 16.3 9.7 11.1 19.7 13.1 13.1 7.7 17.5 38.8 17.2 7.5 16.2 16.8 15.7 40.7 26.1 Recovery 97.0 91.8 103.4 101.0 107.6 92.7 87.7 101.6 46.7 48.5 59.1 58.1 44.7 48.4 0.5 4.2 PG(15:0/15:0) Standard deviation 20.3 21.5 12.6 20.5 14.6 18.2 12.9 18.2 6.4 14.1 7.8 15.0 36.2 21.8 0.5 0.6 CV 38.1 42.0 8.3 23.4 14.2 20.1 34.7 19.1 15.1 21.2 8.5 24.2 50.3 35.0 22.1 40.4 Recovery 53.3 51.1 89.0 90.9 91.1 83.8 98.9 100.5 51.7 58.3 56.1 44.5 45.3 46.5 1.5 0.9 PC(15:0/15:0) Standard deviation 21.4 13.0 9.5 7.8 15.3 21.6 13.6 13.5 2.1 15.6 4.6 4.3 0.9 19.6 0.1 2.1 CV 37.8 24.6 22.0 24.3 15.1 17.7 11.8 17.7 24.4 31.8 8.4 8.4 16.1 21.2 22.9 23.8 Recovery 47.5 44.4 18.6 26.3 91.7 65.5 72.5 81.8 73.6 48.4 54.2 51.1 63.5 67.7 3.8 7.8 PE(15:0/15:0) Standard deviation 4.3 0.7 0.8 4.5 11.5 19.1 8.2 16.1 7.1 15.3 8.0 3.3 11.9 19.8 2.5 1.2 CV 47.0 24.3 22.4 54.8 9.0 21.3 13.2 20.0 17.7 20.4 12.4 6.3 13.9 22.7 19.9 22.7 Recovery 7.4 2.9 6.1 3.3 92.2 69.3 67.5 57.0 65.2 107.8 63.8 61.8 104.5 101.7 13.0 8.0 SM(d18:1/17:0) Standard deviation 19.2 15.1 9.8 30.1 17.2 21.8 13.4 13.3 6.7 9.2 5.2 7.7 8.9 18.1 1.0 1.1 CV 49.9 32.9 33.9 80.6 16.7 26.7 10.7 20.6 18.3 14.9 8.6 13.4 15.8 8.8 26.4 20.9 Recovery 33.8 37.1 31.1 21.0 104.8 67.8 77.8 71.2 50.8 62.6 59.4 59.4 69.2 55.5 3.5 2.8 PS(17:0/17:0) Standard deviation 26.0 5.9 0.6 1.3 13.7 20.7 5.3 8.0 5.8 7.4 7.6 11.4 8.6 36.0 0.9 0.6 CV 40.1 33.8 20.5 40.7 19.2 22.1 23.9 28.6 23.7 31.5 23.8 14.8 17.2 30.8 28.6 35.0 Recovery 20.3 18.4 2.0 2.5 65.5 80.7 19.5 23.9 47.6 48.1 63.6 76.1 63.3 91.4 2.2 2.0 Cer(d18:1/17:0) Standard deviation 5.5 3.3 1.5 4.2 21.7 20.8 13.3 19.2 16.2 18.1 11.5 13.5 17.1 17.7 26.3 10.2 CV 48.5 45.9 51.2 73.4 25.9 35.5 30.7 31.5 46.0 40.3 40.7 65.5 23.6 75.5 36.8 36.3 Recovery 8.5 6.1 2.4 3.6 86.5 64.8 55.2 70.3 91.0 81.5 96.1 78.5 85.4 78.8 82.8 113.9 DG(17:0/0:0/17:0) D5 Standard deviation 15.7 5.6 19.5 22.6 19.6 30.5 23.4 13.1 CV 40.6 25.7 24.4 25.4 26.6 54.0 34.1 47.7 Recovery 25.6 40.1 100.2 96.9 67.2 91.1 96.7 104.6 TG(15:0/15:0/15:0) Standard deviation 4.3 9.4 0.1 20.6 16.0 19.7 23.6 4.3 CV 57.8 70.4 22.0 27.6 50.2 50.7 54.4 31.2 Recovery 7.9 11.0 69.5 50.7 19.8 21.4 45.4 20.7

80 Chapter 3-Results Histogram summarising the data obtained from the lipid recoveries. Samples are compared according to the ESI +/- mode and sample Figure 3.7: preparation method. The error bars show the standard deviation of the replicates recovery.

81 Chapter 3-Results

3.4 Discussion

In this chapter several sample preparation methods for lipid proﬁling were validated using various assessment criteria: throughput, protein removal eﬃciency, selectivity, repeatability and recovery (Table 3.7 and Table 3.8). The number of steps involved in each preparation method varied between two and four for precipitations and extractions respectively. As a result, protocols can be ranked by decreasing throughput as follow: MeOH≈ACN≈IPA > > MeOH/MTBE-H2O≈ MeOH/hexane-

IPA > MeOH/CHCl3≈MeOH/CH2Cl2. Moreover, strong variations in selectivity were observed by multivariate analysis between the different methods, providing evidence of significant loss for some lipid species. In a metabonomic and metabolic phenotyping context, repeatability is one of the most important criteria to improve the reliability of the various analytical assays leading to the identification of potential biomarkers. Performance of analytical methods is evaluated using the US Food and Drug Administration (FDA) (FDA 2013) or European Medicine Agency (EMA) guidances (Medicinal Products for Human Use et al. 2011). For absolute quantification of xeno- biotics the current threshold for repeatability is 15%, but for endogenous biomarker discovery, more flexibility is allowed (currently 30%); it has finally been suggested that a threshold of 20-25% is adequate to evaluate the CV of the assay (Cummings et al. 2008; Cummings et al. 2010). In this respect, precipitation methods have proven to be valuable alternatives to extraction protocols. Protein removal procedures involving acid, heat and organic solvents were exten- sively compared in literature (Want et al. 2006; Fahy et al. 2009; Want et al. 2013; Souverain et al. 2004; Sivaraman et al. 1997; Khan et al. 1991). Generally, denaturation of proteins by acid or heat is limited in reproducibility of the features and metabolome coverage. Want et al. and Bruce et al. studied the impact of different plasma to organic solvent ratios on chromatograms. These studies concluded that organic solvent precipitations are more efficient and more robust compared to acid or heat removal. A plasma to organic solvent ratio of 1:3 was found to be the upper limit in terms of protein removal efficiency (Polson et al. 2003). We therefore selected

82 Chapter 3-Results this ratio for the four precipitation methods. Our current findings expand prior work as we conducted an extensive comparison between extraction and precipitation method (Want et al. 2013; Theodoridis et al. 2012; Bruce et al. 2009; Reis et al. 2013). Folch or Bligh and Dyer protocols have long been the preferred method of plasma sample preparation for LC-MS analyses (Niessen 2006; Bligh et al. 1959; Folch et al. 1957). However, there are concerns about safety issues with extraction solvents and requirement of less toxic solvent for plasma sample preparation. In addition, these results clearly show that liquid-liquid extraction methods are complex to handle and less repeatable than precipitation methods. IPA precipitation enables protein removal in a single step procedure prior to lipid profiling. The presence of remaining metabolites in the monophasic solution does not affect UPLC-MS detection of lipids as the chromatographic gradient starts with aqueous solvent, which results in elution of the polar metabolites.

Table 3.7: Synopsis of the criteria evaluated for the eight sample preparation methods: Precipitations: MeOH, ACN, IPA and IPA combined with ACN and Extractions: MeOH combined with CH2Cl2, CHCl3, hexane/IPA (Hexane) and MTBE/H2O (MTBE).

Methods Advantages Constraints MeOH Broad lipid coverage Loss of DGs and TGs Poor recoveries ACN Two steps - Loss of SMs, DGs and TGs > 30-40% CV<20% Best repeatability < IPA 99% protein 61.1% features 20% - removal Broad lipid coverage

Precipitations Excellent recoveries >60-80% IPA ACN Recoveries ≈70% Loss of TGs CHCl3 - Broad lipid 95% protein Organic phase in bottom CH2Cl2 coverage removal MTBE - Loss of Organic phase Recoveries ≈60% 99% protein phospholipids, CV<25% Multiple steps - lysophospholipids

Extractions in top removal and SMs Hexane - Poor recoveries >20%

Table 3.8: The eight methods were scored against the evaluation criteria. Scores are in the range from 0 to 1: 0 if the method does not meet evaluation criteria, 0.5 if the method partially meets evaluation criteria and 1 if the method fully meets evaluation criteria

MeOH ACN IPA IPA ACN CH2Cl2 CHCl3 Hexane MTBE-H2O pipetting 1 1 1 1 0 0 0.5 0.5 drying 1 1 1 1 0 0 0 0 Simplicity (1 mark each) storage 0 0 0.5 0.5 0.5 0.5 0.5 0.5 cost 1 0 1 0 0 0 0 0 safety 0 0 1 0 0 0 0 0 Protein removal eﬃciency (3 marks) 3 3 3 3 1.5 1.5 3 3 Repeatability (5 marks) 2.5 2.5 5 2.5 0 2.5 2.5 2.5 Lipid coverage (2 marks) 2 1 2 2 1 2 1 1 Lipid recovery (5 marks) 0 0 5 2.5 2.5 2.5 2.5 0 Total scores 11.5 9 20.5 13.5 6.5 10.5 11 8

83 Chapter 3-Results

3.5 Conclusion

Ultimately, taking into account the above-mentioned criteria we found the most suitable sample preparation for lipid profiling was isopropanol precipitation (Table 3.8). Table 3.9 presents a weighted score system for all criteria. High-throughput Phenome centers operate with a certain set of constraints such as the necessity of running thousands of samples per year, and therefore requires simplicity, protein removal efficiency, lipid coverage, repeatability of lipids measurements and recovery efficiency. However, research environments with less emphasis on sample throughput may settle for different analytical settings. These results provide compelling evidence that the isopropanol precipitation method is a straightforward one-step procedure. Isopropanol sample preparation allows the analysis of a wide range of lipids, which is critical for optimal biological interpretation. In conclusion, isopropanol precipitation is a promising sample preparation method for large-scale lipidomic studies and is easily amenable to molecular epidemiology studies involving thousands of samples in high-throughput analytical workflows.

84 Chapter 4

UPLC-MS/MS targeted method for bile acids

4.1 Introduction

Bile acids are major components of bile, synthesised from cholesterol in the hepatocytes that play fundamental roles in many physiological processes. BAs are well known as powerful emulsiﬁers of dietary lipids in the intestine (Tso 1985; Cai et al. 1997), antimicrobial agents (Hofmann et al. 2006; Inagaki et al. 2006), and signalling molecules, regulating their own synthesis and various metabolic pathways (cf. Chap- ter 1) (Parks et al. 1999; Thomas et al. 2008). BAs are known to be modulated by the gut microbiota (Anizan et al. 2011; Chiang 2003; Claus et al. 2008; Ridlon et al. 2006), and perturbations of the circulating BAs pool have been shown to contribute to development of gastrointestinal, hepatic and metabolic diseases (Duboc et al. 2013; Jones et al. 2014; Ryan et al. 2014; Watanabe et al. 2011). Numerous reports have shown the structural diversity of BAs from cholesterol catabolism in the liver to microbial transformations in the intestine (cf. Chapter 1) (Mahato et al. 1994; Russell 2003). On top of their above-mentioned physicochemical properties, BAs are co-metabolised by the host (endogenous metabolism) and by symbiotic microbiota (gut bacterial metabolism), highlighting the complexity of the molecular crosstalk between the human host and its intestinal microbiota. The pool of BAs is made of primary BAs synthesised in the liver which are CA and CDCA, secondary BAs produced mainly in the gut via modiﬁcation of primary BAs (dehy-

85 Chapter 4-Results droxylation, epimerisation and oxidation), and tertiary BAs which are formed in both the liver and gut microbiota via modification of secondary BAs, by processes such as sulfation, glucuronidation, glucosidation and N-acetylglucosaminidation (Marschall et al. 1992). In the intestinal lumen, gut microbiota are free to modulate the hepatic output of BAs through various reactions, which include deconjugation and dehydroxylation at specific sites to form secondary BAs. All BAs are subject to a cycle of absorption, modification in the liver (further conjugation), and excretion back into the gastrointestinal tract in a process known as enterohepatic circulation. BA anabolism and biotransformation is thus a complex iterative process yielding a wide range of molecular variants. The abundance of these BA species also occupies a wide concentration range in humans with respect to biofluid and physiological/pathological phenotype (0.1-3000 nM). Given the biological significance of this diverse pool of molecules, a reliable platform for specific and sensitive analytical measurement is therefore needed. However, the chemical diversity of BAs (Moschetta et al. 2005), the wide range in abundance of BAs in biofluids, and the molecular complexity of the biofluids themselves all pose analytical challenges for sensitive and selective analysis of BAs (Griffiths et al. 2010; Street et al. 1983; Shimada et al. 2001). Within the past decade, LC-MS has been heavily utilised for the separation and detection of BAs in human and animal model biofluid samples (Duboc et al. 2013; Xie et al. 2015; Scherer et al. 2009; Want et al. 2010; Humbert et al. 2012; Steiner et al. 2010; Perwaiz et al. 2001; Sayin et al. 2013; Alnouti et al. 2008; Bathena et al. 2013; Huang et al. 2011; Garc´ıa-Canaveras˜ et al. 2012; Humbert et al. 2012; Sergi et al. 2012; Qiao et al. 2012; Ye et al. 2007; Bobeldijk et al. 2008; Chen et al. 2011; John et al. 2014). The most comprehensive of these methods have a relatively long analytical cycle (> 30 min), precluding rapid sample analysis (Steiner et al. 2010; Bathena et al. 2013; Humbert et al. 2012; Qiao et al. 2012; John et al. 2014). On the other hand, the most rapid methods (< 10 min) have limited BA coverage (Scherer et al. 2009; Chen et al. 2011). Within the past decade, LC-MS has been heavily utilised for the separation and detection of BAs in human and animal model

86 Chapter 4-Results biofluid samples (Duboc et al. 2013; Xie et al. 2015; Scherer et al. 2009; Want et al. 2010; Steiner et al. 2010; Perwaiz et al. 2001; Sayin et al. 2013; Alnouti et al. 2008; Bathena et al. 2013; Huang et al. 2011; Garc´ıa-Canaveras˜ et al. 2012; Humbert et al. 2012; Sergi et al. 2012; Qiao et al. 2012; Ye et al. 2007; Bobeldijk et al. 2008; Chen et al. 2011; John et al. 2014). Very few LC-MS methods report coverage of tertiary BAs such as sulfate conjugates which have recently been implicated in the important role of the gut microbiota in human metabolism (Duboc et al. 2013; Bobeldijk et al. 2008; John et al. 2014; Swann et al. 2011; Ridlon et al. 2014). The analytical foundation of the majority of these methods is reversed-phase separation utilising a C18 stationary phase, which is suitable for retention and separation of the diverse range of hydrophobicity present among BAs (Maekawa et al. 2014). Furthermore, all methods reviewed use methanol, acetonitrile, or a combination thereof as the strong eluent (Cai et al. 1997; Shimada et al. 2001; Inagaki et al. 2006; Ye et al. 2007; Want et al. 2010; Huang et al. 2011; Swann et al. 2011; Humbert et al. 2012; Duboc et al. 2013). Yet, the elution of ubiquitous blood matrix components such as phospholipids and triglycerides is known to benefit from the use of stronger solvents (e.g. 2-propanol, IPA), increased column temperature, or both (Ritchie et al. 2011). The methods reviewed, therefore, risk the accumulation of large amounts of lipid content on the column and consequential carry-over of intact lipids and a steady bleed of FAs during continuous sample analysis. Where present, these effects potentially compromise the BA measurement quality by competing for ionisation in the electrospray process. These effects have been observed within our laboratory when using such methodologies (Want et al. 2010). To combat this, a desirable LC separation should be directly compatible with lipid-rich matrices such as blood and bile, allowing for the analysis of minimally processed samples without sacrificing the excellent retention and separation of BA species provided by a reversed-phase mechanism. The ”dilute-and-shoot” method developed herein features chemical protein precipitation, removal of proteins by centrifugation, and direct analysis of the supernatant. This approach eliminates the potential for BA

87 Chapter 4-Results loss due to pre-processing steps such as SPE and the commonly applied procedure of sample drying prior to reconstitution with water (Want et al. 2010), methanol (Duboc et al. 2013; Shimada et al. 2001), or a combination thereof (Cai et al. 1997; Inagaki et al. 2006; Huang et al. 2011; Humbert et al. 2012; Ye et al. 2007; Swann et al. 2011; Ritchie et al. 2011; Palmer et al. 1971; Cerny et al. 2004; Mostarda et al. 2014). In this manner, selective enrichment or exclusion of BA species across a wide range of hydrophobicity is minimised or circumvented entirely. The method described herein represents substantial development beyond the many methods currently available in the literature. Using UPLC-MS, it combines a high-throughput analytical cycle (15 min) with simple and fast sample preparation, producing an analytical pipeline capable of supporting the rapid analysis of hundreds or thousands of samples. These goals are achieved without sacrificing the coverage of BA species, which is unprecedented in modern UPLC-MS applications. Of the 145 BA species detected by targeted tandem mass spectrometry (MS/MS), 89 were tertiary sulfated BAs synthesised for the development of this assay. Fifty-seven of the 145 BAs were quantitatively measured using 16 commercially-available stable isotope labelled standards. Their measurements were validated according to established FDA criteria for accuracy and precision on intra-/inter-day, linearity, carry-over, stability and matrix effect. Recoveries were evaluated for 16 deuterated BAs spiked in three human biofluids; plasma, serum and urine. In addition, this BA measurement method was validated by application to routine analysis of pre-prandial vs. post-prandial human plasma, serum and urine samples.

88 Chapter 4-Results

4.2 Materials and methods

4.2.1 Materials

Organic solvents (HPLC grade) used for the sulfation and precipitation, and sodium sulfate were obtained from Sigma Aldrich (Dorset, UK). All mobile phases were prepared with LC-MS grade solvents, formic acid and ammonium formate from Sigma Aldrich (Dorset, UK). Table 4.1 shows the 73 BA standards including 36 non- conjugated, 12 conjugated with taurine, 9 conjugated with glycine and 16 deuterated internal standards obtained from Steraloids (Newport, RI) and Medical Isotopes (Pel- ham, USA). Table 4.2 shows the 88 sulfated BA standards that were synthesised including 50 non-conjugated, 23 conjugated with taurine, 15 conjugated with glycine. BA name abbreviations are provided at the abbreviation section in this thesis, as well as in Table 4.1 and Table 4.2 (c.f. 4.2.4 UPLC-MS proﬁling and MS/MS conditions).

4.2.2 Collection of human plasma for targeted analysis

Blood plasma/serum and urine samples were collected from healthy participants (n=20) in either the pre- or post-prandial state. Ten participants were fasted overnight and blood and urine samples were collected after 12h (pre-prandial group). The remaining ten participants were given a high-fat meal and blood and urine samples were collected after 2h (post-prandial group). Serum and EDTA-plasma were centrifuged at 1,000 g and 4°C for 10 min and aliquoted and stored at -80°C. This study was given ethical permission for conduct in the NHS by St Mary’s Research Ethics Committee (Ref 09/H0712/82).

4.2.3 Sample preparation for targeted analysis of human pre-/

post-prandial study

Samples were received on dry ice and stored at -80°C until needed for preparation and analysis. All samples were thawed at 4°C, transferred to 1 mL Eppendorf

89 Chapter 4-Results

96-deepwell plates (Eppendorf, Stevenage UK) and centrifuged at maximum speed in an Eppendorf 5810 R equipped with an A-2-DWP-AT rotor (3,486 x g) for 15 min. Supernatant (50 µL) was transferred to 0.5 mL Eppendorf 96-deepwell plates and 150 µL of ice-cold methanol (1:3 v/v) were added to each sample (Saraﬁan et al. 2014).

Additionally, 10 µL of deuterated internal standards (2.1 µM) in H2O/acetonitrile/2- propanol (10:6:5, v:v:v) were added to the sample prior to the addition of methanol. All plates were heat sealed (Thermo Fisher Scientiﬁc, Hertfordshire UK), homogenised by vortexing for 15 min at 4°C using an Eppendorf MixMate (1400 RPM), and incubated at -20°C for 20 min. All samples were again centrifuged at 4°C (3,486 x g) for 15 min prior to decanting of 100 µL of supernatant to Eppendorf microplates for heat sealing and subsequent analysis.

4.2.4 UPLC-MS proﬁling and MS/MS conditions

BA analysis was performed by ACQUITY UPLC (Waters Ltd, Elstree, UK) coupled to Xevo TQ-S mass spectrometer (for targeted detection application) (Waters, Manchester, UK). The MS system was equipped with an ESI operating in negative ion mode (ESI-). The reversed-phase chromatographic method consisted of a mobile phase system, adapted from existing lipid proﬁling methods (Shockcor et al. 2011), paired with a shorter alkyl chain stationary phase (C8) to facilitate both the separation of BA species and the elution of lipidic matrix content. For this purpose, an ACQUITY BEH C8 column (1.7 µm, 100mm x 2.1 mm) was selected and used at an operating temperature of 60°C. The mobile phase solvent A consisted of a volumetric preparation of 100 mL acetonitrile added to 1L ultrapure water, with a ﬁnal additive concentration of 1mM ammonium acetate and pH adjusted to 4.15 with acetic acid. Mobile phase solvent B consisted of a volumetric preparation of acetonitrile and 2-propanol in a 1:1 mixture. The gradient separation is described in Table 4.3. Critically, the high organic wash step was adjusted in length for the complete elution of observable phospholipids and triglycerides, precluding their accumulation on column. The injection

90 Chapter 4-Results volume of all samples was 5 µL. To minimise injector carry-over, 3 wash cycles of weak (H2O: 2-propanol, 90:10) and strong (2-propanol) solvent preparations were performed simultaneously with sample analysis.

Table 4.1: Chromatographic gradient used for BA proﬁling and targeted analysis. Time(min) Flow Rate %A %B Curve Initial 0.6 90 10 Initial 0.1 0.6 90 10 6 9.25 0.6 65 35 6 11.5 0.65 15 85 6 11.8 0.8 0 100 6 12 0.95 0 100 6 12.1 1 0 100 6 12.4 1 0 100 6 12.45 0.85 45 55 6 12.5 0.85 90 10 6 12.6 0.8 90 10 6 12.7 0.7 90 10 6 12.8 0.6 90 10 6 15 0.6 90 10 6

Mass spectrometry parameters were as follows: capillary voltage was set at 1.5 kV, cone voltage at 60 V, source temperature 150°C, desolvation temperature at 600°C, desolvation gas ﬂow at 1000 L/h, cone gas ﬂow at 150 L/h. BA species yielding characteristic fragments when subjected to collision-induced dissociation were assayed using multiple reaction monitoring (MRM), while those that did not fragment were assayed by selected ion monitoring (SIM). The transitions for each of the BA standards and deuterated internal standards are provided in Table 4.1 and Table 4.2.

91 Chapter 4-Results

Table 4.2: UPLC-MS/MS settings for the 57 BA standards, including 36 non-conjugated, 12 conjugated with taurine, 9 conjugated with glycine and 16 deuterated standards. Individ- ual optimisation of selected ion monitoring (SIM) and multiple reaction monitoring (MRM) transitions, collision energy (CE) and retention time (RT) was implemented. Abbreviations of BAs in Figure 4.3.

Numbers Names and abbreviations SIM (m/z) CE (eV) RT (min) 1 Ursocholanic Acid 359.2955 10 11.58 2 Lithocholenic Acid 373.2748 10 10.62 3 5-Cholenic Acid-3β-ol 373.2748 10 10.13 4 3-Ketocholanic Acid 373.2748 10 10.65 5 Isolithocholic Acid 375.2904 10 10.45 6 Allolithocholic Acid 375.2904 10 10.31 7 3α,12α, 23-Nordeoxycholic Acid 377.2697 10 9.55 8 9(11), (5β)-Cholenic Acid-3α-ol-12-one 387.2540 10 8.13 9 5α-Cholanic Acid-3, 6-dione 387.2540 10 7.76 10 3,7-Diketocholanic Acid 387.2540 10 7.83 11 3,6-Diketocholanic Acid 387.2540 10 8 12 3,12-Diketocholanic Acid 387.2540 10 8 13 8(14),(5β)-Cholenic Acid-3α, 12α-diol 389. 2697 10 9.75 14 5β-Cholenic Acid-7α-ol-3-one 389.2697 10 9.81 15 5α-Cholanic Acid-3α-ol-6-one 389.2697 10 8.13 16 3α-Hydroxy-7 Ketolithocholic Acid 389.2697 10 8.45 17 3α-Hydroxy-12 Ketolithocholic Acid 389.2697 10 8.91 18 Lithocholic acid 375.2904 10 10.78 19 5β-Cholanic Acid-3β, 12α-diol 391.2853 10 8.93 20 Chenodeoxycholic acid 391.2853 10 10.22 21 Deoxycholic acid 391.2853 10 10.33 22 Hyodeoxycholic acid 391.2853 10 8.66 23 Isodeoxycholic Acid 391.2853 10 10.57 24 Murocholic Acid (MuroCA) 391.2853 10 7.33 25 Ursodeoxycholic acid (UDCA) 391.2853 10 8.13 26 3,7,12 Dehydrocholic acid 401.2333 10 3.75 27 3α-Hydroxy-7,12-DiketoCholanic Acid 403.2489 10 4.38 28 3α-Hydroxy-6,7-DiketoCholanic Acid 403.2489 10 8.21 29 5β-Cholanic Acid-3α, 6α-diol-7-one 405.2646 10 6.64 30 3 Dehydrocholic Acid 405.2646 10 7.26 31 12 Dehydrocholic Acid 405.2646 10 6.18 32 α Muricholic Acid 407.2802 10 6.64 33 β Muricholic Acid 407.2802 10 6.72 34 ω Muricholic Acid 407.2802 10 6.52 35 Cholic acid 407.2802 10 8.38 36 Hyocholic acid (HCA) 407.2802 10 7.83

92 Chapter 4-Results

Numbers Names and abbreviations SIM and MRMs (m/z) CE (eV) RT (min) 37 Glycoursocholanic Acid 416.3170 → 74 40 11.1 38 Glycolithocholic Acid (GLCA) 432.3119 → 74 40 9.86 39 Glycoursodeoxycholic Acid (GUDCA) 448.3068 → 74 40 5.23 40 Glycohyodeoxycholic Acid (GHDCA) 448.3068 → 74 40 5.6 41 Glycochenodeoxycholic acid (GCDCA) 448.3068 → 74 40 7.67 42 Glycodeoxycholic acid (GDCA) 448.3068 → 74 40 8.14 43 3,7,12 Glycodehydrocholic acid 458.2548 → 74 40 1.76 44 Glycocholic acid (GCA) 464.3017 → 74 40 5.85 45 Glycohyocholic Acid (GHCA) 464.3017 → 74 40 4.89 46 Taurolithocholic Acid (TLCA) 466.2996 → 80 40 9.03 47 Tauroursocholanic Acid 482.2945 → 80 60 9.15 48 Tauroursodeoxycholic Acid (TUDCA) 498.2894 → 80 60 4.68 49 Taurohyodeoxycholic Acid (THDCA) 498.2894 → 80 60 5 50 Taurochenodeoxycholic acid 498.2894 → 80 60 7 51 Taurodeoxycholic Acid (TDCA) 498.2894 → 80 60 7.45 52 3,7,12 Taurodehydrocholic acid 508.2374 → 80 60 1.45 53 Taurohyocholic Acid (THCA) 514.2843 → 80 60 4.32 54 Tauro-α Muricholic Acid (TαM) 514.2843 → 80 60 3.22 55 Tauro-β Muricholic Acid (TβM) 514.2843 → 80 60 3.22 56 Tauro ω-Muricholic Acid (TωM) 514.2843 → 80 60 3.18 57 Taurocholic acid (TCA) 514.2843 → 80 60 5.2 58 Lithocholic acid-d4 379.5893 10 10.81 59 Deoxycholic acid-13C 395.5886 10 10.34 60 Chenodeoxycholic acid-d4 395.5886 10 10.21 61 Ursodeoxycholic Acid-d4 395.5886 10 8.21 62 Hyodeoxycholic Acid-d5 396.5949 10 8.64 63 Cholic acid-d4 411.5881 10 8.98 64 Glycolithocholic Acid-d4 436.6406 → 74 40 9.96 65 Glycochenodeoxycholic acid-d4 451.5321 → 74 40 7.56 66 Glycodeoxycholic Acid-d4 452.6400 → 74 40 8.28 67 Glycoursodeoxycholic Acid-d4 452.6400 → 74 40 5.37 68 Glycocholic acid-d4 468.6394 → 74 40 7.9 69 Taurolithocholic Acid-d4 486.7209 → 80 60 8.75 70 Taurochenodeoxycholic acid-d4 502.7203 → 80 60 7.28 71 Tauroursodeoxycholic Acid-d5 503.7265 → 80 60 4.93 72 Taurocholic acid-d4 518.7197 → 80 60 5 73 Taurodeoxycholic Acid-d4 501.7124 → 80 60 7.15

93 Chapter 4-Results

Table 4.3: UPLC-MS/MS settings for mono-sulfated BA standards sorted according to sulfation ratio and retention time, including; 44 BAs sulfated in position OH-C3, 15 sulfated in position OH-C6, 21 sulfated in position OH-C7 and 8 sulfated in position OH-C12 sulfated. Individual optimisation of multiple reaction monitoring (MRM) transitions, collision energy (CE) and retention time (RT) was implemented. Abbreviations of BAs in Figure 4.3.

Numbers Names and abbreviations OH group MRMs (m/z) CE (eV) a b c 74 Lithocholenic acid 3-OH 453.2316 → 97 40 8.74 - - 75 5-Cholenic Acid-3β-ol 3-OH 453.2316 → 97 40 7.73 - - 76 Isolithocholic Acid 3-OH 455.2473 → 97 40 8.44 - - 77 Allolithocholic Acid 3-OH 455.2473 → 97 40 8.15 - - 78 Lithocholic acid (LCA-S) 3-OH 455.2473 → 97 40 9.53 - - 79 3α,12α, 23-Nordeoxycholic Acid 3-OH; 12-OH 457.2265 → 97 40 6.4 7.11 - 80 9(11), (5β)-Cholenic Acid-3α-ol-12-one 3-OH 467.2108 → 97 40 6.48 - - 81 5β-Cholenic Acid-7α-ol-3-one 3-OH 469.2265 → 97 40 7.1 - - 82 5α-Cholanic Acid-3α-ol-6-one 3-OH 469.2265 → 97 40 5.92 - - 83 3α-Hydroxy-7 Ketolithocholic Acid 3-OH 469.2265 → 97 40 6 - - 84 3α-Hydroxy-12 Ketolithocholic Acid 3-OH 469.2265 → 97 40 6.04 - - 85 8(14),(5β)-Cholenic Acid-3α, 12α-diol 3-OH; 12-OH 469.2265 → 97 40 6.58 7.49 - 86 Murocholic Acid 3-OH; 6-OH 471.2421 → 97 40 4.9 5.13 - 87 Isodeoxycholic Acid 7-OH; 12-OH 471.2421 → 97 40 7.8 8.06 - 88 5β-Cholanic Acid-3β, 12α-diol 3-OH; 6-OH 471.2421 → 97 40 6.66 6.36 89 Chenodeoxycholic acid 3-OH; 7-OH 471.2421 → 97 40 7.67 7.96 - 90 Hyodeoxycholic acid 3-OH; 6-OH 471.2421 → 97 40 5.62 6.48 - 91 Ursodeoxycholic acid 3-OH; 7-OH 471.2421 → 97 40 5.48 6.53 - 92 Deoxycholic Acid 3-OH; 12-OH 471.2421 → 97 40 7.89 8.22 - 93 3α-Hydroxy-7,12-DiketoCholanic Acid 3-OH 483.2058 → 97 40 2.34 - - 94 3α-Hydroxy-6,7-DiketoCholanic Acid 3-OH 483.2058 → 97 40 5 - - 95 5β-Cholanic Acid-3α, 6α-diol-7-one 3-OH; 6-OH 485.2214 → 97 40 4.65 4.26 96 3 Dehydrocholic Acid 7-OH; 12-OH 485.2214 → 97 40 4.16 - - 97 12 Dehydrocholic Acid 3-OH; 7-OH 485.2214 → 97 40 3.72 3.45 - 98 α Muricholic Acid 3-OH; 6-OH; 7-OH 487.2371 → 97 40 4.17 4.04 5.51 99 ÏĽ Muricholic Acid 3-OH; 6-OH; 7-OH 487.2371 → 97 40 4.42 5.93 4.6 100 β Muricholic Acid 3-OH; 6-OH; 7-OH 487.2371 → 97 40 4.68 5.29 4.34 101 Cholic acid 3-OH; 7-OH; 12-OH 487.2371 → 97 40 5.84 5.54 4.84 102 Hyocholic acid 3-OH; 6-OH; 7-OH 487.2371 → 97 40 4.66 5.95 5.17 103 Glycolithocholic Acid 3-OH 512.2687 → 97 40 6.78 - - 104 Glycoursodeoxycholic Acid 3-OH; 7-OH 528.2636 → 97 40 3.05 4.11 - 105 Glycohyodeoxycholic Acid 3-OH; 6-OH 528.2636 → 97 40 3.01 3.85 - 106 Glycochenodeoxycholic acid-S 3-OH; 7-OH 528.2636 → 97 40 4.66 5.17 - 107 Glycodeoxycholic acid 3-OH; 12-OH 528.2636 → 97 40 5.48 4.77 - 108 Glycocholic acid 3-OH; 7-OH; 12-OH 566.2405 → 97 40 5.95 5.87 5.67 109 Glycohyocholic Acid 3-OH; 6-OH; 7-OH 566.2405 → 97 40 2.74 2.31 3.56 110 Taurolithocholic acid 3-OH 562.2513 → 97 40 5.29 - - 111 Tauro-ursodeoxycholic Acid 3-OH; 7-OH 578.2462 → 97 40 2.26 3.14 - 112 Taurohyodeoxycholic Acid 3-OH; 6-OH 578.2462 → 97 40 2.2 2.92 - 113 Taurochenodeoxycholic acid 3-OH; 7-OH 578.2462 → 97 40 4.01 4.28 114 Taurodeoxycholic Acid 3-OH; 12-OH 578.2462 → 97 40 3.59 4.02 - 115 Taurohyocholic Acid 3-OH; 6-OH; 7-OH 594.2412 → 97 40 1.55 1.91 2.44 116 Tauro-β Muricholic Acid 3-OH; 6-OH; 7-OH 594.2412 → 97 40 1.59 1.34 2.03 117 Tauro-α Muricholic Acid 3-OH; 6-OH; 7-OH 594.2412 → 97 40 1.51 1.61 2.27 118 Tauro ÏĽ-Muricholic Acid 3-OH; 6-OH; 7-OH 594.2412 → 97 40 1.57 1.51 2 119 Taurocholic acid 3-OH; 7-OH; 12-OH 594.2412 → 97 40 2.63 2.68 -

94 Chapter 4-Results

4.2.5 Optimisation of BA sulfation

LCA was selected for the optimisation of the sulfation procedure as it is the most abundant BA in human where sulfation occurs on hydroxyl C3 (Norman et al. 1964; Palmer et al. 1971). Triplicate samples of LCA (pure reference material) were prepared for sixteen sulfation reaction conditions. Sulfur trioxide-pyridine complex was resuspended in CHCl3 (5 mg/mL) or pyridine and added to 20 µL of BA standards (0.5 mg/mL) with or without sodium sulfate. The reaction was stopped after either 1h or 24h at either RT or 55°C by evaporation to dryness (c.f Table 4.4). All samples were stored at -80°C. Prior to use in analysis, all samples were solubilised in

H2O/acetonitrile/2-propanol (10:6:5, v:v:v).

4.2.6 Puriﬁcation of sulfated BAs

Following synthesis, the presence of sulfated BAs was confirmed by UPLC-MS/MS using the chromatographic method described above and multiple reaction monitoring for the [HSO4]- m/z 97 ion. Each of the 46 sulfated BAs was purified by individually tailored chromatographic separations of the reaction mixture using an Acquity UPLC equipped with an XBridge BEH C8 column (3.5 µm, 4.6 mm x 150 mm). The solvents used were the same as described above for the analytical method. The chromatographic separation of each sulfated BA was repeated a number of times and fractions were repeatedly collected using a Fraction Collector III (Waters, Manchester, UK) in order to amass sufficient material for subsequent experiments.

4.2.7 Optimisation of MRM transitions and SIR

Source parameters such as collision energy and capillary voltage were optimised for each standard by direct infusion combined with the UPLC ﬂow rate and appropriate solvents (e.g. UPLC-MS/MS conditions). Source parameters were optimised for glycine, taurine and sulfate fragmentations. These optimisations were automatically performed using Waters IntelliStart software (Waters) and then improved manually.

95 Chapter 4-Results

Cone voltage was set to 60V for all BAs as no signiﬁcant variation in ionisation during the optimisation was observed between 10-60V.

4.2.8 Method validation

The BA targeted UPLC-MS/MS method was validated according to the bioanalytical guidance provided by the FDA (FDA 2013). Linearity was evaluated for each BA over a concentration range of 0.05 nM-5 µM. The limit of detection (LOD) was determined with a signal to noise (S/N) ratio > 3, the lower limit of quantification (LLOQ) was determined using S/N > 5 and <20% of CV, standard deviation divided by the mean), and upper limit of quantification (ULOQ) was established as intensity level below the detector saturation which corresponds to the highest standard concentration in the calibration curve. Matrix effect assessment aimed to detect a potential increase or suppression of BAs ionisation due to the presence of interfering analytes in the samples. The matrix effect was evaluated by comparing integrated peak area of deuterated standards spiked in solvent and spiked in plasma samples for the quality controls; QC1 (10 nM) and QC3 (100 nM). Carry-over was tested by comparison of blanks vs. QC1 and QC5 (5µM). Accuracy and precision of the assay were calculated using the measurement error (expressed in %) and the CV of the assay, respectively. Accuracy and precision were derived from three different concentrations of quality control (QC) prepared in solvent

H2O/acetonitrile/2-propanol (10:6:5, v:v:v). The QC1 (10 nM), QC2 (50 nM), QC3 (100 nM) and QC4 (750nM) precision and accuracy were validated on intra-day (6 replicates analysed on same day) and inter-day (one replicate analysed on each of six diﬀerent days).

calculated concentration Percent error of accuracy = × 100 actual concentration

96 Chapter 4-Results

standard deviation Precision = × 100 mean

Recoveries (%) were evaluated for each deuterated internal standard on six replicates for each bioﬂuid; plasma, serum and urine. Deuterated internal standards were spiked into samples before sample preparation (pre-spiked) and after sample preparation (post-spiked). Recoveries were calculated as follows; area under peaks of pre-spiked-sample divided by the area under peaks of post-spiked samples at QC1 and QC3.

4.2.9 MS data pre-processing

Waters raw data ﬁles were converted to NetCDF format and data were extracted via the XCMS (v1.24.1) package with an R (v2.11) software. MassLynx 4.1 and TargetLynx 4.1 software were used respectively for data acquisition and validation for this high-throughput targeted method for quantiﬁcation of BAs.

4.2.10 Multivariate Statistical Analysis

PCA was carried out on the integrated BAs peaks with Pareto scaling using SIMCA P+ v13 (Umetrics, Ume˚a, Sweden). A standard univariate statistic, the Stu- dent’s t-test, was carried out to establish signiﬁcant variations in BA concentrations observed between the pre- vs. post-prandial participants.

97 Chapter 4-Results

4.3 Results

4.3.1 Bile acid sulfation and puriﬁcation

Sulfation of 46 BA reference standards was conducted to generate sulfated reference materials for use in method development, facilitating a wide detection range of BA species. Sulfation eﬃciency was optimised by evaluation of sixteen reaction conditions selected from previous work (Anizan et al. 2011; Cerny et al. 2004; Sandhoﬀ et al. 1999).

Different reaction conditions were tested such as the choice of solvent (CHCl3 vs. pyridine), reaction time (1h vs. 24h), reaction temperature (22°C vs.. 55°C) and the presence or absence of sodium sulfate. Sulfation efficiency was assessed for sulfated LCA by integration of the area under the peak obtained by monitoring the m/z transition 455.2472 → 97. The results of this procedural optimisation are illustrated in Figure 4.1 and Table 4.4. Analysis of sixteen conditions showed that the optimal sulfation efficiency was obtained with CHCl3, during 24 hours, at 55°C and with sodium sulfate (Figure 4.2).

Figure 4.1: Sulfation scheme with sulfur trioxide-pyridine complex. Example shown for LCA with sulfation on carbon position 3 in the steroid nucleus

Puriﬁcation of the sulfated LCA was performed by HPLC separation with fraction collection. UPLC-MS/MS analysis was implemented to assess the purity of each

- fraction collected by characterisation of two fragments; the sulfate ion [HSO4] m/z 97 and the non-sulfated form [M-H]- (example m/z for LCA 375.2904). This workﬂow was successfully applied for sulfation and puriﬁcation of 46 BAs in total. In addition,

98 Chapter 4-Results sulfation was implemented on BAs without hydroxyl groups on the steroid nucleus (e.g ursocholanic acid) and aimed to conﬁrm that the carboxyl group present on the carbon chain does not provide stable sulfate bonding. BAs were sulfated on hydroxyl groups corresponding to C3, C6, C7 and C12. The elution order and peak intensities observed in the reversed-phase HPLC separation of mono-sulfated BAs was used to determine the sulfate position, as previously reported (C3 < C6 - C7 < C12) (Alnouti et al. 2008).

Figure 4.2: Optimisation of BA sulfation. Sixteen conditions were tested on the LCA to evaluate the eﬀect of time (1h vs. 24h), temperature (RT room temperature vs. 55°C), solvent (chloroform CHCl3 vs. pyridine) and with or without sodium sulfate. The error bars show the standard deviation of the replicates (n=3).

Table 4.4: Percent yield of the 16 sulfation reactions (3 replicates) of LCA modiﬁed into sulfated LCA (LCA-S).

Sulfation conditions CHCl3 Pyridine - Sodium Sulfate - Sodium Sulfate 1h 24 h 1h 24h 1h 24h 1h 24h 1h 24h 1h 24h 1h 24h 1h 24h RT RT 55°C 55°C RT RT 55°C 55°C RT RT 55°C 55°C RT RT 55°C 55°C LCA 96.1 95.8 79.9 6.4 92.2 89.8 83.5 17.2 43.4 49.0 40.3 47.5 41.2 47.1 47.4 44.2 LCA-S 3.9 4.2 20.1 93.6 7.8 10.2 16.5 82.8 56.6 51.0 59.7 52.5 58.8 52.9 52.6 55.8 Standard deviation 3.05 3.2 6.34 2.09 5.38 5.88 6.47 10.7 7.86 9.69 4.69 10.5 8.33 3.1 8.42 22.2

99 Chapter 4-Results

4.3.2 Targeted MS/MS analysis of 145 BAs

In total, 145 BAs, including BA sulfates, were targeted in a single UPLC-MS/MS method operating in a single mode of detection (Figure 4.3, Table 4.1 and Table 4.2). The chromatographic method performed well in retaining all BA species, balancing the overall analysis time with high performance separation. However, the separation of unconjugated muricholic acid sterol isomers, differing only by the OH conformation in α, β, ω,was not achieved due to the early elution of the species. As expected, BA sulfates and also tauro- and glyco-conjugates eluted earlier than their unconjugated counterparts. The combination of fit-for-purpose LC separation and optimised MRM transitions and SIM offered high sensitivity and selectivity for the 145 BA species tested (Table 4.1 and Table 4.2). However, care needs to be taken for the mobile phase A of this chromatographic method which has to be accurately measured and stable if more solvents need to be added during the run (400 samples analysed with 2.5L of mobile phase A) as retention times of the glycine conjugated and unconjugated BA species are pH dependent (Cai et al. 1997). Therefore, this is an important consideration when targeting the MS detection time window to the elution of BA species, as any shift can result in missing the specific peaks.

100 Chapter 4-Results

4.3.3 Method validation of the BA targeted UPLC-MS/MS

method

Optimisation of MRM transitions and SIM aimed to maximise the detected signal for each characteristic ion species or transition. Afterwards, the UPLC-MS/MS method was validated in terms of accuracy and precision on intra/inter-day, linearity, carry-over and matrix eﬀect.

(a) Intra- and Inter-day precision and accuracy

The LODs ranged between 0.05-7.5 nM and linearity was investigated over a wide concentration range of 10000-fold between 0.25 nM and 5 µM respectively LLOQ and ULOQ. This wide concentration range offered the best order of magnitude for quantification as the concentration of BAs spans a wide range of concentrations (1 nM to 3 µM) (Garc´ıa-Canaveras˜ et al. 2012). Validation of the method shows high linear response associated with high mean R2 value of 0.998. Overall, we obtained acceptable accuracy between 81.5-117.1% for intra-day and 81.2-118.9% for inter- day with acceptable precision between 1-15.5% for intra-day and 2.1-19.9% for inter- day analysis (Figure 4.4 and Table 4.5). As can be seen, intra-day precision and accuracy of the 57 BAs were consistent with inter-day precision and accuracy for the increasing concentration of BAs QC1, QC2, QC3 and QC4 compared to previous work that showed significant decreased of the inter-day accuracies (Garc´ıa-Canaveras˜ et al. 2012).

101 Chapter 4-Results Extracted ion chromatograms of SIM (a) and MRM (b-f) UPLC-MS/MS analyses of 145 BA species; unconjugated (a), glyco-conjugated (b), Figure 4.3: tauro-conjugated (c), sulfated unconjugated4.3. (d), sulfated glyco-conjugated (e), sulfated tauro-conjugated (f). For more details see Table 4.2 and Table

102 Chapter 4-Results

The density plots on Figure 4.4 summarise and compare intra- vs. inter-day accuracies. The QC1 accuracy range was found to be 6% higher than QC2, QC3 and QC4 accuracy range. In addition, the QC1 precision range was 8% higher than QC2, QC3 and QC4 precision ranges. The QC1 was close to the noise, which might explain why it displayed a wider range of accuracy and precision values. Similar results have been found in previous work for QC1 where values were approaching the saturation limit (Garc´ıa-Canaveras˜ et al. 2012). Conversely, QC2, QC3 and QC4, which beneﬁt from being far from the noise and the saturation level, show the best accuracy and precision for intra/inter-day analysis.

Figure 4.4: Accuracy distribution of the QC1 (10 nM), QC2 (50 nM), QC3 (100 nM), QC4 (750nM) on intra and inter-day.

103 Chapter 4-Results

Table 4.5: Intra- and inter-day validation of accuracy and precision over 57 BAs with QC1 (10nM), QC2 (50nM), QC3 (100nM) and QC4 (750nM). Limit of detection (LOD), lower limit of quantiﬁcation (LLOQ), upper limit of quantiﬁcation and R2.

INTRA DAY INTER DAY LOD LLOQ ULOQ Accuracy % Precision % Accuracy % Precision % R2 (nM) (nM) (µM) QC1 QC2 QC3 QC4 QC1 QC2 QC3 QC4 QC1 QC2 QC3 QC4 QC1 QC2 QC3 QC4 1 0.1 0.25 2.5 0.9978 104 100.9 102.1 101.9 1.3 4.4 1.8 3.8 100.2 99.5 100.4 96.1 6.3 3.8 3.7 0.9 2 0.5 0.75 5 0.999 108.7 101.9 103 102.8 5.8 3.1 6.8 6.7 93.2 97.4 99.9 96.5 13.3 3.8 8 2.6 3 0.75 1 5 0.998 106.1 100.9 98.4 105.3 1.9 3.6 1.6 2.3 96.6 99.2 104.5 97.5 13.6 3.9 4.9 12.5 4 0.05 0.25 2.5 0.9985 111.8 98.1 101.2 104.4 5.1 5.5 3.2 1.9 114.1 100.9 102 99.3 6 8.9 4.6 9.7 5 0.25 0.5 5 0.999 89.5 98.7 99.9 106.1 5.3 4.3 1.3 4.1 81.2 98.7 100.9 96.7 10.1 5.7 3.7 0.2 6 0.25 0.5 5 0.999 108.8 102.1 102.8 104.4 1.4 4.6 2 2 87.4 100.9 104.4 96.5 10.9 5 6.3 8.2 7 0.25 0.5 2.5 0.999 103.7 98.9 100.4 105.8 2.3 4.2 1.8 3 101.3 99.4 102.7 99.5 7.3 5.6 7.5 10.8 8 0.5 0.75 2.5 0.9994 114.7 102.3 106.2 104.1 4.6 4.8 1.5 1.6 108 106.2 109.4 98.9 14.1 6.6 3 8.4 9 0.5 0.75 2.5 0.9999 105.7 100.6 108.1 104.3 5.3 4.6 1.6 2 102.8 111.1 110 99.2 6.7 8 4.8 8.3 10 0.25 0.5 5 0.9991 94.7 101.8 101.3 103.8 14 3.1 3.7 2.8 103.4 102 104.1 101.8 8.5 8 5.5 0 11 0.5 0.75 5 0.997 85.7 96.8 94.2 107.1 7.7 6.8 3.4 3.2 85.7 98.3 95.5 98.3 15.5 8 5.6 0.4 12 0.5 0.75 5 0.9997 106.5 98.6 100 107.1 12.4 7 4.2 3.2 102.6 95.6 95.5 98.3 19.3 9.5 5.1 0.4 13 2.5 5 5 0.999 92.7 95.1 97.2 104.4 3.2 4 1.4 3.7 92.9 100.4 103.6 94.8 13.1 4.7 8.4 5.3 14 0.25 0.5 2.5 0.9999 117.1 97.7 99.7 103.5 2.3 4.9 1.8 2.4 118.2 101.3 106.5 98 2.1 8.4 4.8 9.6 15 1 2.5 5 0.9985 87.4 100.4 101.4 106.1 6.4 4 2.6 3.4 90.9 91.5 90.7 98.9 13.4 14.4 7.4 6.1 16 1 2.5 2.5 0.9999 108.4 97.5 98.7 103.2 2.8 5.1 2.3 3.2 117.3 103.2 106.4 97.5 2.5 6.5 5 11.2 17 2.5 5 5 0.999 104.7 102.9 100.7 105.6 3.7 3.1 1.2 2.9 96.7 98.6 102.5 96.6 8.9 5.8 7.7 7.5 18 0.25 0.5 5 0.9953 101.7 99.9 100.5 112.1 5.1 4.2 3.7 12.4 99.6 96.9 92.3 98.4 18.8 11.3 5.3 21.7 19 0.25 0.5 2.5 0.9982 104.3 99.2 102.4 103.5 2.1 4.7 2 1.8 111.7 101.2 105.6 97.6 3.3 9.1 4.2 11.2 20 1 2.5 5 0.9999 95.2 97.6 99.2 102.4 2.5 4.9 1.4 8.2 101.3 99.4 102.7 74.9 7.3 5.6 7.5 31.6 21 0.25 0.5 2.5 0.9991 107.4 99.3 104.6 103.3 2.5 4.6 1.9 2.7 103.6 103.1 107.9 97 10.3 2.7 4.1 7.6 22 0.25 0.5 5 0.9999 105.8 97.9 99.7 105.3 2 5.1 2.2 2.5 114.3 103.8 106.9 95.4 3.8 8 4.8 2.6 23 0.25 0.5 5 0.9998 101.3 96.8 97.8 104.4 8.1 5 1.5 5.6 109.2 100.2 103.7 98.4 11.2 5.9 8.1 1 24 0.25 0.5 2.5 0.9905 95.3 100.2 101.8 103.4 7.4 4.1 2.5 3.8 104.6 94.5 91.1 98.2 11.3 16.6 10.6 11.4 25 0.5 1 5 0.9958 93.7 98.8 99.2 102.9 4.5 4.8 3.2 6.7 99.5 94.8 90.3 102.5 13.9 11.7 10.7 3 26 0.25 0.5 5 0.9934 96.9 100.7 105 105.2 5.1 4.4 4.7 2.3 101.5 95.1 92.4 99.6 13.7 15.4 8.8 9.7 27 0.25 0.5 2.5 0.9999 95.3 103.7 97 103.4 3.9 2.9 1.5 2 102.9 98.5 102.2 99.4 9.4 5 9 6.7 28 0.25 0.5 5 0.9998 89.6 101.8 102 103.3 9.7 4.2 1.5 2.4 96 102.9 103.5 97.8 19 2.6 4.4 6.2 29 0.25 0.5 2.5 0.999 100.1 100.8 102.2 104.9 3 4.1 1.5 7.1 96.8 101.2 104.5 105.7 9.5 3.6 6.1 0.1 30 0.1 0.25 5 0.999 102.1 103.6 105.8 107 1 4.1 1.6 3.9 96.8 100.3 104.9 97.7 9.1 5.5 7.1 4.6 31 2.5 5 5 0.9988 91.3 89.4 98.3 106 8.8 5.9 1.9 2.7 100.3 101.6 104 96.9 13.5 6.3 5.7 12.1 32 0.5 0.75 5 0.9998 93.5 100.1 101.6 103.3 3 4.4 2.4 3.8 97 102.5 105.5 101.7 11.1 5 6.9 0.1 33 0.5 0.75 2.5 0.9997 94.2 99 100.9 106.1 3.3 3.6 1.5 2.2 96.1 100.8 104.1 99.4 11.5 4.9 6.8 11.8 34 0.5 0.75 2.5 0.9999 101.4 101.3 100.7 106.3 2.6 4.1 1.8 2 103.7 92.7 93.3 100.1 6 10.3 9.2 11 35 0.5 0.75 5 0.9995 88.6 88.6 98.6 104 11.5 6.3 1.8 4.6 93.3 102 105 103.1 16.1 6.7 5.8 1.7 36 0.75 1 5 0.9998 99.3 105.7 106.9 105 4 2.2 4.9 3.4 105.5 100.1 101.3 94.6 14.1 5.3 9 4.3 37 7.5 10 5 0.9915 104.4 108.4 103.1 106.4 6.6 4.3 5.4 4.2 103.6 102.4 104.5 96.2 10.2 7.1 5.7 5.2 38 5 7.5 5 0.9998 81.5 97.9 111 107.2 7.9 6.6 1 9 92.3 100.6 105.5 94.9 13.4 6.9 5.2 14.4 39 2.5 5 5 0.9998 108.9 108.4 107 102.1 9.3 4.3 2.1 3.5 112.6 101.1 110.1 101.3 14.3 13.3 4.7 3.3 40 2.5 5 5 0.9998 100.9 97.2 97.8 105.7 11.1 4.8 1.5 2.6 115.8 99.7 101.6 99 6 4.9 7.9 12.6 41 2.5 5 5 0.9995 106.7 103.5 110 105.1 15.2 3.8 6.6 5.2 118.9 92.8 102.9 95.5 12.9 5.6 11.3 14.5 42 2.5 5 2.5 0.9998 104.6 103 96.5 107.5 5.3 3.1 4.4 2.8 108.3 99.2 103.9 99.6 7.9 3.7 6.4 10.7 43 5 7.5 5 0.9998 91.1 95.8 112.9 101.4 15.5 6 1.4 1.2 101.4 102.1 105.8 94.9 19.6 5.9 5.2 14.9 44 5 7.5 5 0.9995 96.4 96.2 98.9 102.1 8.3 4.7 5.4 7.2 98.8 102.3 102.9 95.1 14.8 5.2 8.1 9.2 45 2.5 5 5 0.9998 100.5 102.9 107.7 102.6 7.6 6.2 2.7 2.2 107.4 108.4 112.4 101.2 11.1 6.9 3.8 2.1 46 2.5 5 5 0.9993 105.5 101.9 104.3 105.6 9.6 4.9 4.6 7.1 105.4 103.8 94.9 98 13.4 6.6 8.5 3.2 47 5 7.5 5 0.9998 104.5 98.7 98.9 103.4 8.7 4.6 1.7 18.3 95.4 97.9 104.6 99.4 8.8 4.7 6.1 6.7 48 2.5 5 2.5 0.9958 93.2 94 104.6 103.3 7.5 6.9 1.7 20.8 102.5 102.8 104.5 97.8 18.8 7.7 4.5 6.2 49 2.5 5 2.5 0.9996 106.4 99 98.1 104.9 8.3 4.6 3.7 2.3 95.9 100.3 106.8 105.7 10.3 2.7 4.6 0.1 50 2.5 5 5 0.9997 97.2 98.7 100.3 107 14.2 5.5 6.4 2 106.2 102.3 94.2 97.7 16.4 6 7.8 4.6 51 5 7.5 5 0.9978 97.2 93.9 106.5 106 10.8 6.8 1.9 2.4 91.9 107.1 109.9 96.9 15 7.9 3.1 12.1 52 2.5 5 5 0.9999 106.8 99.7 103.9 103.3 4.9 2.7 4.9 7.1 114.6 102.1 109.9 101.7 9.3 4.2 5.8 0.1 53 2.5 5 5 0.9996 106.2 108.2 113.7 101.5 3.8 7.4 1.2 2.3 86.7 115.7 118.1 101.4 11.3 8.8 3.4 0.7 54 2.5 5 5 0.9999 104.9 100 100.4 105.6 7.1 6.1 2.9 2.5 101.8 101.1 105.2 98.6 7.9 2.3 5.8 4.5 55 2.5 5 5 0.999 95.2 99.3 99.2 101.3 12.4 3.9 2.9 1.4 113.8 102.3 105.9 101.1 19.9 6.5 3.2 1.3 56 2.5 5 5 0.9985 97.2 97.8 92.6 91.7 14.9 4.6 3.1 15.1 103.5 91.4 89.3 88.8 18 8.1 6 31.1 57 0.5 1 5 0.9959 103.3 112.3 105.7 101.1 15.1 3.1 5.8 0.8 82.1 105.7 106.7 98.2 19.2 7.4 4.1 7.2

(b) BA recoveries in human serum, plasma and urine

Prior to recovery evaluation, matrix effect and carry-over were measured. As can be seen in Table 4.6 the ionisation signal of deuterated standards in the presence of matrix vs. solvent were between 87.05-111.85% for QC1 and between 79.85-111.37% for QC3. Furthermore, qualitative measurement by combined infusion and injection of the standards confirmed that there was no significant ion suppression or enhancement. As shown in Table 4.7 carry-over of BAs was easily removed by the UPLC system and was negligible. Extraction efficiency of the sample preparation was evaluated by

104 Chapter 4-Results analysis of recoveries and reproducibility (n=6) for each deuterated internal standard at their QC1 and QC3 concentrations (Figure 4.5 and Table 4.8).

Table 4.6: Quantitative matrix eﬀect was evaluated by comparing integrated peak area expressed as a percentage of deuterated standards spiked in solvent and spiked in plasma samples for the QC1 (10 nM) and QC3 (100 nM) (n=3).

Labelled standards QC1 QC3 58 96.19 80.7 59 98.07 81.25 60 107.81 92.22 61 88.99 85.03 62 93.83 80.61 63 91.55 81.03 64 97.09 83.86 65 91.43 111.37 66 106.7 84.46 67 96.8 85.35 68 111.85 87.09 69 91.7 79.85 70 87.05 83.8 71 105.45 86.76 72 94.22 83.06 73 90.38 93.27

Table 4.7: Carry-over was evaluated on the integrated area under the peak of each labelled deuterated standards with the following run sequence; Blank 1 was injected ﬁrst then QC1 (10nM), QC5 (5µM) and Blank 2.

Labelled standards Blank 2 vs. QC5% Blank 2 vs. QC1 % Blank 1 vs. QC1 % 58 0.51 1.94 0.1 59 0.85 4.01 2.35 60 0.08 0.37 0.19 61 0.09 2.24 0.46 62 0.12 0.66 0.29 63 0.09 1.22 0.03 64 0.07 0.4 0.28 65 0.09 0.56 0.07 66 0.02 0.11 0.08 67 0.07 0.38 0.15 68 0.12 0.7 0.16 69 0.08 1.96 0.17 70 0.19 0.97 0.13 71 0.07 0.39 0.16 72 0.04 0.81 0.13 73 0.02 0.13 0.23

105 Chapter 4-Results

Figure 4.5: Recoveries of the 16 deuterated standards evaluated at QC1 (10 nM) and QC3 (100 nM) in human plasma, serum and urine. The error bars show the standard deviation of the replicates recovery (n=6).

Simple protein precipitation is a straightforward one step sample preparation procedure that is known to reduce variation in lipid recoveries (Sarafian et al. 2014). High recoveries of BAs associated with high reproducibility were found in urine between 92.7% and 8.58%. Recovery of the standards in plasma and serum were around 4% less than for urine recoveries, whereas reproducibility around 2% higher than for urine. This difference between plasma/serum and urine can be explained by the presence of high concentrations of proteins in blood, which might affect the BA recoveries. Interestingly, ionisation and recovery of BAs coeluting with other lipid species after 10 min are not affected. Overall, satisfactory results of recoveries were observed for

106 Chapter 4-Results all three bioﬂuids.

4.3.4 Application of the targeted UPLC/MS-MS methodol-

ogy to a human fed/fasted study showing post-prandial

diﬀerences in BA quantiﬁcation and detection

The validated targeted BA method was applied to a pre-/ post-prandial study. BA profiles of plasma, serum and urine human samples were investigated using a combination of univariate and multivariate analysis. As expected, the PCA scores plots for both plasma and serum showed clear separation of samples obtained pre- and post-prandially in PC1, reflecting systematic differences in the BA profile (Figure 4.6). Moreover, samples collected pre-/ post-prandially were more tightly clustered than those obtained following consumption of a fatty meal indicating that the dietary challenge may magnify inter-individual differences in BA synthesis. In total, around 66 BAs were detected in blood and 55 BAs in urine respectively including 15 and 22 sulfated BAs. In contrast, no obvious separation was observed between the pre- and post- prandial groups for the urine samples. A likely explanation for this observation would be that the urine was collected too early, only 2 h after the meal, which might be not long enough for the BAs to be accumulated in any substantial quantity in urine. Since the majority of BA excretion is via the faecal route, the urine should only contain minor concentrations of BAs, and this may also account for the lack of observed variation between the pre- and post-prandial samples. For all three PCA models, the same outlier was observed but no modification in separation occurred on removal of this sample from any model.

107 Chapter 4-Results

Figure 4.6: PCA scores plots of UPLC-MS/MS spectra obtained from pre-/post-prandial human plasma, serum and urine.

108 Chapter 4-Results

The top five most significant BAs in discriminating between the pre- and post- prandial groups are listed in Table 4.7 and were consistent between serum and plasma. However, compared to a previous study, the order of significance was slightly different for the two matrices (Scherer et al. 2009). Higher BA concentrations were observed in serum compared with EDTA-plasma, consistent with general observations in small molecule studies regarding the use of EDTA as an anticoagulant (Vuckovic 2012). GCDCA was the most concentrated BA in blood, yet its measurement remained within the dynamic range of the assay.

Table 4.8: BAs found in plasma and serum that discriminated between pre-prandial vs. post-prandial healthy participants with means and standard deviations of each quantiﬁed BAs. Statistical analysis was performed using 2-tailed Student’ t-test. BA markers Concentration (nM) p-value in plasma Preprandial Postprandial GCDCA 2.11x-5 3300 ∓ 1530 13900 ∓ 4540 GCA 8x-4 23.31 ∓ 7.45 88.47 ∓ 42.16 GDCA 1.7x-3 10.34 ∓ 50.32 111 ∓ 95.71 TCA 2.2x-3 17.81 ∓ 9.59 51.63 ∓ 25.55 GLCA 2.8x-3 12.71 ∓ 4.8 27.24 ∓ 11.32 BA markers Concentration (nM) p-value in serum Preprandial Postprandial GCDCA 6.99-5 2200 ∓ 1367 11800 ∓ 4614 GDCA 1.42-3 8.83 ∓ 6.18 108.47 ∓ 69.66 GCA 2.73-3 17.41 ∓ 6.87 75.04 ∓ 44.76 GLCA 3.76-3 5.36 ∓ 2.48 18.43 ∓ 10.72 TCA 8.36-3 7.71 ∓ 8.11 28.37 ∓ 19.11

109 Chapter 4-Results

4.4 Discussion

To further the investigation of BA functions in humans, an advanced platform for high throughput analysis is essential. In this chapter, the development and application of a 15 min UPLC procedure for the separation of BA species from human bioﬂuid samples requiring minimal sample preparation is described. High-resolution time-of- ﬂight mass spectrometry was applied for sensitive and quantitative targeted analysis of 145 BA species including primary, secondary and tertiary BAs and was validated according to the FDA requirements.

4.4.1 Limits of targeted UPLC-MS/MS method regarding the

BA diversity

Although many of the BAs observed in the urine, serum and plasma samples can be annotated from the 145 BAs characterised in the targeted method, additional signals relating to unidentified BAs (observed by ULPC-MS profiling) enhance the potential for deriving specific BA fingerprints and hence retain the capacity to contribute to understating the underlying mechanistic etiologies of pathologies related to BA metabolism (e.g liver disease). In this example, the wide coverage profiling of primary, secondary and tertiary BAs which allow differential mechanisms of intestinal and hepatic function to be measured within a single chromatogram could have significant translational potential for patients with acute liver failure, cirrhosis or cholestatic liver diseases.

4.4.2 BA sulfation

The C3 and C7 hydroxy groups are the major sites of naturally occurring BA sulfation in humans, while the C3 and C6 hydroxy sites are predominant in rodents (Griﬃths et al. 2010; Hagey et al. 2013). Naturally occurring sulfation on the C12 hydroxy group is still unclear in the literature (Huijghebaert et al. 1984) The in vitro reaction was capable of sulfation at each available hydroxy group, producing all

110 Chapter 4-Results applicable variants of mono-sulfated BAs for purification. The hydroxyl groups present in the sterol nucleus of the BAs offers different sulfation proportions. Poly-sulfated BA products, however, were not observed. As previous studies have shown that liver sulfatase does not allow polysulfation of BAs, their synthesis was not pursued (Loöf¨ et al. 1980).

4.4.3 Variations in BA solubility

Elutions of conjugated BAs were observed to be earlier than the unconjugated forms. This observation demonstrates that addition of sulfate, glycine or taurine to hydroxyls situated on the steroid nucleus significantly change the conformation and solubility of the BA. These results are consistent with previous studies that demonstrate a change in physiological activity due to sulfation, especially regarding the BA transport in the intestine (Palmer et al. 1971). In addition to detection of sulfate conjugates, this chromatographic method offers the possibility to detect and quantify other water soluble BAs such as BA glucuronides. Furthermore, sulfation was implemented in this method as it is the most common modification observed in the BA elimination pathway and a simple one step procedure compared to glucuronidation which involve more complex organic synthesis steps (Mostarda et al. 2014).

4.4.4 Robustness of the targeted UPLC-MS/MS BA method

The UPLC-MS/MS method has been successfully validated according to the validation criteria established by the FDA. The latter system was validated by testing the linearity (LLOQ 0.25-10 nM and ULOQ 2.5-5 µM), precision (≈6.5%) and accuracy (81.2-118.9%) on inter- and intra-day analysis achieving good recovery of BAs (serum/plasma 88% and urine 93%). Tertiary BAs were found in mainly low concentrations in both biofluids (>10nM). Compared to previous reports, less abundant BA forms were included (i.e. sulfated BAs) in this method offering high sensitivity and reliable quantification are guaranteed by the robustness of the chromatographic and

111 Chapter 4-Results mass spectrometry methods (Perwaiz et al. 2001; Ye et al. 2007; Alnouti et al. 2008; Humbert et al. 2012; Bobeldijk et al. 2008; Sayin et al. 2013; Scherer et al. 2009; Want et al. 2010; Steiner et al. 2010; Xie et al. 2015; Chen et al. 2011; Huang et al. 2011; Garc´ıa-Canaveras˜ et al. 2012; Bathena et al. 2013; Qiao et al. 2012; Sergi et al. 2012; Duboc et al. 2013; John et al. 2014).

4.4.5 BA quantiﬁcation in human blood

This study has established BA concentrations in plasma, serum and urine. The values observed in the present study for plasma and serum are in agreement with those reported by Sherer and colleagues (1.710 nM) (Scherer et al. 2009), but differ from those reported by other investigators (450-750 nM) (Perwaiz et al. 2001; Humbert et al. 2012; Garc´ıa-Canaveras˜ et al. 2012; Xie et al. 2015). Subsequent application of the method described here to the analysis of independently obtained blood product samples have yielded results more closely aligned with the latter range, suggesting that the differences observed accurately reflect biological variance. In this thesis, the BA assay was implemented to analyse EDTA-plasma from two human studies. These studies were cohorts of obese participants at pre- vs. post- prandial state and NAFLD/NASH participants, respectively explained in Chapter 5 and Chapter 6.

112 Chapter 4-Results

4.5 Conclusion

Is it well established that BAs undergo multiple modifications during the enterohepatic circulation via interaction with the host or the gut microbiota. As a result, there is great BA diversity and most BAs found in human biofluids are not listed in any database. The parallel development of a non-targeted screening approach for BAs combined with a targeted UPLC-MS/MS BA assay provides a new resource for the profiling, identification and quantification of BAs found in human biofluids. The method described is a reliable platform for BA analysis in human biofluids including those with lipidic matrices. Analysis of plasma, serum, and urine was ef- ficiently performed with a high degree of precision in a rapid analytical cycle, and covering a large number of BA species. Of these species, 89 sulfated BAs have been described, enhancing the capability in the field for tertiary BA analysis. On the basis of these findings, this analytical method should provide new insights into the circulating BA pool regulated by the gut microbiota. Applications in biomarker discovery could furthermore provide guidance for clinical diagnosis and monitoring response to therapy in liver and intestinal diseases.

113 Chapter 5

The post-prandial response is inﬂuenced by increased of visceral fat and disturbed lipid-bile acid metabolism

5.1 Introduction

Obesity is defined as an excess of body fat and occurs when there is an imbalance between energy intake and energy expenditure. The development of obesity depends on genetic and environmental factors (e.g. diet, physical activity, stress) (Després et al. 2006). In obesity, the excess lipids which accumulate in adipocytes with chronic low-grade inflammation have been associated to risk factors (e.g. hypertension, insulin resistance, hyperinsulinemia, glucose intolerance) (Hotamisligil 2006; Kahn et al. 2006; Saltiel et al. 2001). Together these risk factors promotes severe metabolic pathologies such as cardiovascular disease and type 2 diabetes. The combination of these risk factors are strong determinants of metabolic syndrome (cf. Chapter 1). Obesity is assessed by measuring the fat distribution with waist-hip ratio (W/H) ≥0.85 which is a more effective measurement to identify association to risk factors than the BMI (kg/m2) defined as ≥30 in obesity (Dimitriadis et al. 2016). The mechanisms underlying obesity and related outcomes (i.e metabolic syndrome) have been investigated through various nutritional metabonomic studies. Nutritional metabonomic aims to evaluate the role of nutrients in detail with to-

114 Chapter 5-Results tal energy intake considering genetic and environmental factors relying on a dietary approach for a define period of time. In nutritional metabonomics multiple time points are monitored to assess the dynamic of metabolites after a dietary intervention and visualise trends. In this study, a lipid-based diet challenge aimed to investigate metabolic response of obese patients (Martin et al. 2013). Metabolic variation observed in both post-prandial lipemic response (PPLR) and post-prandial glycemic response (PPGR) have been linked to obesity and diseases related to the metabolic syndrome (Martin et al. 2013; Zeevi et al. 2015). PP responses are induced by a dietary challenge in contrast to basal pre-prandial state (i.e. fasting state) and their evaluation provides insights of metabolic changes (e.g. lipid metabolism variations). Introducing personalised diet as a treatment to control normal PPGR was shown to prevent diabetes and metabolic syndrome development (Riccardi et al. 2000; Zeevi et al. 2015). PPLR of circulating triglycerides and cholesterols are exacerbated at both pre-prandial and post-prandial states in participants with obesity and metabolic syndrome (Nakatsuji et al. 2010; Pellis et al. 2012; Tushuizen et al. 2010). In addition, adipose tissue distribution has been reported as an important component of obesity development and is thought to contribute to a range of metabolic disorders such as hypertriglyceridemia, hypertension, high fasting glucose levels and insulin resistance (Saito 2012). Fat distribution can be divided between fat from the upper body and fat from the lower body. It has been suggested that the fat distribution, in the upper body also known as visceral fat, has deleterious effects on metabolic dysfunction related to obesity (Klein et al. 2007). Other ectopic fat accumulation around tissues and organs can induce lipotoxicity and organ malfunctions. In contrast, fat distribution, in the lower body predominantly subcutaneous fat, has been thought to have protective effects (Tran et al. 2008). However, a recent study on mice with lipoatrophic diabetes (no white adipose tissue) has shown that transplantations of fat from both subcutaneous or visceral tissues improved similarly glucose and insulin levels (Gavrilova et al. 2000). Hence, fat distribution appears to be an essential crite- rion when considering an individual’s risks of developing obesity and related metabolic

115 Chapter 5-Results diseases. Also, it has been established that modification of the gut microbiota in obese compared to lean participants has consequences on the post-prandial response (Bäckhed 2011; Delzenne et al. 2011; Turnbaugh et al. 2009). Gut microbiota is known to have an impact on the circulating BA pool (Swann et al. 2011). Changes in post-prandial response in obese people were shown to be modulated by mechanisms related to the gut-host interaction. However, the effect of increased visceral fat on metabolic changes associated to the post-prandial response has not been determined. Untar- geted lipid profiling and targeted BA profiling hold therefore a powerful role in the characterisation of these metabolic disruptions. This chapter aims to determine if visceral obesity is the result of metabolic variation in pre- and post-prandial responses. In that regard, plasma samples from participants with subcutaneous obesity was compared to participants with visceral obesity (n=40 with W/H 0.74 to 0.97 and BMI 29.02 to 43.7). Fat distribution was assessed using anthropometric measurements by CT scan. The cohort was divided in four quartiles defined according to the log10 value of visceral/subcutaneous fat obtained by CT scan. Biochemical processes involved in visceral obesity development at fasting conditions were investigated. Plasma analyses were performed by UPLC-MS lipid profiling and UPLC-MS/MS BA targeted method. The lipid-based dietary challenge was designed to assess lipid and BAs at five different time points at pre-prandial state; -15 min and post-prandial state: 1h, 3h, 6h and 9h (Figure 5.1). After a lipid-based meal, the PPLR and lipids and BA metabolic signature associated to visceral fat were evaluated and compared to subcutaneous fat, in particular: (1) What are the typical PPRLs for lipid classes? (2) Does stratification according to visceral fat entail different PPLR kinetics? (3) Are there baseline fasting parameters affecting the PPLR?

116 Chapter 5-Results

5.2 Materials and methods

5.2.1 Materials

See Chapter 3, section 3.2 and Chapter 4, section 4.2 (Materials and methods).

5.2.2 Participants and experimental design

Forty obese Caucasian women were enrolled in this study. Exclusion criteria were diabetes, pregnancy, antibiotic therapy within 1 month prior to the beginning of the study, any therapy (contraception apart) within the run-in period of 1 week before the visit day, and eating disorders. Baseline corresponds to overnight fasting blood collected at pre-prandial condition (i.e. -15min). At this baseline pre-prandial time- point, detailed clinical biochemistry analyses were performed (Table 5.1) (Martin et al. 2013). BMI was calculated as weight/height squared (kg/m2). Quantiﬁcation of visceral (intra-peritoneal) and subcutaneous fat was measured by CT scan and the cohort was divided accordingly in quartiles. The cohort was divided in four groups according to the log10 value of visceral/ subcutaneous fat ratio (Table 5.1). The lipid-based diet challenge was a milkshake composed as follows; 2%-fat milk, a milkshake powder, double cream and glucose polymer. Of the total energy from the macronutrients, 50% was as carbohydrate, 35% was as fat, and 15% was as protein. After taking the meal, blood samples were collected at four time points as follows; 1h, 3h, 6h and 9h.

Figure 5.1: Design of experiment to evaluate the post-prandial response of 40 participants. The cohort was divided into quartiles according to the amount of visceral fat (i.e. Q1, Q2, Q3 and Q4).

117 Chapter 5-Results

5.2.3 Sample preparation

Plasma samples for lipid proﬁling were aliquoted (50 µL) and cold isopropanol was added (150 µL) to precipitate the proteins. Plasma samples for BA targeted analysis were aliquoted (50 µL) and internal deuterated standard mix (1 µM) was added (150 µL). Samples for lipid proﬁling and BA targeted method were vortexed for 10 min and cold isopropanol was added (150 µL). Afterwards, samples were stored 15 min at -20°C to improve protein precipitation and then centrifuged at 14,000g for 20 min. The supernatant was collected (150 µL) and stored at -80°C awaiting MS analysis.

5.2.4 Lipid proﬁling. Ultra-Performance Liquid Chromatogra-

phy.

See Chapter 3, section 3.2 (Materials and methods).

5.2.5 Lipid proﬁling. Quadrupole-Time-of-Flight Mass Spec-

trometry

See Chapter 3, section 3.2 (Materials and methods).

5.2.6 BA targeted method. Ultra-Performance Liquid Chro-

matography.

See Chapter 4, section 4.2 (Materials and methods).

5.2.7 BA targeted method. Triple Quadrupole Mass Spec-

trometry.

See Chapter 4, section 4.2 (Materials and methods).

5.2.8 MS data preprocessing.

See Chapter 3, section 3.2 and Chapter 4, section 4.2 (Materials and methods).

118 Chapter 5-Results

5.2.9 Structural identiﬁcation.

See Chapter 3, section 3.2 (Materials and methods).

5.2.10 Univariate and multivariate statistical analysis.

Untargeted lipid features and targeted BA measurements were log transformed and normalised using median fold change (Veselkov et al. 2011). Significant metabolite variations associated with visceral or subcutaneous fat profiles were assessed using statistical tests. P-values were determined using Student’s t-test and adjusted for multiple comparisons using Benjamini-Hochberg FRD<5% in MATLAB (R2014a v8.3). For clinical biochemistry measurements, non parametric Mann-Whitney test was employed. Multivariate data analysis was carried out on the XCMS extracted intensities using SIMCA P+ v13 (Umetrics, Ume˚a, Sweden). Prior to model fitting, features were subjected to unit variance scaling. PCA and OPLS-DA were used to evaluate features contributing to the differentiation between subcutaneous vs. visceral obesity. Permutation test was applied with 10,000 permutations to validated OPLS-DA models.

119 Chapter 5-Results

5.3 Results

Previous work on the lipid proﬁling sample preparation method (cf. chapter 3) and the development of the BA targeted assay (cf. chapter 4) were applied to this study in order to emphasise the contribution of visceral fat in obese cohort. Variations of circulating lipids and BAs levels were assessed on pre- and post-prandial response after a diet challenge.

5.3.1 Clinical biochemistry analysis highlighted changes in vis-

ceral fat compared to subcutaneous fat

Clinical biochemistry analyses at pre-prandial state (-15 min before the diet challenge) were implemented on the obese cohort and participants were divided in four groups according to log10 ratio visceral fat over subcutaneous fat (Table 5.1). For this reason, the cohort was stratified according to visceral fat distribution. Statistical analysis on first (Q1) subcutaneous fat vs. last (Q4) visceral fat quartile showed significant differences (p<10-4) of log10 ratio visceral fat over subcutaneous fat. In this study the following results focused mainly on differences associated to Q1 and Q4 which respectively correspond to high level of subcutaneous fat and visceral fat. Average BMI values of Q1 (BMI=34.01) and Q4 (BMI=34.59) were noticed to be similar. However, stratification according to fat deposition between Q1 and Q4 impacts significantly on insulin levels (p<0.015) and homeostasis model assessment values used to assess insulin resistance (HOMA-IR) (p<0.012). Liver functions were assessed by measuring AST, ALT and GGT levels but there was no significant statistical differences between Q1 vs. Q4. Statistical analysis on clinical biochemistry data highlighted significant variations associated to insulin levels and insulin resistance in obese with high levels of visceral fat (Q4).

120 Chapter 5-Results Q4 *** -4 vs. 0.18 0.136 0.926 0.453 0.362 0.343 0.871 0.055 0.255 0.347 0.898 0.985 0.091 0.398 0.137 0.189 0.869 10 0.015 * 0.012 * p-value < Q1 0.92 (4;7) 6.2 (61;81) 0.22 (1;1.8) (Min-Max) 5.52 (26;46) 6.36 (18;40) 12.6 (12;56) Q4 10.62 (13;44) 13.42 (44;86) 0.17 (3.8;4.4) 0.47 (4.6;6.2) 0.75 (2.4;4.9) 0.54 (0.7;2.5) 1.04 (139;142) 30.07 (229;334) 4.19 (29.9;43.7) 0.08 (0.72;0.97) 1.16 (4.55;8.06) 200 (1220;1970) 4.39 (18.81;32.98) 0.062 (-0.47;-0.29) Standard deviation 285 37.6 4.04 5.37 70.3 5.48 1.32 3.47 1.52 24.5 27.1 57.8 1433 -0.39 0.839 139.9 6.122 Mean 34.591 25.439 25.444 (Min-Max) 4.42 (33;45) Q3 9.284 (50;78) 6.944 (17;37) 8.343 (10;35) 4.841 (12;30) 0.251 (3.5;4.4) 0.493 (4.6;6.1) 0.677 (4.3;6.4) 0.248 (0.9;1.7) 0.613 (2.3;4.3) 0.447 (0.7;1.9) 1.581 (138;143) 20.972 (38;104) 2.95 (30.5;40.4) 75.275 (178;421) 0.071 (0.76;0.93) 1.865 (4.48;9.63) 7.218 (16.51;39.8) 0.023 (-0.54;-0.47) 152.494 (1320;1810) Standard deviation 38 24 3.99 5.41 64.8 5.31 1.38 3.34 1.28 23.5 21.1 62.4 1469 -0.52 0.853 141.5 6.063 303.4 Mean 36.996 24.363 2.101 (1;8) (Min-Max) 3.58 (27;39) 6.884 (9;29) Q2 0.289 (1;1.9) 4.477 (13;28) 5.073 (14;31) 3.62 (31;41.3) 11.203 (53;93) 0.182 (3.8;4.4) 0.516 (4.6;6.3) 0.850 (4.2;6.9) 0.883 (2.6;4.7) 1.317 (139;143) 19.751 (35;102) 71.454 (172;410) 0.055 (0.69;0.88) 1.486 (3.03;8.19) 0.026 (-0.64;-0.56) 142.61 (1190;1640) 6.315 (14.95;36.45) Standard deviation 4.1 32.8 5.17 65.2 5.58 1.32 3.56 2.25 21.4 19.2 17.5 71.1 1434 -0.61 0.808 140.8 22.12 4.951 Mean 36.341 263.22 5.8 (10;28) (Min-Max) Q1 4.636 (25;41) 8.969 (55;81) 3.072 (18;26) 11.252 (9;47) 17.645 (27;91) 0.169 (3.8;4.4) 0.335 (4.3;5.4) 0.962 (3.8;6.9) 0.405 (1.1;2.6) 0.917 (2.2;4.8) 0.405 (0.4;1.8) 1.281 (138;142) 39.774 (225;368) 0.062 (0.74;0.93) 8.735 (0.14;8.15) 3.101 (29.02;38.5) 1.911 (0.62;37.52) 0.058 (-0.81;-0.65) 181.937 (1100;1710) Standard deviation 20 3.5 -0.7 33.9 4.05 4.95 65.6 5.52 1.54 1.04 21.4 18.4 57.8 1357 0.798 140.4 4.236 275.2 Mean 34.012 18.601 mol/L) mol/L) µ µ Age Clinical biochemistry data from 40 participants stratiﬁed according to visceral fat (Q1, Q2, Q3, and Q4, n=10 for each quartile). Subcutaneous HOMA-IR Log10ratio ALT (U/L) AST (U/L) GGT (U/L) (visceral fat Glucose (%) BMI (kg/m2) MAP (mmHg) LDL (mmol/L) Insulin(mIU/L) HDL (mmol/L) Waist/Hip (cm) Sodium (mEq/L) subcutaneous fat) Uric acid ( Potassium (mEq/L) Creatinine ( Cholesterol (mmol/L) Calorimetry (kcal/24h) Triglycerides (mmol/L) Table 5.1: obesity (Q1) was compared to visceral obesity (Q4) by Mann Whitney test.

121 Chapter 5-Results

5.3.2 Lipid metabolism modiﬁcations correlate with increased

visceral obesity

To comprehensively characterise the PPLR of participants with high levels of visceral fat (Q4) compared to subject with high levels of subcutaneous fat (Q1), untargeted lipid profiling analysis was implemented. Lipid metabolic profiles from Q1 (subcutaneous) and Q4 (visceral) were compared at each time point by using OPLS-DA models. Each model was calculated by using one predictive component and one or several orthogonal components. The optimal number of orthogonal components was determined by R2Y (goodness of fit) and Q2Y (goodness-of-prediction) statistics. The OPLS-DA score plots for models pre- prandial at -15 min and post-prandial at 1h showed lipid variations associated with increase visceral fat in Q4 (Table 5.2). However, the Q2Y parameter is not significant by permutation testing. Further multivariate analysis with the lipidomic data did not allow to discrimination were observed by comparing the four quartiles and the different time points.

Table 5.2: OPLS-DA analysis to evaluate lipid proﬁles diﬀerences between subcutaneous obesity (Q1) and visceral obesity (Q4).

p-value of permutation Time points PC OC R2X R2Y Q2Y test (n= 10,000) -15 min 1 1 0.231 0.932 0.195 0.427 1h 1 1 0.232 0.988 0.0341 0.501 3h 1 1 0.164 0.909 -0.487 - 6h 1 1 0.278 0.84 -0.715 - 9h 1 1 0.296 0.823 -0.165 -

Student’s t-test was applied and a total of 211 features (n= 913) were observed to vary significantly between Q1 vs. Q4 (p<0.05). These results support the hypothesis that the impaired lipid metabolism is different in increased visceral-obesity (Q4) compared to subcutaneous obesity (Q1). Partial Spearman correlation (without age effect) with p-adjusted values after multiple testing (Benjamini Hochberg) and coefficient of correlation (ρ) of lipid profiles from each time points were assessed to highlight more closely the differences occurring between Q1 and Q4. Interestingly, phosphocholine lipids appeared to be affected at pre-prandial and 9h post-prandial

122 Chapter 5-Results time-points (Figure 5.2), suggesting chronic phospholipid metabolism differences between subcutaneous and visceral obesity: PC (16:0/20:3) was significantly and negatively correlated to increased visceral fat at baseline pre-prandial state (padjBH < 0.007 and ρ=-0.742). Conversely, LPC (16:0) was significantly and positively correlated to increased visceral fat at 9h post-prandial state (padjBH < 0.002 and ρ=0.782). There were no significant correlations observed at 1h, 3h and 6h post-prandial time-points.

Figure 5.2: Comparison of lipid proﬁles associated to increased subcutaneous obesity (Q1) and visceral obesity (Q4) after multiple testing.

Assessment of circulating lipids at pre- and post-prandial response was carried out by comparing participants with subcutaneous fat and with visceral fat. Statistical analysis showed signiﬁcant variations of PC and LPC occurring at pre-prandial state

123 Chapter 5-Results and 9h post-prandial state. TGs and FAs variations were expected as they play a key role in the post-prandial response and composition of lipoproteins circulating (cf. chapter 1.3.1). However, participants with visceral fat does not appear to exacerbate the PPLR after a diet challenge compared to participants with subcutaneous fat. The diﬀerent localisation of fat depot after diet might not be primarily linked to composition of absorbed and circulating lipid.

5.3.3 BA metabolism modiﬁcations correlate with increased

visceral obesity

BAs play a key role in the interaction between host and gut microbiota. BAs aim to emulsify and to absorb lipids, regulate signalling pathways, exhibit anti-microbial activity and are metabolised by the gut microbiota. BA pool composition was evaluated for the 40 participants and variations were inspected in subcutaneous vs. visceral fat. In this study, 145 BA transitions were monitored by applying the UPLC-MS/MS targeted BA assay developed and validated in chapter 4. A total of 49 circulating BAs were detected and quantified in plasma samples from the 40 participants, (Table 5.3). Multivariate data analysis was applied to observe differences between the extreme quartiles, subcutaneous (Q1) and visceral (Q4). The PCA scores plot did not show any discrimination between Q1 and Q4 obese participants. To determine if BAs have different levels in each group, OPLS-DA was applied with group classification Q1 vs. Q4 and validated by permutation test. First an OPLS-DA model was built on all the data points (pre-prandial and post-prandial; 1h, 3h, 6h and 9h) (Figure 5.3). This model showed that Q1 is different from Q4 with R2X=0.389, R2Y=0.646 and Q2Y=0.513 (Figure 5.3.A). However, no time point variation was observed for this model (Figure 5.3.B). Further OPLS-DA models were generated to compare Q1 vs. Q4 at each time point which showed a good discrimination (Figure 5.4 and Table 5.4). Permutation test was implemented and only the model with the data at 1h and 6h post-prandial state were validated (p<0.05)

124 Chapter 5-Results

(Table 5.4). Therefore, no discrimination were observed between the four quartiles (n=40 participants) and the diﬀerent time points.

Table 5.3: Mean BA concentrations in subcutaneous obesity (Q1, n=10) and visceral obesity (Q4, n=10).

Q1 Q4 pre- 1h 3h 6h 9h pre- 1h 3h 6h 9h Ursocholanic acid 2 2.28 2.99 3.45 4 2.01 2.03 2.43 3.25 3.64 5-Cholenic acid-3β-oL 1.36 1.72 1.31 0.72 1.12 1.47 1.19 1.22 1.29 1.32 4-Cholenic acid-3-one 0.39 0.49 0.5 0.23 0.31 0.13 0.14 0.14 0.04 0.17 Isolithocholic acid 3.02 3.79 3.06 2.07 2.65 2.47 2.6 1.22 1.67 1.69 3,7,12 Nordeoxycholic acid 0.08 0.12 0.07 0.06 0.09 0.12 0.13 0.14 0.1 0.15 5β-Cholanic acid-7α-ol-3-one 0.4 0.45 0.52 0.43 0.43 0.57 0.72 0.94 0.92 0.7 7 oxo Lithocholic acid 3.45 2.63 2.35 2.97 3.25 4.21 4.58 7.15 6.22 5.57 Lithocholic acid 21.82 26.02 44.38 11.82 18.51 16.93 14.04 15.46 5.18 19.46 4-Cholenic acid-3, 6-diol 23.55 20.74 17.48 17.11 14.79 75.73 70.69 67.53 78.22 54.08 5β-Cholanic acid-3β, 12α-diol 6.47 9.6 7.11 3.52 3.71 8.05 8.52 10.29 7.75 6.05 3,7,12 Dehydrocholic acid 16.46 13.55 13.69 13.21 11.31 4.39 3.67 3.4 2.03 2.91

Unconjugated 5β-Cholanic acid-3α,7α-diol-12-one 151.13 38.84 29.24 25.77 28.15 32.01 30.75 21.78 58.98 42.45 Chenodeoxycholic acid 42.28 46.88 89.33 97.09 118.95 97 134.55 215.86 337.28 173.19 Ursodeoxycholic acid 18.91 18.66 18.88 19.52 22.83 46.55 55.24 74.21 104.9 53.87 Deoxycholic acid 71.04 101.61 97.45 72.23 87.29 102.94 128.49 169.38 165.92 117.32 ω Muricholic acid 12.94 12.91 11.93 12.86 11.95 13.58 14.77 13.56 16.14 14.46 β Muricholic acid 0.06 0.12 0.14 0 0.18 0.53 0.16 0.56 0.5 0.54 Cholic acid 231.74 119.54 120.37 288.12 472.08 182.55 228.18 236.49 1062.53 732.34 Hyocholic acid 13.27 9.89 8.95 6.71 11.15 10.24 9.43 9.63 18.23 15.67 Glycoursocholanic acid 16.67 18.76 18.53 19.03 21.02 18.46 18.35 18.5 20.13 23.12 Glycolithocholic acid 7.86 18.32 10.36 5.86 3.26 4.3 12.98 6.21 3.95 4.11 Glycoursodeoxycholic acid 47.78 128.86 133.95 137.11 61.77 79.68 195.29 179.95 156.26 84.16 Glycohyodeoxycholic acid 0.07 0.37 0.27 0.09 0.02 0.17 0.5 0.3 0.09 0.06 Glycochenodeoxycholic acid 324.74 1063.57 1011.16 730.4 384.91 406.11 1041.88 935.64 522.65 361.19 Glycodeoxycholic acid 12 49.03 44.9 32.53 16.14 14.82 41.8 35.21 15.66 13.93 Glyco-conjugated Glycohyocholic acid 5.06 10.45 10.08 9.23 7.49 4.22 7.94 8.03 6.73 5.55 Tauroursodeoxycholic acid 8.65 32.84 37.64 33.64 15.5 7.72 29.9 28.77 13.29 14.9 Taurochenodeoxycholic acid 77.82 292.11 269.48 206.93 101.58 46.34 150.98 161.03 81.21 80.33 Taurodeoxycholic acid 48.19 163.45 147.73 135.22 60.42 19.77 76.06 65.82 30.69 31.73 Taurohyocholic acid 2.89 9.75 8.48 5.88 3.11 0.2 1.79 2.07 0.88 1.41 Tauro-β muricholic acid 22.19 47 35.44 26.37 15.92 4.49 9.8 8.89 4.57 6.39 Tauro-α muricholic acid 36.53 83.64 60.08 47.17 25.67 7.2 14.1 15.47 7.87 10.85 Tauro-conjugated Taurocholic acid 145.34 509.81 446.52 318.43 151.14 99.85 256.17 305.47 115.93 232.86 Lithocholenic acid sulfate 0.44 0.54 0.57 0.64 0.59 0.56 0.61 0.7 0.74 0.68 5-cholenic acid-3β-ol sulfate 72.53 75.44 71.14 80.17 70.43 72.5 80.52 66.81 84.86 71.32 Isolithocholic acid sulfate 14.83 28.97 21.09 19.59 22.42 15.72 15.18 18.41 13.49 10.89 3,7,12 nordeoxycholic acid sulfate NA NA 0.01 NA NA 0.01 0 0.01 0.02 0.01 5β -Cholanic acid-7α-ol-3-one sulfate 31.91 33.86 32.92 37.62 34.71 33.14 38.9 36.28 45.64 33.73 Lithocholic acid sulfate 3.89 NA 0.22 NA 0.36 1.09 NA NA NA 0.79 Sulfo-conjugated 4-Cholenic acid-3, 6-diol sulfate 1.79 2.08 1.31 3.26 1.02 5.58 5.61 3.64 7.77 4.75 Isodeoxycholic acid sulfate 1.26 4.33 4.66 3.98 1.09 1.38 6.61 2.97 3.21 0.7 5β-cholanic acid-3β, 12α-diol sulfate 3.62 5.68 5.07 3.49 2.32 4.49 5.95 6.11 6.71 4.51 Chenodeoxycholic acid sulfate 5.85 4.11 3.73 3.74 4.34 3.84 3.71 3.74 5.56 5.36 Ursodeoxycholic acid sulfate 4.78 6.54 4.73 3.74 3.69 2.95 4.1 6.21 5.98 4.71 Glycolithocholic acid sulfate 1220.23 2142 1551.03 1203.95 1432.27 932.21 1283.91 1333.99 1179.26 758.69 Glycoursodeoxycholic acid sulfate 58.34 69.98 40.61 77.32 45.49 104.11 128.87 160.12 184.11 114.4 Glycochenodeoxycholic acid sulfate 14.42 23.29 17.9 20.2 24.79 12.84 23.02 27.44 27.23 19.04 Tauroursodeoxycholic acid sulfate 0.18 0.73 0.17 0.38 0.46 0.43 0.61 1.59 0.5 0.39 Taurodeoxycholic acid sulfate 108.92 158.74 104.57 129.23 145.38 67.58 100.45 184.49 85.37 108.53

125 Chapter 5-Results

Figure 5.3: OPLS-DA score plot of BAs quantiﬁed from plasma according to (A) participants with subcutaneous obesity (Q1) and with visceral obesity (Q4) and to (B) pre- (-15 min) and post-prandial response (1h, 3h, 6h and 9h).

Figure 5.4: OPLS-DA score plots of BAs at 3h and 6h post-prandial state comparing subcutaneous obesity (Q1) to visceral obesity (Q4).

Table 5.4: OPLS-DA analysis to evaluate BA diﬀerences between subcutaneous obesity (Q1) to visceral obesity (Q4).

p-value of permutation Time points PC OC R2X R2Y Q2Y test (n= 10,000) -15 min 1 1 0.94 0.499 0.927 1 1h 1 1 0.0958 0.873 0.521 0.0046 ** 3h 1 1 0.0985 0.824 0.195 0.0948 6h 1 1 0.102 0.779 0.336 0.0355 * 9h 1 1 0.943 0.499 0.934 0.227

126 Chapter 5-Results

BA data from Q1 and Q4 were compared using Student’s t-test (Table 5.5). A total of 11 BAs were observed to exhibit signiﬁcant diﬀerences between the two groups at post-prandial state: 1h, 3h and 6h. Concentrations at each time point for these 11 BAs was plotted (Figure 5.5).

Table 5.5: Mean BAs observed for each time point between Q1 and Q4 were compared using Student’s t-test; p<0.05 (highlighted in red).

pre- 1h 3h 6h 9h Ursocholanic acid 0.968 0.549 0.533 0.809 0.68 5-Cholenic acid-3β-oL 0.865 0.245 0.841 0.18 0.694 4-Cholenic acid-3-one 0.075 0.086 0.079 0.098 0.305 Isolithocholic acid 0.747 0.488 0.087 0.595 0.52 3,7,12 Nordeoxycholic acid 0.627 0.941 0.428 0.622 0.664 5β -Cholanic acid-7α-ol-3-one 0.206 0.073 0.02 0.028 0.144 7 oxo Lithocholic acid 0.545 0.074 0.012 0.056 0.144 Lithocholic acid 0.503 0.162 0.266 0.274 0.897 4-Cholenic acid-3, 6-diol 0.207 0.168 0.073 0.085 0.148 5β-Cholanic acid-3β, 12α-diol 0.667 0.807 0.442 0.019 0.232 3,7,12 Dehydrocholic acid 0.227 0.313 0.309 0.243 0.338

Unconjugated 5β-Cholanic acid-3α, 7α-diol-12-one 0.332 0.635 0.521 0.309 0.436 Chenodeoxycholic acid 0.236 0.077 0.167 0.21 0.476 Ursodeoxycholic acid 0.127 0.041 0.019 0.142 0.1 Deoxycholic acid 0.28 0.534 0.209 0.009 0.396 ω Muricholic acid 0.697 0.268 0.172 0.103 0.171 β Muricholic acid 0.375 0.856 0.478 0.345 0.472 Cholic acid 0.804 0.371 0.123 0.16 0.6 Hyocholic acid 0.733 0.912 0.863 0.177 0.404 Glycoursocholanic acid 0.502 0.887 0.994 0.69 0.449 Glycolithocholic acid 0.093 0.32 0.174 0.208 0.595 Glycoursodeoxycholic acid 0.228 0.313 0.422 0.791 0.484 Glycohyodeoxycholic acid 0.311 0.512 0.791 0.971 0.283 Glycochenodeoxycholic acid 0.534 0.937 0.797 0.162 0.878 Glycodeoxycholic acid 0.601 0.786 0.6 0.156 0.775 Glyco-conjugated Glycohyocholic acid 0.423 0.332 0.239 0.097 0.457 Tauroursodeoxycholic acid 0.792 0.842 0.577 0.206 0.945 Taurochenodeoxycholic acid 0.194 0.081 0.074 0.011 0.652 Taurodeoxycholic acid 0.073 0.105 0.105 0.06 0.224 Taurohyocholic acid 0.079 0.078 0.011 0.018 0.2 Tauro-β muricholic acid 0.125 0.128 0.049 0.051 0.135 Tauro-α muricholic acid 0.128 0.106 0.049 0.067 0.17 Tauro-conjugated Taurocholic acid 0.542 0.134 0.485 0.046 0.681 Lithocholenic acid sulfate 0.089 0.217 0.133 0.196 0.174 5-cholenic acid-3β-ol sulfate 0.996 0.342 0.668 0.43 0.866 Isolithocholic acid sulfate 0.862 0.242 0.65 0.565 0.238 3,7,12 nordeoxycholic acid sulfate 0.343 0.343 0.784 0.343 0.343 5β -Cholanic acid-7α-ol-3-one sulfate 0.595 0.244 0.427 0.081 0.798 Lithocholic acid sulfate 0.364 NA 0.343 NA 0.625 4-Cholenic acid-3, 6-diol sulfate 0.105 0.113 0.024 0.269 0.141 Isodeoxycholic acid sulfate 0.901 0.31 0.245 0.531 0.282 5β-cholanic acid-3β, 12α-diol sulfate 0.499 0.89 0.677 0.101 0.074 Chenodeoxycholic acid sulfate 0.384 0.598 0.984 0.125 0.58

Tauro-conjugated Ursodeoxycholic acid sulfate 0.149 0.203 0.602 0.267 0.521 Glycolithocholic acid sulfate 0.395 0.12 0.551 0.952 0.105 Glycoursodeoxycholic acid sulfate 0.362 0.326 0.09 0.128 0.14 Glycochenodeoxycholic acid sulfate 0.607 0.978 0.263 0.219 0.559 Tauroursodeoxycholic acid sulfate 0.589 0.888 0.161 0.794 0.87 Taurodeoxycholic acid sulfate 0.152 0.401 0.272 0.307 0.492

Unconjugated BAs including two secondary BAs and three tertiary BAs were increased in visceral obesity (Q4) compared to subcutaneous obesity (Q1) (Figure

127 Chapter 5-Results

5.4A). Interestingly, these unconjugated BAs are associated with UDCA and DCA metabolism. However, primary BAs, CA and CDCA were not significantly increased. A similar trend to the unconjugated BAs was observed for the sulfated 4-cholenic acid- 3β, 6β-diol (Figure 5.4B). Conversely, five taurine conjugated BAs were decreased in visceral obesity compared to subcutaneous obesity (Figure 5.4B). Furthermore, five taurine conjugates including primary BAs, TCA and TCDCA and tertiary BA stereoisomers THCA (or TδMCA, 3α, 6α, 7α), TωMCA (3α, 6α, 7β) and TβMCA (3α,6β,7β) were significantly decreased in visceral obesity at 3h to 6h post-prandial time-points. However, these BAs are not significant after multiple testing correction by Benjamini-Hochberg.

Figure 5.5: PPLR curve of BAs with significant time point differences between Q4 and Q1. Student’s t-test was applied to determine significance of (A) unconjugated BAs and (B) conjugated BAs levels.

128 Chapter 5-Results

5.4 Discussion

The work presented in this Chapter offers an in-depth analysis of specific circulating lipids before and after a dietary challenge in participants with subcutaneous (Q1) or visceral (Q4) obesity. Lipids and BAs were characterised to evaluate variations in the two groups (Q1 vs. Q4) at different time points as follows; pre-prandial (-15 min) and post-prandial (1h, 3h, 6h and 9h). The induced post-prandial response aimed to stimulate lipid and BA metabolism to evaluate changes involved in development of subcutaneous fat and visceral fat.

5.4.1 Visceral obesity is associated with impaired insulin levels

and insulin resistance

Clinical biochemistry analysis revealed signiﬁcant increases of insulin and insulin resistance in fasting conditions (i.e. HOMA-IR) in visceral obesity. Insulin is increased in obese participants which is closely associated with the gut microbiota and has been shown to be a risk factor of metabolic syndrome (Cani et al. 2008). Also, subcutaneous fat grafts improve glucose homeostasis and lower metabolic risk compared to visceral fat grafts in mice (Klein et al. 2007). This suggests that visceral rather than subcutaneous fat is associated with the development of insulin resistance. It can be assumed that insulin resistance is one of the underpinning factors linking obesity and metabolic syndrome. However, the triggering event of insulin resistance is a complex mechanism and its understanding requires further investigation.

5.4.2 Visceral obesity is associated with variations of phos-

phocholine levels

By UPLC-MS lipid proﬁling analysis, we observed that visceral obesity has an impact mainly on phosphocholine metabolism. Species such as FAs and TGs were expected to be modiﬁed as they are main components of fat and play an important role in the host-gut interaction (Tremaroli et al. 2012). The variations indicated

129 Chapter 5-Results a signiﬁcant decrease of PC (16:0/20:3) at pre-prandial baseline and a signiﬁcant increase of LPC (16:0) in post-prandial time-points in participants with visceral obesity compared to participants with subcutaneous obesity.

5.4.3 Circulating BA are modiﬁed in visceral obesity

BAs present in the enterohepatic circulation aim to assist the emulsification and the absorption of dietary lipids. BAs can be metabolised by bacterial enzymes which favour formation of secondary and tertiary BAs. Altered BA pool by the gut microbiota might induce critical changes such as the overall BA pool solubility, increase BA pool toxicity and may favour disease development. Here, BA pool composition was assessed in Q1 and Q4 to determine if the gut microbiota had a differential effect on the digestion of lipid nutrients. To offer new insights on the role of BAs, the targeted BA UPLC-MS/MS assay presented and validated in chapter 4 was implemented. In this study, the results indicated that visceral obesity is significantly associated with an increase in unconjugated and sulfated BAs with a decrease in taurine- conjugated BAs in plasma. Dihydroxy BAs, which are UDCA and DCA and their derivatives, dehydroxylated 5β-cholanic acid 7α-ol-3-one and epimerised 5β-cholanic acid-3β, 12α-diol respectively are known as gut microbiota products (Fedorowski et al. 1979; Lee et al. 2013). Modulation of the BA pool has an impact on the gut microbiota function and the tissue function of the host (Swann et al. 2011). Recent evidence suggests that production of DCA increases the risk of hepatocellular carcinoma in obese mice (Yoshimoto et al. 2013). Conversely, UDCA reduces the BA pool hydrophobicity and its increased level may minimise the perturbations induced by DCA (Khare et al. 2008; Serfaty 2012). Taurine conjugates of primary BAs, TCA and TCDCA and tertiary BA stereoisomers derived from DCA and LCA (secondary BAs), THCA, TωMCA and TβMCA were decreased in visceral obesity. These results suggest compromised gut microbiota activity to hydrolyse taurine conjugated BAs which appears to exacerbate the BA pool toxicity and probably impact on the gastrointestinal tract colonisation. Furthermore,

130 Chapter 5-Results taurine metabolism plays a central role in obesity as it ameliorates inflammatory responses (Lin et al. 2013; Murakami 2015). It is also well known that taurine reduces the hydrophobicity of BAs more than glycine. Taurine has a pKa=1.5 for 50-100g/L in water and glycine has a pKa=2.34 for 250g/L in water. In this study, unconjugated BAs were increased and taurine conjugates BAs including TMCA were decreased in visceral obesity compared to subcutaneous obesity. This may indicate that the decreased taurine conjugates might influence processes such as lipid assimilation and fat distribution. Previous mouse studies emphasised the role of TMCA isomers as important gut microbiota intermediate to decrease BA synthesis via FXR activation in germ free mice (Sayin et al. 2013). In our study, the reduction in primary BA was not significant, which indicates that the levels of TMCA in this human cohort for visceral obesity were not sufficient enough to activate the FXR pathway and reduce BA synthesis. In addition, sulfated 4-cholenic acid-3β, 6β-diol (MCA derivate) produced by the action of gut enzymes (7α/β-dehydroxylation) was increased in visceral obesity. Sulfation offers high solubility to the BA and presents less toxic properties than its non-sulfated counterpart by elimination in the urine or faeces. Desulfation has been detected in the intestine and related to gut microbiota activity (Carbonero et al. 2012). Our findings highlight the central role of gut microbiota enzymes (i.e desulfatase and dehydroxylase) and impact of gut microbiota composition in sulfur metabolism (incl. taurine) of BAs.

5.4.4 Key outcomes and drawbacks

Previous work comparing healthy and obese participants, showed that glycine- conjugated levels were significantly different but levels of unconjugated and tauro- conjugated were unchanged (Glicksman et al. 2010). Our findings relate to the emerging recognition of different metabolic adaptations within two ”types of fat deposition” in obesity; subcutaneous and visceral but were not compared to healthy controls. Unconjugated and sulfated BAs are increased while taurine conjugates are

131 Chapter 5-Results down-regulated in participants with visceral obesity. Nevertheless, these data suggest that the BA pool in the enterohepatic circulation is modified in visceral obesity. This outcome can probably result from up- or down- regulated enzymatic function related either to the FXR-liver axis (host) as exacerbated primary BA synthesis was observed in visceral obesity. Or to the gut microbiota as measured levels of tertiary BAs, taurine and sulfate conjugated BAs were significantly different between visceral and subcutaneous obesity. Therefore, information about the host-gut axis related to BA metabolism could have been elaborated by sequencing analysis to obtain informations of the gut microbiota composition and screen for genes expression related to gut microbiota and liver enzymes. This study presented some dietary design limitations and can be improved to obtain more robust results and confirm findings. The PPLR and PPGR are modulated by the amount and composition of dietary fat. The lipid-based diet challenge was suitable for the observation of a significant stratification according to visceral fat for insulinemia and insulin resistance. In this context, the lack of significant PPLR for TG and FFAs is paradoxical for a dietary challenge with high lipid content. There are several explanations for this. The scarcity of significant results could be related to i) the number of obese patients (n=10 per groups) resulted in lack of statistical power, ii) a total fat content in the milkshake that is not sufficient to trigger a significant variation iii) the absence of sample from the standardised meal to evaluate the exact nutrient content or, iiii) the absence of non-obese controls without hyperlipidemia or dyslipidemia.

132 Chapter 5-Results

5.5 Conclusion

This chapter, aimed to evaluate metabolic disturbances that can predict visceral obesity. The objectives here were to demonstrate that the PPLR was influenced by distribution of fat in visceral adipose tissue rather than subcutaneous adipose tissue. The analytical strategy relied on clinical biochemistry analysis and untargeted/targeted metabonomic approaches. PPLR was evaluated by UPLC-MS lipid profiling and by UPLC-MS/MS targeted BA method. Participants with visceral obesity display high levels of insulin and insulin resistance. Distinct responses of lipid and BA metabolism after a diet challenge were observed between subcutaneous and visceral obesity. The assessed PPLR shown that levels of two lipid species, PC (16:0/20:3) and LPC (16:0) were respectively lower at pre-prandial condition and higher at post-prandial conditions after 9h in visceral obesity compared to subcutaneous obesity. Additionally, BA metabolism and catabolism (sulfation) are up-regulated with decrease of tauro-conjugated BAs in visceral obesity which suggests differential underlying physiological processes compared to subcutaneous obesity. For example, impaired gut microbiota hydrolase activity might be related to the decrease of taurine conjugated observed in visceral obesity. This study further highlights the role of host and gut microbiota interaction in different fat deposition in obesity and its associated metabolic disorders such as insulin resistance. Altogether, lipid and BA measurements were implemented on a small population (n=10) and would have to be reproduced and confirmed on a larger population to obtain robust statistical results.

133 Chapter 6

Metabonomic analysis of lipids and bile acids circulating in NAFLD and NASH

6.1 Introduction

As described in Chapter 1 (cf. section 1.2.b) NAFLD is mainly characterised by excessive fat deposition in hepatocytes (Dumas et al. 2014). Due to the rising prevalence of NAFLD, multiple epidemiological studies were carried out to evaluate risk factor variation associated with genetic and environment is critical in disease progression. For example, Asian individuals develop NAFLD with lower BMI threshold than western patients but with similar prevalence (Fan et al. 2007; Das et al. 2010). Although the aetiology mechanism is not fully understood, mice and human studies have demonstrated that risk factors such as obesity, insulin resistance, diabetes and hypertension which can lead to metabolic syndrome are principal drivers of NAFLD (Anstee et al. 2013a; Dowman et al. 2010; Marchesini et al. 2001). The pathogenesis of this liver disease can progress to NASH by the occurrence of hepatocyte injury inducing inflammation or fibrosis (Marchesini et al. 2001; Browning et al. 2004a; Roden 2006). These conditions may lead ultimately to cirrhosis and hepatocellular carcinoma. Two hypotheses have been suggested to explain NAFLD progression to NASH. Initially a ”double hit” hypothesis was proposed (Day et al. 1998). The first hit consists of TG accumulation in hepatocytes and the second hit triggers hepatocyte

134 Chapter 6-Results injury and leads to NASH (e.g. inflammation, fibrosis). Nowadays, a more complex hypothesis is proposed where lipids play a pivotal role in NAFLD. Adipose tissue lypolysis generates increased of circulating FAs which induces excess synthesis of hepatic TGs and DGs responsible of insulin resistance (Perry et al. 2014). In addition, convergent studies highlighted the critical contribution of the gut microbiota which alters multiple host physiological mechanisms via metabolic mediators (Dumas et al. 2014; Mehal 2013). Impaired gut microbiota function in NAFLD/- NASH may affect human metabolism and may modulate lipid homeostasis (Dumas et al. 2014; Beyoglu˘ et al. 2013), endocannabinoid (Muccioli et al. 2010; Cani 2012), triglyceride, BA (Jiang et al. 2015) and FA metabolism (Bäckhed et al. 2004; Hover- stad et al. 1986). This chapter aims to determine metabolic variations in plasma that can explain the transition of NAFLD to NASH. In this context, this challenge was addressed by conducting comprehensive lipid profiling and BA targeted analysis of blood plasma in a human NAFLD cohort. NAFLD severity was assessed using the NAFLD activity score (NAS) based on histological analysis of liver biopsy. In particular, NAFLD was investigated to identify: (1) markers of biochemical processes and (2) lipid/BAs metabolic signature involved in the transition of NAFLD to NASH. The advantages of applying recently developed analytical methods for sample preparation dedicated to lipid profiling (Chapter 3) and to BA targeted quantification (Chapter 4) on UPLC-MS platform are shown in this chapter. The results outlined in this chapter offers new insights into the lipid and BA metabolism in transition to NAFLD to NASH diseases. A total of 132 plasma samples from healthy controls and patients with NAFLD and NASH were analysed. Metabolite variation between each group of patients was evaluated to characterise markers and to identify a metabolic signature associated with the transition of NAFLD to NASH.

135 Chapter 6-Results

6.2 Materials and methods

6.2.1 Materials

See Chapter 3, section 3.2 and Chapter 4, section 4.2 (Materials and methods).

6.2.2 Human plasma samples

This study was performed with a cohort of 132 participants. The disease stage was determined by calculating the NAFLD activity score (NAS), according to degree of steatosis (1 to 3), inflammation (0 to 3), ballooning (0 to 2) and fibrosis (0,1a,1b,1c- 3). The samples were diagnosed as healthy (n=54 with NAS=0), with NAFLD (n=24 with NAS=2.5) and with NASH (n=54 with NAS=4.5). The population ethnicities represented in this data are Caucasian, Asian/Pacific islander, African American and Hispanic. Details of clinical characteristics are presented in Table 6.1. Participants were included if they were between 20-65 years old and with a BMI between 18.5-40 kg/m2. Participants were excluded if they reported that they were consuming alcohol (>20 g/day females, >30 g/day males), enabled to quantify alcohol consumption, actively smoking, using recreational drugs (within previous 2 months), currently participating or having participated in an interventional clinical trial during the last 1 month prior to the beginning of this study, strict vegetarians, pregnant or lactating women and lost weight in the past 1 month (5%). Participants with illness were also excluded, such as type 1 diabetes mellitus, be on transplant medications, or have a history of taking chemotherapy within the preceding year, stage 4 heart failure, post liver transplant, viral, drug related, autoimmune, genetic liver diseases and with terminal illnesses (life expectancy <1 year).

6.2.3 Sample preparation

136 Chapter 6-Results

µL) (Saraﬁan et al. 2014). Samples for lipid proﬁling and the BA targeted method were vortexed for 10 min and cold isopropanol was added (150 µL). Afterwards samples were stored 15 min at –20 °C to improve protein precipitation and then centrifuged at 14 000g for 20 min. The supernatant was collected (150 µL) and stored at –80°C awaiting MS analysis.

6.2.4 Lipid proﬁling. Ultra-Performance Liquid Chromatogra-

phy

See Chapter 3, section 3.2 (Materials and methods).

6.2.5 Lipid proﬁling. Quadrupole-Time-of-Flight Mass Spec-

trometry

See Chapter 3, section 3.2 (Materials and methods).

6.2.6 BA targeted assay. Ultra-Performance Liquid Chromatog-

raphy

See Chapter 4, section 4.2 (Materials and methods).

6.2.7 BA targeted assay. Triple Quadrupole Mass Spectrom-

etry

See Chapter 4, section 4.2 (Materials and methods).

6.2.8 MS Data Preprocessing

See Chapter 3, section 3.2 and Chapter 4, section 4.2 (Materials and methods).

137 Chapter 6-Results

6.2.9 Drift correction

Probabilistic quotient normalisation was applied to quality control samples (QCs) to correct the drift related to analysis time (Dieterle et al. 2006) using MATLAB (R2014a v8.3).

6.2.10 Univariate and multivariate statistical analysis

PCA was carried out on the XCMS extracted intensities using SIMCA P+ v13 (Umetrics, Umea, Sweden). An OPLS-DA model was used to identify the features that contributed to separate controls, NAFLD and NASH individuals. Permutation testing was applied with 1000 permutations. Lipid and BA data were analysed to correct fo multiple testing with a Benjamini-Hochberg false discovery rate FDR<5% using MATLAB (R2014a v8.3). Heatmaps and hierarchical clustering were generated using Spearman’s correlation with centroid linkage.

6.2.11 Lipid structural identiﬁcation

See Chapter 3, section 3.2 (Materials and methods).

138 Chapter 6-Results

6.3 Results

In this study, the metabolic perturbations associated with NAFLD and NASH were evaluated. Global lipid proﬁling and a targeted BA method were implemented on 54 healthy control participants, 24 patients with NAFLD and 54 patients with NASH.

6.3.1 Changes observed in plasma of NAFLD vs. control par-

ticipants

Blood biochemistry data aimed to evaluate levels of biomarkers associated with metabolic syndrome including liver function, renal function, inflammation, diabetes and obesity at fasting condition. As presented in Table 6.1, NAFLD and NASH patients were diagnosed by screening of various clinical biochemistry data. Con- trol individuals were significantly older (age>60 with p<5.33x10-5) than NAFLD and NASH patients. Both male and female were equally recruited for the three different groups. The majority of NAFLD and NASH patients were considered with respectively BMI>25 as overweight (96% and p<7.35x10-13) and BMI>30 as obese (64% and p<2.33x10-32) compared to control participants (overweight 39% and obese 1%). NAFLD patients compared to controls were associated with significantly higher levels of ALT (p<3.26x10-4), AST (p<7.58x10-3), glucose (p<2.66x10-4), insulin (p<0.002), triglycerides (p<0.018) and high-sensitivity C-reactive protein (Hs.CRP) (p<9.44x10-4). Glycated hemoglobin (HbA1c) exhibited a significant increase in NASH patients (p<3.94x10-6) compared to controls. However, noticeable that γ- glutamyl transferase (GGT) levels were lower at advanced disease stage (NASH) compared to early disease stage (NAFLD) (p<0.014). There was also evidence of significant improvement of high density lipoprotein (HDL) in NASH patients (p<3.19x10-7) compared to NAFLD patients. Creatinine, cholesterol and low density lipoprotein (LDL) levels were not significantly disturbed in NAFLD and NASH patients.

139 Chapter 6-Results

Table 6.1: Summary of blood clinical biochemistry data.

Controls NAFLD NASH Normal levels (n=54) (n=24) (n=54) (min;max) (min;max) (min;max) Gender - 46 46 50 Ethnicity - 90 92 83 61.6 ± 13.7 48.9 ± 12.4*** 53.3 ± 10.8*** Age - (20.71;88.1) (31.84;77.08) (26.95;73.07) 24.5 ± 2.5 33.6 ± 5.5*** 33.5 ± 5.5*** BMI (kg/m2) 12.5 - 25 (20.31;38) (24.76;42.03) (19.14;46.47) 23.9 ± 34.7 60.3 ± 39.1*** 87.2 ± 167.1** ALT (U/L) <30-40 (6;188) (17;137) (12;1229) 26.8 ± 20.0 45.0 ± 28.3 85.2 ± 251.8** AST (U/L) <30-40 (14;150) (13;113) (16;1868) 41.9 ± 97.6 125.7 ± 217.0 87.2 ± 91.5* GGT (U/L) <50 (6;596) (13;892) (17;385) 0.9 ± 0.2 0.8 ± 0.2 0.8 ± 0.2 Creatinine(mg/dL) 0.7-1.3 (0.5;1.4) (0.43;1.1) (0.4;1.5) 87.3 ± 9.2 104.5 ± 18.7*** 125.8 ± 57.1***/* Glucose (mg/dL) <100 (49;106) (85;159) (60;370) 6.7 ± 5.1 16.5 ± 13.2** 27.4 ± 24.8***/* Insulin mU/L 5-20 (0.8;38) (2.5;53.5) (1.3;138.2) 5.2 ± 0.5 5.6 ± 1.1 6.2 ± 1.4*** HbA1c % 4-5.9 (4;6.7) (4.3;8.7) (4.5;10.3) 184.0 ± 53.6 198.3 ± 32.1 196.6 ± 40.6*** Cholesterol (mg/dL) <120-180 (86;298) (138;283) (105;292) 53.8 ± 13.2 48.2 ± 13.1 40.4 ± 12.3***/* HDL (mg/dL) 40-60 (26;92) (34;97) (5;79) 109.4 ± 45.3 118.0 ± 24.5 118.5 ± 37.0 LDL (mg/dL) 100-129 (8;198) (76;170) (36;206) 105.5 ± 73.1 163.2 ± 100.8* 201.8 ± 118.4*** Triglycerides (mg/dL) <150 (38;453) (71;472) (58;560) 1.7 ± 2.4 4.2 ± 2.9*** 6.3± 7.1*** Hs.CRP (mg/L) <3 (0.2;10.7) (0.5;11.5) (0.1;33.3) Data are expressed as ± standard error of the mean for each variable with minimum and maximum values. Student’s t-test was applied to evaluate signiﬁcant diﬀerence between control participants vs. NAFLD and vs. NASH patients (*) and between NAFLD and vs. NASH patients (*) p <0.05 **p<0.01 ***p<0.001

Physiological data showed that NAFLD and NASH are characterised by gradual decline of liver functions and other metabolic functions related to the metabolic syndrome and highlighted the complexity of the mechanisms involved in the progression of this disease. The possible effects of risk factors on prevalence to NAFLD/NASH and progression of NAFLD to NASH were evaluated and correlations between the physiological variables were tested (Figure 6.1). BMI and age were identified as potential risk factors as mean values in each group was significantly different from control participants (Table 6.1). The association of BMI and age with NAFLD/NASH and other variables

140 Chapter 6-Results was assessed. As observed in Figure 6.1, variations between groups of the variable diagnosis (i.e. control participants, NAFLD and NASH patients) were driven by BMI. BMI was strongly correlated with NAFLD and NASH patients and was highly signif-

-22 icant (R=0.767, padjBH<6x10 ). The age variable appeared to underestimate the association with NAFLD and NASH (padjBH<0.001). Age was still highly signiﬁcant using adjustment with Benjamini-Hochberg method but depicted a weaker association

-4 to NAFLD (R=-0.331, padjBH<4x10 ). Gender and ethnicity were not signiﬁcantly correlated to the diagnosis of the patients.

Figure 6.1: Correlation structure between physiological variables. Heat-map showing the correlations values and colors are scaled as follows; positive (red) and negative (bleu) correlations. *padjBH<0.05, **padjBH<0.01 and ***padjBH<0.001

The association of BMI and age on NAFLD/NASH progression was assessed. Data presented in the next sections were corrected for BMI, age, gender and ethnicity.

141 Chapter 6-Results

6.3.2 Lipid metabolism modiﬁcation correlates with NAFLD

and NASH

Previous work highlighted the straightforward and robust isopropanol precipitation as the optimal sample preparation method of choice for lipid extraction (cf. Chapter 3). In this study, isopropanol precipitation was applied to assure ideal conditions to decipher the lipidome signature associated with the transition to NAFLD to NASH.

(a) UPLC-MS lipid proﬁling

Features detected in positive mode (n=1,665) and negative mode (n=926) in lipid proﬁling were observed to have higher intensities in NAFLD and NASH than control participants (Figure 6.2). Levels of circulating lipids were exacerbated in NAFLD and NASH compared to control individuals.

Figure 6.2: Lipid proﬁling chromatograms of the three groups; controls (A), NAFLD (B) and NASH (C). The mean average of the combined peak intensities is represented for each group.

The run strategy adopted in this study was previously presented (cf. Chapter 2.4) and QCs were used to assess the robustness of the system. Variability and potential outliers within the three groups were determined by unsupervised PCA analysis in both positive and negative modes. Analysis of the ﬁrst two components revealed that QCs (cf. Chapter 2.4) were shifted during analysis due to run order eﬀect both in positive (Figure 6.3A) and negative modes (Figure 6.3D). Drift correction was

142 Chapter 6-Results implemented to correct that run order eﬀect as observed in Figure 6.3B and Figure 6.3E. The frequency of CVs were plotted to assess the reproducibility of the QCs before (Figure 6.3C) and after drift correction (Figure 6.3F). Corrected QCs were clearly improved and showed high reproducibility (<15%) with 91% features below 15% in positive mode and 82% features below 15% in negative mode. However, after correction for run order eﬀect, it was not possible to separate the control, NAFLD and NASH participants from each other in the PCA scores plots.

Figure 6.3: Drift correction of run order eﬀect occurring in UPLC-MS lipid proﬁling. PCA scores plots and CVs of positive (A, B and C) and negative mode (D, E and F). Raw data were compared before (A and C) and after drift correction (B and D).

(b) OPLS and OPLS-DA models demonstrated systemic differences in lipid profiles between controls, NAFLD and NASH profiles

To further assess the diﬀerences between the disease groups, supervised OPLS and OPLS-DA models were built to explain class disease group membership (e.g. class 1 controls, class 2 NAFLD and class 3 NASH). Cross-validation and permutation tests enabled robustness and validation respectively to be determined for each model (p<0.001, number of permutations=10000). Prior to the modelling step, the data were scaled using univariate scaling which produced the best goodness of ﬁt (R2)

143 Chapter 6-Results and prediction ability (Q2). The OPLS and OPLS-DA parameters were summarised for the ﬁve models that were computed (Table 6.2). Variations between groups were assessed by OPLS and OPLS-DA models. Plasma lipid proﬁles of patients were compared as follows: OPLS with control (1) vs. NAFLD (2) vs. NASH (3) (model 1) and OPLS-DA with control (1) vs. NAFLD (2) vs. NASH (3) (model 2), control (1) vs. NAFLD (2) (model 3), control (1) vs. NASH (3) (model 4), NAFLD (2) vs. NASH (3) (model 5) (Figure 6.4).

Table 6.2: One OPLS model (1) and OPLS-DA models (2,3,4,5) showing the comparison of the control, NAFLD and NASH positive mode lipid proﬁles.

Permutation test Model Group comparisons PC OC R2X R2Y Q2Y p-value 1 Control-NAFLD-NASH 1 1 0.204 0.548 0.38 1x10-4 2 Control vs. NAFLD vs. NASH 1 3 0.306 0.423 0.195 1x10-4 3 Control vs. NAFLD 1 1 0.213 0.504 0.147 1x10-3 4 Control vs. NASH 1 1 0.209 0.638 0.439 1x10-4 5 NAFLD vs. NASH 1 1 0.198 0.346 -0.287 0.700

PC: Predictive Component; OC: Orthogonal Component; R2X: X variables explained by the model; R2Y: Y variables explained by the model; Q2Y: predictive quality of the model showing all models were robust

144 Chapter 6-Results

Figure 6.4: OPLS and OPLS-DA scores plots. OPLS of control participants-NAFLD- NASH samples (A; model 1), OPLS-DA of control participants vs. NAFLD vs. NASH samples (B; model 2), OPLS-DA of control participants vs. NAFLD (C; model 3), OPLS- DA of control participants vs. NASH samples (D; model 4) and OPLS-DA of NAFLD vs. NASH samples (E; model 5).

The ﬁrst model (Model 1) is an OPLS regression model that was computed with the entire data to assess the linearity of the disease as follows; control (1)> NAFLD (2)> NASH (3) and the Y variables was built according to these three groups. Models 2, 3, 4 and 5 are OPLS-DA analysis. The OPLS-DA scores plots obtained from control (1) vs. NAFLD (2) vs. NASH (3) (model 2) and control (1) vs. NASH (3) (model 4) discriminated NASH patients from the control participants along the ﬁrst predictive

145 Chapter 6-Results component. In addition, the robustness of these two models were confirmed by evaluating the R2Y, Q2Y, and validated using permutation tests (p<1x10-4). The best model obtained was the pair-wise OPLS-DA that explained 64% (and predicted 44%) of the lipid variation between Controls and NASH (model 4). Linearity of the disease was explained by 55% and predicted 38% of the variation between Controls, NAFLD and NASH with UPLC-MS variables with the OPLS model (model 1). The three different disease groups hypothesis was explained by 32% which is less variation than with the linear model and only predicts 21% of the variation (model 2). Plasma lipid profiles from NAFLD patients were not discriminated neither from controls participants (model 2) nor NASH patients (model 4). The predictive OPLS-DA score plot from control vs. NASH (model 4) was generated to evaluate a potential lipidomic link between NAFLD patients and the other two groups, NASH and control participants. As presented in Figure 6.5, NAFLD participants were predicted in the model built from NASH and control participants. Such observations suggest that NASH patients induced a greater lipidomic variation (R2Y=0.638) than NAFLD patients (R2Y=0.504).

Figure 6.5: Predictive OPLS-DA score plot of control participants vs. NASH plotted and NAFLD participants were predicted.

146 Chapter 6-Results

In order to visualise the intrinsic similarities and dissimilarities among the control participants, NAFLD and NASH patients, correlation tests with lipidomic data were carried out on all groups. We observed that a total of 1,552 and 476 significant Spear- man’s rank-based correlations (significance threshold at padjBH<0.05) respectively for non-adjusted correlations and for gender, ethnicity, age, BMI-adjusted partial correlations. BMI showed significant variations across control>NAFLD>NASH (Figure 6.6). Over 138 features associated with BMI were shown to change severely the hierarchical clustering of physiological variables and to overestimate differences of NAFLD and NASH to control participants. Gender and ethnicity were found to be confounding factors as both were associated with lipid profiles but as observed previously not to the disease (Figure 6.1). As a consequences partial correlations were derived using gender, ethnicity, age and BMI as covariates, for the subsequent analyses in this chapter. The correlation matrix emphasised several metabolic clusters indicating a significant lipid signature. Comparison of the three groups was implemented to highlight the significant lipid variations occurring between each group as follows; control vs. NAFLD (Figure 6.6A), control vs. NASH (Figure 6.6B) and NAFLD vs. NASH (Figure 6.7). The greatest difference between in lipid profiles was generated by the diagnosis variable that defines the participants as NAFLD, NASH or control (Figure 6.7). Over 304 significant features correlated in NASH vs. controls model (Figure 6.7A), 105 significant features correlated in NAFLD vs. controls model (Figure 6.7B) and 23 of these significant features were highly correlated to both models (Figure 6.7C).

147 Chapter 6-Results

Figure 6.6: Correlations between physiological data of 132 human plasma samples (control, NAFLD and NASH) and features detected in lipid profiling positive mode. Heatmaps and hierarchical clustering were performed using Spearman’s rank based correlation with centroid linkage with data non-adjusted (A) or adjusted for gender, ethnicity, age, and BMI (B). Only features that were significantly different at the padjBH<0.05 level were selected. Red represents highly correlated features and blue showed anti-correlated features.

148 Chapter 6-Results

Figure 6.7: Heatmaps and hierarchical clustering showing 105 (A) and 304 (B) correlations observed in lipid proﬁling positive mode. Partial Spearman’s rank based correlations were performed and adjusted for gender, age, ethnicity and BMI. Only features that were significantly diﬀerent at the padjBH<0.05 level were selected. Red represents highly correlated features and blue showed anti-correlated features.

149 Chapter 6-Results

Further partial correlations (gender, ethnicity, age and BMI-adjusted) were conducted to clarify the lipid modified in the transition to NAFLD to NASH. In order to facilitate the interpretation, structural assignment of the significant features was implemented. A total of 74 significant correlations were observed including 24 correlations from seven structurally assigned lipid species (see structural assignment example in Chapter 2 Methods, and summary of structures listed in Table 6.3). Over 50 correlations were not characterised due to a low ratio of signal to noise or to coeluting lipids or to missing fragmentation spectra or to unknown m/z of peaks detected. Three lipid classes (5 phosphocholines PC, 1 SM and 1 TG) displayed significant correlations with the physiological variables (Figure 6.8). Correlated PCs were structurally composed of 16 to 20 carbons with 1 to 4 unsaturations. The PC (16:0/20:4) has the highest correlation coefficient (ρ=0.451,

-4 padjBH=8x10 ) for the diagnosis variable with higher concentrations observed in NAFLD patients. Choline species (PCs) were strongly correlated to risk factors in both NAFLD and NASH groups which suggested important variations in PC metabolism. TG (16:1/18:2/18:2) level was increased and positively correlated to

NASH (ρ=0.3523, padjBH=0.0291). These ﬁnding might suggest that the metabolism of TG with long FA chains and unsaturations, is altered in NASH. As expected, the two main lipid classes associated with the transition to NAFLD to NASH are PCs and TGs. Clearly, the excess of lipid stored in tissues is spilled over the systemic circulation and/or BA pool helping lipid absorption might be impaired.

150 Chapter 6-Results

Figure 6.8: Heatmaps and hierarchical clustering showing 74 correlations observed in positive mode by lipid proﬁling UPLC-MS. Partial Spearman’s rank based correlations were performed and adjusted for gender, ethnicity, age and BMI. Only features that were significantly diﬀerent at the padjBH<0.05 level were selected. Red represents highly correlated features and blue showed anti-correlated features.

Table 6.3: Characterisation of features intensity that correlate to physiological data obtained from NAFLD vs. NASH patients by UPLC-MS lipid proﬁling. With retention time (RT), p-value adjusted (padjBH).

RT Correlation Structural m/z Ions Physiological data p min adjBH coeﬃcients rho identiﬁcation 788.614 7.81 [M+H] + ALT 0.0004 0.4686 PC (18:0/18:2) 789.367 8.08 [M+H] + ALT 0.0429 0.3395 PC (18:0/18:2) 845.748 14.95 [M+NH4] + ALT 0.0443 -0.3263 TG (16:1/18:2/18:2) 788.614 7.81 [M+H] + AST 0.0243 0.3211 PC (18:0/18:2) 814.938 12.6 [M+H] + AST 0.042 0.316 SM (d18:1/24:1) 845.748 14.95 [M+NH4] + AST 0.0443 -0.314 TG (16:1/18:2/18:2) 763.144 7.56 [M+H] + Cholesterol 0.0263 0.3326 PC(16:0/18:1) 761.848 7.58 [M+H] + Creatinine 0.0069 -0.3958 PC(16:0/18:1) 788.614 7.81 [M+H] + Creatinine 0.0216 -0.3392 PC (18:0/18:2) 763.869 7.57 [M+H] + Creatinine 0.0341 -0.3362 PC(16:0/18:1) 760.017 6.11 [M+H] + Glucose 0.0478 -0.3112 PC(16:0/18:2) 760.017 6.11 [M+H] + HbA1c 0.0209 -0.3628 PC(16:0/18:2) 789.248 8.07 [M+H] + HDL 0.0061 0.3992 PC (18:0/18:2) 762.752 6.11 [M+H] + HDL 0.0448 0.3381 PC(16:0/18:2) 763.869 7.57 [M+H] + Insulin 0.0341 0.3235 PC(16:0/18:1) 762.399 6.11 [M+H] + LDL 0.0125 0.3784 PC(16:0/18:2) 814.938 12.6 [M+H] + LDL 0.042 -0.3254 SM (d18:1/24:1) 808.587 8 [M+Na] + LDL 0.0474 0.3361 PC (18:0/18:2) 784.82 5.86 [M+H] + Diagnosis 0.0008 0.451 PC(16:0/20:4) 876.72 15.01 [M+Na] + Diagnosis 0.0291 0.3523 TG (16:1/18:2/18:2) 815.641 10.36 [M+H] + Triglycerides 0.0105 0.3836 PC (18:0/20:2) 763.144 7.56 [M+H] + Triglycerides 0.0215 0.3619 PC(16:0/18:1)

151 Chapter 6-Results

6.3.3 BA metabolism impaired in NAFLD patients

To explore the potential impaired BA metabolism on transition to NAFLD to NASH, we determined the concentrations of 145 BAs in plasma. Preliminary work showed the robustness of the BA targeted UPLC-MS/MS method and was applied on this study (Chapter 4).

(a) Targeted BA assay

This targeted assay aimed to monitor 145 BAs and 57 of these were detected in plasma samples. Total plasma BA concentrations were significantly higher (p<0.0081) in NAFLD and NASH patients compared to healthy participants. An increase of 12% and 83% NAFLD and NASH patients was respectively observed. These data suggest that the contribution of specific BA species is clearly altered in NAFLD and NASH patients. Further investigations of different BA species were implemented (Table 6.4 and Figure 6.9).

(b) Increase in circulating BA observed in NAFLD and NASH

As shown in Figure 6.9A, proportions of unconjugated BAs varied between the three groups. The level of total primary BAs (CA and CDCA) tended to decrease in NAFLD patients when compared to control participants (CA p<0.014). Signifi- cant increase in primary BAs was observed in NASH patients compared to control participants (CDCA p<0.0166). Consequently, this accumulation of primary BAs in NASH patients led to significant increase of secondary BAs such as DCA in NASH compared to control participants (p<0.0117) and NAFLD (p<0.0025). Furthermore, these findings indicated increased concentrations of glycine and taurine conjugates as well as sulfates in plasma of NASH patients. Indeed, increased concentrations of both primary BAs and their derivates induced an increase in conjugated types such as glyco-conjugates, tauro-conjugates and sulfate in NASH patients (Table 6.4). Glycine conjugates were the most commonly detected BA conjugates found in plasma. GCDCA was the most abundant and significantly

152 Chapter 6-Results increased in both NAFLD (p<3x10-5) and NASH patients (p<8x10-11) compared to control participants. Interestingly, taurine conjugates were significantly increased in NASH compared to NAFLD (p<0.0031) but significantly decreased in NAFLD compared to control participants (p<0.0009) (Figure 6.9A). Moreover, a significant increase of sulfated BAs was observed in NASH compared to control participants (p<0.0029) (Figure 6.9A). The sulfated form of GCDCA show a significant increase in both NAFLD (p<0.026) and NASH patients (p<0.003) compared to control participants. Clearly, these results strongly indicate that NAFLD and NASH patients were exhibiting severe BA metabolism modifications, with disturbed taurine and sulfate pathways.

153 Chapter 6-Results

Figure 6.9: Quantities of BAs species circulating in control participants, NAFLD and NASH patients according to classes; unconjugated BAs, conjugated BAs with taurine, glycine or sulfates (A and B). Error bars refers to the standard deviation divided by the square root and Student’s t-test was applied to evaluate signiﬁcant diﬀerence *p<0.05 ***p<0.001.

154 Chapter 6-Results

Table 6.4: Means of quantiﬁed BA species (n=57) circulating in control participants, NAFLD and NASH patients expressed in nM. Statistically variations are explained with standard deviation divided by the square root and with t-test as follows; control participants vs. NAFLD and vs. NASH patients (*) and between NAFLD and vs. NASH patients (*) p<0.05 **p<0.01 ***p<0.001

Control Steatosis NASH Standard Standard Standard Mean Mean Mean deviation deviation deviation CA 346.47 95.57 285.97 69.51 * 799.04 197.7 CDCA 174.63 37.77 158.53 29.7 393.78 72.24 * UDCA 152.3 27.51 410.35 258.49 201.33 30.79 DCA 113.95 14.72 97.32 24.98 113 19.68 */** 4-Cholenic Acid-3β,6β diol 68.22 16.92 63.09 18.24 100.91 21.98 5-Cholanic Acid-3α 12α diol 61.24 7.72 55.92 15.42 51.55 12.89 HDCA 54.11 5.1 59.19 1.57 80.94 17.41 **/** 12-DehydroCA 52.14 9.6 32.23 6.07 40.72 8.99 LCA 29.92 8.23 25.06 6.65 23.32 4.2 9(11),(5)-Cholenic acid 3 ol 12 one 21.05 2.56 29.65 4.09 24.03 2.13 3α-Hydroxy-12-KetoLCA 18.49 4.37 15.83 5.96 19.68 5.12 AlloLCA 14.47 1.69 10.31 2.56 11.56 2 7 oxo-LCA 13.54 3.92 15.11 4.92 24.71 6.5 Ursocholanic acid 11.05 0.75 11.89 1.02 12.53 0.78 3,7,12-DehydroCA 9.46 1.57 21.81 9.4 15.05 2.97

Unconjugates 5-Cholenic acid 6.37 0.63 5.98 0.64 6.95 0.55 HCA 5.42 1.06 4.39 0.96 10.81 1.81 8(14),5β-Cholenic acid-3α,12α diol 4.04 0.47 2.52 0.39 4.08 0.77 IsoLCA 3.3 0.49 1.86 0.58 1.72 0.34 3-Ketocholanic Acid 3.24 0.62 2.9 0.91 2.51 0.42 αMCA 2.76 0.82 2.13 0.41 3.3 0.44 */* IsoDCA 2.32 0.38 2.4 0.27 2.77 0.27 NorDCA 2.18 0.3 1.58 0.37 1.32 0.2 Lithocholenic acid 2.02 0.44 2.57 1.06 2.34 0.39 5β-Cholanic acid-7α ol 3 one 1.82 0.38 2.15 0.49 2.67 0.46 3-DehydroCA 0.69 0.18 0.73 0.19 0.94 0.16 Diketocholanic acid 0.68 0.06 0.95 0.14 1.01 0.12 GCDCA 526.82 57.35 951.04 75.64 *** 1106.54 57.84 *** GUDCA 168.25 126.31 27.36 14.54 61.61 21.41 GDCA 164.83 109.55 114.25 37.2 116.6 21.64 G-Ursocholanic acid 105.85 49.68 50.16 4.8 60.86 4.12 Glycines GHCA 11.69 5.7 1.81 0.51 7.1 2.36 * GHDCA 4.19 2.09 2.2 0.27 3.86 0.52 ** TCDCA 213.45 111.55 72.63 39.79 264.18 93.96 TDCA 91.45 62.08 20.44 13.26 29.77 11.1 TCA 33.19 21.3 11.77 7.25 55.78 25.18 TαMCA 7.49 3.45 1.42 1.05 11.77 6.28 TβMCA 28.79 8.65 22.65 7.43 72.67 21.54 * TωMCA 29.15 10.06 8.46 6.15 60.09 31.29 Taurines TUDCA 10.54 4.29 12.6 4.99 34.92 8.02 **/* THDCA 8.14 7.31 0.73 0.17 1.16 0.23 THCA 3.74 2.08 0.81 0.43 5.96 2.67 3,7,12-TDehydroCA 1.4 0.44 1.16 0.1 2.69 1.19 GLCA-S 271.19 72.29 344.87 105.2 308.43 48.75 GDCA-S 57.45 15.34 96.98 22.25 89.41 12.94 UDCA-S 43.67 16.9 80.75 29.33 51.79 11.35 GCDCA-S 41.72 11.93 96.01 20.23 * 111.81 19.94 ** 5-Cholanic acid-3α,12α diol-S 37 7.06 59.92 14.88 60.6 15.46 GUDCA-S 33.11 8.85 368.76 302.03 193.3 127.24 4-Cholenic acid-3β,6β diol-S 22 5.2 33.38 7.77 30.73 6.31

Sulfates 5-Cholenic acid-S 17.41 1.61 25.75 3.44 * 43.66 10.62 * IsoLCA-S 13.15 2.13 15.85 3.36 18.48 3.43 TDCA-S 8.9 1.75 16.04 2.67 * 22.98 6.95 TCDCA-S 5.89 2.282734 5.66 1.67 17.72 10.14 IsoDCA-S 5.34 1.09 12.27 2.53 * 20.32 3.48 *** LCA-S 0.98 0.12 1.57 0.32 1.88 0.47

155 Chapter 6-Results

To further validate the signiﬁcance of the variations observed amongst BA species in plasma of control participants, NAFLD and NASH correlations tests were applied. A heatmap correlating UPLC-MS/MS quantiﬁcations of 145 BAs with physiological data from Table 6.1 (i.e diagnosis refers to the three groups: control participants,

NAFLD and NASH patients) was generated with significant correlations (padjBH<0.05) and adjusted for the effect of gender, ethnicity, age and BMI (Figure 6.10). Each heatmap summarises Spearman’s rank based partial correlation values and corresponds to different combinations between groups as follows; control vs. NAFLD vs. NASH (Figure 6.10A), control vs. NAFLD (Figure 6.10B), control vs. NASH (Figure 6.10C) and NAFLD vs. NASH (Figure 6.10D). As shown in Figure 6.10A the main BA signature occurring in NAFLD and NASH is the correlation of conjugated BAs with 4 taurine, 5 glycine or 7 sulfates. In agreement with previous results (Figure 6.9 and Table 6.4) most of the sulfated BAs positively correlate with NAFLD and NASH disease including three of them which were positively correlated with TGs levels. Among these sulfated BA species, TDCA ap-

-6 peared to exhibit the highest correlation (ρ=0.434) and signiﬁcance (padjBH=4.8x10 ) with NAFLD and NASH. In addition, it was apparent that CDCA derivates such as glycine (GCDCA), glycine sulfates (GCDCA-S) and taurine sulfates (TCDCA-S) were strongly associated with NAFLD and NASH. The two strongest positive correlations associated with ALT levels were secondary sulfated BAs and negative correlations were two secondary BAs. These BAs are respectively glycine and taurine sulfates forms of

-6 CDCA (ρ=0.367, padjBH=0.0003 and ρ=0.434, padjBH=4.8x10 ), GLCA and TDCA

(ρ=0.359, padjBH=0.0005 and ρ=0.342, padjBH=0.001). Furthermore, glycine conjugates mainly negatively correlated with insulin levels and taurine conjugates mainly negatively correlates with ALT levels. In Figure 6.10B, a sulfated BA derived from the secondary BA, DCA (i.e. isoDCA) was positively correlated to NAFLD disease. IsoDCA had the highest positive correlation (ρ=0.384, padjBH=0.010). Conversely, negative correlations with NAFLD were

156 Chapter 6-Results associated with another secondary BA, LCA (i.e. isoLCA acid and alloLCA). Conju- gated BAs with taurine and glycine were mainly negatively correlated with ALT and insulin. GLCA had the highest negative correlation (ρ=-0.448, padjBH=0.0008) which confirm results obtained in Figure 6.10A. In Figure 6.10C, the sulfated BA pattern observed in NAFLD patients (Figure 6.10B) was clearly accentuated in NASH patients. These sulfated BA species were positively correlated with ALT, AST, GGT, TGs and glucose levels and negatively correlated with HDL. Conjugated BAs with taurine and glycine were negatively correlated to ALT levels but with lower correlation coefficients and less significance than in NAFLD patients. Similar results were found for the negative correlations of isoLCA with the diagnosis of NASH patients. In Figure 6.10D, hyocholic acid (HCA) is highly correlated to NASH (ρ=0.337, padjBH=0.359) and presented higher concentrations in blood of NASH than NAFLD patients. Significant correlations were obtained between liver enzymes ALT (ρ=0.316, padjBH=0.044) and AST (ρ=0.335, padjBH=0.044) with THCA. Also in this model, significant positive correlation of sulfated 5-cholenic acid with TG levels (ρ=0.382, padjBH=0.012) and negative correlation of TCA with insulin levels (ρ=-0.368, padjBH= 0.018) were observed. Overall, these results provide evidence that altered composition of lipids and BA pool circulating are associated to liver disease when compared to healthy participants. Key biomarkers were found to be significantly elevated in NASH. The BA targeted method aimed to identify HCA as an important biomarker for transition to NAFLD to NASH.

157 Chapter 6-Results NASH (D). 0.05 level were vs. < adjBH NASH (C) and NAFLD vs. NAFLD (B), control vs. NASH (A), control vs. NAFLD vs. Heatmaps and hierarchical clustering dendrograms correlations observed in negative mode with BA targeted UPLC-MS/MS method. Partial spearman correlations wereselected. performed adjusted Red for represents gender, highly ethnicity, correlated age features and and BMI. blue Only showed anti-correlated BA features. with signiﬁcant values at the p Correlations are presented as follows; control Figure 6.10:

158 Chapter 6-Results

6.4 Discussion

The current concept of the NAFLD and NASH is that several metabolic alterations are responsible for the pathogenesis and that disease disorders (e.g. insulin resistance) can be associated with induction of metabolic syndrome factors (Browning et al. 2004a; Marchesini et al. 2001; Roden 2006; Vanni et al. 2010; Kotronen et al. 2008). In the last few years, UPLC-MS based metabonomics has demonstrated potential to understand and predict NAFLD and NASH pathogenesis (Dumas et al. 2014). One of the current challenges lies in the analysis and evaluation of metabolic variations in the transition to stages from NAFLD to NASH to offer predictive metabonomic-based individualised health care. Diagnosis and prognosis is crucial, especially for NASH patients with high morbidity and mortality. In this Chapter, metabolic phenotyping (i.e. metabotyping) (Gavaghan et al. 2002) was used to characterise lipid profiles in plasma from control participants and from NAFLD and NASH patients and revealed association of the lipid profiles with various metabolic and clinical parameters (Table 6.1). Lipid profiling by UPLC-MS was complemented by a BA assay using targeted UPLC-MS/MS method highlighting differential metabolic signatures occurring in the different groups.

6.4.1 NAFLD and NASH diagnosis by metabonomics

NAFLD and NASH are still widely undiagnosed, the gold standard being an invasive technique, liver biopsy, which provides a direct score of disease stage (NAS score). Although if liver biopsy is prone to rare serious complications, it is the most reliable method for diagnosis so far. Information about eventual lipid accumulation and other disease characteristics such as inflammation, ballooning necrosis and fibrosis, are confirmed by liver histology. There are also non-invasive techniques such as transabdominal ultrasound, non-contrast CT scan and magnetic resonance imaging (Reeder et al. 2011; Saadeh et al. 2002; Mendler et al. 1998). However, these techniques present some sensitivity issues in appraising accurately the different stage of the disease. In this context, screening of multiple NAFLD predictors on biofluids

159 Chapter 6-Results

(i.e. plasma, urine) by a non-invasive metabonomic approach can help to decipher the pathogenesis of liver and peripheral tissues. Metabonomics oﬀers a wide range of metabolic and lipid species to evaluate complex physiological mechanisms. However, accurate assessment of NAFLD and NASH might be challenging as the disease can be accompanied by multiple metabolic syndrome factors (e.g. insulin resistance).

6.4.2 Blood clinical data associated with NAFLD and NASH

In this chapter, sixteen physiological factors were measured to evaluate changes between disease and control groups (Table 6.1). Previous assessments reported in the literature have identified BMI but not gender and age as predictor of NAFLD and NASH (Gorden et al. 2015; Zimmermann et al. 2015). However, patients with low BMI may also be at risk of developing NAFLD and NASH. For instance sus- ceptibility of NAFLD and NASH with low BMI has been reported mainly in Eastern countries (Margariti et al. 2012). The results in Table 6.1 outlined that the prevalence of NAFLD and NASH were strongly associated with dyslipidaemia and appear to be exacerbated by hyperglycaemia and hyperinsulinemia, which is in agreement with previous studies (Bugianesi et al. 2005; Gaggini et al. 2013). High levels of Hs.CRP observed in this study are a recognised marker for cardiovascular disease and inflammation and are related to NAFLD and NASH severity (Ndumele et al. 2011). In addition, abnormal levels of liver enzymes AST and ALT are commonly observed in NAFLD and NASH patients and are indicative of liver damage. Here, ALT and AST levels were reported to increase in NAFLD patients and NASH patients. How- ever, several studies have shown that AST and AST as poor predictors in advanced fibrosis (Dyson et al. 2013). NAFLD patients were characterised by an increase of GGT levels which is a frequent indicator of liver dysfunction. GGT is synthesised by hepatocytes and plays an important role in oxidative stress reduction (glutathione synthesis). Several studies showed that GGT levels are associated with risk factors such as cardiovascular and type 2 diabetes (Lee et al. 2007; Reeder et al. 2011). In addition, decrease of GGT levels in NASH patients have been shown to predict

160 Chapter 6-Results improvements in inflammation and fibrosis (Dixon et al. 2006). Clearly, previous studies have shown that the multiplicity of metabolic variations associated with NAFLD and NASH are related to metabolic syndrome factors (Ba- thena et al. 2013; Marchesini et al. 2001). There is a current need for methods of metabolites detection and quantification, as they are keys to the biomarker discovery field. In this thesis, previous work on sample preparation of lipid profiling (Chapter 3) and BA targeted assay development (Chapter 4) by UPLC-MS aimed to offer new insights on lipid metabolism. Biomarker discovery with respect to NAFLD and NASH is a key challenge for metabonomic studies and lipid metabolism evaluation (by lipid profiling and BA targeted assay) is addressed in this chapter.

6.4.3 Limitations of the predictive lipidomic model for NAFLD

In this chapter, comprehension of the NAFLD and NASH onset has been explored by a UPLC-MS based metabonomic approach. This includes univariate and multivariate analysis, which aimed to reduce the dimensionality of the data and highlight the most important variations occurring between control participants, NAFLD and NASH. Although control, NAFLD and NASH were not discriminated in PCA models (Figure 6.3). Implementation of supervised OPLS and OPLS-DA led to the discrimination of lipid profiles from control participants to NASH (Figure 6.4, Figure 6.5 and Table 6.2). The UPLC-MS untargeted lipid profiling method does not distinguish NAFLD from NASH and from control participants. Therefore, in NAFLD increased TGs stocks and lack of lipid export from the liver might contribute to the difficult clustering of NAFLD in multivariate analysis of lipid profiling data. Still, mechanisms behind transition to NAFLD to NASH remain difficult to understand.

6.4.4 Fatty Liver Disease-associated with lipidomic variations

Prior to lipidomic analysis, variation of physiological variables, especially in BMI, between groups was evaluated (Figure 6.1). High BMI values were associated with

161 Chapter 6-Results

NAFLD and NASH patients (Figure 6.6) and partial correlation was applied to avoid the effect of BMI. These observations are in agreement with previous studies conducted in Western and Eastern countries which indicate that both high and low BMI are associated with NAFLD and NASH (Das et al. 2010; Margariti et al. 2012). These findings suggest that NAFLD and NASH share similar lipid patterns to controls participants (Figure 6.7). This result emphasised the link between NAFLD and NASH highlighting the importance of the lipidome in both pathogenesis (Cano et al. 2014). Differences between NASH and NAFLD lipid profiles were assessed and mostly suggested an imbalance in PCs and TGs species (Figure 6.8). PCs were observed to be significantly correlated to NASH. The major fatty acid component identified within PCs was palmitic acid (16:0) which has been shown to promote in vitro inflammatory and fibrogenic response in hepatocytes (Ricchi et al. 2009). Conversely, TGs were correlated to NAFLD. The significant decrease of TGs in NASH was shown to improve hepatic NAFLD but to intensify liver damage and fibrosis (Yamaguchi et al. 2007). These changes in lipid profiles may reflect the fact that lipid absorption is disrupted and BAs might play a fundamental role in defective lipid transport and metabolism in NASH pathogenesis.

6.4.5 Fatty Liver Disease-associated with variations in circu-

lating BA pool

Examination of the circulating BA pool in NAFLD and NASH patients compared to control participants revealed a series of disruptions. Circulating BAs were observed to be signiﬁcantly higher in NASH including drastic increase of primary BAs and sulfated BAs which is in agreement with previous studies (Aranha et al. 2008; Ferslew et al. 2015) (Figure 6.9 and Table 6.4). Considering the complexity of the regulation of BA pool circulating, hypotheses related to impaired liver functions (Russell 2003; Lake et al. 2013) and to altered gut microbiota composition (Mouzaki et al. 2013; Schnabl et al. 2014) could explain the increase in BAs (Figure 6.10).

162 Chapter 6-Results

(a) Impaired BA metabolism related to the host function

A significant increase in primary BAs in NASH reflects altered liver enzymes activities (Kong et al. 2009). The discovery of FXR as a BA receptor identified a pathway to explain the BA signature observed in NASH patients (Watanabe et al. 2006). Activation of FXR is known to inhibit BA synthesis (primary BAs) via the short heterodimer protein (SHP) and target the expression of various genes (Thomas et al. 2010). FXR-deficiency in the mouse has been shown to trigger inflammatory cell infiltration and elevated collagen (i.e. fibrosis) in hepatic cells (Kong et al. 2009). Recent studies showed that hepatic expression of the FXR gene was up-regulated in NASH (Aguilar-Olivos et al. 2014). Consistent with this, the high concentrations of primary BAs reported in this chapter could overcome the FXR effects in NASH and lead to the progression of the disease by inflammation and fibrosis. Consequently, high primary BAs concentrations would trigger the BA sulfation pathway in the liver to increase BA elimination in urine and feces (Alnouti 2009). Importantly, progression of simple NAFLD to NASH was correlated with significant increase of HCA. HCA is one of the muricholic acid (MCA) isomers and synthesised by the gut microbiota from 6α -hydroxylation of CDCA. The four MCAs are named according to the position of two hydroxyls, 6-OH and 7-OH as followed; αMCA 3α, 6β, 7α; βMCA 3α, 6β, 7β; ωMCA 3α, 6α, 7β and δMCA (HCA) 3α, 6α, 7α. MCAs are abundant in rodent and pig studies as they correspond to primary BAs in these animals. However, few studies described the process of MCAs synthesis in human by the liver enzyme cytochrome P450 3A4 (CYP3A4) (Araya et al. 1999; Bodin et al. 2005). CYP3A4 activity induces 6α hydroxylation of CDCA to form HCA and was previously detected at low concentrations (5 to 24 nM) in serum of healthy participants (Perwaiz et al. 2001; Garc´ıa-Canaveras˜ et al. 2012; Bathena et al. 2013). A high levels of HCA have been shown to result from exacerbated activity of hepatic CYP3A4 and correspond to a pathological response in cholestasis and liver disease (Wietholtz et al. 1996; Chen et al. 2014). Higher significant systemic concentrations of taurine conjugates could be asso-

163 Chapter 6-Results ciated with the enzymes that catalyse BA conjugation; BA coenzyme A:amino acid N-acyltransferase (BAAT) and synthase (BACS). Activities of these enzymes have been demonstrated to be increased in NASH liver compared to healthy liver (Lake et al. 2013).

(b) Impaired BA metabolism that are related to the gut microbiota activity

In recent studies, impaired gut microbiota functions related to NAFLD and NASH patients have been observed (Best et al. 2015). In addition, modulation of the gut microbiota with antibiotics has been shown to regulate BA levels (Toda et al. 2009). Interestingly, some bacteria belonging to the Bacteroidetes, Prevotella and Lactobacillus casei taxa are known for their deconjugation activity (Shindo et al. 1989) and have been associated with a decrease in NASH severity which could explain the high proportion of BA conjugates seen in this study (Mouzaki et al. 2013; Okubo et al. 2013). Furthermore, the elevated total sulfated BA proportion may reveal repression of BAs sulfatase activity in the gut microbiota by (Carbonero et al. 2012). Clearly, sulfur metabolism (incl. taurine) is impaired in NASH patients compared to NAFLD patients. A significant increase of HCA plasma concentration correlated with NASH can also be considered as a result from bacterial enzyme activity in the gut microbiota. Clostridium HDCA-1 has been previously isolated and observed as carrying out MCAs conversions to hyodeoxycholic acid (Eyssen et al. 1999). According to these findings, NASH could repress the Clostridium HDCA-1 growth in the gut microbiota by increasing concentration of specific BAs and reducing MCAs and especially HCA degradation. Exploration of gut microbiota by 16S rDNA sequencing combined to metagenomic and metabonomics analysis could be used to help validate this hypothesis.

164 Chapter 6-Results

Figure 6.11: Synopsis of results and hypotheses and the possible role of liver and gut microbiota on lipid and BA metabolism in the transition to NAFLD to NASH.

165 Chapter 6-Results

6.5 Conclusion

This chapter aimed to assess the lipid variability occurring in the transition to NAFLD to NASH. The objectives were to evaluate the impact of NAFLD/NASH on biochemical processes and lipid metabolism. Lipid profiling by UPLC-MS and BA targeted profiling via UPLC-MS/MS assay aimed to characterise lipids and BA signature associated with the NAFLD and NASH compared to control participants. NAFLD to NASH were observed to affect significantly insulin, glucose and HDL levels. These findings demonstrate the important impact of the disease progression to risk factors associated with metabolic syndrome. Lipid profiles demonstrated significant PC and TG variability between NAFLD and NASH patients. In addition, NAFLD and NASH were characterised by increased of unconjugated, tauro-conjugated and sulfated BAs. HCA appeared to be a key biomarker in the transition to NAFLD to NASH. In conclusion these findings highlight profound modifications that can be related to i) exacerbated liver function (i.e. BA synthesis pathway, enzymes) and ii) down- regulated gut microbiota enzymes (i.e. dehydroxylase, sulfatase). Further studies are needed to assess gut microbiota and liver function resulting to defective lipid and BA metabolism in the transition to NAFLD to NASH.

166 Chapter 7

General discussion

The complexity of host-gut microbiota interactions is widely accepted and it is acknowledged that dysregulation of this relationship can lead to disease development such as obesity and fatty liver disease. In this thesis, mass spectrometric methods were developed, optimised, and successfully applied to capture outputs of host-microbial co-metabolism and to further the understanding of molecular mechanisms underpinning obesity and NAFLD diseases, specifically in the areas of lipid and BA metabolism. In this thesis, four main objectives were pursued in order to address important challenges in these areas: 1. Improvement of sample preparation and analysis pipeline for efficient, minimally selective, and high throughput profiling of lipid species including FAs, phospholipids (PC, PE, PS and PG), mono/di/triacylglycerides, sphingomyelins, ceramides and cholesterol esters from human plasma (Chapter 3). 2. Expanding the molecular coverage of host-gut BAs, including glycine, taurine and sulfated conjugates, in targeted quantitative analysis by UPLC-MS/MS (Chapter 4). 3. Application of the developed UPLC-MS methods to elucidate the influence of visceral obesity on postprandial lipemic response (PPLR) in humans (Chapter 5). 4. Application of the developed methods to study the involvement of lipid metabolism in the transition from NAFLD to NASH in humans (Chapter 6). The results presented in this thesis demonstrate that the developed methods contribute to improving our understanding of the host-gut microbiota interaction by

167 Chapter 7-General discussion increasing the molecular coverage of metabolites involved in the disease progression achievable by common lipid proﬁling and BA targeted analytical techniques.

7.1 Key results criteria established for the analytical

pipelines for lipid and BA measurements

7.1.1 Isopropanol precipitation and its suitability for high -

throughput UPLC/MS lipid proﬁling analysis

Chapter 3 aimed to find the optimum method for rapid and minimally selective preparation of plasma for lipid profiling using UPLC-MS. Removal of protein from biofluids such as human blood products is a prerequisite for UPLC-MS methods which use organic solvents in the mobile phase, as failure to do so would result in the denaturation of protein on the column. Beyond this basic need, some researchers perform liquid phase partitioning in order to enrich and/or separate lipids from other sample components (e.g. small molecules and proteins). However, with the basic criteria of protein removal accomplished by either method, the potential added benefit of liquid-liquid extraction must be weighed against the potential introduction of lipid partitioning and therefore the introduction of unintended selectivity. Four methods of liquid-liquid extraction and four methods of protein precipitation were evaluated according to the following set of criteria: protein removal efficiency; selectivity; repeatability; and recovery. Sample preparation by liquid-liquid extraction of plasma appeared to be less straightforward than by chemical protein precipitation methods as it significantly impacted on the repeatability and recovery of lipid species as well as the efficiency of protein removal. Precipitation of proteins with isopropanol (IPA) was found to be 99% efficient, yielding an adequately protein-free sample suitable for LC-MS analysis. This method also produced optimal conditions for lipid solvation yielding a product that was readily amenable to injection for profiling analysis by UPLC-MS. The results

168 Chapter 7-General discussion demonstrate a wide lipid coverage, good repeatability (61.1% of features detected after LC-MS analysis within a CV<20%) and high recoveries (≈83.4%). Simple protein precipitation avoids errors compared to conventional extraction sample preparation methods such as Folch and Bligh-Dyer (Bligh et al. 1959; Folch et al. 1957). When using these extraction methods, it is challenging to cleanly withdraw the lipid- containing organic phase as it requires penetration of the protein layer and protein layer, both of which sit on top of the organic layer. Conversely, biphasic extraction by

MeOH/MTBE-H2O circumvents this problem as the organic phase is less dense and therefore more buoyant, sitting at the top of the centrifuged mixture. However, total metabolite yield may be negatively impacted when using liquid-liquid extractions due to incomplete lipid partitioning between phases (e.g. loss of some more hydrophilic lipids to the aqueous layer), instability of species in the drying step required to exchange sample diluent solvents, and solubilisation in the final LC-MS sample diluent. Generally, the complexity and time required for the multiple steps in liquid-liquid extraction process can induce low repeatability (35.5% of features detected after LC- MS analysis within a CV<20%) and the recovery (≈63.1%). Chemical precipitation of proteins (and complete avoidance of the liquid-liquid extraction) has the distinct advantage of being simple and amenable to high throughput environments. While precipitations with more polar solvents such as methanol are adequate for protein removal, they do not create an environment well suited to the solubilisation of highly hydrophobic lipid species (e.g. TGs). Isopropanol was therefore identified as an ideal solvent, conferring the benefits of simple sample preparation, providing the necessary protein removal, and adequately retaining in solution the lipid content of the original sample.

7.1.2 Targeted UPLC-MS/MS methodology for the detection

and quantiﬁcation of 145 BAs

In Chapter 4, a robust UPLC-MS/MS method was developed and validated to monitor 145 BAs in plasma/serum and urine. The analytical challenge was to op-

169 Chapter 7-General discussion timise the targeted assay taking into account the complexity of BAs structures and many isobaric species found in human biofluids. In order to increase the coverage of important BA conjugate species, a sulfation reaction and HPLC-scale purification were implemented and optimised, generating 88 additional BA species as sulfate conjugates of existing BA standards (n=57). Sulfation is an important process in enterohepatic circulation that regulates the elimination of toxic BAs in the system (Marschall et al. 1992). Moreover, the extent of chromatographic conditions offer a simple ”dilute and shoot” sample preparation process by protein removal with the rapid chromatographic separation (<15 min) detection of a wide range of BA species compared to methods available in the literature (Duboc et al. 2013; Xie et al. 2015; Scherer et al. 2009; Want et al. 2010; Humbert et al. 2012; Steiner et al. 2010; Per- waiz et al. 2001; Sayin et al. 2013; Alnouti et al. 2008; Bathena et al. 2013; Huang et al. 2011; Garc´ıa-Canaveras˜ et al. 2012; Humbert et al. 2012; Sergi et al. 2012; Qiao et al. 2012; Ye et al. 2007; Bobeldijk et al. 2008; Chen et al. 2011; John et al. 2014). Method coverage includes (in order of most hydrophilic to most hydrophobic) sulfate conjugated BAs, tauro-conjugated BAs, glyco-conjugated BAs and unconjugated BAs. In addition, the chromatography was optimised to elute lipids during column washing step (e.g. phospholipids and glycerides) and prevent carry over or matrix effect during the BAs separation and detection. In addition to outstanding coverage, the method was validated according to the FDA recommendations. This analytical method is characterised by a high degree of linearity, with LLOQ 0.25- 10nM and ULOQ 2.5-5nM, precision <10% and accuracy between 81.2-118.9% on intra/inter day and high recovery in plasma/serum >88% and urine >93%. These achievements represent a substantial contribution to BA structural elucidation and to the field for routine measurement of diverse BA species in human biofluids. The improvement of BA coverage for this targeted assay is dependent on commercially available BA standards to offer quantification of other BA conjugates such as glucuronides, glucose and N-acetyl-glucosamine that are found in humans. Moreover, profiling approach is still the best option to obtain a global screening of BA species

170 Chapter 7-General discussion present in a bioﬂuids and can applied using the chromatographic conditions. This targeted assay provides high selectivity and sensitivity for improved high throughput comprehensive analysis of BAs and interpretation of lipid host-gut metabolism (i.e chapter 5 and 6).

7.2 Lipid and BA application of proﬁles to obesity

and liver disease in human population

7.2.1 Key results

UPLC-MS and UPLC-MS/MS based molecular proﬁling approaches were successfully applied within two independent areas of study, detailed in Chapters 5 and 6. They are brieﬂy discussed here.

(a) Contribution of lipid and BA metabolism dysfunction to ectopic fat deposition in obesity

In Chapter 5, fat distribution of obese individuals was found to be associated with a differential lipids and BAs signature to lean individual. Visceral obesity displayed significant increase of LPC (16:0), secondary BAs; UDCA, DCA, 7-oxoLCA, 5β cholanic acid 7α-ol-3-one and 5β cholanic acid 3β, 12α-diol and sulfate conjugate; 4-cholenic acid 3, 6-diol sulfate. Conversely, significant decrease of PC (16:/20:3), taurine BA conjugates; TCA, TCDCA, THCA, TβMCA and TαMCA were observed in visceral obesity compared to subcutaneous obesity. These findings highlight the importance of lipids and BA pool circulating in localisation of fat deposition. Increased of LPC (16:0) might be provided by upregulated phospholipase that catalyse the hydrolysis of the PC, especially PC (16:0/20:3). Many factors might be involved in the observed taurine deficiency in visceral obesity. For example, impaired dietary enzymes (e.g. cysteine dioxygenase) which might promotes increased toxicity of secondary (e.g. DCA) and tertiary BAs circulating and might impact on the BA signalling such as

171 Chapter 7-General discussion deﬁcient activation of TGR5 receptors involved in maintenance of energy expenditure, glycaemia and inﬂammation (Tsuboyama-Kasaoka et al. 2006).

7.2.2 Contribution of lipid and BA metabolism dysfunction to

NAFLD and NASH

In Chapter 6, NASH was associated with altered lipid and BA metabolism compared to NAFLD and control subjects. NAFLD to NASH transition involved a significant increase of TG (16:1/18:2/18:2), total unconjugated BAs, tauro-conjugated BAs and sulfo-conjugated. Targeted BA analysis aimed to identify specific BA species significantly increased in NAFLD to NASH transition such as, unconjugated BAs; DCA, HDCA and αMCA, glyco-conjugated; GHCA and GHDCA, tauro-conjugated BAs; TβMCA and TUDCA. Finally, HCA was found as key BA signature in NASH compared to NAFLD. Significant increase of TG (16:1/18:2/18:2) in NASH compared to NAFLD is characteristic to dyslipidaemia and as observed is accompanied by increased level of HDL. BA signalling through FXR might be exacerbated in NASH compared to NAFLD due to high concentrations of CDCA the most effective activator and impact on energy expenditure, glycaemia and inflammation pathways (Bjursell et al. 2013).

7.2.3 Overall coherence of the markers between application

studies

Application studies were implemented in this thesis to evaluate the impact of new developed analytical techniques on comprehension of lipid metabolism.Results in subcutaneous vs. visceral fat deposition in obesity and NAFLD vs. NASH suggested that liver (host) and gut microbiota play a critical role in BA regulation. Interestingly, a trend was observed in the unconjugated and sulfated BAs, positively correlating with visceral obesity and NASH. Notably, DCA was found to be signiﬁcantly increased in visceral obesity and NASH. This is in agreement with previous results that identiﬁed DCA as cause of obesity and hepatocellular carcinoma (Bernstein et al. 2005; Yoshi-

172 Chapter 7-General discussion moto et al. 2013). DCA is product from CA 7α-dehydroxylation by gut microbiota enzymes. As expected, increased DCA was accompanied by an increase in plasma CA. These results may be interpreted as an up-regulation of primary BA (i.e. CA) synthesis in the liver with consequently increased secondary BA (i.e. DCA) synthesis in the intestine by the gut microbiota. Altogether, unconjugated BAs are increased and metabolic response shift toward activation of BA elimination pathways (i.e. sulfation) by up-regulation of host enzymes or down-regulation of gut enzymes to protect against unconjugated BAs toxicity. Compromised health by obesity or NAFLD development is often evolving with insulin resistance. Consistent with this hypothesis, DCA preponderance was observed in patients with type 2 diabetes (Kuipers et al. 2014). The metabolic information provided by improved analytical methods were critical in reaching these conclusions, and are useful benchmarks for advances in understanding of metabolic pathways associated to metabolic syndrome and related diseases.

7.3 New ﬁndings in obesity and fatty liver disease

The continued development of advanced technologies and methodologies drives the growth of analytical capability, offering comprehensive analysis of numerous analytes to improve prediction and diagnosis of diseases (Nicholson et al. 2012). In this thesis, the application of lipid profiling and BA targeted assays reveals new candidate markers of visceral obesity and transition NAFLD to NASH within plasma and illustrates the relative importance of unconjugated, tauro-conjugated, glyco-conjugated and sulfo-conjugated BAs. In agreement with previous findings, the results obtained in the discussed areas of application highlight strong perturbations of the lipid and BA metabolism in obese and NAFLD patients (Wymann et al. 2008; Dumas et al. 2014). Identification of similar BA variations in obesity and NAFLD can help to understand the possible transition between disease states, including those linked to insulin resistance (Petersen et al. 2006) and global mechanisms behind metabolic syndrome (Alberti et al. 2005). Focus on the BA elimination pathway was characterised by

173 Chapter 7-General discussion quantification of sulfated BAs. This pathway is exacerbated in obesity and NAFLD and has beneficial effect as it prevents cytotoxicity of increased unconjugated BAs to be reabsorbed (Mackie et al. 1997). It is noteworthy that most of the sulfated BAs that were observed to be significantly increased in visceral obesity and in NASH were tertiary BAs. This particularly illustrates the importance of the gut microbiota in regulation of BA metabolism (Ridlon et al. 2006). Several studies, have illustrated the symbiotic interaction existing between the host and gut microbiota through various metabolic messengers (Neves et al. 2015; Nicholson et al. 2005). Results obtained in this thesis suggested that most of the discriminant BAs observed in obesity and NAFLD are known to be intermediates between the gut microbiota and host (Sayin et al. 2013). Moreover, physiological functions of the host and influences of the gut microbiota on these functions can be detrimental for health (Nicholson et al. 2005).

7.4 Strength and weakness of the experiments

Each analytical technique, lipid profiling by UPLC-MS and BA targeted assay by UPLC-MS/MS, provided a distinct set of measurements critical to the understanding of human metabotypes. The analytical methods developed in chapters 3 and 4 aim to depict the impact of obesity and liver diseases on respectively lipid and BA metabolism. Plasma samples were successfully prepared following the IPA precipitation method. This robust and straightforward method offered the best conditions to extract and analyse a wide range of lipid classes. Perhaps as a consequence, the need for identification of previously undescribed lipid species was encountered in both studies (i.e. Chapter5 and Chapter 6). Structural identification was laborious especially for features with low intensity and for those not present in available databases. Feature assignment would have benefited from targeted MS/MS analysis on representative feature-containing samples, and identification determined by comparison of the accurate mass, MS/MS fragmentation pattern, isotope pattern, and LC retention time of

174 Chapter 7-General discussion corresponding lipid standards. In contrast, use of the targeted BA method circumvents the need for metabolite assignment, as the molecular targets were predetermined. The lipid signatures associated with obesity and NAFLD were obtained using the optimised sample preparation and molecular profiling method. However, both studies highlighted few significant lipid classes. The lack of lipid response could be related to several limitations. For example, 1) the limited number of patients may have failed to produce sufficient statistical power and any differences between groups may not have achieved significance, 2) the lipid content of the milkshake was not sufficient (i.e. Chapter 5) 3) more accurate classification of fatty liver disease severity than NAFLD activity score (NAS) (i.e. Chapter 6) or, 4) the high intra-individual variability could induce poor repeatability inside each group of patients. Secondly, the BA signature associated with obesity and NAFLD was obtained by implementation of the BA targeted assay and successful analysis of 145 BAs was achieved. This work provided important advances in terms of BA identification and coverage. However, some BA species are still not characterised, and therefore limit a comprehensive overview of BA metabolism. For instance, glucuronides, N-acetyl- glucosamine, conjugated forms of derivates that are gut products and maybe more were not assessed in the experiments. As these BAs are not available commercially, the synthesis and purification of these BAs needed to be performed in house. Individual variation in gut microbiota composition could impact on consistent BA analysis in a same group of patients. Thirdly, in both studies circulating BAs were quantified from plasma samples. It was unclear if some BAs identified were products of either liver or gut enzymes, and it would be useful to support these findings. For instance, genome-wide (microarrays or RNAseq) and targeted (qPCR) transcriptomic analyses of peripheral organs (liver, adipose tissue, muscle...) would be useful to identify variations in expression of genes encoding for proteins involved in BA metabolism. Therefore, RNA gene sequencing or full metagenomic analysis could provide informations about the bacterial taxa and function of the gut microbiota. Altogether, these data could be relevant for the global

175 Chapter 7-General discussion understanding of host-gut microbiota axis and the BA metabolism impact on obesity and NAFLD. Fourthly, primary and secondary BAs are known to be signalling molecules and pharmacological investigation of tertiary BAs on their ability to activate pathways (e.g. FXR, TGR5) could be key component of the host-gut microbiota axis.

7.5 Preliminary assessment of FXR activation by

the 57 BAs quantiﬁed by the targeted assay

To gain insights on the biological mechanisms impacted by BAs observed in chapters 5 and 6, a preliminary assessment of the pharmacological impact of BAs on FXR signalling was performed, using optimised cell-based assays (FXR overexpressed in CHO cells). Control were integrated to this study such as synthetic positive control GW4064, positive control CDCA which is the most potent FXR ligands (Wang et al. 2006; Makishima et al. 1999) and negative control UDCA. Ten out of the 57 BAs tested were observed to be able to activate FXR (Figure 7.1 and Figure 7.2). Unfor- tunately, half maximal eﬀective BA concentration (EC50) couldn’t be determined as activation did not reach a plateau. However, these results provide interesting informations on BA species able to activate FXR. It appears that some BAs associated to obesity and NAFLD signature are able to activate FXR. For instance, DCA common BA associated to both studies, 5β cholenic acid 7α ol 3 one associated to visceral fat deposition and αMCA associated to NAFLD. Although these results are preliminary and would require careful assay optimisation, combining multiplexed BA quantiﬁcation with comprehensive pharmacological assessment of the FXR agonists a promising research avenue and may be particularly useful for treatment of obesity or NAFLD associated disorders.

176 Chapter 7-General discussion Ten BAs out of the 57 activated FXR overexpressed in CHO cells. (A) Controls; synthetic positive control GW4064, positive control CDCA and negative control UDCA, (B) BAs able to activate FXR. Figure 7.1:

177 Chapter 7-General discussion

Figure 7.2: Top ten BAs able to activate FXR are ranked according to CDCA (100%) and the last concentration tested (10-4M).

7.6 Future works and perspectives

In this thesis, a UPLC-MS based lipid profiling approach was specifically developed and applied to study subtle disease progression phenotypes in metabolic syndrome, exemplified by the identification of PPLR between subcutaneous and visceral obesity, or lipid signature of the NAFLD to NASH transition. This work illustrates that profiling and targeted method can be used to identify biomarkers predicting disease progression over time and is promising for individualised treatments. If these markers are confirmed in human studies, they could be used to anticipate altered responses related to metabolic syndrome. It also shows the critical role of host-gut interaction and impact on lipid and BA metabolism. Further in vitro and in vivo analyses are required to validate this hypothesis. However, the human gut ecosystem and its functions (i.e. enzymes activity) are still under extensive investigation in the literature. Finally, several circulating BAs were found to predict the severity of fat distribution (i.e. subcutaneous vs. visceral) and NAFLD to NASH transition. These signatures could also present a great potential to assess at early stage alteration of liver or

178 Chapter 7-General discussion gut entities. However, the significant gut microbial ecosystem and genetic variability between humans have to be taken into account to validate these BAs as effective predictors. To conclude, these results shown the importance of newly developed and optimised analytical methods. In this thesis, the new concept of BAs metabolism as predictors of disease related to obesity and fatty liver disease and as key intermediates of host-gut microbiota interactions offer novel perspectives in marker discovery.

179 References

Adams, L. A. et al. (2005). “The natural history of nonalcoholic fatty liver disease: a population-based cohort study”. Gastroenterology 129.1, pp. 113–121. Aguilar-Olivos, N. E. et al. (2014). “The nuclear receptor FXR, but not LXR, up- regulates bile acid transporter expression in non-alcoholic fatty liver disease.” An- nals of hepatology 14.4, pp. 487–493. Alberti, K. G. M. et al. (2005). “The metabolic syndrome a new worldwide definition”. The Lancet 366.9491, pp. 1059–1062. Alnouti, Y. (2009). “Bile acid sulfation: a pathway of bile acid elimination and detox- ification”. Toxicological sciences 108.2, pp. 225–246. Alnouti, Y. et al. (2008). “Quantitative-profiling of bile acids and their conjugates in mouse liver, bile, plasma, and urine using LC–MS/MS”. Journal of Chromatogra- phy B 873.2, pp. 209–217. Angulo, P. et al. (1999). “Independent predictors of liver fibrosis in patients with nonalcoholic steatohepatitis”. Hepatology 30.6, pp. 1356–1362. Anizan, S. et al. (2011). “Screening of 4-androstenedione misuse in cattle by LC– MS/MS profiling of glucuronide and sulfate steroids in urine”. Talanta 86, pp. 186– 194. Anstee, Q. M. et al. (2013a). “Progression of NAFLD to diabetes mellitus, cardiovascular disease or cirrhosis”. Nature Reviews Gastroenterology and Hepatology 10.6, pp. 330–344. Anstee, Q. M. et al. (2013b). “The genetics of NAFLD”. Nature Reviews Gastroen- terology and Hepatology 10.11, pp. 645–655.

180 References

Aragozzini, F. et al. (1985). “A study on the mechanism of the epimerization at C-3 of chenodeoxycholic acid by Clostridium perfringens”. Biochemical journal 230.2, pp. 451–455. Aranha, M. M. et al. (2008). “Bile acid levels are increased in the liver of patients with steatohepatitis”. European journal of gastroenterology & hepatology 20.6, pp. 519–525. Araya, Z. et al. (1999). “6α-Hydroxylation of taurochenodeoxycholic acid and lithocholic acid by CYP3A4 in human liver microsomes”. Biochimica et Biophysica Acta (BBA)-Molecular and Cell Biology of Lipids 1438.1, pp. 47–54. Aron-Wisnewsky, J. et al. (2012). “The importance of the gut microbiota after bariatric surgery”. Nature Reviews Gastroenterology and Hepatology 9.10, pp. 590– 598. Bäckhed, F. (2011). “Programming of host metabolism by the gut microbiota”. An- nals of Nutrition and Metabolism 58.Suppl. 2, pp. 44–52. Bäckhed, F. et al. (2004). “The gut microbiota as an environmental factor that regulates fat storage”. Proceedings of the National Academy of Sciences of the United States of America 101.44, pp. 15718–15723. Bäckhed, F. et al. (2005). “Host-bacterial mutualism in the human intestine”. science 307.5717, pp. 1915–1920. Barsh, G. S. et al. (2000). “Genetics of body-weight regulation”. Nature 404.6778, pp. 644–651. Bateman, K. P. et al. (2007). “MSE with mass defect filtering for in vitro and in vivo metabolite identification”. Rapid Communications in Mass Spectrometry 21.9, pp. 1485–1496. Bathena, S. P. R. et al. (2013). “The profile of bile acids and their sulfate metabolites in human urine and serum”. Journal of Chromatography B 942, pp. 53–62. Begley, M. et al. (2005). “The interaction between bacteria and bile”. FEMS microbiology reviews 29.4, pp. 625–651.

181 References

Beller, M. et al. (2010). “Lipid droplets: a dynamic organelle moves into focus”. FEBS letters 584.11, pp. 2176–2182. Bernstein, H. et al. (2005). “Bile acids as carcinogens in human gastrointestinal cancers”. Mutation Research/Reviews in Mutation Research 589.1, pp. 47–65. Best, N. van et al. (2015). “The gut microbiota of nonalcoholic fatty liver disease: current methods and their interpretation”. Hepatology international 9.3, pp. 406– 415. Beyoglu,˘ D. et al. (2013). “The metabolomic window into hepatobiliary disease”. Journal of hepatology 59.4, pp. 842–858. Bird, S. S. et al. (2011). “Serum lipidomics profiling using LC–MS and high-energy collisional dissociation fragmentation: Focus on triglyceride detection and characterization”. Analytical chemistry 83.17, pp. 6648–6657. Bjursell, M. et al. (2013). “Ageing Fxr deficient mice develop increased energy expenditure, improved glucose control and liver damage resembling NASH”. PloS one 8.5, e64721. Bligh, E. G. et al. (1959). “A rapid method of total lipid extraction and purification”. Canadian journal of biochemistry and physiology 37.8, pp. 911–917. Bobeldijk, I. et al. (2008). “Quantitative profiling of bile acids in biofluids and tissues based on accurate mass high resolution LC-FT-MS: compound class targeting in a metabolomics workflow”. Journal of Chromatography B 871.2, pp. 306–313. Bodin, K. et al. (2005). “Novel pathways of bile acid metabolism involving CYP3A4”. Biochimica et Biophysica Acta (BBA)-Molecular and Cell Biology of Lipids 1687.1, pp. 84–93. Bouchard, C. et al. (1990). “The response to long-term overfeeding in identical twins”. New England Journal of Medicine 322.21, pp. 1477–1482. Bradford, M. M. (1976). “A rapid and sensitive method for the quantitation of micro- gram quantities of protein utilizing the principle of protein-dye binding”. Analytical biochemistry 72.1, pp. 248–254.

182 References

Breier, M. et al. (2014). “Targeted metabolomics identifies reliable and stable metabolites in human serum and plasma samples”. PloS one 9.2, e89728. Bronk (1999). “Human metabolism”. Browning, J. D. et al. (2004a). “Molecular mediators of hepatic steatosis and liver injury”. Journal of Clinical Investigation 114.2, p. 147. Browning, J. D. et al. (2004b). “Prevalence of hepatic steatosis in an urban population in the United States: impact of ethnicity”. Hepatology 40.6, pp. 1387–1395. Bruce, S. J. et al. (2008). “Evaluation of a protocol for metabolic profiling studies on human blood plasma by combined ultra-performance liquid chromatography/mass spectrometry: From extraction to data analysis”. Analytical biochemistry 372.2, pp. 237–249. Bruce, S. J. et al. (2009). “Investigation of human blood plasma sample preparation for performing metabolomics using ultrahigh performance liquid chromatography/mass spectrometry”. Analytical chemistry 81.9, pp. 3285–3296. Bugianesi, E. et al. (2005). “Insulin resistance in non-diabetic patients with nonalcoholic fatty liver disease: sites and mechanisms”. Diabetologia 48.4, pp. 634– 642. Cai, X. et al. (1997). “Analysis of the solubilization of steroids by bile salt micelles”. Journal of pharmaceutical sciences 86.3, pp. 372–377. Cani, P. D. (2012). “Crosstalk between the gut microbiota and the endocannabinoid system: impact on the gut barrier function and the adipose tissue”. Clinical Microbiology and Infection 18.s4, pp. 50–53. Cani, P. D. et al. (2008). “Role of gut microflora in the development of obesity and insulin resistance following high-fat diet feeding”. Pathologie Biologie 56.5, pp. 305–309. Cano, A. et al. (2014). “Deciphering non-alcoholic fatty liver disease through metabolomics”. Biochem Soc Trans 42.5, pp. 1447–1452. Carbonero, F. et al. (2012). “Microbial pathways in colonic sulfur metabolism and links with health and disease”. Frontiers in physiology 3.

183 References

Cequier-Sánchez, E. et al. (2008). “Dichloromethane as a solvent for lipid extraction and assessment of lipid classes and fatty acids from samples of different natures”. Journal of Agricultural and Food Chemistry 56.12, pp. 4297–4303. Cerny, I. et al. (2004). “Synthesis of [19-2H3]-analogs of dehydroepiandrosterone and pregnenolone and their sulfates”. Steroids 69.3, pp. 161–171. Chen, J. et al. (2014). “The role of CYP3A4 in the biotransformation of bile acids and therapeutic implication for cholestasis”. Annals of translational medicine 2.1. Chen, T. et al. (2011). “Serum and urine metabolite profiling reveals potential biomarkers of human hepatocellular carcinoma”. Molecular & Cellular Proteomics 10.7, pp. M110–004945. Chiang, J. Y. (2003). “III. Bile acids and nuclear receptors”. American Journal of Physiology-Gastrointestinal and Liver Physiology 284.3, G349–G356. Cinti, S. et al. (2005). “Adipocyte death defines macrophage localization and function in adipose tissue of obese mice and humans”. Journal of lipid research 46.11, pp. 2347–2355. Claus, S. P. et al. (2008). “Systemic multicompartmental effects of the gut microbiome on mouse metabolic phenotypes”. Molecular systems biology 4.1, p. 219. Coleman, J. P. et al. (1987). “Biosynthesis of a novel bile acid nucleotide and mechanism of 7 alpha-dehydroxylation by an intestinal Eubacterium species.” Journal of Biological Chemistry 262.10, pp. 4701–4707. Cortes, J. et al. (1990). “An unusually large multifunctional polypeptide in the erythromycin-producing polyketide synthase of Saccharopolyspora erythraea.” Nature 348.6297, pp. 176–178. Corvera, S. et al. (2014). “Adipose tissue angiogenesis: impact on obesity and type- 2 diabetes”. Biochimica et Biophysica Acta (BBA)-Molecular Basis of Disease 1842.3, pp. 463–472. Couillard, C. et al. (1998). “Postprandial triglyceride response in visceral obesity in men.” Diabetes 47.6, pp. 953–960.

184 References

Cummings, J. et al. (2008). “Biomarker method validation in anticancer drug development”. British journal of pharmacology 153.4, pp. 646–656. Cummings, J. et al. (2010). “Fit-for-purpose biomarker method validation for application in clinical trials of anticancer drugs”. British journal of cancer 103.9, pp. 1313–1317. Das, K. et al. (2010). “Nonobese population in a developing country has a high prevalence of nonalcoholic fatty liver and significant liver disease”. Hepatology 51.5, pp. 1593–1602. Day, C. P. et al. (1998). “Steatohepatitis: a tale of two ”hits”?” Gastroenterology 114.4, pp. 842–845. De Boever, P. et al. (1999). “Bile salt deconjugation by Lactobacillus plantarum 80 and its implication for bacterial toxicity”. Journal of applied microbiology 87.3, pp. 345–352. Delzenne, N. M. et al. (1992). “Comparative hepatotoxicity of cholic acid, deoxycholic acid and lithocholic acid in the rat: in vivo and in vitro studies”. Toxicology letters 61.2, pp. 291–304. Delzenne, N. M. et al. (2011). “Targeting gut microbiota in obesity: effects of prebiotics and probiotics”. Nature Reviews Endocrinology 7.11, pp. 639–646. Després, J.-P. et al. (2006). “Abdominal obesity and metabolic syndrome”. Nature 444.7121, pp. 881–887. Dieterle, F. et al. (2006). “Probabilistic quotient normalization as robust method to account for dilution of complex biological mixtures. Application in 1H NMR metabonomics”. Analytical chemistry 78.13, pp. 4281–4290. Dietrich, P. et al. (2014). “Non-alcoholic fatty liver disease, obesity and the metabolic syndrome”. Best Practice & Research Clinical Gastroenterology 28.4, pp. 637– 653. Dimitriadis, K. et al. (2016). “Waist circumference compared with other obesity parameters as determinants of coronary artery disease in essential hypertension: a 6-year follow-up study”. Hypertension Research.

185 References

Dixon, J. B. et al. (2006). “Weight loss and non-alcoholic fatty liver disease: falls in gamma-glutamyl transferase concentrations are associated with histologic improvement”. Obesity surgery 16.10, pp. 1278–1286. Doerner, K. C. et al. (1997). “Assessment of fecal bacteria with bile acid 7 alpha- dehydroxylating activity for the presence of bai-like genes.” Applied and environmental microbiology 63.3, pp. 1185–1188. Dowman, J. K. et al. (2010). “Pathogenesis of non-alcoholic fatty liver disease”. Qjm 103.2, pp. 71–83. Duboc, H. et al. (2013). “Connecting dysbiosis, bile-acid dysmetabolism and gut inflammation in inflammatory bowel diseases”. Gut 62.4, pp. 531–539. Dulloo, A. G. et al. (2010). “Body composition phenotypes in pathways to obesity and the metabolic syndrome”. International journal of obesity 34, S4–S17. Dumas, M.-.-.-E. et al. (2014). “Metabolic phenotyping and systems biology approaches to understanding metabolic syndrome and fatty liver disease”. Gastroen- terology 146.1, pp. 46–62. Duportet, X. et al. (2012). “The biological interpretation of metabolomic data can be misled by the extraction method used”. Metabolomics 8.3, pp. 410–421. Dyson, J. K. et al. (2013). “Non-alcoholic fatty liver disease: a practical approach to diagnosis and staging”. Frontline gastroenterology, flgastro–2013. Eckel, R. H. et al. (2005). “The metabolic syndrome”. The Lancet 365.9468, pp. 1415– 1428. Edenharder, R. et al. (1985). “12 beta-dehydrogenation of bile acids by Clostrid- ium paraputrificum, C. tertium, and C. difficile and epimerization at carbon-12 of deoxycholic acid by cocultivation with 12 alpha-dehydrogenating Eubacterium lentum.” Applied and environmental microbiology 49.4, pp. 964–968. Edenharder, R. et al. (1984). “Epimerization, oxidation and reduction of bile acids by Eubacterium lentum”. Systematic and applied microbiology 5.3, pp. 287–298.

186 References

Edenharder, R. et al. (1989). “Partial purification and characterization of an NAD- dependent 3 beta-hydroxysteroid dehydrogenase from Clostridium innocuum.” Ap- plied and environmental microbiology 55.6, pp. 1656–1659. Eliasson, M. et al. (2012). “Strategy for optimizing LC-MS data processing in metabolomics: a design of experiments approach”. Analytical chemistry 84.15, pp. 6869– 6876. Elkins, C. A. et al. (2001). “Genes encoding bile salt hydrolases and conjugated bile salt transporters in Lactobacillus johnsonii 100-100 and other Lactobacillus species”. Microbiology 147.12, pp. 3403–3412. Eriksson, L. et al. (2004). “Using chemometrics for navigating in the large data sets of genomics, proteomics, and metabonomics (gpm)”. Analytical and bioanalytical chemistry 380.3, pp. 419–429. Eyssen, H. et al. (1999). “Formation of hyodeoxycholic acid from muricholic acid and hyocholic acid by an unidentified gram-positive rod termed HDCA-1 isolated from rat intestinal microflora”. Applied and environmental microbiology 65.7, pp. 3158– 3163. Fahy, E. et al. (2005). “A comprehensive classification system for lipids”. Journal of lipid research 46.5, pp. 839–862. Fahy, E. et al. (2009). “Update of the LIPID MAPS comprehensive classification system for lipids”. Journal of lipid research 50.Supplement, S9–S14. Fan, J.-G. et al. (2007). “Metabolic syndrome and non-alcoholic fatty liver disease: Asian definitions and Asian studies”. Hepatobiliary Pancreat Dis Int 6.6, pp. 572– 578. FDA (2013). “Bioanalytical method”. Health and Human Services, Center for Drug Evaluation and Research 695 and Center for Veterinary Medicine: Rockville, MD. Fedorowski, T. et al. (1979). “Transformation of chenodeoxycholic acid and ursodeoxycholic acid by human intestinal bacteria”. Gastroenterology 77.5, pp. 1068– 1073.

187 References

Fenn, J. B. et al. (1989). “Electrospray ionization for mass spectrometry of large biomolecules”. Science 246.4926, pp. 64–71. Ferrari, A. et al. (1977). “Activity on bile acids of a Clostridium bifermentans cell-free extract”. FEBS letters 75.1, pp. 163–165. Ferslew, B. C. et al. (2015). “Altered bile acid metabolome in patients with nonalcoholic steatohepatitis”. Digestive diseases and sciences 60.11, pp. 3318–3328. Fletcher, A. (2013). “The cell membrane and receptors”. Anaesthesia & Intensive Care Medicine 14.9, pp. 417–421. Floch, M. H. et al. (1972). “The effect of bile acids on intestinal microflora”. The American journal of clinical nutrition 25.12, pp. 1418–1426. Folch, J. et al. (1957). “A simple method for the isolation and purification of total lipids from animal tissues”. J biol chem 226.1, pp. 497–509. Fonville, J. M. et al. (2010). “The evolution of partial least squares models and related chemometric approaches in metabonomics and metabolic phenotyping”. Journal of Chemometrics 24.11-12, pp. 636–649. Fracanzani, A. L. et al. (2008). “Risk of severe liver disease in nonalcoholic fatty liver disease with normal aminotransferase levels: a role for insulin resistance and diabetes”. Hepatology 48.3, pp. 792–798. Frenia, M. L. et al. (1993). “Methanol inhalation toxicity”. Annals of emergency medicine 22.12, pp. 1919–1923. Fukiya, S. et al. (2009). “Conversion of cholic acid and chenodeoxycholic acid into their 7-oxo derivatives by Bacteroides intestinalis AM-1 isolated from human feces”. FEMS microbiology letters 293.2, pp. 263–270. Gaggini, M. et al. (2013). “Non-alcoholic fatty liver disease (NAFLD) and its con- nection with insulin resistance, dyslipidemia, atherosclerosis and coronary heart disease”. Nutrients 5.5, pp. 1544–1560. Garc´ıa-Canaveras,˜ J. C. et al. (2012). “Targeted profiling of circulating and hepatic bile acids in human, mouse, and rat using a UPLC-MRM-MS-validated method”. Journal of lipid research 53.10, pp. 2231–2241.

188 References

Gault, C. R. et al. (2010). “An overview of sphingolipid metabolism: from synthesis to breakdown”. Sphingolipids as Signaling and Regulatory Molecules. Springer, pp. 1–23. Gavaghan, C. et al. (2002). “Physiological variation in metabolic phenotyping and functional genomic studies: use of orthogonal signal correction and PLS-DA”. FEBS letters 530.1, pp. 191–196. Gavrilova, O. et al. (2000). “Surgical implantation of adipose tissue reverses diabetes in lipoatrophic mice”. Journal of Clinical Investigation 105.3, p. 271. Gika, H. et al. (2011). “Sample preparation prior to the LC-MS-based metabolomic- s/metabonomics of blood-derived samples”. Bioanalysis 3.14, pp. 1647–1661. Gilliland, S. et al. (1977). “Deconjugation of bile acids by intestinal lactobacilli.” Applied and Environmental Microbiology 33.1, pp. 15–18. Glicksman, C. et al. (2010). “Postprandial plasma bile acid responses in normal weight and obese subjects”. Annals of clinical biochemistry 47.5, pp. 482–484. Gorden, D. L. et al. (2015). “Biomarkers of NAFLD progression: a lipidomics approach to an epidemic”. Journal of lipid research 56.3, pp. 722–736. Griﬃths, W. J. et al. (2010). “Bile acids: analysis in biological ﬂuids and tissues”. Journal of lipid research 51.1, pp. 23–41. Grundy, S. M. (2008). “Metabolic syndrome pandemic”. Arteriosclerosis, thrombosis, and vascular biology 28.4, pp. 629–636. Hagey, L. R. et al. (2013). “Microbial biotransformations of bile acids as detected by electrospray mass spectrometry”. Advances in Nutrition: An International Review Journal 4.1, pp. 29–35. Hammad, S. M. et al. (2010). “Blood sphingolipidomics in healthy humans: impact of sample collection methodology”. Journal of lipid research 51.10, pp. 3074–3087. Hara, A. et al. (1978). “Lipid extraction of tissues with a low-toxicity solvent”. Ana- lytical biochemistry 90.1, pp. 420–426.

189 References

Hayakawa, S. et al. (1970). “7α-dehydroxylation of cholic acid by Clostridium bifermentans strain ATCC 9714 and Clostridium sordellii strain NCIB 6929”. FEBS letters 6.2, pp. 131–133. Hertle, E. et al. (2014). “The complement system in human cardiometabolic disease”. Molecular immunology 61.2, pp. 135–148. Hirano, S. et al. (1982a). “Characterization of NADP-dependent 7 beta-hydroxysteroid dehydrogenases from Peptostreptococcus productus and Eubacterium aerofaciens.” Applied and environmental microbiology 43.5, pp. 1057–1063. Hirano, S. et al. (1982b). “Enhancement of the 7 alpha-dehydroxylase activity of a gram-positive intestinal anaerobe by Bacteroides and its significance in the 7- dehydroxylation of ursodeoxycholic acid.” Journal of lipid research 23.8, pp. 1152– 1158. Hofmann, A. F. et al. (2006). “How bile acids confer gut mucosal protection against bacteria”. Proceedings of the National Academy of Sciences of the United States of America 103.12, pp. 4333–4334. Hotamisligil, G. S. (2006). “Inflammation and metabolic disorders”. Nature 444.7121, pp. 860–867. Hoverstad, T. et al. (1986). “Short-chain fatty acids in germfree mice and rats.” The Journal of nutrition 116.9, pp. 1772–1776. Hoyles, L. et al. (2010). “Gastrointestinal Tract: Fat Metabolism in the Colon”. Hand- book of Hydrocarbon and Lipid Microbiology. Springer, pp. 3111–3118. Huang, A. et al. (1990). “Identification of three clostridia strains capable of bio- transforming chenodeoxycholic acid to ursodeoxycholic acid”. Fudan Xuebao Ziran Kexueban 29, pp. 180–188. Huang, J. et al. (2011). “Simultaneous characterization of bile acids and their sulfate metabolites in mouse liver, plasma, bile, and urine using LC–MS/MS”. Journal of pharmaceutical and biomedical analysis 55.5, pp. 1111–1119.

190 References

Huijghebaert, S. M. et al. (1982). “Isolation of a bile salt sulfatase-producing Clostrid- ium strain from rat intestinal microflora.” Applied and environmental microbiology 43.1, pp. 185–192. Huijghebaert, S. et al. (1984). “Specificity of bile salt sulfatase activity in man, mouse and rat intestinal microflora”. Journal of steroid biochemistry 20.4, pp. 907–912. Humbert, L. et al. (2012). “Bile acid profiling in human biological samples: comparison of extraction procedures and application to normal and cholestatic patients”. Journal of Chromatography B 899, pp. 135–145. Inagaki, T. et al. (2006). “Regulation of antibacterial defense in the small intestine by the nuclear bile acid receptor”. Proceedings of the National Academy of Sciences of the United States of America 103.10, pp. 3920–3925. Isaac, G. et al. (2011). “Lipid separation using UPLC with charged surface hybrid technology”. Waters App note. Ishikawa, M. et al. (2014). “Plasma and serum lipidomics of healthy white adults shows characteristic profiles by subjectsfffdfffdfffd gender and age”. PloS one 9.3, e91806. Jiang, C. et al. (2015). “Intestinal farnesoid X receptor signaling promotes nonalcoholic fatty liver disease”. The Journal of clinical investigation 125.1, p. 386. John, C. et al. (2014). “A liquid chromatography-tandem mass spectrometry-based method for the simultaneous determination of hydroxy sterols and bile acids”. Journal of Chromatography A 1371, pp. 184–195. Jones, M. L. et al. (2014). “The human microbiome and bile acid metabolism: dysbiosis, dysmetabolism, disease and intervention”. Expert opinion on biological therapy 14.4, pp. 467–482. Jørgenrud, B. et al. (2015). “The influence of sample collection methodology and sample preprocessing on the blood metabolic profile”. Kahn, S. E. et al. (2006). “Mechanisms linking obesity to insulin resistance and type 2 diabetes”. Nature 444.7121, pp. 840–846.

191 References

Karpe, F. et al. (2011). “Fatty acids, obesity, and insulin resistance: time for a reeval- uation”. Diabetes 60.10, pp. 2441–2449. Katajamaa, M. et al. (2007). “Data processing for mass spectrometry-based metabolomics”. Journal of chromatography A 1158.1, pp. 318–328. Kawamata, Y. et al. (2003). “AG protein-coupled receptor responsive to bile acids”. Journal of Biological Chemistry 278.11, pp. 9435–9440. Kawamoto, K. et al. (1989). “Puriﬁcation and characterization of a new hydrolase for conjugated bile acids, chenodeoxycholyltaurine hydrolase, from Bacteroides vulgatus”. Journal of biochemistry 106.6, pp. 1049–1053. Khan, K. et al. (1991). “Quantifying intermediary metabolites in whole blood after a simple deproteinization step with sulfosalicylic acid.” Clinical chemistry 37.5, pp. 728–733. Khare, S. et al. (2008). “Ursodeoxycholic acid suppresses Cox-2 expression in colon cancer: roles of Ras, p38, and CCAAT/enhancer-binding protein”. Nutrition and cancer 60.3, pp. 389–400. Kimura, H. et al. (1994). “Oxidation of 3-, 7-, and 12-hydroxyl Groups of Cholic Acid by an Alkalophilic Bacillus sp.” Bioscience, biotechnology, and biochemistry 58.6, pp. 1002–1006. Klein, S. et al. (2007). “Association for Weight Management and Obesity Prevention; NAASO, The Obesity Society; American Society for Nutrition; American Diabetes Association. Waist circumference and cardiometabolic risk: a consensus statement from Shaping America\’s Health: Association for Weight Management and Obe- sity Prevention; NAASO, The Obesity Society; the American Society for Nutrition; and the American Diabetes Association”. Am J Clin Nutr 85, pp. 1197–1202. Kobashi, K. et al. (1978). “A new hydrolase speciﬁc for taurine-conjugates of bile acids”. Journal of biochemistry 84.2, pp. 495–497. Kole, M. et al. (1985). “Conversion of chenodeoxycholic acid to ursodeoxycholic acid by Clostridium absonum in culture and by immobilized cells”. FEMS microbiology letters 28.1, pp. 69–72.

192 References

Kong, B. et al. (2009). “Farnesoid X receptor deﬁciency induces nonalcoholic steatohepatitis in low-density lipoprotein receptor-knockout mice fed a high-fat diet”. Journal of Pharmacology and Experimental Therapeutics 328.1, pp. 116–122. Kopelman, P. G. (2000). “Obesity as a medical problem”. Nature 404.6778, pp. 635– 643. Kotronen, A. et al. (2008). “Fatty liver a novel component of the metabolic syndrome”. Arteriosclerosis, thrombosis, and vascular biology 28.1, pp. 27–38. Kuipers, F. et al. (2014). “Beyond intestinal soap bile acids in metabolic control”. Nature Reviews Endocrinology 10.8, pp. 488–498. Lai, L. et al. (2009). “Methodological considerations in the development of HPLC-MS methods for the analysis of rodent plasma for metabonomic studies”. Molecular Biosystems 6.1, pp. 108–120. Lake, A. D. et al. (2013). “Decreased hepatotoxic bile acid composition and altered synthesis in progressive human nonalcoholic fatty liver disease”. Toxicology and applied pharmacology 268.2, pp. 132–140. Lee, D. S. et al. (2007). “Gamma Glutamyl Transferase and Metabolic Syndrome, Cardiovascular Disease, and Mortality Risk The Framingham Heart Study”. Arte- riosclerosis, thrombosis, and vascular biology 27.1, pp. 127–133. Lee, J.-Y. et al. (2013). “Contribution of the 7β-hydroxysteroid dehydrogenase from Ruminococcus gnavus N53 to ursodeoxycholic acid formation in the human colon”. Journal of lipid research 54.11, pp. 3062–3069. Lenz, E. M. et al. (2007). “Analytical strategies in metabonomics”. Journal of proteome research 6.2, pp. 443–458. Lepercq, P. et al. (2004). “Epimerization of chenodeoxycholic acid to ursodeoxycholic acid by Clostridium baratii isolated from human feces”. FEMS microbiology letters 235.1, pp. 65–72. Lin, S. et al. (2013). “Taurine improves obesity-induced inﬂammatory responses and modulates the unbalanced phenotype of adipose tissue macrophages”. Molecular nutrition & food research 57.12, pp. 2155–2165.

193 References

Loöf,¨ L. et al. (1980). “Partial purification of a human liver sulphotransferase ac- tive towards bile salts”. Biochimica et Biophysica Acta (BBA)-Lipids and Lipid Metabolism 617.2, pp. 192–204. Lundeen, S. G. et al. (1992). “Multiple forms of bile salt hydrolase from Lactobacillus sp. strain 100-100.” Journal of bacteriology 174.22, pp. 7217–7220. Ma, K. et al. (2006). “Farnesoid X receptor is essential for normal glucose homeostasis”. Journal of Clinical Investigation 116.4, p. 1102. Macdonald, I. A. et al. (1979). “Bile salt 3α-and 12α-hydroxysteroid dehydrogenases from Eubacterium lentum and related organisms”. Applied and environmental microbiology 37.5, pp. 992–1000. Macdonald, I. et al. (1975). “Convenient non-chromatographic assays for the microbial deconjugation and 7α-OH bioconversion of taurocholate”. Applied microbiology 30.4, pp. 530–535. Macdonald, I. et al. (1982). “Formation of ursodeoxycholic acid from chenodeoxycholic acid by a 7 beta-hydroxysteroid dehydrogenase-elaborating Eubacterium aerofaciens strain cocultured with 7 alpha-hydroxysteroid dehydrogenase-elaborating organisms.” Applied and environmental microbiology 44.5, pp. 1187–1195. Mackie, R. et al. (1997). “Gastrointestinal Ecosystems and Fermentations”. Gastroin- testinal Microbiology 1, pp. 319–379. Maekawa, M. et al. (2014). “Tandem mass spectrometric characterization of bile acids and steroid conjugates based on low-energy collision-induced dissociation”. Steroids 80, pp. 80–91. Mahato, S. B. et al. (1994). “Advances in microbial biotechnology of bile acids”. Biotechnology advances 12.2, pp. 357–391. Makishima, M. et al. (1999). “Identification of a nuclear receptor for bile acids”. Science 284.5418, pp. 1362–1365. Marchesini, G. et al. (2001). “Nonalcoholic fatty liver disease a feature of the metabolic syndrome”. Diabetes 50.8, pp. 1844–1850.

194 References

Margariti, E. et al. (2012). “Non-alcoholic fatty liver disease may develop in individuals with normal body mass index”. Annals of Gastroenterology 25.1, p. 45. Marschall, H. et al. (1992). “Bile acid N-acetylglucosaminidation. In vivo and in vitro evidence for a selective conjugation reaction of 7 beta-hydroxylated bile acids in humans.” Journal of Clinical Investigation 89.6, p. 1981. Martin, A. et al. (1941). “A new form of chromatogram employing two liquid phases: A theory of chromatography. 2. Application to the micro-determination of the higher monoamino-acids in proteins”. Biochemical Journal 35.12, p. 1358. Martin, F. et al. (2013). “Topographical body fat distribution links to amino acid and lipid metabolism in healthy non-obese women”. PloS one 8.9, e73445. Masuda, N. (1981). “Deconjugation of bile salts by Bacteroides and Clostridium”. Microbiology and immunology 25.1, pp. 1–11. Matyash, V. et al. (2008). “Lipid extraction by methyl-tert-butyl ether for high- throughput lipidomics”. Journal of lipid research 49.5, pp. 1137–1146. Mazumder, I. et al. (1993). “Bile acid transformations by Alcaligenes recti”. Steroids 58.2, pp. 79–86. McAuliﬀe, O. et al. (2005). “Genetic analysis of two bile salt hydrolase activities in Lactobacillus acidophilus NCFM”. Applied and environmental microbiology 71.8, pp. 4925–4929. Medicinal Products for Human Use, C. for et al. (2011). “Guideline on bioanalytical method validation”. European Medicines Agency. Mehal, W. Z. (2013). “The Gordian Knot of dysbiosis, obesity and NAFLD”. Nature Reviews Gastroenterology and Hepatology 10.11, pp. 637–644. Mendler, M.-H. et al. (1998). “Dual-energy CT in the diagnosis and quantiﬁcation of fatty liver: limited clinical value in comparison to ultrasound scan and single- energy CT, with special reference to iron overload”. Journal of hepatology 28.5, pp. 785–794. Mofrad, P. et al. (2003). “Clinical and histologic spectrum of nonalcoholic fatty liver disease associated with normal ALT values”. Hepatology 37.6, pp. 1286–1292.

195 References

Morotomi, M. et al. (1990). “Production of diacylglycerol, an activator of protein kinase C, by human intestinal microflora”. Cancer research 50.12, pp. 3595–3599. Moschetta, A. et al. (2005). “A phylogenetic survey of biliary lipids in vertebrates”. Journal of lipid research 46.10, pp. 2221–2232. Mostarda, S. et al. (2014). “Glucuronidation of bile acids under flow conditions: design of experiments and Koenigs–Knorr reaction optimization”. Organic & biomolecular chemistry 12.47, pp. 9592–9600. Mouzaki, M. et al. (2013). “Intestinal microbiota in patients with nonalcoholic fatty liver disease”. Hepatology 58.1, pp. 120–127. Muccioli, G. G. et al. (2010). “The endocannabinoid system links gut microbiota to adipogenesis”. Molecular systems biology 6.1, p. 392. Mukherjee, E. et al. (1993). “Transformation of cholic acid by Arthrobacter simplex”. Steroids 58.10, pp. 484–490. Murakami, S. (2015). “Role of taurine in the pathogenesis of obesity”. Molecular nutrition & food research. Nair, P. P. et al. (1965). “The cleavage of bile acid conjugates by cell-free extracts from Clostridium perfringens”. Life sciences 4.19, pp. 1887–1892. Nakatsuji, H. et al. (2010). “Dysregulation of glucose, insulin, triglyceride, blood pressure, and oxidative stress after an oral glucose tolerance test in men with abdominal obesity”. Metabolism 59.4, pp. 520–526. Ndumele, C. E. et al. (2011). “Hepatic steatosis, obesity, and the metabolic syndrome are independently and additively associated with increased systemic inflammation”. Arteriosclerosis, thrombosis, and vascular biology 31.8, pp. 1927–1932. Neves, A. L. et al. (2015). “The microbiome and its pharmacological targets: therapeutic avenues in cardiometabolic diseases”. Current opinion in pharmacology 25, pp. 36–44. Ng, M. et al. (2014). “Global, regional, and national prevalence of overweight and obesity in children and adults during 1980–2013: a systematic analysis for the Global Burden of Disease Study 2013”. The Lancet 384.9945, pp. 766–781.

196 References

Nicholson, J. K. et al. (1999). “’Metabonomics’: understanding the metabolic responses of living systems to pathophysiological stimuli via multivariate statistical analysis of biological NMR spectroscopic data”. Xenobiotica 29.11, pp. 1181– 1189. Nicholson, J. K. et al. (2002). “Metabonomics: a platform for studying drug toxicity and gene function”. Nature reviews Drug discovery 1.2, pp. 153–161. Nicholson, J. K. et al. (2005). “Gut microorganisms, mammalian metabolism and personalized health care”. Nature Reviews Microbiology 3.5, pp. 431–438. Nicholson, J. K. et al. (2008). “Systems biology: metabonomics”. Nature 455.7216, pp. 1054–1056. Nicholson, J. K. et al. (2012). “Metabolic phenotyping in clinical and surgical environments”. Nature 491.7424, pp. 384–392. Niessen, W. M. (2006). Liquid chromatography-mass spectrometry. CRC Press. Norman, A. et al. (1964). “Metabolites of lithocholic acid-24-C14 in human bile and feces”. The Journal of laboratory and clinical medicine 63.6, pp. 986–1001. Nygren, H. et al. (2011). “Liquid chromatography-mass spectrometry (LC-MS)-based lipidomics for studies of body fluids and tissues”. Metabolic Profiling. Sprin- ger, pp. 247–257. Okubo, H. et al. (2013). “Lactobacillus casei strain Shirota protects against nonalcoholic steatohepatitis development in a rodent model”. American Journal of Physiology-Gastrointestinal and Liver Physiology 305.12, G911–G918. Oreˇsiˇc, M. et al. (2008). “Lipidomics: A new window to biomedical frontiers”. Trends in biotechnology 26.12, pp. 647–652. Owen, R. W. et al. (1983). “Biotransformation of chenodeoxycholic acid by Pseu- domonas species NCIB 10590 under anaerobic conditions.” Journal of lipid research 24.9, pp. 1109–1118. Palmer, R. H. et al. (1971). “Bile acid sulfates. I. Synthesis of lithocholic acid sulfates and their identification in human bile”. Journal of lipid research 12.6, pp. 671–679.

197 References

Parks, D. J. et al. (1999). “Bile acids: natural ligands for an orphan nuclear receptor”. Science 284.5418, pp. 1365–1368. Pellis, L. et al. (2012). “Plasma metabolomics and proteomics profiling after a postprandial challenge reveal subtle diet effects on human metabolic status”. Metabolomics 8.2, pp. 347–359. Penney, N. et al. (2015). “The role of bile acids in reducing the metabolic complications of obesity after bariatric surgery: a systematic review”. International Journal of Obesity. Perry, R. J. et al. (2014). “The role of hepatic lipids in hepatic insulin resistance and type 2 diabetes”. Nature 510.7503, pp. 84–91. Perwaiz, S. et al. (2001). “Determination of bile acids in biological fluids by liquid chromatography-electrospray tandem mass spectrometry”. Journal of lipid research 42.1, pp. 114–119. Petersen, K. F. et al. (2006). “Increased prevalence of insulin resistance and nonalcoholic fatty liver disease in Asian-Indian men”. Proceedings of the National Academy of Sciences 103.48, pp. 18273–18277. Phan, C. et al. (2001). “Intestinal lipid absorption and transport.” Frontiers in bioscience: a journal and virtual library 6, pp. D299–319. Phillips, L. K. et al. (2008). “The link between abdominal obesity and the metabolic syndrome”. Current hypertension reports 10.2, pp. 156–164. Pietiläinen, K. H. et al. (2007). “Acquired obesity is associated with changes in the serum lipidomic profile independent of genetic effects–a monozygotic twin study”. PloS one 2.2, e218. Pietiläinen, K. H. et al. (2011). “Association of lipidome remodeling in the adipocyte membrane with acquired obesity in humans”. PLoS biology 9.6, e1000623. Plumb, R. S. et al. (2006). “UPLC/MSE; a new approach for generating molecular fragment information for biomarker structure elucidation”. Rapid Communications in Mass Spectrometry 20.13, pp. 1989–1994.

198 References

Plumb, R. et al. (2004). “Ultra-performance liquid chromatography coupled to quadrupole-orthogonal time-of-flight mass spectrometry”. Rapid Communications in Mass Spectrometry 18.19, pp. 2331–2337. Pollex, R. L. et al. (2006). “Genetic determinants of the metabolic syndrome”. Nature clinical practice Cardiovascular medicine 3.9, pp. 482–489. Polson, C. et al. (2003). “Optimization of protein precipitation based upon effective- ness of protein removal and ionization effect in liquid chromatography–tandem mass spectrometry”. Journal of Chromatography B 785.2, pp. 263–275. Qiao, L. et al. (2001). “Deoxycholic acid (DCA) causes ligand-independent activation of epidermal growth factor receptor (EGFR) and FAS receptor in primary hepatocytes: inhibition of EGFR/mitogen-activated protein kinase-signaling module enhances DCA-induced apoptosis”. Molecular biology of the cell 12.9, pp. 2629– 2645. Qiao, X. et al. (2012). “A tandem mass spectrometric study of bile acids: Interpre- tation of fragmentation pathways and differentiation of steroid isomers”. Steroids 77.3, pp. 204–211. Quehenberger, O. et al. (2010). “Lipidomics reveals a remarkable diversity of lipids in human plasma”. Journal of lipid research 51.11, pp. 3299–3305. Raetz, C. R. et al. (2002). “Lipopolysaccharide endotoxins”. Annual review of biochemistry 71, p. 635. Rainville, P. D. et al. (2007). “Novel application of reversed-phase UPLC-oaTOF- MS for lipid analysis in complex biological mixtures: A new tool for lipidomics”. Journal of Proteome Research 6.2, pp. 552–558. Reeder, S. B. et al. (2011). “Quantitative assessment of liver fat with magnetic resonance imaging and spectroscopy”. Journal of magnetic resonance imaging 34.4, pp. 729–749. Reis, A. et al. (2013). “A comparison of five lipid extraction solvent systems for lipidomic studies of human LDL”. Journal of lipid research 54.7, pp. 1812–1824.

199 References

Riccardi, G. et al. (2000). “Dietary treatment of the metabolic syndrome the optimal diet”. British Journal of Nutrition 83.S1, S143–S148. Ricchi, M. et al. (2009). “Differential effect of oleic and palmitic acid on lipid accumulation and apoptosis in cultured hepatocytes”. Journal of gastroenterology and hepatology 24.5, pp. 830–840. Ridlon, J. M. et al. (2006). “Bile salt biotransformations by human intestinal bacteria”. Journal of lipid research 47.2, pp. 241–259. Ridlon, J. M. et al. (2014). “Bile acids and the gut microbiome”. Current opinion in gastroenterology 30.3, p. 332. Rinella, M. E. (2015). “Nonalcoholic fatty liver disease: a systematic review”. Jama 313.22, pp. 2263–2273. Ritchie, M. et al. (2011). Rapid Lipid Profiling of Serum by Reverse Phase UPLC- Tandem Quadrupole MS. Roberts, L. D. et al. (2008). “A matter of fat: an introduction to lipidomic profiling methods”. Journal of Chromatography B 871.2, pp. 174–181. Roden, M. (2006). “Mechanisms of disease: hepatic steatosis in type 2 diabetes-pathogenesis and clinical relevance”. Nature Reviews Endocrinology 2.6, pp. 335– 348. Russell, D. W. (2003). “The enzymes, regulation, and genetics of bile acid synthesis”. Annual review of biochemistry 72.1, pp. 137–174. Ryan, K. K. et al. (2014). “FXR is a molecular target for the effects of vertical sleeve gastrectomy”. Nature 509.7499, pp. 183–188. Saadeh, S. et al. (2002). “The utility of radiological imaging in nonalcoholic fatty liver disease”. Gastroenterology 123.3, pp. 745–750. Saito, I. (2012). “Epidemiological evidence of type 2 diabetes mellitus, metabolic syndrome, and cardiovascular disease in Japan”. Circulation Journal 76.5, pp. 1066– 1073. Saltiel, A. R. et al. (2001). “Insulin signalling and the regulation of glucose and lipid metabolism”. Nature 414.6865, pp. 799–806.

200 References

Sandhoff, R. et al. (1999). “Determination of cholesterol at the low picomole level by nano-electrospray ionization tandem mass spectrometry”. Journal of lipid research 40.1, pp. 126–132. Sarafian, M. H. et al. (2014). “Objective Set of Criteria for Optimization of Sam- ple Preparation Procedures for Ultra-High Throughput Untargeted Blood Plasma Lipid Profiling by Ultra Performance Liquid Chromatography–Mass Spectrome- try”. Analytical chemistry 86.12, pp. 5766–5774. Sayin, S. I. et al. (2013). “Gut microbiota regulates bile acid metabolism by reducing the levels of tauro-beta-muricholic acid, a naturally occurring FXR antagonist”. Cell metabolism 17.2, pp. 225–235. Scherer, M. et al. (2009). “Rapid quantification of bile acids and their conjugates in serum by liquid chromatography–tandem mass spectrometry”. Journal of Chro- matography B 877.30, pp. 3920–3925. Schnabl, B. et al. (2014). “Interactions between the intestinal microbiome and liver diseases”. Gastroenterology 146.6, pp. 1513–1524. Seidell, J. C. et al. (1997). “Assessing obesity: classification and epidemiology”. British medical bulletin 53.2, pp. 238–252. Serfaty, L. (2012). “Chemoprevention of colorectal cancer with ursodeoxycholic acid: pro”. Clinics and research in hepatology and gastroenterology 36, S53–S60. Sergi, M. et al. (2012). “Analysis of bile acids profile in human serum by ultrafiltration clean-up and LC-MS/MS”. Chromatographia 75.9-10, pp. 479–489. Shimada, K. et al. (2001). “Gas chromatography and high-performance liquid chromatography of natural steroids”. Journal of Chromatography A 935.1, pp. 141– 172. Shindo, K. et al. (1988). “Deconjugation of bile acids by anaerobic bacteria isolated from the small intestine of cirrhotic patients”. Current therapeutic research 44.5, pp. 699–704. Shindo, K. et al. (1989). “The deconjugation ability of bacteria isolated from the jejunal fluids in the blind loop syndrome with high 14 CO 2 excretion using the

201 References

breath analysis technique and thin-layer chromatography”. Life sciences 45.24, pp. 2275–2283. Shockcor, J. et al. (2011). “Analysis of intact lipids from biological Matrices by UPLC/High Definition”. Waters Application note. Sivaraman, T. et al. (1997). “The mechanism of 2, 2, 2-trichloroacetic acid-induced protein precipitation”. Journal of protein chemistry 16.4, pp. 291–297. Smith, C. A. et al. (2005). “METLIN: a metabolite mass spectral database”. Thera- peutic drug monitoring 27.6, pp. 747–751. Smith, C. A. et al. (2006). “XCMS: processing mass spectrometry data for metabolite profiling using nonlinear peak alignment, matching, and identification”. Analytical chemistry 78.3, pp. 779–787. Sommer, U. et al. (2006). “LC-MS-based method for the qualitative and quantitative analysis of complex lipid mixtures”. Journal of lipid research 47.4, pp. 804–814. Souverain, S. et al. (2004). “Protein precipitation for the analysis of a drug cocktail in plasma by LC–ESI–MS”. Journal of pharmaceutical and biomedical analysis 35.4, pp. 913–920. Staudinger, J. L. et al. (2001). “The nuclear receptor PXR is a lithocholic acid sensor that protects against liver toxicity”. Proceedings of the National Academy of Sciences 98.6, pp. 3369–3374. Steiner, C. et al. (2010). “Quantification of the 15 major human bile acids and their precursor 7α-hydroxy-4-cholesten-3-one in serum by liquid chromatography– tandem mass spectrometry”. Journal of Chromatography B 878.28, pp. 2870– 2880. Stellwag, E. et al. (1979). “7alpha-Dehydroxylation of cholic acid and chenodeoxycholic acid by Clostridium leptum.” Journal of lipid research 20.3, pp. 325–333. Street, J. et al. (1983). “The quantitative estimation of bile acids and their conjugates in human biological fluids.” Journal of lipid research 24.5, pp. 491–511.

202 References

Sutherland, J. D. et al. (1985). “Bile acid induction of 7 alpha-and 7 beta-hydroxysteroid dehydrogenases in Clostridium limosum.” Journal of lipid research 26.3, pp. 344– 350. Sutherland, J. D. et al. (1987). “Oxidation of primary bile acids by a 7α-hydroxysteroid dehydrogenase elaborating Clostridium bifermentans soil isolate”. Canadian journal of microbiology 33.8, pp. 663–669. Swann, J. R. et al. (2011). “Systemic gut microbial modulation of bile acid metabolism in host tissue compartments”. Proceedings of the National Academy of Sciences 108.Supplement 1, pp. 4523–4530. Taira, K. et al. (1999). “Delayed post-prandial lipid metabolism in subjects with intra- abdominal visceral fat accumulation”. European journal of clinical investigation 29.4, pp. 301–308. Takamine, F. et al. (1995). “Isolation and Characterization of Bile Acid 7-Dehydroxyla- ting Bacteria from Human Feces”. Microbiology and immunology 39.1, pp. 11– 18. Takeuchi, Y. et al. (1980). “A comparative study on the neurotoxicity of n-pentane, n-hexane, and n-heptane in the rat.” British journal of industrial medicine 37.3, pp. 241–247. Tanaka, H. et al. (2000). “Bile salt hydrolase of Biﬁdobacterium longum-biochemical and genetic characterization”. Applied and environmental microbiology 66.6, pp. 2502–2512. Tautenhahn, R. et al. (2008). “Highly sensitive feature detection for high resolution LC/MS”. BMC bioinformatics 9.1, p. 504. Tenneson, M. E. et al. (1977). “The anaerobic side-chain cleavage of bile acids by Escherichia coli isolated from human faeces”. Biochemical Society Transactions 5.6, pp. 1758–1760. Theodoridis, G. A. et al. (2012). “Liquid chromatography–mass spectrometry based global metabolite proﬁling: a review”. Analytica chimica acta 711, pp. 7–16.

203 References

Thomas, A. M. et al. (2010). “Genome-wide tissue-specific farnesoid X receptor binding in mouse liver and intestine”. Hepatology 51.4, pp. 1410–1419. Thomas, C. et al. (2008). “Targeting bile-acid signalling for metabolic diseases”. Nature reviews Drug discovery 7.8, pp. 678–693. Toda, T. et al. (2009). “Antibiotics suppress Cyp3a in the mouse liver by reducing lithocholic acid-producing intestinal flora.” Yakugaku zasshi: Journal of the Pharmaceutical Society of Japan 129.5, pp. 601–608. Torres, D. M. et al. (2008). “Diagnosis and therapy of nonalcoholic steatohepatitis”. Gastroenterology 134.6, pp. 1682–1698. Traber, M. G. et al. (1999). “Molecular mechanisms of vitamin E transport”. Annual review of nutrition 19.1, pp. 343–355. Tran, T. T. et al. (2008). “Beneficial effects of subcutaneous fat transplantation on metabolism”. Cell metabolism 7.5, pp. 410–420. Tremaroli, V. et al. (2012). “Functional interactions between the gut microbiota and host metabolism”. Nature 489.7415, pp. 242–249. Trygg, J. et al. (2007). “Chemometrics in metabonomics”. Journal of proteome research 6.2, pp. 469–479. Tso, P. (1985). “Gastrointestinal digestion and absorption of lipid”. Adv. Lipid Res 21, pp. 143–186. Tsuboyama-Kasaoka, N. et al. (2006). “Taurine (2-aminoethanesulfonic acid) deficiency creates a vicious circle promoting obesity”. Endocrinology 147.7, pp. 3276– 3284. Tulipani, S. et al. (2012). “Comparative analysis of sample preparation methods to handle the complexity of the blood fluid metabolome: When less is more”. Ana- lytical chemistry 85.1, pp. 341–348. Turnbaugh, P. J. et al. (2006). “An obesity-associated gut microbiome with increased capacity for energy harvest”. nature 444.7122, pp. 1027–131. Turnbaugh, P. J. et al. (2009). “The core gut microbiome, energy balance and obesity”. The Journal of physiology 587.17, pp. 4153–4158.

204 References

Tushuizen, M. E. et al. (2010). “Postprandial lipid and apolipoprotein responses following three consecutive meals associate with liver fat content in type 2 diabetes and the metabolic syndrome”. Atherosclerosis 211.1, pp. 308–314. Unger, R. H. et al. (2010). “Lipid homeostasis, lipotoxicity and the metabolic syndrome”. Biochimica et Biophysica Acta (BBA)-Molecular and Cell Biology of Lipids 1801.3, pp. 209–214. Van Deemter, J. J. et al. (1956). “Longitudinal diffusion and resistance to mass transfer as causes of nonideality in chromatography”. Chemical Engineering Sci- ence 5.6, pp. 271–289. Van Eldere, J. et al. (1996). “Tauroconjugation of cholic acid stimulates 7 alpha- dehydroxylation by fecal bacteria.” Applied and environmental microbiology 62.2, pp. 656–661. Vanni, E. et al. (2010). “From the metabolic syndrome to NAFLD or vice versa?” Digestive and liver Disease 42.5, pp. 320–330. Veselkov, K. A. et al. (2011). “Optimized preprocessing of ultra-performance liquid chromatography/mass spectrometry urinary metabolic profiles for improved information recovery”. Analytical chemistry 83.15, pp. 5864–5872. Vuckovic, D. (2012). “Current trends and challenges in sample preparation for global metabolomics using liquid chromatography–mass spectrometry”. Analytical and bioanalytical chemistry 403.6, pp. 1523–1548. Wan, C. et al. (2008). “Coupling of cholesterol-rich lipid phases in asymmetric bilayers”. Biochemistry 47.7, pp. 2190–2198. Wang, Y. et al. (2006). “Matrix-assisted laser desorption/ionization high-energy collision-induced dissociation of steroids: analysis of oxysterols in rat brain”. Analytical chemistry 78.1, pp. 164–173. Want, E. J. et al. (2006). “Solvent-dependent metabolite distribution, clustering, and protein extraction for serum profiling with mass spectrometry”. Analytical Chemistry 78.3, pp. 743–752.

205 References

Want, E. J. et al. (2010). “Ultra performance liquid chromatography-mass spectrometry profiling of bile acid metabolites in biofluids: application to experimental toxicology studies”. Analytical chemistry 82.12, pp. 5282–5289. Want, E. J. et al. (2013). “Global metabolic profiling of animal and human tissues via UPLC-MS”. Nature protocols 8.1, pp. 17–32. Watanabe, M. et al. (2004). “Bile acids lower triglyceride levels via a pathway involving FXR, SHP, and SREBP-1c”. Journal of Clinical Investigation 113.10, p. 1408. Watanabe, M. et al. (2006). “Bile acids induce energy expenditure by promoting intracellular thyroid hormone activation”. Nature 439.7075, pp. 484–489. Watanabe, M. et al. (2011). “Lowering bile acid pool size with a synthetic farnesoid X receptor (FXR) agonist induces obesity and diabetes through reduced energy expenditure”. Journal of Biological Chemistry 286.30, pp. 26913–26920. Weisberg, S. P. et al. (2003). “Obesity is associated with macrophage accumulation in adipose tissue”. Journal of clinical investigation 112.12, p. 1796. White, B. et al. (1983). “Regulation of bile acid 7-dehydroxylase activity by NAD+ and NADH in cell extracts of Eubacterium species VPI 12708.” Journal of lipid research 24.1, pp. 20–27. WHO (2009). “Unhealthy diets and physical activity”. — (2015). “Obesity and overweight”. Wietholtz, H. et al. (1996). “Stimulation of bile acid 6α-hydroxylation by rifampin”. Journal of hepatology 24.6, pp. 713–718. Winslow, S. et al. (1978). “Health aspects of chloroform-a review”. Drug and chemical toxicology 1.3, pp. 259–275. Wishart, D. S. et al. (2007). “HMDB: the human metabolome database”. Nucleic acids research 35.suppl 1, pp. D521–D526. Wold, S. et al. (1987). “Principal component analysis”. Chemometrics and intelligent laboratory systems 2.1, pp. 37–52. Wolf, C. et al. (2008). “Lipidomics: Practical aspects and applications”. Progress in lipid research 47.1, pp. 15–36.

206 References

Wolk, A. et al. (2001). “Fatty acid composition of adipose tissue and serum lipids are valid biological markers of dairy fat intake in men”. The Journal of nutrition 131.3, pp. 828–833. Wymann, M. P. et al. (2008). “Lipid signalling in disease”. Nature Reviews Molecular Cell Biology 9.2, pp. 162–176. Xie, G. et al. (2015). “Profiling of Serum Bile Acids in a Healthy Chinese Population Using UPLC–MS/MS”. Journal of proteome research 14.2, pp. 850–859. Yamaguchi, K. et al. (2007). “Inhibiting triglyceride synthesis improves hepatic steatosis but exacerbates liver damage and fibrosis in obese mice with nonalcoholic steatohepatitis”. Hepatology 45.6, pp. 1366–1374. Yang, Y. et al. (2013). “New sample preparation approach for mass spectrometry- based profiling of plasma results in improved coverage of metabolome”. Journal of Chromatography A 1300, pp. 217–226. Ye, L. et al. (2007). “High-performance liquid chromatography–tandem mass spectrometry for the analysis of bile acid profiles in serum of women with intrahepatic cholestasis of pregnancy”. Journal of Chromatography B 860.1, pp. 10–17. Yoshimoto, S. et al. (2013). “Obesity-induced gut microbial metabolite promotes liver cancer through senescence secretome”. Nature 499.7456, pp. 97–101. Yu, J. et al. (2013). “Obesity, insulin resistance, NASH and hepatocellular carcinoma”. Seminars in cancer biology. Vol. 23. 6. Elsevier, pp. 483–491. Yun, J. W. et al. (2009). “Abnormal glucose tolerance in young male patients with nonalcoholic fatty liver disease”. Liver International 29.4, pp. 525–529. Zeevi, D. et al. (2015). “Personalized Nutrition by Prediction of Glycemic Responses”. Cell 163.5, pp. 1079–1094. Zhang, Y. et al. (2006). “Activation of the nuclear receptor FXR improves hyperglycemia and hyperlipidemia in diabetic mice”. Proceedings of the National Academy of Sciences of the United States of America 103.4, pp. 1006–1011. Zimmermann, E. et al. (2015). “Body mass index in school-aged children and the risk of routinely diagnosed non-alcoholic fatty liver disease in adulthood: a prospective

207 References

study based on the Copenhagen School Health Records Register”. BMJ open 5.4, e006998. Zivkovic, A. M. et al. (2009). “Eﬀects of sample handling and storage on quantitative lipid analysis in human serum”. Metabolomics 5.4, pp. 507–516.

208 Part I

Appendix:Copyrights

209 29/01/2015 Rightslink® by Copyright Clearance Center

Title: Objective Set of Criteria for Optimization of Sample If you're a copyright.com Preparation Procedures for Ultra user, you can login to High Throughput Untargeted RightsLink using your Blood Plasma Lipid Profiling by copyright.com credentials. Ultra Performance Liquid Already a RightsLink user or want to learn more? Chromatography–Mass Spectrometry Author: Magali H. Sarafian, Mathieu Gaudin, Matthew R. Lewis, et al Publication: Analytical Chemistry Publisher: American Chemical Society Date: Jun 1, 2014 Copyright © 2014, American Chemical Society

PERMISSION/LICENSE IS GRANTED FOR YOUR ORDER AT NO CHARGE

This type of permission/license, instead of the standard Terms & Conditions, is sent to you because no fee is being charged for your order. Please note the following:

Permission is granted for your request in both print and electronic formats, and translations. If figures and/or tables were requested, they may be adapted or used in part. Please print this page for your records and send a copy of it to your publisher/graduate school. Appropriate credit for the requested material should be given as follows: "Reprinted (adapted) with permission from (COMPLETE REFERENCE CITATION). Copyright (YEAR) American Chemical Society." Insert appropriate information in place of the capitalized words. Onetime permission is granted only for the use specified in your request. No additional uses are granted (such as derivative works or other editions). For any other uses, please submit a new request.

210 https://s100.copyright.com/AppDispatchServlet#formTop 1/1 08/09/2015 Rightslink® by Copyright Clearance Center

Title: Bile Acid Profiling and Quantification in Biofluids using If you're a copyright.com UltraPerformance Liquid user, you can login to Chromatography Tandem Mass RightsLink using your Spectrometry copyright.com credentials. Already a RightsLink user or Author: Magali H. Sarafian, Matthew R. want to learn more? Lewis, Alexandros Pechlivanis, et al Publication: Analytical Chemistry Publisher: American Chemical Society Date: Sep 1, 2015 Copyright © 2015, American Chemical Society

PERMISSION/LICENSE IS GRANTED FOR YOUR ORDER AT NO CHARGE

This type of permission/license, instead of the standard Terms & Conditions, is sent to you because no fee is being charged for your order. Please note the following:

211 https://s100.copyright.com/AppDispatchServlet#formTop 1/1 Part II

Appendix:Published articles

212 Article

pubs.acs.org/ac

Objective Set of Criteria for Optimization of Sample Preparation Procedures for Ultra-High Throughput Untargeted Blood Plasma Lipid Proﬁling by Ultra Performance Liquid Chromatography−Mass Spectrometry † † ‡ § ∥ † Magali H. Saraﬁan, Mathieu Gaudin, , Matthew R. Lewis, Francois-Pierre Martin, Elaine Holmes, † § † Jeremy K. Nicholson,*, , and Marc-Emmanuel Dumas*, † Imperial College London, Section of Biomolecular Medicine, Division of Computational Systems Medicine, Department of Surgery and Cancer, Sir Alexander Building, Exhibition Road, South Kensington, London SW7 2AZ, U.K. ‡ Technologie Servier, 25 Rue Eugenè Vignat, 45000 Orleans,́ France § Imperial College London, MRC NIHR National Phenome Centre, Division of Computational Systems Medicine, Department of Surgery and Cancer, IRDB Building, Du Cane Road, London W12 0NN, U.K. ∥ NestléInstitute of Health Sciences SA, 1015 Lausanne, Switzerland

*S Supporting Information

ABSTRACT: Exploratory or untargeted ultra performance liquid chromatography−mass spectrometry (UPLC−MS) profiling offers an overview of the complex lipid species diversity present in blood plasma. Here, we evaluate and compare eight sample preparation protocols for optimized blood plasma lipid extraction and measurement by UPLC−MS lipid profiling, including four protein precipitation methods (i.e., methanol, acetonitrile, isopropanol, and isopropanol− acetonitrile) and four liquid−liquid extractions (i.e., methanol combined with chloroform, dichloromethane, and methyl-tert butyl ether and isopropanol with hexane). The eight methods were then benchmarked using a set of qualitative and quantitative criteria selected to warrant compliance with high-throughput analytical workflows: protein removal efficiency, selectivity, repeatability, and recovery efficiency of the sample preparation. We found that protein removal was more efficient by precipitation (99%) than extraction (95%). Additionally, isopropanol appeared to be the most straightforward and robust solvent (61.1% of features with coefficient of variation (CV) < 20%) while enabling a broad coverage and recovery of plasma lipid species. These results demonstrate that isopropanol precipitation is an excellent sample preparation procedure for high-throughput untargeted lipid profiling using UPLC−MS. Isopropanol precipitation is not limited to untargeted profiling and could also be of interest for targeted UPLC−MS/MS lipid analysis. Collectively, these data show that lipid profiling greatly benefits from an isopropanol precipitation in terms of simplicity, protein removal efficiency, repeatability, lipid recovery, and coverage.

etabolic profiling and phenotyping provide a global MS profiling, nonselectivity during sample preparation is critical M understanding of complex metabolism using two major for successful analysis of these two molecular classes since analytical techniques: nuclear magnetic resonance spectroscopy sample preparation issues may impact the final quality of the − (NMR) and mass spectrometry (MS).1 3 Because of the data. However, sample preparation is key yet often overlooked complex composition of blood, the characterization of this step for the successful UPLC−MS based lipid profiling and can biofluid greatly benefits from the multidimensional separation have a strong impact on the quality of subsequent spectral data. and sensitive detection provided by ultra-high performance Given the importance of lipidomics in systems biology and in − variousdiseaseareassuchasobesityandmetabolic liquid chromatography coupled to MS (UPLC MS). UPLC − columns can operate with higher flows and pressures, which syndrome,11 14 there is a strong analytical need for developing directly results in shorter acquisition times for similar resolution robust and efficient methods fit-for-purpose in large-scale to HPLC.4 However, because of the large diversity in physicochemical properties of the metabolites found in plasma, Received: January 23, 2014 polar metabolites, and lipids are often analyzed separately by Accepted: May 12, 2014 − two different LC−MS methods.5 10 For untargeted UPLC− Published: May 12, 2014

213 Article

pubs.acs.org/ac

Bile Acid Profiling and Quantification in Biofluids Using Ultra- Performance Liquid Chromatography Tandem Mass Spectrometry † # † ‡ # † § Magali H. Sarafian, , Matthew R. Lewis, , , Alexandros Pechlivanis, Simon Ralphs, § ⊥ † † Mark J. W. McPhail, Vishal C. Patel, Marc-Emmanuel Dumas, Elaine Holmes, † and Jeremy K. Nicholson*, † Imperial College of London, Division of Computational Systems Medicine, Department of Surgery and Cancer, Sir Alexander Building, Exhibition Road, South Kensington, London SW7 2AZ, United Kingdom ‡ Imperial College of London, MRC-NHR National Phenome Centre, Department of Surgery and Cancer, IRDB building, Du Cane Road, London W12 0NN, United Kingdom § Imperial College of London, Department of Hepatology, St. Mary’s Hospital, Paddington, London, United Kingdom ⊥ King’s College London, Institute of Liver Sciences, Hospital NHS Foundation Trust, Division of Transplantation Immunology and Mucosal Biology, MRC Centre for Transplantation, London, United Kingdom

*S Supporting Information

ABSTRACT: Bile acids are important end products of cholesterol metabolism. While they have been identified as key factors in lipid emulsification and absorption due to their detergent properties, bile acids have also been shown to act as signaling molecules and intermediates between the host and the gut microbiota. To further the investigation of bile acid functions in humans, an advanced platform for high throughput analysis is essential. Herein, we describe the development and application of a 15 min UPLC procedure for the separation of bile acid species from human biofluid samples requiring minimal sample preparation. High resolution time-of-flight mass spectrometry was applied for profiling applications, elucidating rich bile acid profiles in both normal and disease state plasma. In parallel, a second mode of detection was developed utilizing tandem mass spectrometry for sensitive and quantitative targeted analysis of 145 bile acid (BA) species including primary, secondary, and tertiary bile acids. The latter system was validated by testing the linearity (lower limit of quantification, LLOQ, 0.25−10 nM and upper limit of quantification, ULOQ, 2.5−5 μM), precision (≈6.5%), and accuracy (81.2−118.9%) on inter- and intraday analysis achieving good recovery of bile acids (serum/plasma 88% and urine 93%). The ultra performance liquid chromatography−mass spectrometry (UPLC-MS)/MS targeted method was successfully applied to plasma, serum, and urine samples in order to compare the bile acid pool compositional difference between preprandial and postprandial states, demonstrating the utility of such analysis on human biofluids.

ile acids (BAs) are major components of bile, synthesized acid), secondary BAs produced mainly in the gut via B from cholesterol in the hepatocytes of the liver and play modification of primary BAs, and tertiary BAs which are fundamental roles in many physiological processes. BAs are formed in both the liver and gut via modification of secondary well-known as powerful emulsifiers of dietary lipids in the BAs, such as sulfation, glucuronidation, glucosidation, and N- intestine,1,2 antimicrobial agents,3,4 and signaling molecules acetylglucosaminidation.13 In the intestinal lumen, gut micro- regulating their own synthesis. BAs are modulated by gut biota are free to modulate the hepatic output through various − microbiota,5 8 and perturbations of the circulating BAs pool reactions, which include deconjugation and dehydroxylation at have been shown to contribute to development of liver and specific sites to form secondary BAs. All BAs are subject to a intestinal diseases.9,10 cycle of absorption, modification in the liver (further Numerous reports have shown the structural diversity of BAs conjugation), and excretion back to the gastrointestinal tract from cholesterol catabolism in the liver to microbial trans- in a process known as enterohepatic circulation. BA anabolism formations in the intestine.11,12 BAs traverse the boundary and biotransformation is thus a complex iterative process between endogenous metabolism and symbiotic gut bacterial metabolism, acting as a strong link between humans and their Received: April 24, 2015 intestinal microbiota. The pool of BAs is comprised of primary Accepted: September 1, 2015 BAs synthesized in the liver (cholic acid and chenodeoxycholic Published: September 1, 2015

214

Available online at www.sciencedirect.com

ScienceDirect

The microbiome and its pharmacological targets:

therapeutic avenues in cardiometabolic diseases

1 1

Ana Luisa Neves , Julien Chilloux , Magali H Saraﬁan,

Mohd Badrin Abdul Rahim, Claire L Boulange´ and

Marc-Emmanuel Dumas

Consisting of trillions of non-pathogenic bacteria living in a people with a dramatic impact on mortality, morbidity

symbiotic relationship with their mammalian host, the gut and quality of life [1]. These factors (including impaired

microbiota has emerged in the past decades as one of the key glucose tolerance, dyslipidemia, arterial hypertension,

drivers for cardiometabolic diseases (CMD). By degrading insulin resistance and central obesity) are epidemiologi-

dietary substrates, the gut microbiota produces several cally clustered — the presence of at least three of ﬁve of

metabolites that bind human pharmacological targets, impact these symptoms corresponding to the ‘metabolic syn-

subsequent signalling networks and in ﬁne modulate host’s drome’ clinical diagnosis [1]. Although many pharmaco-

metabolism. In this review, we revisit the pharmacological logical mechanisms have been suggested, the underlying

relevance of four classes of gut microbial metabolites in CMD: causes of CMD and its potential therapeutic avenues

short-chain fatty acids (SCFA), bile acids, methylamines and remain to be fully explored. With the advent of high-

indoles. Unravelling the signalling mechanisms of the microbial– throughput methodologies (metagenomics, metabolo-

mammalian metabolic axis adds one more layer of complexity to mics), the gut microbiome emerged as one of the key

the physiopathology of CMD and opens new avenues for the drivers for CMD [2]. The gut ecosystem, as well as its

development of microbiota-based pharmacological therapies. individual members, was shown to contribute to the host

Address metabolism [3 ]. A lower bacterial gene count (LGC) is

Division of Computational and Systems Medicine, Department of associated to increased adiposity, insulin resistance and

Surgery and Cancer, Imperial College London, Exhibition Road,

dyslipidemia [4 ] and dietary intervention can improve

London SW7 2AZ, UK

both bacterial gene richness and clinical metabolic out-

comes [5 ]. Patients with type 2 diabetes (T2D) also

Corresponding author: Dumas, Marc-Emmanuel

([email protected]) show speciﬁc compositional and functional changes in

their metagenomes [6 ].

These authors contributed equally to this review.

Current address: Metabometrix Ltd, Bio-incubator, Prince Consort

With the increasing number of clinical studies reporting

Road, South Kensington, London SW7 2BP, UK.

associations between the composition of the gut micro-

Chemical compounds studied in this article

biota and CMD outcomes, one question arises — how are

Acetate (PubChem CID: 176) Butyrate (PubChem CID: 264) Cholic

these changes in microbial ecology translated into phar-

acid (PubChem CID: 221493) Chenodeoxycholic acid (PubChem CID:

10133) Deoxycholic acid (PubChem CID: 222528) Indole-3- macological messages to the mammalian host? Consisting

propionate (PubChem CID: 3744) 3-indoxylsulfate (PubChem CID: of trillions of non-pathogenic bacteria living in a symbi-

10258) Propionate (PubChem CID: 1032) Trimethylamine (PubChem

otic relationship with their host, gut microbiota produces

CID: 1146) Trimethylamine-N-oxide (PubChem CID: 1145)

several signalling molecules (e.g., LPS, peptidoglycans,

but also metabolites) that bind host proteins and impact

Current Opinion in Pharmacology 2015, 25:36–44

signalling networks, therefore playing a central role as

This review comes from a themed issue on Endocrine and metabolic

chemical messengers in the microbial–mammalian cross-

diseases

talk [7]. The identiﬁcation of the pharmacological targets

Edited by Kevin G Murphy

and signalling pathways of these metabolites is key to a

For a complete overview see the Issue and the Editorial better understanding the molecular crosstalk supporting

Available online 28th October 2015 the microbial–mammalian metabolic axis — and provides

http://dx.doi.org/10.1016/j.coph.2015.09.013 a suitable framework for the discovery of the mechanistic

basis of these associations. In this context, ﬁne mapping of

the microbial signalling metabolome and its host molec-

ular targets opens up novel pharmacological avenues for

microbiome interventions.

The microbiome interacts with its host

Introduction through microbial metabolites

Cardiometabolic diseases (CMD) present a complex array In this review, we shall present four classes of gut micro-

of interrelated risk factors affecting more than one billion bial metabolites impacting host molecular mechanisms

Current Opinion in Pharmacology 2015, 25:36–44 www.sciencedirect.com

215 Part III

Appendix:Bile acid standards

216 1- Ursocholanic Acid 2- Lithocholenic Acid COOH COOH H3C H3C m/z 359.2955 m/z 373.2748

CH3 H CH3 H

CH3 CH3 H H H H

HO H H

3- 5-Cholenic Acid-3ß-ol 4- Ketocholanic Acid H C COOH H C COOH m/z 373.2748 3 m/z 373.2748 3 CH H CH3 H 3

CH CH3 3 H H H

HO O

5- Isolithocholic Acid 6- Allolithocholic Acid H C COOH COOH m/z 375.2904 3 m/z 375.2904 H3C CH H 3 CH3 H

CH 3 CH3 H H H H

HO HO H H 7- 3α,12α, 23-Nordeoxycholic Acid 8- 9(11), (5β)-Cholenic Acid-3α-ol COOH H3C m/z 377.2697 H3C -12-one COOH m/z 387.2540 O OH CH3 H CH3

CH3 CH3

HO HO H H 9- 5α-Cholanic Acid-3, 6-dione 10- 3,7-Diketocholanic Acid COOH H C COOH H3C m/z 387.2540 3 m/z 387.2540 CH H CH3 H 3

CH CH3 3 H H

O O H O O 217 11- 3,6-Diketocholanic Acid 12- 3,12-Diketocholanic Acid H C COOH COOH m/z 387.2540 3 m/z 387.2540 H3C O CH H 3 CH3 H

CH 3 CH3 H H H H O H O O H 13- 8(14),(5β)-Cholenic Acid-3α, 14- 5β-Cholenic Acid-7α-ol-3-one COOH H C COOH 12α-diol H3C m/z 389.2697 3 m/z 389.2697 HO CH H CH3 H 3

CH CH3 3 H H H

HO O OH H H 15- 5α-Cholanic Acid-3α-ol-6-one 16- 3α-Hydroxy-7 Ketolithocholic COOH H C COOH H3C m/z 389.2697 3 Acid m/z 389.2697 CH H CH3 H 3

CH CH3 3 H H H H H

HO HO O H H O 17- 3α-Hydroxy-12 Ketolithocholic 18- Lithocholic acid COOH H C Acid H3C m/z 375.2904 3 m/z 389.2697 O CH H COOH CH3 H 3

CH3 CH3 H H H H H

HO H HO H 19- 5β-Cholanic Acid-3β, 12α-diol 20- Chenodeoxycholic Acid COOH COOH m/z 391.2853 H3C m/z 391.2853 H3C OH H CH3 CH3 H

CH 3 CH3 H H H H H

HO OH H HO H 218 21- Deoxycholic Acid 22- Hyodeoxycholic acid COOH COOH m/z 391.2853 H3C m/z 391.2853 H3C

OH H CH3 CH3

CH CH3 3 H H H

HO HO H H OH 23- Isodeoxycholic Acid 24- Murocholic Acid m/z 391.2853 COOH m/z 391.2853 COOH H3C H3C OH H CH3 H CH3

CH3 H CH3 H H H

HO H OH OH 25- Ursodeoxycholic acid 26- 3,7,12 Dehydrocholic acid COOH H3C COOH m/z 391.2853 m/z 401.2333 H3C O H CH3 CH3 H

CH3 H CH3 H H H H OH HO O H H O 27- 3α-Hydroxy-7,12-DiketoCholanic 28- 3α-Hydroxy-6,7-DiketoCholanic COOH COOH Acid H3C Acid H3C m/z 403.2389 O m/z 403.2489 H CH3 H CH3

CH3 CH3 H H

HO HO O H O H O 29- 5β-Cholanic Acid-3α, 6α-diol-7 30- 3 Dehydrocholic Acid COOH -one H3C m/z 405.2646 COOH H3C m/z 405.2646 OH CH3 H CH3 H

CH3 CH3 H H HO H O O HO H OH 219 31- 12 Dehydrocholic Acid 32- α Muricholic Acid O COOH H3C m/z 405.2646 H3C m/z 407.2802 OH O CH3 H CH3 H

CH3 CH3 H H H H H HO OH HO OH H H OH 33- β Muricholic Acid 34- ω Muricholic Acid O O m/z 407.2802 H C m/z 407.2802 H3C 3 OH OH

CH3 H CH3 H

CH3 CH3 H H H H H H

OH HO OH HO H H OH OH 35- Cholic acid 36- Hyocholic acid COOH H3C m/z 407.2802 H3C m/z 407.2802

OH CH3 H CH3 O

OH CH3 H CH3 H H

HO OH HO OH H H OH 37- Glyco-ursocholanic Acid 38- Glycolithocholic Acid O O m/z 416.3170 m/z 432.3119 H C OH H3C OH 3 NH NH O H O CH3 H CH3

CH CH3 H 3 H H H H H

H HO H 39- Glycoursodeoxycholic Acid 40- Glycohyodeoxycholic Acid O O m/z 448.3068 m/z 448.3068 H C OH H3C OH 3 NH NH O CH H O CH3 H 3

CH CH3 H 3 H H H H H

OH HO HO H H 220 OH 41- Glycochenodeoxycholic Acid 42- Glycodeoxycholic acid O O m/z 448.3068 m/z 448.3068 H3C OH H3C OH NH NH O OH O CH3 H CH3 H

CH3 H CH3 H H H H H

OH HO H HO H 43- 3,7,12 Glycodehydrocholic acid 44- Glycocholic Acid O O m/z 458.2548 m/z 464.3017 H C OH H3C OH 3 NH NH O OH O O CH H CH3 H 3

CH CH3 H 3 H H H H H

HO OH O H O H 45- Glycohyocholic Acid 46- Taurolithocholic Acid O O O m/z 464.3017 H C OH m/z 482.2945 3 NH H C S OH 3 NH H O CH3 O CH3 H

CH 3 H CH3 H H H H H

OH HO HO H H OH 47- Tauro-ursocholanic Acid 48- Tauro-ursodeoxycholic Acid O O O m/z 466.2996 m/z 498.2894 O H3C S OH H C S OH NH 3 NH H O O CH3 CH3 H

CH CH3 H 3 H H H H H HO OH H H 49- Taurohyodeoxycholic Acid 50- Taurochenodeoxycholic Acid O O m/z 498.2894 O m/z 498.2894 O H C S OH H C S OH 3 NH 3 NH H O O CH3 CH3 H

CH 3 H CH3 H H H H H

OH HO H HO H OH 221 51- Taurodeoxycholic Acid 52- 3,7,12 Taurodehydrocholic acid O O m/z 498.2894 O m/z 508.2374 O H3C S OH NH H3C S OH NH OH H O O CH3 H O CH3

CH3 CH H 3 H H H H H HO H O H O 53- Taurohyocholic Acid 54- Tauro-α Muricholic Acid O O O m/z 514.2843 m/z 514.2843 O H3C S OH H C S OH NH 3 NH H O H O CH3 CH3

CH CH3 H 3 H H H H H

OH HO OH HO H H OH OH 55- Tauro-β Muricholic Acid 56- Tauro ω-Muricholic Acid O O m/z 514.2843 O m/z 514.2843 O H C S OH H C S OH 3 NH 3 NH H O H O CH3 CH3

CH CH 3 H 3 H H H H H

OH OH HO H HO H OH OH 57- Taurocholic Acid O m/z 514.2843 O H C S OH 3 NH OH H O CH3

CH3 H H H

HO H OH

222