Assessing and Evaluating Biomarkers and Chemical Markers by Targeted and Untargeted Mass Spectrometry-Based Metabolomics

MIAMI UNIVERSITY The Graduate School

Certificate for Approving the Dissertation

We hereby approve the Dissertation

Kundi Yang

Candidate for the Degree

Doctor of Philosophy

______Dr. Michael Crowder, Director

______Dr. Neil Danielson, Reader

______Dr. Rick Page, Reader

______Dr. Scott Hartley, Reader

______Dr. Xin Wang, Graduate School Representative

ABSTRACT

ASSESSING AND EVALUATING BIOMARKERS AND CHEMICAL MARKERS BY TARGETED AND UNTARGETED MASS SPECTROMETRY-BASED METABOLOMICS

Kundi Yang

The features of high sensitivity, selectivity, and reproducibility make liquid chromatography coupled mass spectrometry (LC-MS) the mainstream analytical platform in metabolomics studies. In this dissertation work, we developed two LC-MS platforms with electrospray ionization (ESI) sources, including a LC-triple quadrupole (LC-QQQ) MS platform for targeted metabolomics analysis and a LC-Orbitrap MS platform for untargeted metabolomics analysis, to assess and evaluate either biomarkers or chemical markers in various sample matrices. The main focus of targeted metabolomics platform was to study human gut microbiota and how the gut microbiome responds to therapies. We utilized a LC-QQQ targeted platform for this focus and conducted three projects: (1) to evaluate the effects of four nutrients, mucin, bile salts, inorganic salts, and short chain fatty acids, on the gut microbiome in terms of their impact to the microbial metabolic profiles in vitro, (2) to examine the ability of Lactobacillus acidophilus (LA) to ferment black tea extract (BTE) and to investigate the enhancement of antimicrobial ability of the fermented BTE against Escherichia coli due to the increasing abundant of phenolic compounds in the fermentation process, (3) to determine the modulating effects of the exposure to gastrointestinal (GI) ultrafine particles (UFPs) on gut microbial composition and functions that may lead to a systematic evaluation of the impact of UFPs on host health by using an in vivo murine model. Two additional studies were accomplished by using an LC-Orbitrap untargeted/hybrid platform. In one study we used both LC-QQQ and LC-Orbitrap approaches to analyze polar metabolites from micro anaerobic bacterial culture matrices, to systematically differentiate four Lactobacillus species based on their metabolic profiles. In the other project, the LC-Orbitrap was incorporated with a Fourier-transform ion cyclotron resonance mass spectrometry (FT-ICR MS) system to determine whether bourbons could be “fingerprinted” by their chemical compositions.

ASSESSING AND EVALUATING BIOMARKERS AND CHEMICAL MARKERS BY TARGETED AND UNTARGETED MASS SPECTROMETRY-BASED METABOLOMICS

A DISSERTATION

Presented to the Faculty of

Miami University in partial

fulfillment of the requirements

for the degree of

Doctor of Philosophy

Department of Chemistry and Biochemistry

Kundi Yang

The Graduate School Miami University Oxford, Ohio

2020

Dissertation Director: Dr. Michael Crowder

Kundi Yang

2020

TABLE OF CONTENTS

TABLE OF CONTENTS ...... iii LIST OF TABLES ...... vi LIST OF FIGURES ...... vii DEDICATION ...... x ACKNOWLEDGEMENTS ...... xi CHAPTER 1 ...... 1 Introduction ...... 1 1.1 Metabolomics and its applications ...... 1 Targeted/untargeted approaches in metabolomics ...... 1 Mass spectrometry and MS-based metabolomics ...... 3 Metabolomics in human gut microbiota research ...... 4 Metabolomics in food science ...... 6 1.2 Specific aims ...... 7 CHAPTER 2 ...... 16 Evaluating the Impact of Four Major Nutrients on Gut Microbial Metabolism by a Targeted Metabolomics Approach ...... 16 2.1 Introduction ...... 18 2.2 Materials and Methods ...... 19 2.3 Results and Discussion ...... 22 Targeted metabolic profiling differentiates the gut microbial growth in LCIS and HCIS ... 22 Targeted metabolic profiling differentiates the gut microbial metabolic profiles with other nutrients in a defined inorganic salts level ...... 25 Metabolic pathway impact analyses ...... 29 2.4 Conclusions ...... 33 CHAPTER 3 ...... 37 Metabolomics Study Reveals Enhanced Inhibition and Metabolic Dysregulation in Escherichia coli Induced by Lactobacillus acidophilus-Fermented Black Tea Extract ...... 37 3.1 Introduction ...... 39 3.2 Materials and Methods ...... 40 3.3 Results and Discussion ...... 44 CHAPTER 4 ...... 60

iii Ultrafine Particles Altered Gut Microbial Population and Metabolic Profiles in a Sex-Specific Manner in an Obese Mouse Model ...... 60 4.1 Introduction ...... 62 4.2 Materials & Methods ...... 63 4.3 Results ...... 67 UFP-Induced Changes in the Structures of Gut Microbial Communities ...... 67 UFP-induced changes in the metabolic profile of the gut microbiome ...... 71 UFP-induced metabolic pathway alteration ...... 74 4.4 Discussion ...... 80 4.5 Conclusions ...... 82 ASSOCIATED CONTENT ...... 83 CHAPTER 5 ...... 91 Rapid Differentiation of Lactobacillus Species via Metabolic Profiling ...... 91 5.1 Introduction ...... 93 5.2 Materials and method ...... 94 5.3 Results ...... 96 5.4 Discussion ...... 105 5.5 Conclusions ...... 106 CHAPTER 6 ...... 111 Analysis of Barrel-Aged Kentucky Bourbon Whiskey by Ultrahigh Resolution Mass Spectrometry ...... 111 6.1 Introduction ...... 113 6.2 Materials and Methods ...... 114 6.3 Results and Discussion ...... 116 FT-ICR MS analysis of different aged bourbon samples ...... 116 HPLC-MS analysis of different aged bourbon samples ...... 123 Different chemical compositions of single barrel bourbons ...... 125 6.4 Conclusions ...... 129 CHAPTER 7 ...... 134 Conclusions ...... 134 APPENDIX A ...... 138 Supporting information for Chapter 2 ...... 138 APPENDIX B ...... 151 Supporting information for Chapter 4 ...... 151 APPENDIX C ...... 155

iv Supporting information for Chapter 5 ...... 155 APPENDIX D ...... 165 Supporting information for Chapter 6 ...... 165

LIST OF TABLES

Table 2.1 Experiment design and group information...... 22

Table 3.1 Targeted phenolic compounds and their detection parameters...... 48 Table 3.2 Example of targeted metabolites and their detection parameters...... 49

Table 6.1 Representative compounds that have been detected from LC-MS, which showed significant different abundances among aged bourbon whiskey samples...... 121

Table S2.1 Detailed list of the overview of metabolic profiles of all studied samples with the various nutritional enrichments...... 142 Table S2.2 Metabolites that have fold change larger than 2 in terms of their abundant in inorganic salts comparison...... 146 Table S2.3 Metabolites that have fold change larger than 2 in terms of their abundant in mucin comparison...... 149

Table S4.1 Mice cecum microbial metabolic pathways identified in Figure 4.5...... 153 Table S4.2 Host metabolic pathways (using plasma metabolites) identified in pathway analysis (Figure 4.7)...... 154

Table S5.1 Targeted compounds detected by LC-QQQ-MS...... 156 Table S5.2 Untargeted compounds detected by LC-Orbi-MS. The compounds were matched with Chemspider database...... 161

Table S6.1 Characteristics of bourbon samples used in this study. Table includes number of samples (n), percent alcohol by volume (%ABV), pH values (mean and standard deviation), and colors of sample...... 165 Table S6.2 Twelve compounds initially abundant in the unaged spirit but are at much lower levels in the 6-year aged bourbon...... 166 Table S6.3 Number of elemental formulae detected from different bourbon samples by FT-ICR MS...... 167

LIST OF FIGURES

Figure 1.1 General classification of metabolomics...... 3

Figure 2.1 Workflow of this study...... 23 Figure 2.2 The overview of metabolic profiles of all studied samples with the various nutritional enrichments...... 27 Figure 2.3 Detailed PCA score plots of metabolomics data from low vs. high concentration inorganic salts supplemented gut microbial cultures...... 28 Figure 2.4 Major metabolic pathways of the two studied nutrient factors that impacted metabolic pathways during this study...... 30 Figure 2.5 Pathway analysis of mucin group (L-3)...... 32

Figure 3.1 A. The schematic of workflow used in this study. B. Inhibitory effect of BTE treatment to E. coli with or w/o Lactobacillus acidophilus (LA) fermentation...... 45 Figure 3.2 Measurements of phenolic compound concentrations...... 47 Figure 3.3 SEM images showing the possible intracellular oxidative stress induced cell death. . 50 Figure 3.4 Heatmap presentation of intracellular metabolic profiles...... 53 Figure 3.5 Box plots of nine example metabolites showing significant difference from different treatment groups and control group...... 54 Figure 3.6 Partial least square – discriminant analysis (PLS-DA) approach differentiates the E. coli control group, the BTE treatment group and fermented BTE treatment group based on their metabolic profiles...... 55

Figure 4.1 The relative abundance of bacterial levels in mice cecum samples...... 69 Figure 4.2 The difference in mean proportion of microbial function...... 70 Figure 4.3 Gut microbial metabolic profiling of female and male mice based on the metabolomic analysis of cecal samples...... 73 Figure 4.4 Host metabolic profile difference in both female and male mice and the most contributed metabolites in plasma samples of female mice...... 76 Figure 4.5 Gut microbial metabolic pathway analysis on cecal samples...... 77 Figure 4.6 Connections among metabolic pathways that matched with microbial functional prediction result. Females (Figure 4.6A) and males (Figure 4.6B)...... 78

vii Figure 4.7 Host metabolism pathway analysis on plasma samples...... 79

Figure 5.1 Overview of polar metabolites platform...... 98 Figure 5.2 Triple quadrupole MS and Orbitrap MS data analyses workflow...... 99 Figure 5.3 A representative spectral comparison of the same compound between targeted and untargeted platforms...... 101 Figure 5.4 Analytical performance of both MS-Metabolomics workflows...... 102 Figure 5.5 Overview of data quality...... 103 Figure 5.6 MS intensity distribution of four anaerobic bacterial strains from targeted (a) and untargeted (b) metabolomics workflow...... 103 Figure 5.7 PLS-DA analysis of targeted and untargeted metabolomics study...... 104 Figure 5.8 Heat map of targeted and untargeted metabolomics study...... 104

Figure 6.1 FT-ICR-MS spectra of bourbon samples aged (top to bottom) from 0, 2, 4, and 6 years in oak barrels...... 117 Figure 6.2 FT-ICR MS analyses...... 120 Figure 6.3 HPLC-MS/MS (Orbitrap) analysis of different aged bourbons...... 122 Figure 6.4 Van Krevelen diagrams (H/C vs. O/C atomic ratio) of the three 8-year bourbons that aged parallel in three different barrels...... 126 Figure 6.5 LC-MS/MS analyses on ‘single barrel’ bourbons...... 128

Figure S2.1 Comparison between high and low concentration of inorganic salts...... 138 Figure S2.2 Heatmap of (a) low concentration inorganic salt groups and (b) high concentration inorganic salt groups...... 139 Figure S2.3 (a) PCA plot showed L-3 group clearly different from L-4 group. (b) and (c) are 5- hydroxymethyluracil and indole-3-acetaldehyde showed lower concentrations in mucin treated groups...... 140 Figure S2.4 Bile salts and SCFAs effect without mucin treatment in low concentration inorganic salts medium...... 141

Figure S4.1 PLS-DA plot of B0 vs B20 in (a) female obese mice and (b) male obese mice. .... 151 Figure S4.2 Intestinal and plasma pyridoxal levels detected from female obese mice...... 151 Figure S4.3 Plasma Myo-inositol detected from male obese mice. The box and whisker plots summarize the normalized values...... 152

viii Figure S5.1 Zoomed mass spectrums of leucine/isoleucine in targeted and untargeted platform...... 155

Figure S6.1 An example of ion with m/z of 300.99989 was observed in maturated samples but not in non-matured whiskey sample and was identified as the deprotonated ellagic acid (C14H5O8)...... 168 Figure S6.2 Metabolic patterns of 2, 4 and 6-year bourbons...... 169 Figure S6.3 Box plots of ten characteristic compounds showing significant differences from bourbon whiskies that aged differently (2, 4 and 6 years)...... 170 Figure S6.4 Loading plot of the first two principal components...... 171

DEDICATION

I dedicate this dissertation to my parents. I would not be here without their endless support and love.

ACKNOWLEDGEMENTS

I am incredibly grateful for all the amazing support from so many people during this journey. I would first like to thank my advisors, Dr. Michael W. Crowder and Dr. Jiangjiang (Chris) Zhu, who helped me most to pursue my Ph.D. degree. During the four and half years of my study at Miami, both of them devoted so much time and efforts to educate me. Dr. Crowder and Dr. Zhu are great examples of chemistry scientists to me, helping me whenever I was struggling along with my journey. I want to acknowledge my committee members: Dr. Neil Danielson, Dr. Rick Page, Dr. Scott Hartley, and Dr. Xin Wang for taking the time to provide guidance on my research and offer valuable suggestions. I would especially like to thank Dr. Neil Danielson and Dr. Xin Wang for their kind help in my learning and troubleshooting of the instruments. I would like to thank my colleagues, both current and former. It was a terrific experience to know and work with you. I appreciate all the help from you about my research projects and the after-lab adventures we had together in life. I wish you all the best. Last but not least, I want to thank my family and friends, especially my parents, Zhongqing Yang and Yuru Fang, for their unending and unconditional love and support. Your encouragement was the strongest power that supported me moving forward even when I wanted to give up.

xi CHAPTER 1 Introduction 1.1 Metabolomics and its applications The term ‘metabolomics’ has been defined as the measurement of metabolite concentrations in microorganisms, cells, and tissues, and the concentrations of the metabolites are correlated with gene expression and protein activities.[1] As an “-omic” science in systems biology, metabolomics have been used to determine the abundance of metabolites that exist in endogenous and exogenous biological systems.[2] For such reasons, the metabolomic fingerprint, which represents the expression pattern of individual metabolites, reflects the idiosyncratic biological environment of the subject from which a sample was obtained.[3] In many ways, metabolomics shows its advantage as a disease diagnostic tool because the metabolic dynamics reflect responses of living systems to stimuli from the biological environments.[4] For example by applying 16S rRNA gene sequencing combined with mass spectrometry-based metabolomics profiling, Gao et al. revealed that mice, which were exposed to diazinon, exhibited a perturbation in their gut microbiome and metabolic functions in a sex-specific manner.[5] Another work reported by Li et al. utilized lipid metabolomics and found that ingestion of ultrafine particles led to an alteration in gut microbiota in association with increased atherogenic lipid metabolites.[6] Metabolomics can also be thought of downstream transcriptomics, genomics, and proteomics because the metabolic profile is likely a more definitive indicator of the disease process, and metabolomics can be used to diagnose human diseases, predict the potential outcomes of diagnoses, and evaluate toxicities of treatments.[4, 7, 8]

Targeted/untargeted approaches in metabolomics Targeted and untargeted methods are two major approaches to conduct metabolomics studies (Figure 1.1). The targeted approach is hypothesis-driven, while the untargeted approach does not require a prior hypothesis and follows a discovery-based strategy. To be specific, targeted analyses focus on specified metabolites and involve detection of these metabolites in a given sample. In contrast, untargeted metabolomics analyses involve the detection of as many metabolites as possible to obtain patterns or chemical fingerprints without necessarily identifying nor quantifying the specific metabolites.[9] In most cases, targeted metabolomics focus on a limited number of analytes.[10] For example, Yang et al. detected 177 out of 234 targeted

1 metabolites from Lactobacillus acidophilus, and the detected metabolites could be used to differentiate this bacterial strain from others.[11] In a different report, 221 metabolites were identified from multiple bacterial samples, and 82 of the compounds were detected in three strains.[12] A previous targeted metabolomic study indicated that up to 500 compounds could be quantified and identified.[13] Since targeted metabolomic analyses focus on metabolites of interest with most cases requiring identification and quantification of as many metabolites within a specific group, higher purities of the samples and selective extraction of metabolites are always required.[14] Unlike targeted metabolomic approaches that are limited by coverage of detected metabolites, untargeted metabolomic approaches evaluate all signals, which are either above the limit of detection of the instrument or above a pre-determined detection level. Therefore, untargeted metabolomic approaches achieve a wider range of metabolite detection. For example, Roullier et al. detected around 6,000 potential m/z values by utilizing a Fourier-transform ion cyclotron resonance mass spectrometry (FT-ICR-MS) metabolomic approach, and 791 compounds were identified after database annotation.[15] Another untargeted work by Kew et al. used high resolution time-of-flight (TOF) MS to successfully fingerprint Scotch whiskies with around 700 metabolites detected.[16] Since the aim of untargeted metabolomics is to characterize the biological sample by analyzing a wide range of unknowns based on their metabolic profiles, chemical fingerprints from different groups are compared and overlaid to find out those with statistically-significant differences, and the corresponding chemical identifications of those signals can often be made based on database searches.[17] Because there is a potential to discover novel biomarkers, this untargeted metabolomic approach can offer insights into metabolic pathways and interactions.[18, 19] These insights have been reported in drug development,[20] food analyses, and epidemiological studies.[21]

Figure 1.1 General classification of metabolomics.

Mass spectrometry and MS-based metabolomics Nuclear magnetic resonance (NMR) and mass spectrometry (MS) are the two most popular platforms utilized in metabolomics research. Compared to NMR-based metabolomics, MS-based approaches allow for metabolite limits of detection down to picomolar levels, as compared to the nanomolar levels reported for NMR-based approaches.[22] Moreover, direct complementary separation techniques make the chromatography-separation coupled, MS-based metabolomics the mainstream in recent years.[23] Early on, gas chromatography (GC) was the only separative technology that could be coupled to mass spectrometry. The constraint of GC-based separative MS detection is obvious; that is, only a small set of biological molecules can be volatilized, for example, some volatile organic compounds (VOCs).[24] For this reason, macromolecules, such as peptides, proteins, and nucleic acids, were very difficult to detect by using GC/MS. However, the development of liquid chromatography (LC) coupled with atmospheric pressure ionization mass spectrometry (API-MS) improved this situation. This technique has advantages as it exhibits good sensitivity, high dynamic range, and versatility.[25] The diversity of MS analyzers also makes API-MS a strong technique to be applied to different studies. For example, ion trap, triple quadrupoles (QQQ), time-of-flight (TOF), Orbitrap, and Fourier transform ion cyclotron resonance (FT-ICR) MS analyzers have been used to analyze complex biological samples.[26-28]

3 Compared to GC-MS-based metabolomics, the greatest advantage of LC-MS-based metabolomics is that a more global chemical/biochemical profile can be provided because a larger coverage of compounds can be achieved.[29] LC-coupled MS with an electrospray ionization (ESI) source is one of the most common platforms that has been utilized in both targeted and untargeted metabolomics.[10, 30] Targeted LC-MS is a technique that used to detect known, pre-selected metabolites reliably and quantitatively.[29] The stationary phase of LC columns and the solvents can be used to optimally elute the analytes of interest.[31] A commonly-used LC/MS column is a reversed phase (RP) analytical column (2.1 or 4.6 mm internal diameter) to separate the analytes before they are ionized.[32, 33] However, RP columns cannot be used to retain small, polar metabolites, and the use of hydrophilic interaction chromatography (HILIC) columns are often used to separate these compounds.[34, 35] The MS analyzer chosen in targeted analyses needs to have the capability to collect metabolite information at a high scan rates.[29] QQQ or linear ion trap instruments are chosen as the MS analyzers in most situations because they can be operated in the selected reaction monitoring (SRM) mode (or multiple reaction monitoring mode, MRM).[36] This scanning mode exhibits high selectivity by isolating the precursor ions, then breaking them into different fragments with optimized collision energies.[37, 38] A major difference between targeted and untargeted LC/MS is that they use different MS analyzers.[29] In an untargeted LC/MS analysis, MS analyzers are required to be able to analyze complex biological matrix with high mass accuracies, while maintaining reasonable scan rates.[29] TOF and QTOF are two types of instruments that meet these requirements.[29] FT-ICR offers the highest mass accuracies of all mass spectrometers (sub-ppm accuracies), while Orbitraps produce similar though slightly lower accuracies.[39-41] Both FT-ICR and Orbitrap instruments are usually configured as hybrid mass spectrometers, with ion trap or quadrupole mass analyzers preceding the main detectors.[29]

Metabolomics in human gut microbiota research Gut microbiota is an assortment of microorganisms inhabiting the length and width of the mammalian gastrointestinal tract.[42] In humans, the initial individual gut microbiota is provided by the mother.[43] Subsequent changes of microbial communities are influenced by various determinants, such as topographical factors, diet, disease, and usage of antibiotics.[44] The interactions between gut microbiota and host immune system can affect various biochemical

4 signaling pathways, which act upon many organs of the human body.[45] Many metabolic reactions are modulated by these signaling pathways, which in turn can ultimately lead to the production of bile acids, choline, and short-chain fatty acids (SCFAs) that are essential for host health.[1] Gut microbiota play an important role in the human body. Previous studies have reported that gut microbiota participate in several physiological functions, which include maintaining intestinal functions,[46] harvesting energy,[47] and helping with host immune system regulation.[48] It also has been shown that the microbial composition affects host-pathogen interactions, and altered compositions can cause diseases in certain circumstances.[49] In addition to the varied physiological functions that gut microbiota influence, several metabolically- important functions, such as short chain fatty acid (SCFAs) production, bile acid biotransformation, and amino acids synthesis, are also impacted notably.[50] Many diseases, such as obesity, diabetes, cancer, allergy, inflammatory bowel disease (IBD), metabolic disorders, and neuropathology, may develop when the symbiotic balance between gut microbiota and gastrointestinal (GI) tract is disrupted.[51, 52] Well-developed metabolomic approaches have become powerful tools to investigate the relationships between the environment, genetics, and health, and these approaches have been used to identify clinical biomarkers for human diseases.[53] Many studies have applied metabolomic approaches to probe microbiota, and most of these studies aimed to unravel the important metabolic pathways in the gut by exploring disease-related metabolites.[52] For example, Li et al. successfully detected 77 metabolites from an in vitro gut microbial culture by using a mass spectrometry-based metabolomics approach.[54] In another study, Li et al. reported that ingestion of ambient ultrafine particles alter gut microbiota in association with increased atherogenic lipid metabolites.[6] In other words, gut microbiota affects the metabolism of the host; therefore, the study of gut microbiota and host metabolic pathways can be used to discriminate between healthy and unhealthy experimental subjects.[55] Colorectal cancer threatens the health of a million people every year, and this disease can be predicted by monitoring alterations in the gut ecosystem.[56] By studying intestinal metabolites, which includes amino acids, bile acids, and SCFAs, in healthy and diseased patients, one can understand the association between these compounds and colorectal cancer by linking the specific metabolites to certain functional metabolic pathways.[57-59]

Metabolomics in food science Foods are complex materials and have been studied by chemists, microbiologists, and physicists.[60] However, food adulteration has become one of the most common risks to overall food safety.[61] In addition, there are reports of food fraud/substitution in which more expensive, harder to find ingredients are replaced with cheaper, poorer quality ingredients.[62] A noteworthy example of this event occurred in the Dominican Republic when cheap, poorly-prepared alcoholic beverages were sold as “brand” name beverages, and these substituted beverages were reported to cause the deaths of at least nine people (https://www.businessinsider.com/dominican-illicit- alochol-may-have-killed-tourists-report-history-2019-6). This conclusion has been called into question though (https://www.usatoday.com/story/travel/2019/10/18/dominican-republic-deaths- fbi-testing-shows-no-tainted-alcohol/4024849002/). Often the adulterated or substituted foods are not analyzed for quality. Therefore the authentication of certain foods can have substantial impact on food and human safety.[63] Iodine and saponification values on edible oils have been used as classic analytical methods for food authentication during the past decades.[64] However, new methodologies, such as metabolomics, are becoming appealing because they showed advantages as they are more economical and less time consuming in food authentication applications.[63] Metabolomics with all its advantages has been utilized in many aspects of food science, such as authenticity,[65] food safety,[66] and bioactivity evaluation.[67] Generally, based on the specific objective of the analyses, the aims of food metabolomic studies can include (1) to discriminate samples, (2) to identify and quantitatively analyze samples based on their metabolic profiles, and (3) to predict classification/identification schemes by creating and utilizing statistical models.[14] The purpose of discriminate analyses is to uncover differences between sample populations. Roullier et al. successfully discriminated whiskies matured in bourbon barrels and sherry barrels by using a metabolomic approach.[15] Another work reported the use of a ultrahigh- performance liquid chromatography-orbitrap mass spectrometry-based metabolomic approach to analyze golden rums.[65] Although in some circumstances metabolomics-based food authentication approaches showed obvious advantages over traditional methods,[68] metabolomics-based approaches have not been recognized as viable analytical methods by regulatory agencies for use in these issues.[63] Indeed, a lot more studies are still needed to draw valid conclusions and to help develop these approaches in food authentication and identification.

1.2 Specific aims

This dissertation describes the development and use of MS-based metabolomic approaches to address several different questions. The dissertation is divided into 5 experimental chapters:

(1) Chapter 2 describes efforts to identify the specific shifts in metabolic profiles in response to inorganic salts, mucin, bile salts, and short chain fatty acids supplementation to Gifu Anaerobic medium (GAM) growth of a human-derived gut microbial population. This study was used to investigate the mechanism of nutrient-bacteria interactions and to suggest an optimal environment for in vitro gut microbiome cultures.

(2) Chapter 3 describes the combination of high-performance liquid chromatography−tandem mass spectrometry (HPLC−MS/MS)-based targeted compound detection and L. acidophilus (LA) fermentation to investigate the enhanced inhibitory capability of LA-fermented black tea extract (BTE) and its effect on Escherichia coli. To better understand the molecular level events in E. coli cells during the fermented BTE treatment, both intracellular levels of phenolic compounds and metabolites were detected and compared to those found in E. coli with a non-fermented BTE treatment.

(3) Chapter 4 describes efforts to determine whether the exposure to ultrafine particles (UFP) generated from combustion of two fuels can alter the gut microbiota community, along with the gut metabolic profiles and host metabolism. We also investigated sex-specific responses in the mouse gastrointestinal (GI) tract after exposure to UFPs.

(4) Chapter 5 describes efforts to develop targeted/untargeted hybrid metabolomics methods and fully reveal the metabolic differences in closely-related Lactobacillus species/strains. In this study, we demonstrated the ability of differentiating Lactobacillus acidophilus, Lactobacillus fermentum, Lactobacillus reuteri, and Lactobacillus delbrueckii.

7 (5) Chapter 6 describes efforts to develop an untargeted metabolomics-based approach for characterizing bourbons that matured in barrels for different times by utilizing Fourier-transform ion cyclotron resonance (FT-ICR) and LC-ultrahigh resolution mass spectrometry (LC-Orbitrap MS) techniques.

These studies nicely reveal the power of MS-based metabolomics to analyze complex mixtures. We utilized LC-MS platforms for metabolomic studies for different purposes. An LC- coupled QQQ system was used for all targeted metabolomics research, which focused on revealing the interactions between the host metabolism and external stimuli in in vivo or in vitro systems. A LC-coupled linear ion trap, Orbitrap system was applied for untargeted metabolomics studies, and we hybridized LC-QQQ and LC-Orbitrap systems to establish an approach, which better differentiate some similar bacterial strains. Moreover, FT-ICR and LC-Orbitrap systems were used for food classification and authentication purposes, and in our study, American bourbon whiskies were analyzed to investigate the importance of time of maturation in oak barrels.

8 References [1] Nicholson JK, W., Opinion: Understanding 'global' systems biology: Metabonomics and the continuum of metabolism. Nature Reviews Drug Discovery, 2003. 2(8): p. 668-676. [2] Chen, L., Zhong, F., and Zhu, J., Bridging Targeted and Untargeted Mass Spectrometry- Based Metabolomics via Hybrid Approaches. Metabolites, 2020. 10(9): p. 348. [3] Beger, R.D., Dunn, W., Schmidt, M.A., Gross, S.S., Kirwan, J.A., Cascante, M., Brennan, L., Wishart, D.S., Oresic, M., and Hankemeier, T., Metabolomics enables precision medicine:“a white paper, community perspective”. Metabolomics, 2016. 12(9): p. 149. [4] Emwas, A.-H.M., Salek, R.M., Griffin, J.L., and Merzaban, J., NMR-based metabolomics in human disease diagnosis: applications, limitations, and recommendations. Metabolomics, 2013. 9(5): p. 1048-1072. [5] Gao, B., Bian, X., Mahbub, R., and Lu, K., Sex-Specific Effects of Organophosphate Diazinon on the Gut Microbiome and Its Metabolic Functions. Environ Health Perspect, 2017. 125(2): p. 198-206. [6] Li, R., Yang, J., Saffari, A., Jacobs, J., Baek, K.I., Hough, G., Larauche, M.H., Ma, J., Jen, N., Moussaoui, N., Zhou, B., Kang, H., Reddy, S., Henning, S.M., Campen, M.J., Pisegna, J., Li, Z., Fogelman, A.M., Sioutas, C., Navab, M., and Hsiai, T.K., Ambient Ultrafine Particle Ingestion Alters Gut Microbiota in Association with Increased Atherogenic Lipid Metabolites. Sci Rep, 2017. 7: p. 42906. [7] Spratlin, J.L., Serkova, N.J., and Eckhardt, S.G., Clinical applications of metabolomics in oncology: a review. Clinical cancer research, 2009. 15(2): p. 431-440. [8] Hart, C.D., Tenori, L., Luchinat, C., and Di Leo, A., Metabolomics in breast cancer: current status and perspectives, in Novel biomarkers in the continuum of breast cancer. 2016, Springer. p. 217-234. [9] Monton, M.R.N. and Soga, T., Metabolome analysis by capillary electrophoresis–mass spectrometry. J Chromatograph A, 2007. 1168(1-2): p. 237-246. [10] Lu, W., Bennett, B.D., and Rabinowitz, J.D., Analytical strategies for LC–MS-based targeted metabolomics. J Chromatograph B, 2008. 871(2): p. 236-242. [11] Yang, K., Xu, M., Zhong, F., and Zhu, J.J., Rapid differentiation of Lactobacillus species via metabolic profiling. J Microbiol Methods, 2018. 154: p. 147-155.

9 [12] Xu, M., Zhong, F., and Zhu, J., Evaluating metabolic response to light exposure in Lactobacillus species via targeted metabolic profiling. J Microbiol Methods, 2017. 133: p. 14-19. [13] Mazzini, F.N., Cook, F., Gounarides, J., Marciano, S., Haddad, L., Tamaroff, A.J., Casciato, P., Narvaez, A., Mascardi, M.F., and Anders, M., Plasma and stool metabolomic biomarkers of non-alcoholic fatty liver disease in Argentina. medRxiv, 2020. [14] Cevallos-Cevallos, J.M., Reyes-De-Corcuera, J.I., Etxeberria, E., Danyluk, M.D., and Rodrick, G.E., Metabolomic analysis in food science: a review. Trends Food Sci Tech, 2009. 20(11-12): p. 557-566. [15] Roullier-Gall, C., Signoret, J., Hemmler, D., Witting, M.A., Kanawati, B., Schäfer, B., Gougeon, R.D., and Schmitt-Kopplin, P., Usage of FT-ICR-MS Metabolomics for characterizing the chemical signatures of barrel-aged whisky. Frontiers Chemistry, 2018. 6: p. 29. [16] Kew, W., Mackay, C.L., Goodall, I., Clarke, D.J., and Uhrín, D.a., Complementary Ionization Techniques for the Analysis of Scotch Whisky by High Resolution Mass Spectrometry. Anal Chem, 2018. 90(19): p. 11265-11272. [17] Dudzik, D., Barbas-Bernardos, C., García, A., and Barbas, C., Quality assurance procedures for mass spectrometry untargeted metabolomics. a review. J Pharm Biomed Anal, 2018. 147: p. 149-173. [18] Frederich, M., Pirotte, B., Fillet, M., and De Tullio, P., Metabolomics as a challenging approach for medicinal chemistry and personalized medicine. J Med Chem, 2016. 59(19): p. 8649-8666. [19] Bujak, R., Daghir-Wojtkowiak, E., Kaliszan, R., and Markuszewski, M.J., PLS-based and regularization-based methods for the selection of relevant variables in non-targeted metabolomics data. Frontiers Mol Biosci, 2016. 3: p. 35. [20] Garcia-Perez, I., Posma, J.M., Gibson, R., Chambers, E.S., Hansen, T.H., Vestergaard, H., Hansen, T., Beckmann, M., Pedersen, O., and Elliott, P., Objective assessment of dietary patterns by use of metabolic phenotyping: a randomised, controlled, crossover trial. Lancet Diabetes Endocrinol, 2017. 5(3): p. 184-195.

10 [21] Tzoulaki, I., Ebbels, T.M., Valdes, A., Elliott, P., and Ioannidis, J.P., Design and analysis of metabolomics studies in epidemiologic research: a primer on-omic technologies. Am J Epidemiol, 2014. 180(2): p. 129-139. [22] Orians, K.J. and Boyle, E.A., Determination of picomolar concentrations of titanium, gallium and indium in sea water by inductively coupled plasma mass spectrometry following an 8-hydroxyquinoline chelating resin preconcentration. Anal Chim Acta, 1993. 282(1): p. 63-74. [23] Zhang, A., Sun, H., Wang, P., Han, Y., and Wang, X., Modern analytical techniques in metabolomics analysis. Analyst, 2012. 137(2): p. 293-300. [24] Roux, A., Lison, D., Junot, C., and Heilier, J.-F., Applications of liquid chromatography coupled to mass spectrometry-based metabolomics in clinical chemistry and toxicology: A review. Clin Biochem, 2011. 44(1): p. 119-135. [25] McEwen, C.N. and McKay, R.G., A Combination Atmospheric Pressure Lc/Ms: Gc/Ms Ion Source: Advantages of Dual Ap-Lc/Ms: Gc/Ms Instrumentation. J Am Soc Mass Spectrom, 2005. 16(11): p. 1730-1738. [26] Yang, K., Somogyi, A., Thomas, C., Zhang, H., Cheng, Z., Xu, S., Miller, C., Spivey, D., Blake, C., and Smith, C., Analysis of Barrel-Aged Kentucky Bourbon Whiskey by Ultrahigh Resolution Mass Spectrometry. Food Anal Meth, 2020: p. 1-11. [27] Yang, K., Xu, M., Zhong, F., and Zhu, J.J., Rapid differentiation of Lactobacillus species via metabolic profiling. J Microbiol Meth, 2018. 154: p. 147-155. [28] Zhong, F., Xu, M., Metz, P., Ghosh-Dastidar, P., and Zhu, J., A quantitative metabolomics study of bacterial metabolites in different domains. Anal Chim Acta, 2018. 1037: p. 237- 244. [29] Viant, M.R. and Sommer, U., Mass spectrometry based environmental metabolomics: a primer and review. Metabolomics, 2013. 9(1): p. 144-158. [30] Godzien, J., Ciborowski, M., Martínez-Alcázar, M.a.P., Samczuk, P., Kretowski, A., and Barbas, C., Rapid and reliable identification of phospholipids for untargeted Metabolomics with LC–ESI–QTOF–MS/MS. J Proteome Res, 2015. 14(8): p. 3204-3216. [31] Berthod, A., Friesen, J.B., Inui, T., and Pauli, G.F., Elution− extrusion countercurrent chromatography: theory and concepts in metabolic analysis. Anal Chem, 2007. 79(9): p. 3371-3382.

11 [32] Plumb, R.S., Johnson, K.A., Rainville, P., Shockcor, J.P., Williams, R., Granger, J.H., and Wilson, I.D., The detection of phenotypic differences in the metabolic plasma profile of three strains of Zucker rats at 20 weeks of age using ultra‐performance liquid chromatography/orthogonal acceleration time‐of‐flight mass spectrometry. Rapid Commun Mass Spectrom:, 2006. 20(19): p. 2800-2806. [33] Lai, L., Michopoulos, F., Gika, H., Theodoridis, G., Wilkinson, R.W., Odedra, R., Wingate, J., Bonner, R., Tate, S., and Wilson, I.D., Methodological considerations in the development of HPLC-MS methods for the analysis of rodent plasma for metabonomic studies. Molecular BioSystems, 2009. 6(1): p. 108-120. [34] Colbourne, J.K., Pfrender, M.E., Gilbert, D., Thomas, W.K., Tucker, A., Oakley, T.H., Tokishita, S., Aerts, A., Arnold, G.J., and Basu, M.K., The ecoresponsive genome of Daphnia pulex. Science, 2011. 331(6017): p. 555-561. [35] Gika, H.G., Theodoridis, G.A., and Wilson, I.D., Hydrophilic interaction and reversed‐ phase ultra‐performance liquid chromatography TOF‐MS for metabonomic analysis of Zucker rat urine. J Separation Sci, 2008. 31(9): p. 1598-1608. [36] de Graaf, E.L., Altelaar, A.M., van Breukelen, B., Mohammed, S., and Heck, A.J., Improving SRM assay development: a global comparison between triple quadrupole, ion trap, and higher energy CID peptide fragmentation spectra. J Proteome Res, 2011. 10(9): p. 4334-4341. [37] Han, J., Datla, R., Chan, S., and Borchers, C.H., Mass spectrometry-based technologies for high-throughput metabolomics. Bioanalysis, 2009. 1(9): p. 1665-1684. [38] Wei, R., Li, G., and Seymour, A.B., High-throughput and multiplexed LC/MS/MRM method for targeted metabolomics. Anal Chem, 2010. 82(13): p. 5527-5533. [39] Brown, S.C., Kruppa, G., and Dasseux, J.L., Metabolomics applications of FT‐ICR mass spectrometry. Mass Spectrometry Rev, 2005. 24(2): p. 223-231. [40] Junot, C., Madalinski, G., Tabet, J.-C., and Ezan, E., Fourier transform mass spectrometry for metabolome analysis. Analyst, 2010. 135(9): p. 2203-2219. [41] Kind, T. and Fiehn, O., Advances in structure elucidation of small molecules using mass spectrometry. Bioanal Rev, 2010. 2(1-4): p. 23-60. [42] Sekirov, I., Russell, S.L., Antunes, L.C.M., and Finlay, B.B., Gut microbiota in health and disease. Physiolog Rev, 2010. 90(3): p. 859-904.

12 [43] Torrazza, R.M. and Neu, J., The developing intestinal microbiome and its relationship to health and disease in the neonate. J Perinatol, 2011. 31(1): p. S29-S34. [44] Nicholson, J.K., Holmes, E., Kinross, J., Burcelin, R., Gibson, G., Jia, W., and Pettersson, S., Host-gut microbiota metabolic interactions. Science, 2012. 336(6086): p. 1262-1267. [45] Kim, K.-A., Gu, W., Lee, I.-A., Joh, E.-H., and Kim, D.-H., High fat diet-induced gut microbiota exacerbates inflammation and obesity in mice via the TLR4 signaling pathway. PLoS One, 2012. 7(10): p. e47713. [46] Hooper, L.V. and Gordon, J.I., Commensal host-bacterial relationships in the gut. Science, 2001. 292(5519): p. 1115-1118. [47] Bäckhed, F., Ding, H., Wang, T., Hooper, L.V., Koh, G.Y., Nagy, A., Semenkovich, C.F., and Gordon, J.I., The gut microbiota as an environmental factor that regulates fat storage. Proc Nat Acad Sci, 2004. 101(44): p. 15718-15723. [48] Kau, A.L., Ahern, P.P., Griffin, N.W., Goodman, A.L., and Gordon, J.I., Human nutrition, the gut microbiome and the immune system. Nature, 2011. 474(7351): p. 327-336. [49] Vrieze, A., Holleman, F., Zoetendal, E., De Vos, W., Hoekstra, J., and Nieuwdorp, M., The environment within: how gut microbiota may influence metabolism and body composition. Diabetologia, 2010. 53(4): p. 606-613. [50] Putignani, L., Del Chierico, F., Vernocchi, P., Cicala, M., Cucchiara, S., Dallapiccola, B., and Group, D.S., Gut microbiota dysbiosis as risk and premorbid factors of IBD and IBS along the childhood–adulthood transition. Inflammatory Bowel Diseases, 2016. 22(2): p. 487-504. [51] Holmes, E., Li, J.V., Athanasiou, T., Ashrafian, H., and Nicholson, J.K., Understanding the role of gut microbiome–host metabolic signal disruption in health and disease. Trends Microbiol, 2011. 19(7): p. 349-359. [52] Vernocchi, P., Del Chierico, F., and Putignani, L., Gut microbiota profiling: metabolomics based approach to unravel compounds affecting human health. Frontiers Microbiol, 2016. 7: p. 1144. [53] Moco, S., Collino, S., Rezzi, S., and Martin, F.-P.J., Metabolomics perspectives in pediatric research. Pediatric Res, 2013. 73(2): p. 570-576.

13 [54] Li, H., Xu, M., and Zhu, J.J., Headspace Gas Monitoring of Gut Microbiota Using Targeted and Globally Optimized Targeted Secondary Electrospray Ionization Mass Spectrometry. Anal Chem, 2018. 91(1): p. 854-863. [55] Del Chierico, F., Vernocchi, P., Bonizzi, L., Carsetti, R., Castellazzi, A.M., Dallapiccola, B., de Vos, W., Guerzoni, M.E., Manco, M., and Marseglia, G.L., Early-life gut microbiota under physiological and pathological conditions: the central role of combined meta-omics- based approaches. J Proteomics, 2012. 75(15): p. 4580-4587. [56] Ashktorab, H., Kupfer, S.S., Brim, H., and Carethers, J.M., Racial disparity in gastrointestinal cancer risk. Gastroenterol, 2017. 153(4): p. 910-923. [57] Louis, P., Hold, G.L., and Flint, H.J., The gut microbiota, bacterial metabolites and colorectal cancer. Nature Rev Microbiol, 2014. 12(10): p. 661-672. [58] Hirayama, A., Kami, K., Sugimoto, M., Sugawara, M., Toki, N., Onozuka, H., Kinoshita, T., Saito, N., Ochiai, A., and Tomita, M., Quantitative metabolome profiling of colon and stomach cancer microenvironment by capillary electrophoresis time-of-flight mass spectrometry. Cancer Res, 2009. 69(11): p. 4918-4925. [59] Yachida, S., Mizutani, S., Shiroma, H., Shiba, S., Nakajima, T., Sakamoto, T., Watanabe, H., Masuda, K., Nishimoto, Y., and Kubo, M., Metagenomic and metabolomic analyses reveal distinct stage-specific phenotypes of the gut microbiota in colorectal cancer. Nature Medicine, 2019. 25(6): p. 968-976. [60] Granato, D., Putnik, P., Kovačević, D.B., Santos, J.S., Calado, V., Rocha, R.S., Cruz, A.G.D., Jarvis, B., Rodionova, O.Y., and Pomerantsev, A., Trends in chemometrics: Food authentication, microbiology, and effects of processing. Comprehensive Rev Food Sci Food Safety, 2018. 17(3): p. 663-677. [61] Handford, C.E., Campbell, K., and Elliott, C.T., Impacts of milk fraud on food safety and nutrition with special emphasis on developing countries. Comprehensive Rev Food Sci Food Safety, 2016. 15(1): p. 130-142. [62] Ehmke, M.D., Bonanno, A., Boys, K., and Smith, T.G., Food fraud: economic insights into the dark side of incentives. Australian J Agric Res Econ, 2019. 63(4): p. 685-700. [63] Cubero-Leon, E., Peñalver, R., and Maquet, A., Review on metabolomics for food authentication. Food Res Internat, 2014. 60: p. 95-107.

14 [64] Endo, Y., Tagiri‐Endo, M., and Kimura, K., Rapid determination of iodine value and saponification value of fish oils by near‐infrared spectroscopy. J Food Sci, 2005. 70(2): p. C127-C131. [65] Belmonte-Sánchez, J.R.l., Romero-González, R., Arrebola, F.J., Vidal, J.L.M.n., and Garrido Frenich, A., An Innovative Metabolomic Approach for Golden Rum Classification Combining Ultrahigh-Performance Liquid Chromatography–Orbitrap Mass Spectrometry and Chemometric Strategies. J Agric Food Chem, 2019. 67(4): p. 1302-1311. [66] Castro-Puyana, M. and Herrero, M., Metabolomics approaches based on mass spectrometry for food safety, quality and traceability. Trends Anal Chem, 2013. 52: p. 74- 87. [67] Stewart, D., McDougall, G.J., Sungurtas, J., Verrall, S., Graham, J., and Martinussen, I., Metabolomic approach to identifying bioactive compounds in berries: advances toward fruit nutritional enhancement. Mol Nutrit Food Res, 2007. 51(6): p. 645-651. [68] Danezis, G.P., Tsagkaris, A.S., Camin, F., Brusic, V., and Georgiou, C.A., Food authentication: Techniques, trends & emerging approaches. Trends Anal Chem, 2016. 85: p. 123-132.

15 CHAPTER 2 Evaluating the Impact of Four Major Nutrients on Gut Microbial Metabolism by a Targeted Metabolomics Approach

Kundi Yang1, Mengyang Xu2 and Jiangjiang Zhu*3, 4

1. Department of Chemistry & Biochemistry, Miami University, Oxford, OH, USA, 45056 2. Cell, Molecular & Structural Biology Program, Miami University, Oxford, OH, USA, 45056 3. Department of Human Sciences, The Ohio State University, Columbus, OH, USA, 43210 4. James Comprehensive Cancer Center, The Ohio State University, Columbus, OH, USA 43210

Contributions to the chapter. Kundi Yang and Mengyang Xu conducted the bacterial experiments and extracted the metabolites. The data analyses were performed by Kundi Yang. This chapter was written by Kundi Yang. The written document was edited by Jiangjiang Zhu and Michael Crowder.

16 Abstract Gut microbiome plays fundamental roles in host physiology, and gut microbial metabolism is important to the host-microbiome homeostasis. As major contributors to gut microbial metabolism, the medium nutritional components are essential to in vitro gut microbiome growths, and four nutrients, namely inorganic salts, bile salts, short chain fatty acids (SCFAs), and mucin, have gained particular attention because their significant variation found in different growth environments and their ability to modulate the gut microbial population and functions. However, a systematic study is lacking to evaluate the effects of these four nutrients on gut microbiome in terms of their impact to the microbial metabolic profiles. To fill this gap of knowledge, we applied a mass spectrometry-based targeted metabolomic approach to study the regulation effects of these four medium components on in vitro cultured gut microbiota. Our results show inorganic salts and mucin had the greatest impact on the gut microbiome metabolic profile compared to the other components studied, with gut microbial cultures grown with low concentration inorganic salts and mucin-supplemented medium demonstrating greater numbers of metabolites detected. We also applied metabolic pathway impact analysis that revealed several significantly-impacted metabolic pathways during the comparison of different medium supplements, which could further assist our understanding of the overall impacts of certain critical nutrients on gut microbial metabolism. In summary, this pilot study serves as a first attempt to evaluate individual nutritional components in their contribution to gut microbial metabolic functions.

Key words: Gut microbiome, metabolomics, targeted mass spectrometry, medium nutrition.

17 2.1 Introduction The human gut microbiome has been linked to both human health and disease.[1-3] As a consequence, targeting the gut microbiota using drug therapeutics has become popular as well as increased interest in understanding how gut microbiome responds to the therapies.[4, 5] In vivo animal and human models have been used to study the function and dynamics of host-associated microbiomes.[6-8] However, logistical constraints limited the use of these models, and thus, compounds were generally tested using cultured bacterium and microbiome.[9, 10] Instead of culturing single bacteria in vitro, culturing the whole gut microbiome can reveal the direct interaction between gut microbiome and nutritional components, and it is an approach that helps to provide invaluable evidence to guide further nutritional studies that probe nutrient-gut microbe interactions. Previous studies suggested that the interaction between the microbiota and the host are often mediated by bacterial metabolites, such as amino acids, short chain fatty acids (SCFAs), and vitamins.[6, 11] In order to investigate the metabolic capacity of gut microbiota and the metabolic interactions among different microbiome communities, an in vitro culture-based model system is essential for preliminary studies. However, the complex composition of culture medium has been rarely defined.[12] Consequently, a severe limitation on mechanistic investigations of community functions, for example, identifying cross-fed metabolites or correlating functional metabolites to their producers, is obstructing the progress in studying the gut microbiome.[13] Nutritional components in the culture medium are key determinants for the performance of in vitro culture systems of gut microbiomes.[5] Gifu Anaerobic medium (GAM) is recommended as a general culture medium for cultivation and isolation of anaerobic bacteria and to test their susceptibilities to antibiotics other than sulfa drugs. Composition of GAM broth includes yeast, beef and liver extracts, peptone, inorganic salts, and bile salts. Previous studies performed by other researchers have demonstrated how modifications of nutrient content can alter the microbial communities and increase the diversity of the gut bacterial community.[14, 15] However, there is still a lack of a systematic evaluation of how some vital nutrients impact the metabolome of the gut microbiome. Four nutrients, inorganic salts, bile salts, SCFAs, and mucin, have caught particularly attention since they can vary significantly in different growth environments.[5] Therefore, it can be helpful to understand the holistic impact of these nutrients on the gut

18 microbiome and their metabolites, and the gained knowledge could be used as scientific evidence for choosing the optimal medium compositions for culturing gut microbiota in vitro. In this study, we designed experiments to test the impact of four medium components (inorganic salts, mucin, bile salts, and SCFAs) on the metabolic profiles of cultured human gut microbiome by utilizing a targeted metabolomic approach. High-performance liquid chromatography-triple quadrupole mass spectrometry (HPLC-QQQ-MS) was used to analyze the metabolic differences of gut microbiome incubated with different medium components. The goal of this study was to reveal the influence of several essential nutritional components on the metabolome of the complex gut microbiome cultured anaerobically in vitro. The discovery work performed can be used to further improve our understanding of gut microbial metabolism and their potential impact to human host.

2.2 Materials and Methods

Human colon microbiota Human colon microbiota were cultured from the fecal samples of two healthy volunteers by following a gut bacterial isolation and enrichment procedure published earlier.[16] The fecal content was vortexed and suspended in phosphate buffered saline (PBS) pre-reduced with 0.1% cysteine. The diluted suspension was plated on Gifu Anaerobic (GAM) Agar plates (HiMedia Laboratories LLC, West Chester, PA, USA) for 48 h at 37 °C in an anaerobic environment inside a type A vinyl anaerobic chamber (COY lab, Grass Lake, MI, USA). A sterilized wire loop was used to inoculate multiple colonies from the agar plate to GAM broth and incubated for 48 hours under the same conditions as stated above.

Experimental design In this study, experiments were designed for the evaluation of the influence of different medium components towards gut microbiome when cultured in vitro.[5] Briefly, all experimental groups were assigned to one or multiple supplements of the four factors (inorganic salts, mucin, bile salts, and short chain fatty acids) during the gut microbial growth in GAM to evaluate one- factor and multi-factor interactions. A group of cultures grown in base GAM medium were included and served as a baseline comparison. Gut microbiomes were donated from volunteers

19 and were cultured for 48 hours under anaerobic conditions in four replicates in each treatment group, to give a total of 64 cultured samples in this study. Optical density (OD) values of each cultured sample were measured at 595 nm. All samples were extracted by a cold methanol metabolite extraction method (see below) and analyzed by LC-MS/MS system for target metabolites analysis. The metabolite levels were then normalized by the OD values of each biological replicate, log-transferred, and then subjected to autoscaling to minimize possible bias from skewed distribution from the raw data.

Culture medium preparation Sixteen varieties of freshly-prepared culture media were included in this study and are listed in Table 2.1. All media were made and modified from GAM broth, which contains 2.5 g L- 1 proteose peptone, 3.4 g L-1 digested serum, 1.25 g L-1 yeast extract, 0.55 g L-1 beef extract, 0.75 -1 -1 -1 -1 -1 g L dextrose, 0.625 g L KH2PO4, 0.75 g L NaCl, 1.25 g L starch, 0.075 g L L-cysteine hydrochloride, 0.075 g L-1 sodium thioglycolate, 2 mL L-1 Tween 80, 5 mg L-1 Hemin, and 10 μL L-1 vitamin K1. The low concentration inorganic salts (LCIS) groups were prepared referencing to basal culture medium (BCM) with modification, which were prepared by adding 0.08 g L-1 -1 -1 -1 K2HPO4, 0.01 g L MgSO4, 0.01 g L CaCl2 and 2 g L NaHCO3 to the GAM medium;[5] the -1 high concentration inorganic salts (HCIS) groups were prepared by adding 0.8 g L K2HPO4, 0.1 -1 -1 -1 - g L MgSO4, 0.1 g L CaCl2, and 20 g L NaHCO3 to the GAM medium.[17] Bile salts (0.5 g L 1) were supplemented to media 1, 2, 6, and 8 in both LCIS and HCIS groups; 4 g L-1 mucin was supplemented to media 1, 2, 3, and 7 in both LCIS and HCIS groups; and media 1, 5, 6, and 7 in both LCIS and HCIS groups were supplemented with a SCFA mixture (Sigma-Aldrich, St. Louis, MO, USA) to reach the final concentration of 1 mM.[12]

Metabolites extraction Metabolite extractions of intracellular contents of bacterial cells were prepared using a cold methanol method following the sampling protocol from our previous work[18]. Briefly, 1 mL of bacterial culture was subjected to two rounds of rapid centrifugation and a PBS wash. A portion of 250 μL methanol plus 50 μL of C13N15 labeled internal standards mixture was added to the cell pellet in the tube, and the samples were vortexed vigorously for 2 min. After incubation at -20 °C for 20 mins, a 150 μL portion of the supernatant was collected and dried by using a Speedvac

20 system and centrifugation. Samples were then reconstituted by using 250 μL of a 1:1 mixture of acetonitrile and Ultrapure water and loaded into liquid chromatography vials for analysis.

Metabolic profiling Both positive and negative mode detections were performed on a Thermo Scientific TSQ Quantiva triple quadrupole mass spectrometer equipped with an electrospray ionization source, which was coupled to a Thermo Scientific Ultimate 3000 high performance liquid chromatography (HPLC) system equipped with a 2.1×150 mm, 2.5 µm hydrophilic interaction chromatography (HILIC) amide column (Waters Corporation, Milford, MA, USA). Metabolite separation was completed on an HPLC by eluting the extracted bacterial intracellular samples through the column. The reconstituted samples were gradient-eluted at 0.300 mL/min using solvents A (5 mM ammonium acetate in 90% water / 10% acetonitrile + 0.2% acetic acid) and B (5 mM ammonium acetate in 90% acetonitrile / 10% water + 0.2% acetic acid). The auto-sampler temperature was kept at 4 °C, the column compartment was kept at 40 °C, and the separation time for each sample was 20 min. Retention time and selected reaction monitoring (SRM) transition of targeted metabolites were established by running pure standards (purchased from Sigma, Saint Louis, MO, USA and IROA Technology, Boston, MA) and collecting the tandem mass spectrum (MS/MS), so the orthogonal information of retention time and two pairs of SRM transitions can be used to 13 15 confidently detect and identify targeted compounds. Stable C and N universally-labeled amino acid mix (Cambridge Isotope, Tewksbury, MA, USA) was used as internal standards for quality control purposes. Pooled quality control samples were also tested in between every ten samples to monitor the instrument stability.

Statistical analyses The Quanbrowser module of Xcalibur 4.0 was used to manually process targeted metabolite profiling data. Acquired peak intensities were normalized to corresponding optical density values at 635 nm wavelength of the gut microbial culture. MetaboAnalyst 4.0 was used for statistical analyses (http://www.metaboanalyst.ca/). Peak intensities were subjected to a log transformation and auto-scaling to achieve an approximately normalized distribution. ANOVA module, a principal component analysis module, and heatmap module, as well as pathway analysis module, were used for data analysis and visualization.

2.3 Results and Discussion

Targeted metabolic profiling differentiates the gut microbial growth in LCIS and HCIS Detailed experiment design of the multiple medium components used to supplement with the base medium are shown in Table 2.1.

Table 2.1 Experiment design and group information. In groups of low (L) and high (H) concentration inorganic salts, groups 1, 2, 3, and 7 were treated with mucin; groups 1, 2, 6, and 8 were treated with bile salt and groups 1, 5, 6, and 7 were treated with short-chain fatty acid mix.

The OD results showed that the gut microbiomes treated with LCIS grew better than those treated with HCIS, as the mean OD value of L1, L2, L3, and L7 were 0.642, 0.647, 0.742 and 0.671, respectively. The ODs of the remaining groups (L4-6 and L8) were all above 0.3. However, gut microbiomes treated with HCIS showed relatively lower OD values, even the top three OD values (0.456, 0.386 and 0.340 correspond to H1, H8 and H2, respectively) were significantly lower than those with the same nutrition combinations in the LCIS groups. The workflow is shown in Figure 2.1. A total of 122 polar metabolites were detected based on our targeted method with 84 out of 122 found in high concentration inorganic salts groups and 112 out of 122 can be found in low concentration inorganic salts groups. We first evaluated the average coefficient of variation (CV) for all of the detected metabolites, and an average CV value of 14.87% indicated a good reproducibility of our experiments. A list of detected metabolites in this study can be found in Supplementary Table 2.1 (Table S2.1).

Figure 2.1 Workflow of this study. The multi-components experimental design was used to evaluate the gut microbiome cultured with base media enriched with four nutritional components: (1) inorganic salts, (2) bile salts, (3) SCFAs and (4) mucin. Post inoculation, the intracellular metabolites of each sample were extracted and analyzed by a HPLC-QQQ-MS metabolomicplatform and followed by multivariate statistical analyses.

We then applied statistical analyses on all samples to tease out major and minor nutritional factors that can shape the overall metabolic profiles of cultured gut microbes. Based on the ANOVA test, 115 out of the 122 detected metabolites were significantly-different among the tested groups. Figure 2.2a shows the PCA plot of all 16 test groups, with the x-axis and y-axis st nd representing the 1 and 2 principal components (PC1 and PC2) that contribute most to the characteristic major metabolic impacts of the tested medium components, which resulted in 71.1% variation summarized by PC1 and 8.3% for PC2. The colors represent different group sets, while the ovals represent the 95% confidential level. Figure 2.2b demonstrates the metabolic profile of all sample groups involved in this study. Each column represents one biological replicate of gut microbial samples with corresponding nutrition supplementation, while each row represents the individual metabolite. The color indicates the concentration level of each metabolite, as dark red

23 means the highest abundance, and dark blue indicates the lowest abundance of a metabolite in the sample. The heat map analysis reveals a significantly-different metabolic profile of gut microbiome when treated with different concentrations of inorganic salts. As seen in Figure 2.2a, two major clusters of samples, determined by metabolic profiles generated by LCIS and HCIS, were clearly separated. The effects of the other factors involved in this study, for instance, mucin, bile salts, and short chain fatty acids, were less dominant in this PCA because of the tremendous influence by inorganic salts. This overviewed metabolic profile indicated a greater impact of medium inorganic salts on the gut microbiota in terms of metabolite abundance than other factors tested in this study. To analyze the detailed metabolite features of inorganic salts assignments, we conducted a comparison between L-4 and H-4 groups, which were only assigned differently in the concentrations of inorganic salts. As seen in Figure S1, the LCIS groups are significantly-different from the HCIS groups, and a detailed list of metabolites that have significant differences (p-value <0.01 and fold change (FC) >2) in comparison with LCIS and HCIS are shown in Table S2. Although the widely-held notion indicated that most bacteria need inorganic salts for good growth,[17, 19, 20] low-concentration inorganic salts results in more metabolites detected than high-concentration in this study. As our results show, medium with low inorganic salts concentration can support gut bacteria growth which in turn resulted in a significantly-larger number (23% increase) of metabolites that can be detected above the detection threshold. On the other hand, a previous study pointed out that inorganic salts content may induce an environment of oxidative stress because low concentrations of such substrate in medium causes an inadequate redox potential in the medium.[5] For example, metabolites such as 5-hydroxymethyluracil and indole-3-acetate, detected in our study can barely be detected with a higher inorganic salt concentration medium (Figure S1b and S1c). 5-Hydroxymethyluracil is a significant product of radiation damage or the chemical oxidants generated by activated polymorphonuclear neutrophils,[21] while indole-3-acetate has been reported as a substrate that can induce the lipid peroxidation.[22] These results make it difficult to ascertain whether inorganic salts in medium is good for gut microbiota. Study of the specific requirements of inorganic salts on the gut bacterial strains may be needed for better understanding of the roles that inorganic salts play in different gut bacteria growth environment.

24 Targeted metabolic profiling differentiates the gut microbial metabolic profiles with other nutrients in a defined inorganic salts level In order to investigate the metabolic impact of nutrient factors other than inorganic salts on the gut microbiome, we conducted statistical analyses in constant inorganic salts content (either low or high concentration inorganic salts). As seen in Figures 2.3a and 2.3b, the PCA analyses were applied separately for the two defined inorganic salts concentrations, and two major clusters were separated in each set. Sample group L-1, L-2, L-3, and L-7, which assigned with mucin in the medium were grouped together based on their metabolic profiles, while L-4, L-5, L-6, and L- 8, the samples without mucin in the medium, were grouped. The same pattern in terms of metabolic profile can be found in HCIS set. The metabolic profile heatmap of these two sets of the experiment can be seen in Figure S2. The results indicated a greater impact of mucin in medium than bile salts and short chain fatty acids. To analyze the detailed metabolite features of mucin assignments, we conducted a comparison between L-3 and L-4 groups, which were only different in the mucin content with a low inorganic concentration to minimize the effect of inorganic salts towards the resulting metabolic profile (Figure S3a). A detailed list of metabolites that have significant differences (p‐value <0.01 and fold change (FC) >2) in comparison of groups with and without mucin in the medium is shown in Table S3. As seen in Figure S3a, mucin-enriched group L-3 is significantly-different from the L-4 group, which contained no mucin. Shown in Figure S3b and S3c, 5-hydroxymethyluracil, and indole-3-acetaldehyde, which is the precursor of indole-3-acetate and can be oxidized into indole-3-acetate via tryptophan metabolism, are significantly-lower in their abundance in the mucin-assigned groups. This result indicates a less oxidative stress environment for gut microbiome with the supplementation of mucin in the medium. Mucins contain abundant carbohydrate side chains,[23] and three of the common carbohydrates galactose, mannose, and N-acetylglucosamine, are metabolized by the gut microbiome.[24-26] Lack of mucin in the medium may lead to carbon-starvation of gut microbiome when cultured in vitro. Our results suggested the mucin-enriched environment is preferred for gut microbiome culturing. Samples in low concentration inorganic salts medium and without mucin were analyzed separately to investigate the other two nutritional factors (i.e., bile salts and SCFAs), which included group L-4, L-5, L-6, and L-8, and the metabolic profile heatmap can be seen in Figure S4 (top 50 metabolites). All four groups were clustered in their own patterns. However, 6/50 metabolites were significantly-increased in SCFAs groups, while 31/50 metabolites were

25 decreased significantly compared to group L-4. On the other hand, bile salts caused an increase in 13/50 metabolites and a decrease in 11/50 metabolites when compared to group L-4, which contained neither bile salts nor SCFAs. The rest of the metabolites remained unchanged. It has been reported by Tramontano and her colleagues in 2018 that medium without SCFAs can be a benefit for a large number of gut bacterial strains.[12] They reported such nutrient inhibitory effects may be caused by intracellular accumulation of toxic intermediates or pH imbalance.[12] In our study, the results indicated the bile salts and SCFAs component in the medium have minor impacts towards the metabolome of gut microbiome when culturing in vitro compared to the major metabolic impacts generated by inorganic salts and mucin. Our results stand with Tramontano’s conclusions but are interpreted in a metabolomic way. However, further investigation and detailed comparison will be needed to make a clear conclusion of how bile salts and SCFAs in medium impact the growth environment of gut microbiota.

a b

Figure 2.2 The overview of metabolic profiles of all studied samples with the various nutritional enrichments. (a) Principle component analysis (PCA) with all samples in this study, as highlighted by the two major ovals, the low vs. high inorganic salts contents executed as a major differential factor in separating the LCIS vs. HCIS samples. The shadow indicated 95% confidence region for the classification. (b) Heatmap of combined metabolic profiling. Each column represents one biological sample; each row represents one targeted metabolite. The detailed list of metabolites can be seen in Table S1 (supporting information). Color represents relative metabolite concentration in comparison of the same metabolite from different samples after data normalization

27 a b Figure 2.3 Detailed PCA score plots of metabolomics data from low vs. high concentration inorganic salts supplemented gut microbial cultures. Shadow indicates 95% confidence region for the classification. (a) Samples from low concentration inorganic salts groups. (b) Samples from high concentration inorganic salts groups.

Metabolic pathway impact analyses Furthermore, we conducted metabolic pathway impact analyses to categorize individual metabolites into the context of connected metabolic pathway networks. All detected metabolites were included in the metabolic pathway analysis, so the broader coverage of extensive metabolic networks could be achieved. The pathway analyses of 112 metabolites detected in ICIS groups and 84 metabolites detected in HCIS groups are shown in Figure 2.4a and 2.4b. The major metabolic pathways of the different concentration-dependent inorganic salts impacted the gut microbiome metabolic profile, where the x-axis is the metabolic pathway impact value (from pathway topology analysis, which is calculated as the sum of the importance measures of the matched metabolites normalized by the sum of the importance measures of all metabolites in each pathway), and the y- axis is the statistical-significance (represented by p-value) of the impacted pathways between culture in the high or low abundance of inorganic salts. The dot size corresponds to the x-axis value, and the dot color corresponds to the y-axis value. Sévin and colleagues reported that metabolite groups, such as amino acid and various CoA thioesters, were significantly-affected by inorganic salts in culture medium.[27] Our findings are consistent with Sévin’s study. Metabolic pathways, such as alanine, aspartate, and glutamate, arginine and proline metabolism, biotin metabolism, citrate cycle (TCA cycle), and tryptophan metabolism, among others, were identified as significantly-impacted metabolic pathways. It can be observed that in Figure 2.4 the biotin metabolism and tryptophan metabolic pathways from the two groups tend to have similar pathway impact factors and p-values, while the other three labeled pathways showed different responses to LCIS versus HCIS treatments. Important metabolites among these significantly-impacted metabolic pathways were compared between groups. As seen in Figure 2.4c, the levels of aspartic acid, oxaloacetic acid, pyruvate, proline, and glutamic acid were found significantly-altered when the media was treated with LCIS compared to HCIS.

a A b

B D B

A C E D

E C

Aspartic acid Oxaloacetic acid Pyruvate Proline Glutamic acid c (p=8.6E-13) (p=6.9E-13) (p=8.7E-11) (p=1.4E-4) (p=8.3E-5)

Figure 2.4 Major metabolic pathways of the two studied nutrient factors that impacted metabolic pathways during this study. (a) Low concentration inorganic salts groups. (b) High concentration inorganic salts groups. The x-axis is the metabolic pathway impact value, and the y-axis is the statistical significance (represented by p-value) of the impacted pathways between culture in the high or low abundance of inorganic salts. The dot sizes correspond to the x-axis values, and the dot colors correspond to the y-axis values. Metabolic pathways were labeled alphabet letters as: A. Alanine, aspartate, and glutamate metabolism; B. Arginine and proline metabolism; C. Biotin metabolism; D. Citrate cycle (TCA cycle) and E. Tryptophan metabolism. (c) The metabolites were selected based on pathway analysis data using the comparison between L-4 and H-4 groups.

We also conducted metabolic pathway impact analyses to investigate the pathway impact induced by mucin content in a medium because it appeared to be a second factor that impacts the gut microbiome significantly. All sample groups from the LCIS medium-fed bacteria were used for comparisons and pathway analysis in order to minimize the effects that the inorganic salts had on metabolic profiles (L3 vs L4). As seen in Figure 2.5, alanine, aspartate, and glutamate metabolism, arginine and proline metabolism, TCA cycle, and tryptophan metabolism are still the most impacted metabolic pathways. Aspartic acid, glutamic acid, and proline are the three important metabolites among these highly-impacted pathways, and all of them decreased in abundance significantly when the media was treated with mucin. Such phenomena may be due to biosynthetic regulation by gut bacteria, in which resources are redirected towards stress responses or these amino acids are used as precursors for other processes that involve in osmoprotection.[27]

31 Aspartic acid (p=6.0E-4) B

Glutamic acid (P=4.1E-5)

D E Proline C (p=7.8E-4)

Figure 2.5 Pathway analysis of mucin group (L-3). The x-axis is the metabolic pathway impact value (from pathway topology analysis, which is calculated as the sum of the importance measures of the matched metabolites normalized by the sum of the importance measures of all metabolites in each pathway), and the y-axis is the statistical significance (represented by p-value) of the impacted pathways between culture with or without the mucin. The dot sizes correspond to the x- axis values, and the dot colors correspond to the y-axis values. Metabolic pathways were labeled alphabet letters as: A. Alanine, aspartate, and glutamate metabolism; B. Arginine and proline metabolism; C. Biotin metabolism; D. Citrate cycle (TCA cycle) and E. Tryptophan metabolism. The metabolites on the right were selected based on pathway analysis data using the comparison between L-3 and L-4 groups.

2.4 Conclusions In this study, we systematically investigated the effects of multiple medium components to the metabolome of human gut microbiome using a targeted metabolic profiling approach. We identified the specific shifts in metabolic profiles in response to inorganic salts, mucin, bile salts, and short chain fatty acids supplementation to GAM medium growth of human-derived gut microbial population. While this work is only a proof-of-concept study, we have demonstrated that inorganic salts and mucin are indeed major nutritional factors to be considered more significant than bile salts and SCFAs for future metabolic studies of the gut microbiome, and a mucin-rich environment is optimal for reducing environmental oxidative stress to in vitro gut microbiome cultures.

ASSOCIATED CONTENT Supporting Information The supporting information is available free of charge at ACS website https://urldefense.com/v3/__http://pubs.acs.org__;!!KGKeukY!kgM6T_GSIhgVi_Xk2RfGiL9N BEo8LbnNYs6FThi7KvN4AYT2V6WjtVdupUpCEPAVVIU$ This material is available in Appendix A.

Conflict of Interest The authors declare no competing financial interest.

References [1] Hooper, L.V., Littman, D.R., and Macpherson, A.J., Interactions between the microbiota and the immune system. Science, 2012. 336(6086): p. 1268-1273. [2] Sharon, G., Sampson, T.R., Geschwind, D.H., and Mazmanian, S.K., The central nervous system and the gut microbiome. Cell, 2016. 167(4): p. 915-932. [3] Zeevi, D., Korem, T., Zmora, N., Israeli, D., Rothschild, D., Weinberger, A., Ben-Yacov, O., Lador, D., Avnit-Sagi, T., and Lotan-Pompan, M., Personalized nutrition by prediction of glycemic responses. Cell, 2015. 163(5): p. 1079-1094. [4] Jia, W., Li, H., Zhao, L., and Nicholson, J.K., Gut microbiota: a potential new territory for drug targeting. Nature Rev Drug Dis, 2008. 7(2): p. 123. [5] Li, L., Zhang, X., Ning, Z., Mayne, J., Moore, J.I., Butcher, J., Chiang, C.-K., Mack, D., Stintzi, A., and Figeys, D., Evaluating in Vitro Culture Medium of Gut Microbiome with Orthogonal Experimental Design and a Metaproteomics Approach. J Proteome Res, 2017. 17(1): p. 154-163. [6] Bäumler, A.J. and Sperandio, V., Interactions between the microbiota and pathogenic bacteria in the gut. Nature, 2016. 535(7610): p. 85. [7] Zhang, X., Zhao, Y., Xu, J., Xue, Z., Zhang, M., Pang, X., Zhang, X., and Zhao, L., Modulation of gut microbiota by berberine and metformin during the treatment of high-fat diet-induced obesity in rats. Sci Rep, 2015. 5: p. 14405. [8] Kang, D.-W., DiBaise, J.K., Ilhan, Z.E., Crowell, M.D., Rideout, J.R., Caporaso, J.G., Rittmann, B.E., and Krajmalnik-Brown, R., Gut microbial and short-chain fatty acid profiles in adults with chronic constipation before and after treatment with lubiprostone. Anaerobe, 2015. 33: p. 33-41. [9] Zhang, X., Ning, Z., Mayne, J., Deeke, S.A., Li, J., Starr, A.E., Chen, R., Singleton, R., Butcher, J., and Mack, D.R.J., In vitro metabolic labeling of intestinal microbiota for quantitative metaproteomics. Anal Chem, 2016. 88(12): p. 6120-6125. [10] Haiser, H.J., Gootenberg, D.B., Chatman, K., Sirasani, G., Balskus, E.P., and Turnbaugh, P.J., Predicting and manipulating cardiac drug inactivation by the human gut bacterium Eggerthella lenta. Science, 2013. 341(6143): p. 295-298.

34 [11] O’Mahony, S.M., Clarke, G., Borre, Y., Dinan, T., and Cryan, J., Serotonin, tryptophan metabolism and the brain-gut-microbiome axis. Behavioural Brain Res, 2015. 277: p. 32- 48. [12] Tramontano, M., Andrejev, S., Pruteanu, M., Klünemann, M., Kuhn, M., Galardini, M., Jouhten, P., Zelezniak, A., Zeller, G., and Bork, P., Nutritional preferences of human gut bacteria reveal their metabolic idiosyncrasies. Nature Microbiol, 2018. 3(4): p. 514. [13] Ponomarova, O. and Patil, K.R., Metabolic interactions in microbial communities: untangling the Gordian knot. Current Opinion In Microbiology, 2015. 27: p. 37-44. [14] McDonald, J.A., Fuentes, S., Schroeter, K., Heikamp-deJong, I., Khursigara, C.M., de Vos, W.M., and Allen-Vercoe, E., Simulating distal gut mucosal and luminal communities using packed-column biofilm reactors and an in vitro chemostat model. J Microbiol Methods, 2015. 108: p. 36-44. [15] McDonald, J.A., Schroeter, K., Fuentes, S., Heikamp-deJong, I., Khursigara, C.M., de Vos, W.M., and Allen-Vercoe, E., Evaluation of microbial community reproducibility, stability and composition in a human distal gut chemostat model. J Microbiol Methods, 2013. 95(2): p. 167-174. [16] Goodman, A.L., Kallstrom, G., Faith, J.J., Reyes, A., Moore, A., Dantas, G., and Gordon, J., Extensive personal human gut microbiota culture collections characterized and manipulated in gnotobiotic mice. PNAS, 2011: p. 201102938. [17] Duncan, S.H., Hold, G.L., Harmsen, H.J., Stewart, C.S., and Flint, H.J., Growth requirements and fermentation products of Fusobacterium prausnitzii, and a proposal to reclassify it as Faecalibacterium prausnitzii gen. nov., comb. nov. Int J Syst Evol Micr, 2002. 52(6): p. 2141-2146. [18] Yang, K.D., M. L. Zhu, J., Metabolomics Study Reveals Enhanced Inhibition and Metabolic Dysregulation in Escherichia coli Induced by Lactobacillus acidophilus- Fermented Black Tea Extract. J Agric Food Chem, 2018. 66(6): p. 1386-1393. [19] Rodriguez-Valera, F., Ruiz-Berraquero, F., and Ramos-Cormenzana, A., Short communication isolation of extremely halophilic bacteria able to grow in defined inorganic media with single carbon sources. Microbiology, 1980. 119(2): p. 535-538. [20] Cook, A.M., Daughton, C.G., and Alexander, M., Phosphonate utilization by bacteria. J Bacteriol, 1978. 133(1): p. 85-90.

35 [21] Demple, B. and Harrison, L., Repair of oxidative damage to DNA: enzymology and biology.Annu Rev Biochem, 1994. 63(1): p. 915-948. [22] Candeias, L.P., Folkes, L.K., Porssa, M., Parrick, J., and Wardman, P., Enhancement of lipid peroxidation by indole-3-acetic acid and derivatives: substituent effects. J Free Radical Research, 1995. 23(5): p. 403-418. [23] Derrien, M., Vaughan, E.E., Plugge, C.M., de Vos, W.M., Akkermansia muciniphila gen. nov., sp. nov., a human intestinal mucin-degrading bacterium. Microbiology Society, 2004. 54(5): p. 1469-1476. [24] Rampelli, S., Candela, M., Turroni, S., Biagi, E., Collino, S., Franceschi, C., O'Toole, P.W., and Brigidi, P., Functional metagenomic profiling of intestinal microbiome in extreme ageing. Aging (Albany NY), 2013. 5(12): p. 902. [25] Ibrahim, M. and Anishetty, S., A meta-metabolome network of carbohydrate metabolism: Interactions between gut microbiota and host.Biochem bioph Res Co, 2012. 428(2): p. 278-284. [26] Rodríguez-Díaz, J., Rubio-del-Campo, A., and Yebra, M.J., Lactobacillus casei ferments the N-acetylglucosamine moiety of fucosyl-α-1, 3-N-acetylglucosamine and excretes l- fucose. Appl. Environ. Microbiol., 2012. 78(13): p. 4613-4619. [27] Sevin, D.C., Stählin, J.N., Pollak, G.R., Kuehne, A., and Sauer, U., Global metabolic responses to salt stress in fifteen species. PLoS One, 2016. 11(2): p. e0148888.

36 CHAPTER 3 Metabolomics Study Reveals Enhanced Inhibition and Metabolic Dysregulation in Escherichia coli Induced by Lactobacillus acidophilus-Fermented Black Tea Extract

Kundi Yang1, Matthew L. Duley2 and Jiangjiang Zhu1*

1Department of Chemistry and Biochemistry, Miami University, Oxford, OH, USA 45056 2Center for Advanced Microscopy and Imaging, Miami University, Oxford, OH, USA 45056

Contributions to the chapter. The bacterial experiments, metabolites extraction, and LC/MS experiments were completed by Kundi Yang. Matthew L. Duley helped with the SEM experiment. The data analyses were performed by Kundi Yang under the supervision of Jiangjiang Zhu. This chapter was written by Kundi Yang and edited by Jiangjiang Zhu and Michael Crowder.

37 Abstract This study examined the ability of Lactobacillus acidophilus (LA) to ferment black tea extract (BTE) and the enhancement of Escherichia coli cellular uptake of phenolic compounds when these bacteria were incubated with fermented BTE. The inhibitory effects of BTE to E. coli bacteria with and without fermentation were compared. Several intracellular phenolic compounds as well as metabolic profiles of E. coli with and without treatments were also detected and evaluated using a high-performance liquid chromatography−tandem mass spectrometry-based approach. Our results showed that of the three concentrations from the non-fermented BTE treatment, only the extract from the 25 mg/mL solution of tea leaves solution could inhibit E. coli survival, while LA-fermented BTE extract from 5, 10, and 25 mg/mL tea leaves solutions all inhibited E. coli growth significantly. Intracellular concentrations of (+)-catechin-3-gallate/(−)- epicatechin-3-gallate and (+)-catechin/(−)-epicatechin were significantly-higher when E. coli was treated with fermented BTE in comparison to non-fermented BTE. Scanning electron microscopy images indicated that the intracellular phenolic compounds inhibited E. coli growth by increasing endogenous oxidative stress. Metabolic profiles of E. coli were also investigated to understand their metabolic response when treated with BTE, and significant metabolic changes of E. coli were observed. Metabolic profile data were further analyzed using partial least squares discriminant analysis to distinguish the fermented BTE treatment group from the control group and the non- fermented BTE treatment group. The results indicated a large-scale E. coli metabolic dysregulation induced by the fermented BTE. Our findings showed that LA fermentation can be an efficient approach to enhance phenolic inhibition of bacterial cells through increased endogenous oxidative stress and dysregulated metabolic activities.

Keywords: Black tea extract, phenolic compounds, E. coli, metabolic profiling, bacteria inhibitory effect.

38 3.1 Introduction Functional foods, such as tea drinks, have been known for several decades to have beneficial health effects, although the overall clinical study outcomes are still debatable.[1-4] Tea consumption has been associated with a decreased incidence of cancer, reduced diabetic incidence, assistance with weight control, support of cognitive function, improved antihyperglycemic effect, and antibacterial activities.[5-7] Phenolic compounds from tea extract, including extracts from green tea, black tea, yellow tea, and others, have long been known to make primary contributions to these beneficial health effects. Of these choices, black tea is one of the world’s most popular beverages, which is a polyphenol-rich, aqueous infusion of dried leaves from the plant Camellia sinensis.[1] Over the past decade the polyphenolic constituents of black tea have been studied extensively to explore their biological properties. However, the antibacterial aspect of black tea extract (BTE) has not been fully understood in comparison to the well-studied green tea extract. Meanwhile, the gut bacteria population is known to modulate the utilization of food components, and some probiotic bacteria, such as the Lactobacillus species, can assist human intestinal food process/digestion/fermentation and increase the absorption of nutritional components.[8] The existence of these probiotic bacteria is essential for a healthy gut environment, because they can also act as a first line of defense when pathogenic bacteria invade their territory.[9, 10] Recently, Zhao and Shah examined lactic acid bacteria for the metabolism of tea phenolics to enhance their cellular uptake in colon cancer cells.[11] They also observed that the increase in the cellular antioxidant activity of tea samples after fermentation, particularly in BTE, was in alignment with a bacterial-altered and increased total phenolic composition. While both the bacterial inhibition by phenolic compounds and the function of Lactobacillus acidophilus in fermentation of black tea extract (LA BTE) have been explored separately,[12, 13] the synergistic effects between these two factors are still unclear, and their underlying mechanism in bacteria inhibition has not been fully studied. In this study, we combined several biotechnologies, high- performance liquid chromatography−tandem mass spectrometry (HPLC−MS/MS)-based targeted compound detection and L. acidophilus (LA) fermentation, to investigate the enhanced inhibitory capability of LA-fermented BTE to affect the bacterium Escherichia coli. To better understand the molecular level events inside E. coli cells during the fermented BTE treatment, both intracellular levels of phenolic compounds and metabolites were detected and compared to those found in E. coli with a nonfermented BTE treatment.

39 3.2 Materials and Methods Bacterial strain and growth conditions The model bacterial strain used in this study was E. coli K12. The bacteria were grown on a Difco Luria−Bertani (LB) agar plate (BD Diagnostics, Franklin Lakes, NJ, U.S.A.) and incubated for 24 h at 37 °C. A single colony was selected and transferred to liquid LB broth by using a bioloop. An overnight culture was made by growing bacteria in 5 mL LB broth medium placed in a Thermo Scientific Nunc 50 mL conical sterile polypropylene centrifuge tube, and the culture was incubated at 37 °C for 24 h to reach a stationary phase. Four testing cultures were made as replicates by transferring 50 μL of the overnight culture to 5 mL of freshly-prepared LB broth for subsequent incubation under the same conditions for another 24 h. L. acidophilus ATCC 4356 was purchased from the American Type Culture Center (ATCC) and was used to ferment BTE and metabolize phenolic compounds. The growth method for this strain was similar to our previous work with a minor modification.[14] Briefly, the bacterial strain was grown on Lactobacilli de Man, Rogosa, and Sharpe (MRS) agar (BD Diagnostics, Franklin Lakes, NJ, U.S.A.) for 24 h at 37 °C in an anaerobic chamber (Coy Laboratory Products, Inc., Grass Lake, MI, U.S.A.) following the instructions of the manufacturer. A single colony was selected by using a bioloop and inoculated in MRS broth medium before incubation for 24 h at 37 °C. Four testing cultures were made as replicates by transferring 50 μL of overnight culture to 5 mL of freshly-prepared MRS broth for subsequent incubation under the same conditions for another 24 h to reach stationary phase.

Preparation of BTE Black tea was purchased from a local grocery store. BTE was obtained from dried black tea leaves according to a protocol described previously but with some modification.[11] Briefly, dried black tea leaves (2.5 g) were crushed and then added to a 70% methanol/30% water solution (v/v) in the ratio of 1:20 (w/v). This suspension was placed in a 60 °C water bath for 30 min. The extract was filtered through a Büchner funnel equipped with a 0.4 μmfilter and dried. The final BTE stock was made by reconstitution with 50 mL of ultrapure water to reach the concentration of 50 mg/mL (representing 50 mg of black tea dry leaves originally extracted by 1 mL of solvent). The final BTE stock was diluted to BTE values equivalent to the extract of 25 and 10 mg/mL dry

40 leaves, respectively. These BTE solutions were then sterilized by filtration with 0.2 μm filters and stored at 4 °C.

Fermentation of BTE and co-incubation of fermented BTE with E. coli BTEs were fermented according to a previously-published method with minor modification.[11] Briefly, L. acidophilus ATCC 4356 was inoculated at 50% (v/v) into BTEs of 50, 20, and 10 mg/mL dry tea leaves, and then incubation was allowed at 37 °C for 48 h. Representing the control, nonfermented BTEs were also incubated with MRS medium without LA under the same conditions. Fermented BTE products were collected by filtering the fermented bacterial/BTEs medium with a 0.2 μm filter. The fermented BTE was applied to E. coli K12 culture in the ratio of 1:1 (v/v). Non-fermented BTEs were applied to E. coli K12 directly in the ratio of 1:1 (v/v), and the MRS broth was applied as a negative control group. All three BTE concentrations and controls were replicated 4 times. Plate counts were applied before and after co-incubation to obtain the detailed concentration of the bacterial culture.

Measurement of the Total Phenolic Compound Concentration The total phenolic compound concentration was analyzed according to the Folin−Ciocalteu method with slight modification.[15] Briefly, a 500 μL volume of BTE (50 mg/mL) was added to a 2 mL centrifuge tube, which contained 500 μL of MRS broth, and the tube then vortexed for 30 s. A 100 μL volume of this diluted BTE (25 mg/mL) was transferred to 900 μL of distilled water in a 15 mL test tube, followed by the addition of 0.5 mL of the 1 M Folin−Ciocalteu reagent, and the contents were thoroughly mixed. A 3 mL volume of 7.5% Na2CO3 solution was added after a 3 min interval. Fermented BTE samples were measured using the same protocol, except for no dilution with MRS broth at the beginning. A microplate reader (Biotek ELx808) was used to make all absorbance measurements at 630 nm. A gallic acid aqueous stock solution was used to prepare a standard curve by using the same method. All measurements were carried out in triplicate. The total polyphenol sample content was calculated as the equivalent concentration of gallic acid in mg/mL.

Sample Preparation for HPLC−MS/MS

41 Intracellular metabolites and phenolic compounds from each biological replicate were extracted using a cold methanol extraction approach as previously reported.[16] Briefly, a 200 μL volume of the co-incubated culture was transferred by pipet from a 96-well plate to a 2 mL centrifuge tube. This step was followed by centrifugation and a PBS wash; these steps were repeated 3 times. A 250 μL volume of methanol was added to the cell pellet in the tube, and the sample was mixed vigorously for 1 min on a vortex machine. A 150 μL volume of extracted supernatant was collected after centrifugation at 14,000 rpm for 5 min and then dried. A 50 μL volume of the isotopically-labeled amino acid mixture (spiking) solution prepared in 50% acetonitirile/50% ultrapure water was used to reconstitute the sample. The final samples were loaded into liquid chromatography vials for analysis.

Targeted HPLC−MS/MS method for phenolic compound detection and metabolic profiling The targeted metabolite and phenolic compound detection approach applied in this study was similar to our previous work.[14] Briefly, a Thermo Scientific TSQ Quantiva triple quadruple mass spectrometer equipped with an electrospray ionization (ESI) source, applied for both positive and negative mode compound detection, was coupled with a Thermo Scientific Ultimate 3000 high performance liquid chromatograph equipped with an amide hydrophilic interaction chromatography (HILIC) column having dimensions of 2.1 x 150 mm and a particle size of 2.5 μm (Waters Corporation, Milford, MA, U.S.A.). The extracted bacterial intracellular reconstituted samples were injected on the column for gradient elution separation at 0.300 mL/min using solvents A (5 mM ammonium acetate in 90% water/10% acetonitrile + 0.2% acetic acid) and B (5 mM ammonium acetate in 90% acetonitrile/10% water + 0.2% acetic acid). The autosampler temperature was kept at 4 °C; the column compartment was set at 40 °C; and the separation time for each sample was 20 min. The retention time and selected reaction monitoring (SRM) transitions of targeted metabolites and phenolic compound (or compound pairs) were established by running pure standards (purchased from Sigma, St. Louis, MO, U.S.A., and IROA Technology, Boston, MA, U.S.A.) and collecting the tandem mass spectra (MS/MS). Therefore, the orthogonal information on the retention time and two pairs of SRM transitions could be used to confidently detect and identify targeted compounds. The stable isotopically-labeled amino acid mix (20 amino acids; U-13C, 97−99%; U-15N, 97−99%; catalog number CNLM-6696-1) was purchased from Cambridge Isotope Laboratories (Tewksbury, MA, U.S.A.) and used for quality control during MS

42 runs. This method has been checked and validated monthly to ensure its performance. When biological samples were run, pooled quality control samples were also tested in between every 10 samples to monitor instrument stability.

Scanning electron microscope (SEM) assay

The test culture of E. coli K12 was prepared as described above. A 100 μL volume of the test culture was mixed with LB broth at a 1:20 (v/v) ratio for incubation at 37 °C for 12 h. Three treatments were applied after preincubation; E. coli plus sterilized water was used as the control group, while E. coli plus 25 mg/mL BTE and E. coli plus 25 mg/mL BTE plus 2 mM acetylcysteine (NAC) were used as testing groups. Each treatment was incubated at 37 °C for 180 min. For initial fixation, each bacterial suspension was mixed 1:1 (v/v) with a 2.5% (v/v) glutaraldehyde and 2.0% (v/v) paraformaldehyde in 2xPBS buffer solution. The fixed bacterial suspension was then filtered through a 0.45 μm nylon transfer membrane (GVS North America, Sanford, ME, U.S.A.), and an approximately 20 mm square was cut from the center of the filter paper. Ross optical lens tissue was folded into a small envelope, and the 20 mm square was inserted to protect it during processing. From this point on, all of the samples remained in the envelopes.

Envelopes were placed in 2% (w/v) osmium tetroxide in doubly-distilled (dd)H2O overnight at room temperature. Samples were then washed with ddH2O 4 times for 45 min each, and then they proceeded through an ethanol serial dehydration, basically 3 washes in 100% electron microscopy (EM)-grade ethanol before being critical point-dried in a Samdri-780A (Tousimis Corp.,

Rockville, MD, U.S.A.). After drying, the filter papers were removed from the paper envelopes. The bacteria were carefully scraped from the paper surface onto a carbon adhesive tab mounted on an aluminum stub with a new razor blade that had been cleaned with 100% acetone. Samples were then coated with approximately 20 nm of gold in a Denton Desk II sputter coater (Denton Vacuum, LLC, Moorestown, NJ, U.S.A.). Samples were imaged in a Zeiss 35VP scanning electron microscope (Carl Zeiss AG, Thornwood, NY, U.S.A.) at 4 keV with an 8 mm working distance.

Data processing and statistical analyses Xcalibur 4.0 was used to process the raw data, and the data processing method was similar to that described in our previous work.[14, 17] Briefly, the raw data were reprocessed, integrated manually, and then exported as an Excel file. Data normalization was completed using a viable

43 count of testing culture, and both uni- and multivariate statistical analyses were conducted using JMP Pro12 (SAS Institute, Cary, NC, U.S.A.) and MetaboAnalyst 3.0.[18]

3.3 Results and Discussion In this study, the BTE was obtained following the extraction approach reported previously.[11] The BTEs were then split into two groups, with one set of samples mixed with LA and another set mixed with the MRS medium used for LA growth. After LA fermentation, both types of BTE samples were filtered through 0.22 μm filters, and then the filtered products were used to incubate with E. coli to test for growth inhibition. A schematic of our workflow is shown in Figure 3.1A. A comparison between the E. coli cultures incubated with BTE (with or without LA fermentation) was then conducted, and the results can be seen in Figure 3.1B. A concentration- dependent inhibitory effect of E. coli can be clearly observed for the LA-fermented BTE treatment, with all three tested concentrations having statistically-significant decreases in comparison to the control group. As for the BTE without LA fermentation, only the highest concentration of BTE used for incubation showed a significant-decrease of E. coli growth. It is known that black tea contains various salts, which can cause changes in osmolarity in the growth environment, and these changes in osmolarity may lead to a bacterial membrane leakage when the bacteria are treated with BTE.[19] However, osmolarity measurements, conducted before and after fermentation of BTE by testing freezing points, showed no significant differences between those two groups, indicating the similar osmolarities with or without BTE/FBTE treatment to the control group. Also, our focus was whether the fermentation of BTE changed the bacterial inhibitory function; whether the osmolarity of BTE played an important role in the inhibition process needs further study and was not be pursued further.

44 Figure 3.1 A. The schematic of workflow used in this study. B. Inhibitory effect of BTE treatment to E. coli with or w/o Lactobacillus acidophilus (LA) fermentation. C05, C10, and C25 stands for different concentrations of BTE treatment without LA fermentation (BTE extracted from 5 mg/mL; 10 mg/mL, and 25 mg/mL dry leaves, respectively); D05, D10, and D25 stands for different concentrations of fermented BTE (extracted from 5 mg/mL; 10 mg/mL, and 25 mg/mL, then fermented with LA). All treatment groups were compared to untreated control group (UN). * p<0.05; ** p<0.005.

We then examined the concentration of total phenolic compounds from the BTE with and without LA fermentation. The LA-fermented BTE contained a significantly-higher total amount of phenolic compounds compared BTE samples that were not fermented (Figure 3.2A), with a ∼30% increase of bioavailable phenolic compounds detected in the fermented BTE samples. After treatment of the E. coli with both types of BTEs, the intracellular levels of five major phenolic compounds (or compound pairs), namely, caffeine, (+)-catechin/(−)-epicatechin, gallocatechin, catechin-3-gallate/epicatechin-3-gallate, and gallocatechin-3-gallate/epigallocatechin-3-gallate, was measured using our HPLC−MS/MS approach. These three compound pairs were considered because we could not differentiate the isomers using our current chromatographic conditions. The detailed detection parameters used for these compounds can be seen in Table 3.1. When the intracellular levels of these phenolic compounds from E. coli are compared, two of them (gallocatechin-3-gallate/epigallocatechin-3-gallate and gallocatechin) cannot be detected as a result of a relatively low signal intensities. Of the three compounds (compound pairs), the catechin- 3-gallate/epicatechin-3-gallate and (+)-catechin/(−)-epicatechin pairs were detected at significantly-higher levels in samples with the fermented BTE treatment than with the non- fermented BTE treatment (panels B and C of Figure 3.2). This result indicates that the LA fermentation increased the amount of bioavailable total phenolic compounds, resulting in a higher intracellular phenolic compound uptake and lower E. coli growth. However, we also observed the decreased concentration of caffeine in E. coli cells from the fermented BTE treatment (Figure 3.2D), which could possibly suggest the metabolic degradation of caffeine by bacteria, as discussed by Summers and co-workers in a previous study.[20]

46 Figure 3.2 Measurements of phenolic compound concentrations. A. Comparison of BTE total phenolic compound concentrations with and w/o 48h fermentation by LA (B-D). A. E. coli intracellular phenolic compounds comparison post BTE treatment with and w/o LA fermentation. B. (+)-Catechin-3-gallate/(-)-Epicatechin-3-gallate; C. (+)-Catechin/(-)-Epicatechin; D. Caffeine;. C05, C10, C25 stands for different concentrations of BTE treatment without LA fermentation (BTE extracted from 5 mg/mL; 10 mg/mL, and 25 mg/mL dry leaves, respectively); D05, D10, and D25 stands for different concentrations of fermented BTE (extracted from 5 mg/mL; 10 mg/mL, and 25 mg/mL, then fermented with LA). T-test was applied on groups in same concentration. ** p<0.005; *** p<0.001.

Table 3.1 Targeted phenolic compounds and their detection parameters.

Retention Collision Collision RF Precursor Product Product Phenolic compounds time Polarity Energy Energy Lens (m/z) 1(m/z) 2(m/z) (min) 1 (V) 2 (V) (V) Caffeine 1.33 positive 195.34 138.05 110.11 22.89 19.30 81.17 (+)-Catechin/(-)- 1.44 positive 291.06 139.05 123.05 15.51 16.88 70.54 Epicatechin Gallocatechin 1.99 positive 307.06 151.05 139.05 15.51 10.25 72.02 Catechin-3- gallate/Epicatechin-3- 1.28 positive 443.10 273.07 123.11 16.17 10.25 73.51 gallate Gallocatechin-3- gallate/Epigallocatechin- 1.43 positive 459.10 288.75 139.00 21.33 10.25 79.19 3-gallate

It is well-known that the bacterial metabolic activities can change during incubation with inhibitory drugs or compounds.[17, 21, 22] Therefore, we also examined differences in intracellular metabolic profiles, and metabolite data were obtained from either LA-fermented or non-fermented BTE treated E. coli groups. The substantial group difference of 90 detected metabolites is shown in Figure 3.4. In this heatmap, each column represents one biological replicate of an E. coli sample, and each row represents one targeted metabolite. The comparison of BTE treatment with and without LA fermentation clearly indicates dramatic metabolic dysregulation when additional phenolic compounds became available after LA fermentation. Metabolic profiles from both treatments were also compared to those of the control samples, and differences in their metabolic profile differences were observed. More specific comparisons of several representative metabolites were conducted and are plotted in Figure 3.5. The detailed detection parameters used for these compounds can be seen in Table 3.2. As demonstrated, several amino acids, such as creatinine, phenylalanine, valine, and betaine, displayed clear differences from the fermented BTE treatment group compared to the non-fermented BTE group. Other metabolites, such as lactose, NADP, and inosine 5′-triphosphate, which are known important indicators of energy metabolism,[23-25] were found in different levels in the two BTE treatment groups. The levels of cytidine 5′-diphosphocholine, an intermediate in the generation of phosphatidylcholine from choline,[26] which is a common biochemical process in cell membranes,

48 significantly-increased in the fermented BTE-treated groups. This finding suggested the greater need for this group of bacteria to generate or repair the cell membrane in the presence of higher levels of phenolic compounds.

Table 3.2 Example of targeted metabolites and their detection parameters.

Retention Collision Collision RF Precursor Product Product Compound Time Polarity Energy Energy Lens (m/z) 1(m/z) 2(m/z) (min) 1(V) 2(V) (V) N-acetyl- 1.08 Positive 222.64 191.07 207.07 28.30 15.76 55 mannosamine

lactose 1.32 Positive 361.03 342.97 352.03 10.25 10.86 84

phenylalanine 1.86 Positive 166.15 103.07 120.11 27.65 13.79 54

Betaine 2.04 Positive 118.15 58.17 59.17 23.40 19 80

valine 2.1 Positive 118.15 72.11 100.11 13.69 13.39 56

creatine 2.36 Positive 132.12 86.11 114.11 10.25 10.25 39

inosine 5'- 2.66 Positive 508.94 262.95 344.94 24.61 10.25 89 triphosphate cytidine 5'- 3.2 Positive 490.18 264.90 379 25.93 13.03 103 diphosphocholine Following the univariate analyses, which only compares individual metabolites from the metabolic profiles, we also conducted multivariate statistical analyses using partial least squares discriminant analysis (PLS-DA). This approach was to generate an overview of the entire metabolic profile from each sample in a respective group and to confirm whether the fermented and non-fermented BTE systematically dysregulated the metabolic activities of E. coli at the metabolome level. As seen in Figure 3.6A, three major clusters, determined by the metabolic profiles observed in the E. coli control group, the fermented BTE-treated group, and the non- fermented BTE-treated group, were clearly separated. The ovals with different colors indicate the 95% confidence level for the separation. Not surprisingly, the BTE concentration-dependent variation in E. coli growth inhibition in either group was not significant enough to result in separation. Figure 3.6B also lists the top 15 metabolites that contribute the most to the group separation achieved in Figure 3.6A, as measured by variance importance projection (VIP) scores.

49 Glucosamine, 4-guanidinobutanoate, and glutaric acid were the top three metabolites that were significantly-different in the three comparison groups. On the basis of the color bar on the right side of the figure, all three metabolites were detected at higher levels in the non-fermented BTE- treated cultures than in the fermented BTE treatment group or in the control group.

Figure 3.3 SEM images showing the possible intracellular oxidative stress induced cell death. A. E. coli only (Control group). B. E. coli treated with black tea extraction from 25 mg/mL tea leaves. C. E. coli treated with black tea extraction from 25 mg/mL tea leaves plus 2 mM NAC. Increased cell debris can be observed in B and such effect was reduced in C, at least partially due to the antioxidant effect from NAC.

Extensive work has been carried out to study the beneficial health effects of tea extracts, including antibacterial function,[2, 27, 28] anticancer function,[29] and gut microbiome modulation.[30] Nonetheless, the underlying mechanism for antibacterial activity is not yet fully understood. Our present studies suggest that the inhibition of E. coli growth depends on the intracellular concentration of phenolic compounds, and we posit that metabolic dysregulation could be induced by oxidative stress caused by phenolic compounds, such as catechins. Several recent studies have explored the endogenous oxidative stress induced by phenolic compounds, such as using epigallocatechin gallate to inhibit E. coli growth [27] or investigating the reduced quorum sensing and biofilm formation through Rosa rugosa tea phenolic extract.[31] To test the hypothesis that intracellular phenolic compounds cause increased oxidative stress and lower E. coli growth, we conducted E. coli growth tests using a control group (mock dose with sterilized water), a BTE-treated group, and a BTE + NAC (N-acetylcysteine)-treated group. Our results

50 indicated that NAC treatment could partially rescue the BTE-induced inhibited cell growth via cell count measurement (data not shown). SEM experiments were also conducted to examine the cell morphologies. As shown in Figure 3.3, the control group (E. coli treated with sterile water; Figure 3.3A) appears to contain healthy, rod-shaped cells. On the other hand, the cells from the BTE- treated E. coli K12 group (Figure 3.3B) had visibly-damaged cell walls. The cells from the BTE- NAC-treated group showed cell damage but at a reduced level than the cells with BTE treatment alone (Figure 3.3C). In our study, total bioavailable phenolic compounds were made increasingly accessible through LA fermentation, a process that could happen in a healthy human host or in a human gut with additional probiotic supplement intake. Meanwhile, intracellular phenolic compounds catechin-3-gallate/epicatechin-3-gallate and (+)-catechin/(−)-epicatechin were significantly increased after 48 h of LA fermentation, which increased the possibility for the catechins to reach desired intracellular targets and execute inhibitory function to E. coli. We also suspect that the increased phenolic compounds inside bacteria cells could enhance the oxidative stress, and therefore, could contribute to the stronger inhibition of E. coli proliferation. Future follow-up studies will be needed to confirm this working hypothesis. Another important phenolic compound, epigallocatechin-3-gallate, is also frequently linked to antimicrobial activities from a primarily green tea extract origin. In our BTE study, we were unable to detect an intracellular level of epigallocatechin-3-gallate, probably as a result of the excessive loss of this compound during the black tea production process.[6] A recent review also discussed the antimicrobial effects by three out of four main catechins found in green tea, including (−)-epicatechin-3-gallate, (−)-epigallocatechin, and (−)- epigallocatechin-3-gallate.[32] The known mechanisms behind this antimicrobial function include the damage of the bacterial cell membrane (such as catechins binding to the bacterial lipid bilayer cell membrane, therefore inhibiting the ability of bacteria to bind to each other to form biofilms and to bind to host cells), inhibition of fatty acid synthesis (as a result of the inhibition of specific reductases in bacterial type II fatty acid synthesis and inhibition of bacterial production of toxic metabolites), and inhibition of functional enzymes, such as tyrosine phosphatase and cysteine proteinases.[32] Several of these reported mechanisms involved the disruption of metabolic enzymes, which strongly suggested that the downstream production of metabolites from these enzymes could possibly have been influenced as well. To further understand this possible influence, we conducted HPLC−MS/MS-based targeted metabolic profiling to obtain systematic metabolic

51 information. Our results have shown that massive metabolic profile changes can be observed from fermented BTE-treated E. coli, and different classes of metabolites, such as amino acids, sugars, and nucleotides, were detected at significantly different levels than in the nonfermented BTE treatment group (Figures 3.4 and 3.5). The multivariate statistical analysis approach PLS-DA also summarized the group difference by projecting an individual biological replicate into different clusters on its score plot, which was primarily based on the contribution of a group of significantly altered metabolites from different groups (Figure 3.6). Clear separation of different treatment groups via their metabolic profiles by PLS-DA strongly suggested that the fermented BTE treatment systematically disturbed the metabolic activities of the treated E. coli population and inhibited their growth via enhanced oxidative stress. Our results highlighted that the BTE fermentation by LA can contribute to increased intracellular phenolic compounds, such as catechin-3-gallate/epicatechin-3-gallate and (+)-catechin/(−)-epicatechin, and therefore dysregulate massive metabolic activities inside E. coli cells, inhibiting growth and proliferation. While the findings were exciting, we acknowledge the need for further follow-up studies to seek a better understanding of specific metabolic pathway interruptions and metabolic enzyme inhibition induced by the intracellular phenolic compounds. Our study provided firsthand evidence for fermented BTE-induced metabolic dysregulation, which should pave the way for future studies in exploring the interactions between intracellular phenolic compounds and bacterial metabolic activities and how these interactions influence bacteria survival. In summary, our study used state-of-the-art bioanalytical technology, MS-targeted metabolite measurement and phenolic compound detection, to systematically investigate the potential enhancement of LA bacteria to the accessibility of bioavailable phenolic compounds from BTE and its consequential bacterial inhibitory effect to E. coli. Sensitive and specific detection of phenolic compounds and metabolites enabled a molecular level understanding of the bacterial inhibitory effect by fermented BTE. We will continue this line of research in the future to further elucidate the enhanced antibacterial effect of LA-fermented BTE and explore its capability in other model bacteria systems to prove the broader utility of this synergistic effect.

Figure 3.4 Heatmap presentation of intracellular metabolic profiles from E. coli control group (UN), BTE treatment group (C05, C10, C25) and fermented BTE treatment group (D05, D10, D25). Each column represented one biological replicate, and each row represent one targeted metabolite detected in this study.

53 Figure 3.5 Box plots of nine example metabolites showing significant difference from different treatment groups and control group. C05, C10, C25 stands for different concentrations of BTE treatment without LA fermentation (BTE extracted from 5 mg/mL; 10 mg/mL, and 25 mg/mL dry leaves, respectively); D05, D10, and D25 stands for different concentrations of fermented BTE (extracted from 5 mg/mL; 10 mg/mL, and 25 mg/mL, then fermented with LA). All treatment groups were compared to untreated control group (UN).

54 Figure 3.6 Partial least square – discriminant analysis (PLS-DA) approach differentiates the E. coli control group, the BTE treatment group and fermented BTE treatment group based on their metabolic profiles. (A). PLS-DA score plot demonstrated the clear separation of three clusters of samples. (B). Top fifteen metabolites that contribute significantly to the separation of these three groups by PLS-DA approach, the contribution can be quantified by variance importance projection (VIP) scores listed at the x-axis. Color represents the relative concentration of each metabolite from different sample groups.

ASSOCIATED CONTENT Acknowledgement This work was supported by Miami University (Startup fund to JZ). The authors thank Dr. Richard E. Edelmann from the Center for Advanced Microscopy and Imaging at Miami University for his assistance to our SEM experiments.

Conflict of interest No conflict of interest was declared by the authors.

56 References

[1] Mujtaba, T. and Dou, Q.P., Black tea polyphenols inhibit tumor proteasome activity. In Vivo, 2012. 26(2): p. 197-202. [2] Hatano, T., Kusuda, M., Inada, K., Ogawa, T.-o., Shiota, S., Tsuchiya, T., and Yoshida, T., Effects of tannins and related polyphenols on methicillin-resistant Staphylococcus aureus. Phytochemistry, 2005. 66(17): p. 2047-2055. [3] Taylor, P.W., Hamilton-Miller, J.M., and Stapleton, P.D., Antimicrobial properties of green tea catechins. Food Science and Technology Bulletin, 2005. 2: p. 71. [4] Stapleton, P.D., Shah, S., Anderson, J.C., Hara, Y., Hamilton-Miller, J.M., and Taylor, P.W., Modulation of β-lactam resistance in Staphylococcus aureus by catechins and gallates.Int J Antimicrob Ag, 2004. 23(5): p. 462-467. [5] Ruxton, C.H., The health effects of black tea and flavonoids. Nutrition & Food Science, 2009. [6] Han, M., Zhao, G., Wang, Y., Wang, D., Sun, F., Ning, J., Wan, X., and Zhang, J., Safety and anti-hyperglycemic efficacy of various tea types in mice. Sci Rep, 2016. 6(1): p. 1-12. [7] Beresniak, A., Duru, G., Berger, G., and Bremond-Gignac, D., Relationships between black tea consumption and key health indicators in the world: an ecological study. BMJ open, 2012. 2(6). [8] Gerritsen, J., Smidt, H., Rijkers, G.T., and de Vos, W.M., Intestinal microbiota in human health and disease: the impact of probiotics. Genes & Nutrition, 2011. 6(3): p. 209-240. [9] Laparra, J.M. and Sanz, Y., Interactions of gut microbiota with functional food components and nutraceuticals. Pharmacol Res, 2010. 61(3): p. 219-225. [10] Tuohy, K.M., Probert, H.M., Smejkal, C.W., and Gibson, G.R., Using probiotics and prebiotics to improve gut health. Drug Discov Today, 2003. 8(15): p. 692-700. [11] Zhao, D. and Shah, N.P., Lactic acid bacterial fermentation modified phenolic composition in tea extracts and enhanced their antioxidant activity and cellular uptake of phenolic compounds following in vitro digestion. J Funct Food, 2016. 20: p. 182-194. [12] Toda, M., Okubo, S., Hiyoshi, R., and Shimamura, T., The bactericidal activity of tea and coffee. Lett Appl Microbiol, 1989. 8(4): p. 123-125.

57 [13] Marhamatizadeh, M.H., Ehsandoost, E., and Gholami, P., The influence of green tea (Camellia sinensis L.) extract on characteristic of probiotic bacteria in milk and yoghurt during fermentation and refrigerated storage. Intern J Farm Allied Sci, 2013. 2(17): p. 599-606. [14] Xu, M., Zhong, F., and Zhu, J., Evaluating metabolic response to light exposure in Lactobacillus species via targeted metabolic profiling. J Microbiol Methods, 2017. 133: p. 14-19. [15] Pothitirat, W., Chomnawang, M.T., Supabphol, R., and Gritsanapan, W., Comparison of bioactive compounds content, free radical scavenging and anti-acne inducing bacteria activities of extracts from the mangosteen fruit rind at two stages of maturity. Fitoterapia, 2009. 80(7): p. 442-447. [16] Zhong, F., Xu, M., Bruno, R.S., Ballard, K.D., and Zhu, J., Targeted high performance liquid chromatography tandem mass spectrometry-based metabolomics differentiates metabolic syndrome from obesity. Exp Biol M, 2017. 242(7): p. 773-780. [17] Schelli, K., Rutowski, J., Roubidoux, J., and Zhu, J., Staphylococcus aureus methicillin resistance detected by HPLC-MS/MS targeted metabolic profiling. J Chromatogr B, 2017. 1047: p. 124-130. [18] Xia, J., Sinelnikov, I.V., Han, B., and Wishart, D.S., MetaboAnalyst 3.0—making metabolomics more meaningful. Nucleic Acid Res, 2015. 43(W1): p. W251-W257. [19] Gao, Q., Liou, L.-C., Ren, Q., Bao, X., and Zhang, Z., Salt stress causes cell wall damage in yeast cells lacking mitochondrial DNA. Microb Cell, 2014. 1(3): p. 94. [20] Summers, R.M., Mohanty, S.K., Gopishetty, S., and Subramanian, M., Genetic characterization of caffeine degradation by bacteria and its potential applications. Microb Biotechnol, 2015. 8(3): p. 369-378. [21] Schelli, K., Zhong, F., and Zhu, J., Comparative metabolomics revealing Staphylococcus aureus metabolic response to different antibiotics. Microb Biotechnol, 2017. 10(6): p. 1764-1774. [22] Allison, K.R., Brynildsen, M.P., and Collins, J.J., Metabolite-enabled eradication of bacterial persisters by aminoglycosides. Nature, 2011. 473(7346): p. 216-220.

58 [23] Otto, R., Vije, J., ten Brink, B., Klont, B., and Konings, W.N., Energy metabolism in Streptococcus cremoris during lactose starvation. Arch Microbiol, 1985. 141(4): p. 348- 352. [24] Xiao, W., Wang, R.-S., Handy, D.E., and Loscalzo, J., NAD (H) and NADP (H) redox couples and cellular energy metabolism. Antioxid Redox Sign, 2018. 28(3): p. 251-272. [25] Tolkovsky, A. and Suidan, H., Adenosine 5'-triphosphate synthesis and metabolism localized in neurites of cultured sympathetic neurons. Neuroscience, 1987. 23(3): p. 1133- 1142. [26] Adibhatla, R.M., Hatcher, J.F., and Dempsey, R.J., Cytidine‐5′‐diphosphocholine affects CTP‐phosphocholine cytidylyltransferase and lyso‐phosphatidylcholine after transient brain ischemia. J Neurosci Res, 2004. 76(3): p. 390-396. [27] Xiong, L.-G., Chen, Y.-J., Tong, J.-W., Huang, J.-A., Li, J., Gong, Y.-S., and Liu, Z.-H., Tea polyphenol epigallocatechin gallate inhibits Escherichia coli by increasing endogenous oxidative stress. Food Chem, 2017. 217: p. 196-204. [28] Steinmann, J., Buer, J., Pietschmann, T., and Steinmann, E., Anti‐infective properties of epigallocatechin‐3‐gallate (EGCG), a component of green tea. Brit J Pharmacol, 2013. 168(5): p. 1059-1073. [29] Du, G.-J., Zhang, Z., Wen, X.-D., Yu, C., Calway, T., Yuan, C.-S., and Wang, C.-Z., Epigallocatechin Gallate (EGCG) is the most effective cancer chemopreventive polyphenol in green tea. Nutrients, 2012. 4(11): p. 1679-1691. [30] Moco, S., Martin, F.o.-P.J., and Rezzi, S., Metabolomics view on gut microbiome modulation by polyphenol-rich foods. J Proteome Res, 2012. 11(10): p. 4781-4790. [31] Zhang, J., Rui, X., Wang, L., Guan, Y., Sun, X., and Dong, M., Polyphenolic extract from Rosa rugosa tea inhibits bacterial quorum sensing and biofilm formation. Food control, 2014. 42: p. 125-131. [32] Reygaert, W.C., The antimicrobial possibilities of green tea. Front Microbiol, 2014. 5: p. 434.

59 CHAPTER 4 Ultrafine Particles Altered Gut Microbial Population and Metabolic Profiles in a Sex-Specific Manner in an Obese Mouse Model

1# 1# 1 2 1 3 Kundi Yang , Mengyang Xu , Jingyi Cao , Qi Zhu , Monica Rahman , Britt Holmén , Naomi K. Fukagawa4, and Jiangjiang Zhu 5, 6*

1. Department of Chemistry and Biochemistry, Miami University, Oxford, OH, 45056 2. Department of Biology, Miami University, Oxford, OH, 45056 3. School of Engineering, University of Vermont, Burlington, VT 05405 4. USDA ARS Beltsville Human Nutrition Research Center, Beltsville, MD 20705 USA 5. Department of Human Sciences, The Ohio State University, Columbus, OH, 43210 6. James Comprehensive Cancer Center, The Ohio State University, Columbus, OH 43210 # These two authors contributed equally to this work

Contributions to the chapter. Mengyang Xu and Monica Rahman helped with mice husbandry. Mengyang Xu, Jingyi Cao, and Qi Zhu helped with mice sample (organs, plasma, cecum) collection. The ultrafine particle was provided by Dr. Britt Holmén’s lab. The data analyses were performed by Kundi Yang and Mengyang Xu. This chapter was written by Kundi Yang and Mengyang Xu and edited by Jiangjiang Zhu, Britt Holmén, Naomi K. Fukagawa, and Michael Crowder.

Submitted to Scientific Reports (last status update 09/14/2020)

60 Abstract Emerging evidence has highlighted the need for scientists, physicians and other healthcare professionals, and policy-makers to understand the potential adverse effects of components in air pollution, such as ultrafine particles (UFPs), on overall human health through the complex interactions with the human gut microbiota. In this study, to determine the modulating effects of gastrointestinal (GI) UFPs exposure on gut microbial composition and functions that may lead to a systematic evaluation of the impact of UFPs on host health, an in vivo murine model of obesity in both sexes was used to investigate the gut microbial population and diversity changes during UFP exposure over ten days. Multi-omics approaches, including targeted metabolomics and microbiome analysis, were applied to evaluate the UFP-related effects on the gut microbial population and its function. Due to the rising use of biodiesel fuel in our society, and the paucity of information on the health effects of biodiesel exhaust particles, UFPs generated from the combustion of both petrodiesel (B0) and a petrodiesel/biodiesel blend (80:20 v/v, B20) in a representative light-duty diesel engine were compared. We discovered that UFP exposure is associated with changes in the relative abundance of the microbes and their metabolites in the host gut. Furthermore, plasma metabolites from the obese mice were differentially-affected by the fuel type used to generate the UFPs (B0 vs. B20). Bacterial cellular oxidative stress and bacterial metabolic signatures, such as the decreased concentration of nucleotides and lipids and increased concentrations of carbohydrate, energy and vitamin metabolites, were detected. Sex-specific differences in the gut microbial population and their metabolic profiles in response to UFP exposure were also observed.

Key Words: Ultrafine Particles, obese model, gut microbiota, sex-specific response, mass spectrometry-based metabolomics.

61 4.1 Introduction Obesity has been reported to be associated with diabetes, atherosclerosis, and hypertension,[1] and approximately sixty-five percent of adults in the US and more than a hundred billion people all over the world are overweight or obese.[2] Substantial epidemiologic evidence also suggests that obesity causes a state of chronic subclinical inflammation, which mediates most of the systemic complications associated with obesity.[3] Obesity is always accompanied by many metabolic problems, such as blood pressure elevation, insulin resistance, and atherogenic dyslipidemia, and these problems often contribute to obesity-related tissue injury.[4] Particulate matter (PM) from a variety of sources is ubiquitous in ambient urban and indoor air. Based on epidemiological studies, increased levels of ambient particulate matter, with an aerodynamic diameter less than 10 µm and 2.5 µm (PM10 and PM2.5), are associated with increased inflammation.[5, 6] The underlying mechanism contributing to the adverse health effects of PM exposure is thought to be related to the production of reactive oxygen species (ROS).[7, 8] However, a considerable knowledge gap about the health effects of particles less than 100 nm, known as ultrafine particles (UFPs), exists. Increasingly, recent studies provided evidence that these submicron-scale particles exhibit significantly-different physicochemical properties than the larger PM, and UFPs cause adverse health effects through different mechanisms.[9] One of the most important sources of PM in ambient air is diesel engine exhaust.[10] In order to lower the generation of regulated emissions, such as PM, NOx, and CO, efforts have been made to improve the design of diesel engines or to use “cleaner” fuels. However even with improved engines, UFPs are still produced and pose health risks.[11-13] The most commonly-used commercial biodiesel blend in the U.S. is 20% soybean biodiesel (B20). Unfortunately, there are relatively few studies on the composition of PM from the exhaust when this fuel is used in diesel engines. Nonetheless, the combustion products of B20 have been reported.[14, 15] In fact, lower amounts of CO, PM mass, UFPs, and polycyclic aromatic hydrocarbons (PAH) are produced when biodiesel is used than when petrodiesel is used.[16, 17] However, there is debate about whether UFPs from biodiesel combustion can induce inflammation more readily than UFPs from petrodiesel combustion. Some researchers have suggested that the higher levels of polar organic compounds in biodiesels, as compared to petrodiesels, results in higher levels of ROS being produced.[18] In contrast, other investigators

62 have reported reduced ROS production and DNA damage when biodiesel was used, when compared to petrodiesel.[11] Inhalation of UFPs is considered to be the primary way that these compounds enter humans, making the lungs a primary target. However, the gastrointestinal tract is also directly impacted by UFPs.[18] The intestine is exposed to inhaled UFPs by mucociliary transport from the lung to the gastrointestinal tract.[19, 20] UFPs can also be ingested when on food or in drinks. There are reports that UFPs can contribute to gastrointestinal diseases, such as inflammatory bowel disease (IBD).[21, 22] Potential mechanisms for UFP ingestion to cause gastrointestinal diseases include immune activation, systemic inflammation, and intestinal microbiota modulation.[19] Although recent studies suggest that UFP exposure impacts the gut, there are no studies examining whether UFPs affect the gut microbiota. While many metabolic disorders, characterized by low- grade inflammation, are associated with obesity,[23, 24] recent studies suggest that the exposure to traffic-related air pollution, including UFPs, may play an important role in development and exacerbation of metabolic diseases.[25-27] Evidence shows that gut microbiota can be different among lean, obese, and type 2 diabetic patients, and this difference may play an important role in our understanding of the pathogenesis of metabolic diseases.[28, 29] Since B20 and B0 (petrodiesel) particles differ in composition and cause different amounts of inflammation when inhaled, we wanted to address whether oral exposure to UFPs causes changes in the gut microbiome. In this study, we utilized an in vivo murine model of obesity and exposed male and female mice to UFPs generated from the combustion of B0 and B20 diesel fuels. Genomic and metabolomic analyses of mouse cecal samples were performed, and the metabolites that potentially contribute to the connection between the intestinal microbiome and host metabolome were evaluated.

4.2 Materials & Methods Mouse husbandry, exposure to UFP, and sample collection UFPs were generated, collected, and characterized by the University of Vermont Transportation Air Quality laboratory, and the particle stock suspension was prepared at the final concentration of 0.35 mg/ml before the oral administration to the tested animals, as previously reported.[16, 31] Three-week-old male and female C57B6-specific, pathogen-free mice were purchased from the Jackson Lab (Bar Harbor, ME, USA) and housed in a clean, pathogen-free

63 environment under controlled lighting (12 h light: dark; lights on at 6 a.m. and lights off at 6 p.m.) and temperature (21 °C). The animals were housed in pairs for 4 weeks and then housed individually for the rest of the experiment. The animal experimental protocol was approved by the Institutional Animal Care and Use Committee (IACUC) at Miami University in accordance with the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals. A total of 36 obese mice (18 males and 18 females) were divided into 6 groups (3 groups of n=6 for each sex). After a week of acclimation, the thirty-six 4-week old mice were fed a high-fat diet (45 kcal% Fat, Research Diets, Inc Indianapolis, Indiana, USA) to induce the obese phenotype. Two additional groups of mice (n=6 for each sex) were fed a normal diet (10 kcal% Fat, Research Diets, Inc Indianapolis, Indiana, USA) and were used as control group for weight difference monitoring. Mouse body weight was measured on a weekly basis until significant differences were found between the high-fat diet group and the normal diet group, and the differences were maintained for an additional week. After ~6 weeks on the respective diets, both females and males in the high- fat diet group exhibited significant body weight increases when compared to the normal diet groups. Because the focus of this study was to investigate the impact of UFPs on obese subjects, only obese groups were used for further data analyses. All mice used in this experiment were at the age of ~10 weeks when treatments were initiated. Three groups of obese mice from each sex were treated with PBS, B0, and B20 by oral gavage without sedation to achieve precise enteric administration; 2 mL of each suspension (~0.70 mg PM mass) was administered to the mice every other day for a total of 10 days (i.e., 5 administrations per mouse for a total of 10 ml of the suspension over 10 days). Fecal samples were collected every three days, and body weight (BW) was measured weekly throughout the experiment. Mice were anesthetized by isoflurane followed by decapitation after the 10-day treatment period. The whole blood sample was collected in a 2 mL tube containing 10 µL heparin. The sample was placed on ice and centrifuged, and the obtained plasma was transferred into a new 2 mL centrifuge tube then immediately transferred to -80 oC freezer for long term storage. Cecal contents were collected on ice and immediately stored at - 80 °C until further analyses.

DNA isolation and 16S ribosomal DNA sequencing 16S ribosomal DNA sequencing was used for metagenome analysis by following a modified version of a previously-published protocol.[30, 31] DNA extraction from cecal content

64 was performed using the Fast DNA Spin Kit for Feces (MP Biomedicals). Briefly, 500 mg of cecal samples were mixed with 825 µl of sodium phosphate buffer, 122 µl of MT buffer, and 0.5 ml of 0.1 mm zirconia/silica beads (BioSpec Products, Bartlesville, OK, USA). Samples were homogenized with a bead beater (BioSpec Products, Bartlesville, OK, USA) for 3 min at a setting of 6.0 m/s for 40 secs. The DNA recovered from these samples was assessed using a NanoDrop and gel electrophoresis. Purified genomic DNA samples were subjected to 16S rRNA sequencing at the Miami University Center for Bioinformatics and Functional Genomics (CBFG). The 515f/806r primer set and the GoTaq® Hot Start Colorless Master Mix (Promega) was used for PCR amplification of the 16S rRNA V4 region. Each DNA sample was amplified using a reverse primer tagged individually with a unique 12-base Golay barcode. Agarose gel electrophoresis was conducted for the quality check of PCR products. The amplified 16s rRNA library and Amplicons were purified by using the SequalPrep Normalization Plate kit (Thermo Fisher, Waltham, MA, USA). Purified products were quantified by KAPA Library Quantification Kit Illumina Platforms (Kapa Biosystems, Wilmington, MA, USA). An Illumina Next Generation Sequencing (NGS) MiSeq platform was used for amplicon sequencing.

Bioinformatics analysis of genomic data Illumina-generated fastq files were demultiplexed, processed, quality filtered, and analyzed by Quantitative Insights into Microbial Ecology (QIIME, V 1.9.1 and V 2018.4).[32] The DADA2 software package in QIIME 2 [33] was used for de-noising paired-end Illumina sequenced fastq files, dereplicating them, and filtering chimeras. Operational Taxonomic Unit (OTU) and taxonomic binning of classified sequences based on a pre-trained SILVA database version 128 (at 97 % sequence similarity) [34] were used for identifying the taxonomic classification of the sequence reads by using a feature classifier plugin.[35] In addition, the QIIME2 taxa bar plot command was used for viewing the taxonomic composition of the samples. A QIIME 2 plugin, qi- longitudinal was performed for Alpha and beta-diversity analyses at a sampling depth of 4000.[36] Principal coordinate analyses (PCoA) were performed based on Jaccard distances in QIIME2. Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUST, Galaxy V1.1.1) was used to predict microbial function.[37] A closed-reference OTU- picking strategy in QIIME 1 was used for assigning the sequencing reads to species equivalent operational taxonomic units (OTUs) (at 97% sequence similarity) using GreenGenes

65 (gg_13_08).[38] The output OTU table was normalized to 16S rRNA gene copy number from known bacterial genomes in Integrated Microbial Genomes. Gene function was then predicted based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) database.

Metabolite extraction Fecal metabolite extraction was performed using a modified cold methanol extraction method. Briefly, one single pellet of mice feces from each individual was weighed and recorded for use in normalization procedures. The weighed feces (15~20 mg) were homogenized with 500 µL phosphate buffered saline (PBS), followed by the addition of 500 µL cold MeOH and mixing. The mixture was then quenched at -20 oC for at least 20 min, followed by the addition of 50 µL isotopically-labeled amino acid mixture as internal standards (Cambridge Isotope Laboratories, Inc.). The samples were vortexed vigorously for 1 min, and 450 µL of extracted supernatant was collected after centrifugation at 14,000 rpm for 5 minutes. The resulting pellet was dried in a vacuum concentrator. A mixture of 50% acetonitrile combined with 50% Ultrapure water was used to reconstitute the sample. The final samples were loaded into liquid chromatography vials for analysis.

Metabolic profiling The targeted metabolic profiling method used in this study is similar to previously- published work.[39-41] Retention time and selected reaction monitoring (SRM) transition of targeted metabolites were established by running pure standards (purchased from Sigma, Saint Louis, MO, USA and IROA Technology, Boston, MA) and collecting the tandem mass spectrum (MS/MS), so the orthogonal information of retention time and two pairs of SRM transitions could be used to confidently detect and identify targeted compounds. Briefly, a Thermo Scientific TSQ Quantiva triple quadrupole mass spectrometer equipped with an electrospray ionization source was applied to both positive and negative mode compound detection. The mass spectrometer was coupled to a Thermo Scientific Ultimate 3000 high-performance liquid chromatography (HPLC) equipped with a hydrophilic interaction chromatography (HILIC) column (Waters Corporation, Milford, MA, USA), 2.1×150 mm, amide 2.5 µm. The auto-sampler temperature was kept at 4 °C, the column compartment was set at 40 °C, and the separation time for each sample was 20 min. The reconstituted fecal samples were gradient-eluted at 0.300 mL/min using solvents A (5 mM

66 ammonium acetate in 90% water / 10% acetonitrile + 0.2% acetic acid) and B (5 mM ammonium acetate in 90% acetonitrile / 10% water + 0.2% acetic acid). Stable 13C and 15N universally-labeled amino acid mix (Cambridge Isotope, Tewksbury, MA) was used as an internal standard for quality control purposes. Pooled quality control samples were included after ten samples were run to monitor the instrument stability.

Statistical analyses Data were analyzed by Statistical Analysis of Metagenomic Profiles (STAMP, V2.1.3) with unclassified reads removed. Welch’s t-test with a p-value filter of P ≤ 0.05 was used to identify the differences between groups. The Quanbrowser module of Xcalibur 4.0 was used to manually process targeted metabolite profiling data. Acquired peak intensities were normalized by the weight of cecal samples. MetaboAnalyst 4.0 was used for statistical analysis of metabolites (http://www.metaboanalyst.ca/). Peak intensities were subjected to a log transformation and auto- scaling to achieve an approximately normal distribution. ANOVA module, a partial least squares- discriminant analysis module, and a heatmap module were used for data analysis and visualization.

4.3 Results UFP-Induced Changes in the Structures of Gut Microbial Communities To assess the gut microbial communities in the treatment groups, 16S ribosomal DNA sequencing was utilized. A total of 4477 OTU was identified from the sequence results, and the relative abundance of bacteria at different taxonomic levels was assigned based on the OTU representative sequences against the SILVA 128 database. Beta diversity results were scrutinized using principal coordinate analysis that measured bacterial community relatedness, which represented the variation in species composition among different treatment groups.[42] Three dimensional PCoA plots (Figures 4.1A and 4.1B) were used to compare microbial community structure, based on Jaccard distance in female and male mice. Each axis represents one of the three primary principal components (PC) and represents the OTU variation. In female groups, 17.30%, 11.90%, and 10.13% of the variation were explained by PC1, PC2, and PC3, respectively. In male groups, 13.47%, 11.61%, and 10.42% were explained by PC1, PC2, and PC3, respectively. The colors represent different group identities, of which red represents petrodiesel (B0), blue 20% soy biodiesel blend (B20), and orange the control (PBS) group. Each treatment group included six

67 individual mice; however, due to the low DNA sequencing readings of a few samples in some of the groups, not all six samples from each group were used for statistical analyses (3

68 Figure 4.1 The relative abundance of bacterial levels in mice cecum samples. PCoA plots based on OTU of mice cecum samples. The results were generated based on Jaccard Distance matrix. Females (A) and males (B). The colors represent different group identities, of which red stands for B0, blue stands for B20, and orange stands for PBS group. Genus levels were compared to the PBS group, and the percent change was given in bar chart for both females (C) and males (D). Genera are G1. Bifidobacteriaceae; G2. Coriobacteriaceae; G3. Bacteroidaceae; G4. Bacteroidales S24-7 group; G5. Streptococcaceae; G6. Clostridiaceae 1; G7. Lachnospiraceae; G8. Peptostreptococcaceae; G9. Ruminococcaceae and G10. Erysipelotrichaceae. Firmicutes/Bacteriodetes represent the phyla level difference after UPF administration in females and males (E). * P<0.05.

69 Two bacterial divisions, the Firmicutes and the Bacteroidetes, were previously reported to dominate the distal gut microbiota from all groups (80.7% and 16.9% of all sequences, respectively) in both mice and humans.[43] Data obtained from animal models revealed consistent differences with a significant increase of the Firmicutes and a decrease of the Bacteroidetes levels in obese mice, which could promote adiposity, or alternatively, indicate a host-mediated adaptive response to energy uptake/storage.[44] In the present study, Figure 4.1E shows the calculated ratio of Firmicutes to Bacteroidetes (F/B) using relative abundances. The x-axis represents the group information, and the y-axis represents the ratio of Firmicutes to Bacteroidetes. Using the pairwise t-test, a significant difference between the B20 and PBS group was observed, but no significant difference was seen between B0 and PBS groups in female obese mice, and the average F/B ratio of B0 was greater than that in the PBS groups. Accumulating evidence suggests that an increase of F/B ratio and some bacterial function (Bifidobacteria) is related to the increase of various opportunistic pathogens and some endotoxins-producing Gram-negative bacteria.[45-47] The results from this study suggest that exposure to UFPs, especially B20, alters the gut bacterial community and may cause adverse health effects in the host. Notably, no significant differences in responses to UFP were found among male obese mice groups.

Figure 4.2 The difference in mean proportion of microbial function. between (A) female PBS and B0, (B) female PBS and B20, (C) male PBS and B0, and (D) male PBS and B20 along with the associated 95% confidence interval of this effect size and the p-value less than 0.05.

Distinguishing features for differentiating responses to UFPs treatments compared to control To further analyze the functional profiles of microbial communities, the functional composition of the metagenome data collected in this study was predicted by PICRUST

70 (phylogenetic investigation of communities by reconstruction of unobserved states) using Greengenes reference databases. The known bacterial gene counts obtained through the integrated microbial genomes database and comparative analysis system were used to predict the gene content of a given metagenomic sample.[48] Welch’s t-test with a p-value of less than 0.05 was used for statistical analyses. Gene-functional data were inputted into the STAMP software package. Figure 4.2 shows the level 3 functional categories of PBS-treated obese mice that were significantly-different from either B0 or B20 groups. Within each sex, the B0, B20, and PBS treatments are illustrated in blue, orange, and green, respectively. The 95% confidence level is shown as a dashed line in the right part of the plot, and the enrichment of each functional category is shown as a circle of the corresponding color. The p-value of each functional category is listed as a column on the right next to the 95% confidence level. Consistent with the changes in the gut microbiome community structures, the mean proportion of microbial function between PBS control groups and treatment groups was also significantly-different (p<0.05), where many of the key metabolic pathways were identified as altered by UFPs. It is interesting to note that although the microbial communities may be more significantly-altered in female obese mice post-UFP exposure compared to male obese mice, based on this analysis, many of the predicted metabolic functional alternations were also observed in the male obese mice post-UFP exposure. This observation emphasizes the importance of analyzing the gut microbial metabolic features together with genomic information.

UFP-induced changes in the metabolic profile of the gut microbiome In an effort to further probe the effects of UFP on the gut, we further characterized the metabolomes of the gut microbes and the host. Our targeted metabolomics analyses show that the UFP administration perturbed the metabolic profiles of the gut microbiome in both female and male obese mice, and the overall metabolic profiles of mice from the different treatment groups are shown in Figure 4.3. A total of 135 targeted polar metabolites were detected among all cecal samples in both sexes. As shown in Figure 4.3, the metabolic profiles of UFP-treated groups in both females and male obese mice can be differentiated from the PBS control groups based on the different metabolite intensities detected, and the overall profile can be largely separated by using the first two components of PLS-DA results (Figure 4.3A and 4.3B). If the data from the PBS groups are removed, as they were clearly different than the UFP treated groups, distinguishing

71 group separation can also be observed when only comparing the B0 and B20 groups (Figure S4.1a and S4.1b). Figures 4.3C and 4.3D show the metabolic profile (top 75 based on ANOVA) heatmap of mice cecal samples, with females shown in Figure 4.3C and males shown in Figure 4.3D. The horizontal axis represents the group information, while the vertical axis represents individual metabolites. The color indicates the expression level of each metabolite, with dark red as the highest abundance and dark blue the lowest abundance of the metabolite in the sample. The heat map analysis was applied to the top 75 metabolites that differed significantly to make the heat map more readable. The result reveals that the metabolic profiles of females were generally clustered in their treatment groups, whereas differences between groups were more difficult to discern in males.

Figure 4.3 Gut microbial metabolic profiling of female and male mice based on the metabolomic analysis of cecal samples. Within each sex, the B0, B20, and PBS treatments are illustrated in blue, orange, and green, respectively. Figures 3A and 3C show females PLS-DA plot and heatmap, while Figure 3B and 3D show the results from the male groups. The color within each heatmap indicates the expression level of each metabolite; the dark red indicates the highest abundance and dark blue indicates the lowest abundance of the metabolite in the sample.

The host metabolic profile was also assessed by the detection of plasma metabolites through our targeted metabolomics platform. As shown in Figure 4.4A-B, the PLS-DA plots indicate good differentiation of the metabolic profiling of female obese mice from the two UFPs treatment groups and the PBS control, while less distinctive groups were observed in male obese mice. These results were consistent with the mice gut microbiome metabolites data, which collectively showed less alteration in males in terms of both host and gut microbiome metabolic profiles post-UFPs exposure. The clear differentiation among female groups was supported by several strongly-differentially expressed metabolites after UFPs exposure. The top four metabolites that contributed to the clustering (based on ANOVA) are shown in Figure 4.4C-F. Clearly, after exposure to UFPs (especially B20 administration), glutathione was significantly decreased in the plasma of female mice, while pyridoxal, putrescine, and theophylline increased significantly in female hosts. There were no significant changes observed for B0 groups, relative to the PBS control. One noteworthy metabolite, theophylline, is an exogenous component, and further experiments are needed to investigate why the levels of this metabolite changed in this study.

UFP-induced metabolic pathway alteration In order to link all of the metabolites detected into the context of a connected metabolic pathway network, we conducted metabolic pathway impact analyses. All detected metabolites of each sex were included in the metabolic pathway analysis, so a broader coverage of metabolic networks could be achieved. Figure 4.5 shows the major metabolic pathways of gut microbial bacteria that were impacted by UFPs exposure. The x-axis is the metabolic pathway impact value (from pathway topology analysis,which is calculated as the sum of the importance measures of the matched metabolites normalized by the sum of the importance measures of all metabolites in each pathway), while the y-axis is statistical significance (represented by log p-value) of the impacted pathway from exposure groups compared to the control group. The dot size corresponds to the pathway impact value, and the dot color corresponds to the p-value. It is interesting to note that metabolic pathways, such as those for alanine, aspartate, and glutamate, glycolysis or gluconeogenesis, as well as for butanoate, which are marked as metabolic pathways b, c, and f in Figure 4.5, were also found as matched with the functional prediction results from Figure 4.2 and

74 were identified as significantly-impacted metabolic pathways. All of the metabolic pathways that were significantly-changed in pathway analyses of cecal microbial bacteria, as well as matched to functional prediction analyses of genus level, were aligned as shown in Figure 4.6A for males and 4.6B for females; solid arrows indicate direct connections among metabolic pathways and metabolites, while dashed arrows indicate indirect connections (multiple intermediate steps were excluded to save space). Host metabolic pathways analyses were also conducted. As shown in Figure 4.7A-D, pathway analyses showed the important metabolic pathways impacted by UFP administration. Pairwise comparison of female PBS vs B0 (Figure 4.7A), female PBS vs B20 (Figure 4.7B), male PBS vs B0 (Figure 4.7C), and male PBS vs B20 (Figure 4.7D) were conducted. Interestingly, all four analyses had similar metabolic pathways that were most impacted. Detailed metabolic pathways information is listed in Table S4.2, and metabolic pathways in bold font were significantly-altered, appearing repetitively in female and male groups after UFPs exposure. An overview pathway connection plot was generated based on these major impacted metabolic pathways among PBS control and UFPs treatment groups (Figure 4.7E). Several amino acid metabolic pathways, together with energy metabolic pathways (e.g., glycolysis) were the primary pathways impacted by UFP exposure during the comparison of host metabolism.

75 Figure 4.4 Host metabolic profile difference in both female and male mice and the most contributed metabolites in plasma samples of female mice. The PLS-DA plots showed the metabolic profile of B0, B20, and PBS groups for females (Figure 4.4A) and males (Figure 4.4B). Metabolites that contributed most to the significantly-different clustering of female mice based on their p values were shown in Figure 4.4C-F.

Figure 4.5 Gut microbial metabolic pathway analysis on cecal samples. Female PBS vs B0 (Figure 4.5A), female PBS vs B20 (Figure 4.5B), male PBS vs B0 (Figure 4.5C), and male PBS vs B20 (Figure 4.5D). Some representative metabolic pathways were assigned arbitrary lower-case letters in alphabetical order, and the full names are shown in Table S4.1.

77 Figure 4.6 Connections among metabolic pathways that matched with microbial functional prediction result. Females (Figure 4.6A) and males (Figure 4.6B).

78 Figure 4.7 Host metabolism pathway analysis on plasma samples. Female PBS vs B0 (Figure 4.7A), female PBS vs B20 (Figure 4.7B), male PBS vs B0 (Figure 4.7C), and male PBS vs B20 (Figure 4.7D). Detailed metabolic pathways information is listed in Table S2. Significantly-altered metabolites and their pathways were summarized in the pathway tree plot (Figure 4.7E).

4.4 Discussion Ultrafine particles from multiple sources can be taken up by the intestine and may impact the gastrointestinal tract.[22] Previous studies have shown associations between traffic density and obesity,[49] but there are few studies that examined a link between UFP exposure and obesity. Moreover, previous studies have seldomly focused on the effects of UFP exposure to gut microbiome and host metabolism. In the present work, we provide the first evidence that UFP exposure can alter the metabolic profiles of the host as well as the gut microbiota community in the distal intestine. The findings suggest new insights into the relationship between air pollution and the gastrointestinal responses of obese rodents and by extrapolation obese humans. During the evaluation of gut microbial community after exposure to the two different types of UFPs, the PCoA of 16S DNA analyses helped us to investigate the differences between UFP exposure group and non-exposure group in both sexes, and the results indicate clear distinctions. In addition, at the bacterial phyla level, Firmicutes, Bacteroidetes, and Actinobacteria were the major phyla identified that were impacted by UFP exposure. As in humans, Firmicutes and Bacteroidetes are the most abundant bacterial phyla in these study animals, whereas actinobacteria are present at lower abundance in mice.[44] Upon the oral exposure to the UFP treatments, a slight decrease in the relative abundance of both Firmicutes and Bacteroidetes for obese female mice was observed. Several studies have reported that a decrease in abundance of several taxa within the Firmicutes phylum is related to structural imbalances and dysbiosis that occur in inflammatory bowel disease.[50] A significant decrease in the abundance of Bacteroidetes in germ-free mice colonized with human microbiota was reported upon a dietary shift to a high-fat, high-sugar “Western” diet.[51] However, the same level of change in the relative abundance of both Firmicutes and Bacteroidetes was not found in the male mouse obesity model used in this study. Despite the reduction in abundance of both Firmicutes and Bacteroidetes after exposure to UFPs, there was a significant increase in the Firmicutes to Bacteroidetes ratio in the female B20 treatment group. In previous studies, the Firmicutes/Bacteroidetes ratio has been reported to be positively- correlated with the obese phenotype independent of diet.[44] Based on our results, a disadvantageous shift in gut microbial communities may occur relative to the UFP exposure in females, but not in male mice. This result is consistent with previous findings that sex-specific

80 effects on the gut microbiome may occur when animals were exposed to the pesticide, organophosphate diazinon.[52] Several distinguishing features of the biological functions of the microbiota associated with exposure to UFPs can be identified in this study. There were eleven gut microbial functional pathways that contributed a large fraction of total significant function counts among the female obese mice. Reduced functions of gut microbes in the B0-exposed female mice included amino acid, sugar and nucleotide sugar metabolism, and streptomycin biosynthesis, while butanoate metabolism, propanoate metabolism, and lysine biosynthesis were elevated in these samples in comparison to the same sex PBS control samples. In addition, other elevated functions included cysteine and methionine metabolism and higher levels DNA replication proteins were associated with the exposure to B20 compared to the PBS control in females. Other functions such as glycan degradation, pentose phosphate pathway, cyanoamino acid metabolism, and glycerolipid metabolism were reduced after B20 exposure in female obese mice. In the screening of hub genes and pathways in colorectal cancer with microarray technology, butanoate metabolism and DNA replication were included in the 34 functions that were significantly over-represented in colorectal cancer.[53] Among the 34 functions, DNA replication is the most significantly over-represented pathway. The investigators of this prior study further explored the interactions in the colorectal cancer-related pathways, and among which, P53 signaling pathway, cytokine-cytokine receptor interaction, and type I diabetes mellitus were ranked as the three most important interactions. Therefore, we speculate that frequent oral exposure to UFPs in female obese mice may be associated with a higher risk for colorectal cancer development later in life. On the other hand, there are 26 functional pathways that contributed significant counts among male samples (Figure 2). Only one of these functions, the DNA replication, was also significantly expressed in female samples. Intestinal microbiota metabolites are important to host-microbe interactions and consequently are key determinants of the health or disease of the intestinal tract.[54] Gut microbiota varies in populations and is regulated by factors that regulate host metabolism, immunity, and inflammatory responses.[55-57] In addition, gut microbial metabolites also play an important role in controlling the gut microbiota population and diversity. For example, cholesterol is produced in the intestine and converted to bile acids, and this process is critical for cholesterol homeostasis [58] and the prevention of the accumulation of toxic metabolites in the liver and other

81 organs.[59] In this study, the results showed an obvious perturbation in gut metabolic profiles of female obese mice compared to males after the administration of UFPs. For example, females showed a significant increase in the pyridoxal levels in the intestine and plasma (Figure S4.2). It has been shown that increased levels of pyridoxal are associated with a variety of inflammatory disease conditions, including rheumatoid arthritis, inflammatory bowel disease, and cardiovascular disease.[60] Allantoin, another representative metabolite, showed greater abundance after both UFP treatments in females but was undetectable in the female PBS control group. A previous study reported that this metabolite can mitigate inflammation by inducing the reduction in the levels of IL (interleukin)-4 and IL-5.[61] However, a sex-specific difference existed, as insignificant differences in both intestinal microbiome and plasma were observed when comparing UFP treatments to the PBS control group in males in our study. This result may be related to the similar metabolic profiles between the male B0 treatment and the male PBS treatment or the big relative standard deviation (RSD%) of some metabolic measurements in the male B0 samples. We then compared the male B20 groups and the male PBS groups, and statistically-significant perturbations in metabolic profiles were observed in plasma. For example, myo-inositol, the most significantly- changed metabolite in the host metabolic comparison, was significantly-decreased in B20 treatment groups compared to PBS control groups in male plasma (Figure S4.3). The decreasing level of myo-inositol can be associated with inflammation induced by UFPs.[62, 63] Consistent with the F/B ratio result from the gut microbes, the results of the host metabolite comparison also indicated a higher impact of ingested UFPs in the gastrointestinal tract in females than males.

4.5 Conclusions In this study, we combined metabolomics and 16s rRNA sequencing techniques to study the effect of ultrafine particles from the combustion of different types of diesel fuel on the gut microbial community and their metabolic processes in an obese mice model. The results showed that the gut microbiome composition, as well as their metabolic profiles, were altered after direct exposure to two different types of ultrafine particles in the intestinal tract of both C57BL6 male and female mice. Our genomic results predicted the possible functional impacts of ultrafine particles on the mouse gut microbiome, and our metabolomics analyses validated some of the functional predictions. Both genomic and metabolomics results suggested differences by exposure obese mice to B0 and B20. Interestingly, different levels of perturbation on the gut microbiome

82 were observed between females and males, and stronger perturbation effects on females were observed based on data from both 16s rRNA gene sequencing and metabolic profiling, which suggested female obese mice may be associated with a higher risk for metabolic dysfunction. These results could decipher the role of the gut microbiome in modulating the gastrointestinal system and the host metabolic process in response to the UPF ingestion from environmental exposure. The changes of gut microbiome and metabolic profile post-UFP ingestion in this obese mice model could serve as useful preliminary data to the future investigation of the potential sex- specific health effects of UFPs exposure in humans, and the perturbation caused by different sources of UFPs may also contribute to the understanding of biological effects of UFP ingestion from different fuel types.

ASSOCIATED CONTENT Supporting information This material is available in Appendix B.

Acknowledgments This study was partially supported by the National Institute of General Medical Sciences of the National Institutes of Health under Award Number R35GM133510 (to J.Z.). The authors would like to thank Dr. Yanxue Han (UVM) for quantifying raw impinger UFP concentrations.

83 References

[1] Zanella, M.T., Kohlmann Jr, O., and Ribeiro, A.B., Treatment of obesity hypertension and diabetes syndrome. Hypertension, 2001. 38(3): p. 705-708. [2] Boutagy, N.E., McMillan, R.P., Frisard, M.I., and Hulver, M.W., Metabolic endotoxemia with obesity: Is it real and is it relevant? Biochimie, 2016. 124: p. 11-20. [3] Deng, T., Lyon, C.J., Bergin, S., Caligiuri, M.A., and Hsueh, W.A., Obesity, inflammation, and cancer. Annu Rev Pathol-Mech, 2016. 11: p. 421-449. [4] Zhang, X., Li, Z.-L., Woollard, J.R., Eirin, A., Ebrahimi, B., Crane, J.A., Zhu, X.-Y., Pawar, A.S., Krier, J.D., and Jordan, K.L., Obesity-metabolic derangement preserves hemodynamics but promotes intrarenal adiposity and macrophage infiltration in swine renovascular disease. Am J Physiol-Renal, 2013. 305(3): p. F265-F276. [5] Pope III, C.A., Bhatnagar, A., McCracken, J.P., Abplanalp, W., Conklin, D.J., and O’Toole, T., Exposure to fine particulate air pollution is associated with endothelial injury and systemic inflammation. Circ Res, 2016. 119(11): p. 1204-1214. [6] Zhou, X., A Cohort Study of Air-Pollution and Childhood Obesity Incidence. 2016, University of Southern California. [7] Fukagawa, N.K., Li, M., Poynter, M.E., Palmer, B.C., Parker, E., Kasumba, J., and Holmén, B.A., Soy biodiesel and petrodiesel emissions differ in size, chemical composition and stimulation of inflammatory responses in cells and animals. Environ Sci Technol, 2013. 47(21): p. 12496-12504.

[8] Holmén, B.A., Rukavina, B., Kasumba, J., and Fukagawa, N.K., Reactive oxidative species and speciated particulate light-duty engine emissions from diesel and biodiesel fuel blends. Energ Fuel, 2017. 31(8): p. 8171-8180. [9] Buckley, A., Warren, J., Hodgson, A., Marczylo, T., Ignatyev, K., Guo, C., and Smith, R., Slow lung clearance and limited translocation of four sizes of inhaled iridium nanoparticles. Part Fibre Toxicol, 2017. 14(1): p. 5. [10] Mohankumar, S. and Senthilkumar, P., Particulate matter formation and its control methodologies for diesel engine: A comprehensive review. Renew Sustain Energ Revi, 2017. 80: p. 1227-1238.

84 [11] Hemmingsen, J.G., Møller, P., Nøjgaard, J.K., Roursgaard, M., and Loft, S., Oxidative stress, genotoxicity, and vascular cell adhesion molecule expression in cells exposed to particulate matter from combustion of conventional diesel and methyl ester biodiesel blends. Environ Sci Technol, 2011. 45(19): p. 8545-8551. [12] Jalava, P.I., Tapanainen, M., Kuuspalo, K., Markkanen, A., Hakulinen, P., Happo, M.S., Pennanen, A.S., Ihalainen, M., Yli-Pirilä, P., and Makkonen, U., Toxicological effects of emission particles from fossil-and biodiesel-fueled diesel engine with and without DOC/POC catalytic converter. J Inhal Toxico, 2010. 22(sup2): p. 48-58. [13] Swanson, K.J., Kado, N.Y., Funk, W.E., Pleil, J.D., Madden, M.C., and Ghio, A.J., Release of the pro-inflammatory markers by BEAS-2B cells following in vitro exposure to biodiesel extracts. Open Toxico J, 2009. 3(1): p. 8-15. [14] Kasumba, J., Fukagawa, N.K., and Holmén, B.A., Fuel Composition Effects on Carbonyls and Quinones in Particulate Matter from a Light-Duty Diesel Engine Running Biodiesel Blends from Two Feedstocks. Energ Fuel, 2019. 33(2): p. 1133-1145. [15] Kasumba, J. and Holmén, B.A., Nonpolar organic compound emission rates for light-duty diesel engine soybean and waste vegetable oil biodiesel fuel combustion. Energ Fuel, 2016. 30(11): p. 9783-9792. [16] Durbin, T.D., Collins, J.R., Norbeck, J.M., Smith, M.R., Effects of biodiesel, biodiesel blends, and a synthetic diesel on emissions from light heavy-duty diesel vehicles. Environ Sci Technol, 2000. 34(3): p. 349-355. [17] Lapuerta, M., Rodríguez-Fernández, J., and Agudelo, J.R., Diesel particulate emissions from used cooking oil biodiesel. Bioresource Technol, 2008. 99(4): p. 731-740. [18] Li, R., Yang, J., Saffari, A., Jacobs, J., Baek, K.I., Hough, G., Larauche, M.H., Ma, J., Jen, N., Moussaoui, N., Zhou, B., Kang, H., Reddy, S., Henning, S.M., Campen, M.J., Pisegna, J., Li, Z., Fogelman, A.M., Sioutas, C., Navab, M., and Hsiai, T.K., Ambient Ultrafine Particle Ingestion Alters Gut Microbiota in Association with Increased Atherogenic Lipid Metabolites. Sci Rep, 2017. 7: p. 42906. [19] Möller, W., Häußinger, K., Winkler-Heil, R., Stahlhofen, W., Meyer, T., Hofmann, W., and Heyder, J., Mucociliary and long-term particle clearance in the airways of healthy nonsmoker subjects. J Appl Physiol, 2004. 97(6): p. 2200-2206.

85 [20] Möller, W., Häußinger, K., Ziegler-Heitbrock, L., and Heyder, J., Mucociliary and long- term particle clearance in airways of patients with immotile cilia. Resp Res, 2006. 7(1): p. 10. [21] Lomer, M.C., Thompson, R.P., and Powell, J.J., Fine and ultrafine particles of the diet: influence on the mucosal immune response and association with Crohn’s disease. P Nutr Soc, 2002. 61(1): p. 123-130. [22] Powell, J., Harvey, R., Ashwood, P., Wolstencroft, R., Gershwin, M., and Thompson, R., Immune potentiation of ultrafine dietary particles in normal subjects and patients with inflammatory bowel disease. J Autoimmun, 2000. 14(1): p. 99-105. [23] Matarese, G., Mantzoros, C., and Cava, A.L., Leptin and adipocytokines: bridging the gap between immunity and atherosclerosis. Curr Pharm Design, 2007. 13(36): p. 3676-3680. [24] Zimmet, P., Alberti, K.G.M., and Ríos, M.S., A new International Diabetes Federation (IDF) worldwide definition of the metabolic syndrome: the rationale and the results. 2005, Elsevier Doyma. [25] Jerrett, M., McConnell, R., Wolch, J., Chang, R., Lam, C., Dunton, G., Gilliland, F., Lurmann, F., Islam, T., and Berhane, K., Traffic-related air pollution and obesity formation in children: a longitudinal, multilevel analysis. Environ Health, 2014. 13(1): p. 49. [26] Thiering, E., Markevych, I., Brüske, I., Fuertes, E., Kratzsch, J., Sugiri, D., Hoffmann, B., von Berg, A., Bauer, C.-P., and Koletzko, S., Associations of residential long-term air pollution exposures and satellite-derived greenness with insulin resistance in German adolescents. Environ Health Persp, 2016. 124(8): p. 1291. [27] Toledo‐Corral, C., Alderete, T., Habre, R., Berhane, K., Lurmann, F., Weigensberg, M., Goran, M., and Gilliland, F., Effects of air pollution exposure on glucose metabolism in Los Angeles minority children. Pediatr Obes, 2018. 13(1): p. 54-62. [28] Makki, K., Deehan, E.C., Walter, J., and Bäckhed, F., The Impact of Dietary Fiber on Gut Microbiota in Host Health and Disease. Cell Host Microbe, 2018. 23(6): p. 705-715. [29] Hartstra, A.V., Bouter, K.E.C., Bäckhed, F., and Nieuwdorp, M., Insights Into the Role of the Microbiome in Obesity and Type 2 Diabetes. Diabetes Care, 2015. 38(1): p. 159-165. [30] Caporaso, J.G., Lauber, C.L., Walters, W.A., Berg-Lyons, D., Lozupone, C.A., Turnbaugh, P.J., Fierer, N., and Knight, R., Global patterns of 16S rRNA diversity at a depth of millions of sequences per sample. P Natl A Sci, 2011. 108(Supplement 1): p. 4516-4522.

86 [31] Caporaso, J.G., Lauber, C.L., Walters, W.A., Berg-Lyons, D., Huntley, J., Fierer, N., Owens, S.M., Betley, J., Fraser, L., and Bauer, M., Ultra-high-throughput microbial community analysis on the Illumina HiSeq and MiSeq platforms. ISME J, 2012. 6(8): p. 1621-1624. [32] Caporaso, J.G., Kuczynski, J., Stombaugh, J., Bittinger, K., Bushman, F.D., Costello, E.K., Fierer, N., Pena, A.G., Goodrich, J.K., and Gordon, J.I., QIIME allows analysis of high- throughput community sequencing data. Nat Methods, 2010. 7(5): p. 335-336. [33] Callahan, B.J., McMurdie, P.J., Rosen, M.J., Han, A.W., Johnson, A.J.A., and Holmes, S.P., DADA2: high-resolution sample inference from Illumina amplicon data. Nature Methods, 2016. 13(7): p. 581. [34] Quast, C., Pruesse, E., Yilmaz, P., Gerken, J., Schweer, T., Yarza, P., Peplies, J., and Glöckner, F.O., The SILVA ribosomal RNA gene database project: improved data processing and web-based tools. Nucleic Acids Res, 2012. 41(D1): p. D590-D596. [35] Bokulich, N.A., Kaehler, B.D., Rideout, J.R., Dillon, M., Bolyen, E., Knight, R., Huttley, G.A., and Caporaso, J.G., Optimizing taxonomic classification of marker-gene amplicon sequences with QIIME 2’s q2-feature-classifier plugin. Microbiome, 2018. 6(1): p. 90. [36] Bokulich, N., Zhang, Y., Dillon, M., Rideout, J.R., Bolyen, E., Li, H., Albert, P., and Caporaso, J.G., q2-longitudinal: a QIIME 2 plugin for longitudinal and paired-sample analyses of microbiome data. bioRxiv, 2017: p. 223974. [37] Langille, M.G., Zaneveld, J., Caporaso, J.G., McDonald, D., Knights, D., Reyes, J.A., Clemente, J.C., Burkepile, D.E., Thurber, R.L.V., and Knight, R., Predictive functional profiling of microbial communities using 16S rRNA marker gene sequences. Natr Biotechn, 2013. 31(9): p. 814. [38] DeSantis, T.Z., Hugenholtz, P., Larsen, N., Rojas, M., Brodie, E.L., Keller, K., Huber, T., Dalevi, D., Hu, P., and Andersen, G.L., Greengenes, a chimera-checked 16S rRNA gene database and workbench compatible with ARB. Appl Environ Microb, 2006. 72(7): p. 5069-5072. [39] Zhu, J., Djukovic, D., Deng, L., Gu, H., Himmati, F., Abu Zaid, M., Chiorean, E., and Raftery, D., Targeted serum metabolite profiling and sequential metabolite ratio analysis for colorectal cancer progression monitoring. Anal Bioanal Chem, 2015. 407(26): p. 7857- 7863.

87 [40] Zhu, J., Djukovic, D., Deng, L., Gu, H., Himmati, F., Chiorean, E.G., and Raftery, D., Colorectal Cancer Detection Using Targeted Serum Metabolic Profiling. J Proteom Res, 2014. 13(9): p. 4120-4130. [41] Schelli, K., Rutowski, J., Roubidoux, J., and Zhu, J., Staphylococcus aureus Methicillin Resistance Detected by HPLC-MS/MS Targeted Metabolic Profiling. J Chromatogr B, 2016. [42] Legendre, P., Studying beta diversity: ecological variation partitioning by multiple regression and canonical analysis. J Plant Ecol, 2008. 1(1): p. 3-8. [43] Wen, L., Ley, R.E., Volchkov, P.Y., Stranges, P.B., Avanesyan, L., Stonebraker, A.C., Hu, C., Wong, F.S., Szot, G.L., and Bluestone, J.A., Innate immunity and intestinal microbiota in the development of Type 1 diabetes. Nature, 2008. 455(7216): p. 1109. [44] Ley, R.E., Bäckhed, F., Turnbaugh, P., Lozupone, C.A., Knight, R.D., and Gordon, J.I., Obesity alters gut microbial ecology. P Natl A Sci, 2005. 102(31): p. 11070-11075. [45] Larsen, N., Vogensen, F.K., van den Berg, F.W., Nielsen, D.S., Andreasen, A.S., Pedersen, B.K., Al-Soud, W.A., Sørensen, S.J., Hansen, L.H., and Jakobsen, M., Gut microbiota in human adults with type 2 diabetes differs from non-diabetic adults. PloS one, 2010. 5(2): p. e9085. [46] Qin, J., Li, Y., Cai, Z., Li, S., Zhu, J., Zhang, F., Liang, S., Zhang, W., Guan, Y., and Shen, D., A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature, 2012. 490(7418): p. 55. [47] Wu, X., Ma, C., Han, L., Nawaz, M., Gao, F., Zhang, X., Yu, P., Zhao, C.a., Li, L., Zhou, A., Wang, J., Moore, J.E., Cherie Millar, B., and Xu, J., Molecular Characterisation of the Faecal Microbiota in Patients with Type II Diabetes. Curr Microbiol, 2010. 61(1): p. 69- 78. [48] Markowitz, V.M., Chen, I.-M.A., Palaniappan, K., Chu, K., Szeto, E., Grechkin, Y., Ratner, A., Jacob, B., Huang, J., and Williams, P., IMG: the integrated microbial genomes database and comparative analysis system. Nucleic Acids Res, 2011. 40(D1): p. D115- D122. [49] Wei, Y., Zhang, J., Li, Z., Gow, A., Chung, K.F., Hu, M., Sun, Z., Zeng, L., Zhu, T., and Jia, G., Chronic exposure to air pollution particles increases the risk of obesity and

88 metabolic syndrome: findings from a natural experiment in Beijing. FASEB J, 2016. 30(6): p. 2115-2122. [50] Kostic, A.D., Xavier, R.J., and Gevers, D., The microbiome in inflammatory bowel disease: current status and the future ahead. Gastroenterology, 2014. 146(6): p. 1489-99. [51] Turnbaugh, P.J., Ridaura, V.K., Faith, J.J., Rey, F.E., Knight, R., and Gordon, J. The effect of diet on the human gut microbiome: a metagenomic analysis in humanized gnotobiotic mice. Sci Transl Med, 2009. 1(6): p. 6ra14-6ra14. [52] Gao, B., Bian, X., Mahbub, R., and Lu, K., Sex-Specific Effects of Organophosphate Diazinon on the Gut Microbiome and Its Metabolic Functions. Environ Health Persp, 2017. 125(2): p. 198-206. [53] Wang, Y., Zheng, T., Screening of hub genes and pathways in colorectal cancer with microarray technology. J Proteom Res, 2014. 20(3): p. 611-618. [54] Trompette, A., Gollwitzer, E.S., Yadava, K., Sichelstiel, A.K., Sprenger, N., Ngom-Bru, C., Blanchard, C., Junt, T., Nicod, L.P., and Harris, N.L., Gut microbiota metabolism of dietary fiber influences allergic airway disease and hematopoiesis. Nat Med, 2014. 20(2): p. 159. [55] Org, E., Mehrabian, M., and Lusis, A., Unraveling the environmental and genetic interactions in atherosclerosis: central role of the gut microbiota. Atherosclerosis, 2015. 241(2): p. 387-399. [56] Shulzhenko, N., Morgun, A., Hsiao, W., Battle, M., Yao, M., Gavrilova, O., Orandle, M., Mayer, L., Macpherson, A.J., and McCoy, K.D., Crosstalk between B lymphocytes, microbiota and the intestinal epithelium governs immunity versus metabolism in the gut. Nat Med, 2011. 17(12): p. 1585. [57] Tang, W.W. and Hazen, S.L., The contributory role of gut microbiota in cardiovascular disease. J Clinical Invest, 2014. 124(10): p. 4204-4211. [58] Brunham, L.R., Kruit, J.K., Iqbal, J., Fievet, C., Timmins, J.M., Pape, T.D., Coburn, B.A., Bissada, N., Staels, B., and Groen, A.K., Intestinal ABCA1 directly contributes to HDL biogenesis in vivo. J Clinical Invest, 2006. 116(4): p. 1052-1062. [59] Chiang, J.Y., Bile acid metabolism and signaling. Compr Psychiat, 2013. 3(3): p. 1191- 1212.

89 [60] Paul, L., Ueland, P.M., and Selhub, J., Mechanistic perspective on the relationship between pyridoxal 5'-phosphate and inflammation. Nutr Rev, 2013. 71(4): p. 239-244. [61] Lee, M.Y., Lee, N.H., Jung, D., Lee, J.-A., Seo, C.-S., Lee, H., Kim, J.-H., and Shin, H.K., Protective effects of allantoin against ovalbumin (OVA)-induced lung inflammation in a murine model of asthma. Int Immunopharmacol, 2010. 10(4): p. 474-480. [62] Han, Y., Jung, H.W., Lee, J.Y., Kim, J.S., Kang, S.S., Kim, Y.S., and Park, Y.K., 2, 5- dihydroxyacetophenone isolated from Rehmanniae Radix Preparata inhibits inflammatory responses in lipopolysaccharide-stimulated RAW264. 7 macrophages. J Med Food, 2012. 15(6): p. 505-510. [63] Kim, H.W., Kim, J.H., An, H.S., Park, K.K., Kim, B.K., and Park, T., Myo-inositol restores the inflammation-induced down-regulation of taurine transport by the murine macrophage cell line, RAW 264.7. Life Sci, 2003. 73(19): p. 2477-2489.

CHAPTER 5 Rapid Differentiation of Lactobacillus Species via Metabolic Profiling

Kundi Yang, Mengyang Xu, Fanyi Zhong and Jiangjiang Zhu*

Department of Chemistry and Biochemistry, Miami University, Oxford, OH, USA, 45056

Contributions to the chapter. The bacterial experiments, metabolites extraction and LC/MS experiments were completed by Kundi Yang. Matthew L. Duley helped with the SEM experiment. The data analyses were performed by Kundi Yang under the supervision of Jiangjiang Zhu. This chapter was written by Kundi Yang and edited by Jiangjiang Zhu and Michael Crowder.

91 Abstract Lactobacillus, the major genus of lactic acid bacteria, plays functional roles in the human body; for example, these bacteria convert sugars to lactic acid. They are significant microbiota that can be found at a number of human body sites, such as in the digestive system, urinary system, and genital system. A number of Lactobacillus species are often used as probiotics and can benefit host health when administered in adequate amounts. Due to their diverse functional characteristics, it is essential to have identification and high-resolution typing techniques to support the need in health and nutritional research of Lactobacillus species. In this study, we took advantages of both targeted and untargeted metabolomic technologies by using a triple quadrupole mass spectrometer (MS) in combination with a linear ion trap-Orbitrap hybrid MS, to investigate their capability and performance in deciphering the subtle metabolic differences in four closely-related Lactobacillus species/strains. First, we evaluated the selected reaction monitoring (SRM) and high-resolution MS data for metabolite quantitation. Then the acquired data quality was further evaluated via the number of metabolites detected, the coefficient variation (CV) distributions, and signal intensity distributions. The established platforms were eventually applied to differentiate four Lactobacillus species in identical growth conditions. The proposed workflow demonstrated the capability of targeted and untargeted metabolomics in differentiating closely-related bacterial strains/species.

Keywords: Lactobacillus species, bacterial metabolism, triple quadrupole mass spectrometry, Orbitrap mass spectrometry.

92 5.1 Introduction The Lactobacillus species are important groups of bacteria to control human intestinal microflora and preserve it in a normal state [1]. During the past decade, there has been a great interest to use Lactobacillus bacteria of human origin to treat human gastrointestinal disorders [2]. For example, Lactobacillus acidophilus can alter the level of human fecal, bacterial enzymes, such as decreased fecal bacterial β-glucuronidase and nitroreductase activities [3]. Some researchers found oral intake of another type of Lactobacillus species, Lactobacillus fermentum, can affect the immunologic response of an anti-influenza vaccine [4]. Lactobacillus reuteri has revealed efficacy in reducing both gingivitis and plaque in patients with moderate to severe gingivitis [5]. Another study also proved that exogenously-administered Lactobacillus delbrueckii stimulates expansion of T-cells and acidophilic granulocytes in the mucosa, without affecting integrity of the gut [6]. Thus, the identification of closely-related Lactobacillus species is crucial to the health effects studies of consuming these species of probiotics. It has long been recognized that the physiological and biochemical identification of Lactobacillus can be complicated and ambiguous [7]. Existing techniques such as DNA-DNA hybridization can be time-consuming and labor-intensive, and it improved our knowledge on the taxonomic relationships of Lactobacillus species [8]. In addition, the similarity in nutrition requirements among Lactobacillus species hampers the identification by traditional methods [9]. Therefore, a rapid and reliable method for identifying these four Lactobacillus species is needed. Metabolomics is a systems biology technique that has been widely applied in many biomedical applications, ranging from disease diagnostics to treatment monitoring [10]. During the past few decades, metabolomics and its related approaches have been developed at a dramatic rate. Mass spectrometry (MS)-based metabolomics has been developed to investigate metabolic characteristics, which help researchers to distinguish between the closely-related pathogenic and non-pathogenic bacterial strains [11]. The advantages, such as high analysis speed, sensitivity, and specificity make it possible for applying MS in bacteria detection and identification [12]. Selected reaction monitoring (SRM) mode analysis using triple quadrupole mass spectrometers has been the major technology for quantitation of small molecule detection in biological matrices because of its high selectivity and sensitivity [13]. This technology requires each compound of interest to be optimized on their tandem mass spectrometric parameters, include the correct precursor ions and optimization of collision energies used for collision-induced dissociation (CID). This process

93 is typically time-consuming and could be problematic when detecting compounds similar in both retention time and mass over charge ratios (m/z). [14]. On the other hand, a high-resolution mass spectrometer (HR-MS) is able to distinguish compounds with m/z differences < 5ppm, and CID is not essential for a HR-MS in analyte detection. High specificity of the HR-MS approach by applying an orbitrap mass spectrometer can be achieved thanks to the incredible resolution of instrument [13]. This approach also detects all the ionizable metabolites, instead of only analyzing selected ions in targeted lists, which may lead to the loss of useful information. A previous targeted metabolomics study [15] from our group used a HPLC-triple quadrupole mass spectrometer (LC-QQQ-MS) to evaluate the metabolic response of Lactobacillus strains to light exposure based on their metabolite patterns via targeted metabolic profiling. However, the limited number of metabolites detected in the past may not fully disclose the metabolic differences in closely-related Lactobacillus species/strains. Herein, we analyzed for targeted metabolites using a triple quadrupole mass spectrometer and untargeted metabolites using a linear ion trap-Orbitrap mass spectrometer (LC-Orbi-MS) in an effort to differentiate the metabolic profile of four closely- related Lactobacillus species/strains.

5.2 Materials and method Bacterial strain and growth condition Model bacterial strains, Lactobacillus acidophilus (ATCC 4356), Lactobacillus fermentum (ATCC 9338), Lactobacillus reuteri (ATCC 55730), and Lactobacillus delbrueckii (ATCC 11842), were purchased from American Type Culture Center (ATCC). The growth method for these strains was modified from our previous work [15]. Briefly, the bacterial strains were grown on Lactobacilli MRS agar (BD Diagnostics, Franklin Lakes, New Jersey) in an anaerobic chamber (Coy Laboratory Products Inc, Grass Lake, MI) at 37 °C for 24 h following manufacturer's instructions. A single colony was used to inoculate 5 mL MRS broth medium, and the culture was incubated for 24 h at 37 °C. Five testing cultures of each bacterial strain were prepared as replicates by transferring 50 µL of the overnight culture to 5 mL of MRS broth, and the cultures were incubated for another 24 h.

Metabolite extraction and sample preparation for LC-MS work

94 The extraction of intracellular metabolites was achieved by using a modified cold methanol extraction method as previously reported [15, 16]. Briefly, rapid bacteria/medium separation of each biological replicate was achieved by centrifugation for 5 min. The supernatant was discarded, and the resulting pellets were washed three times with 1 mL of phosphate-buffered saline (PBS). To the cell pellets, 250 µL of methanol and 50 µL of isotopically-labeled amino acid mixture, which was used as an internal standard (Cambridge Isotope Laboratories, Tewksbury, MA), were added, and the samples were mixed vigorously for 1 min. After storage at -20 oC for 20 min, 150 µL of the extracted supernatant was collected after centrifugation at 14,000 rpm for 5 minutes, and the supernatant was then dried on a vacuum concentrator. A mixture of 50 % acetonitrile and 50 % ultrapure water was used to reconstitute the sample. The final samples were loaded into liquid chromatography vials for analyses.

Targeted LC-MS/MS method for metabolic profiling The method to detect targeted metabolites was modified from our previous work [17]. Both positive and negative mode detections were performed on a Thermo Scientific TSQ Quantiva triple quadrupole mass spectrometer equipped with an electrospray ionization source, which was coupled to a Thermo Scientific Ultimate 3000 high performance liquid chromatography (HPLC) system equipped with a 2.1×150 mm, 2.5 µm hydrophilic interaction chromatography (HILIC) amide column (Waters Corporation, Milford, MA, USA). Metabolite separation was completed on an HPLC by eluting the extracted bacterial intracellular samples through the column. The reconstituted samples were gradient-eluted at 0.300 mL/min using solvents A (5 mM ammonium acetate in 90% water / 10% acetonitrile + 0.2% acetic acid) and B (5 mM ammonium acetate in 90% acetonitrile / 10% water + 0.2% acetic acid). The auto-sampler temperature was kept at 4 °C, the column compartment was set to 40 °C, and the separation time for each sample was 20 min. Retention time and SRM transition of targeted metabolites were established by running pure standards (purchased from Sigma, Saint Louis, MO, USA and IROA Technology, Boston, MA) and collecting the tandem mass spectrum (MS/MS), so the orthogonal information of retention time and two pairs of SRM transitions could be used to confidently detect and identify targeted compounds. Pooled quality control samples were also tested in between every ten samples to monitor the instrument stability.

95 Untargeted LC-HRMS method for metabolic profiling The untargeted metabolite analysis was performed following the procedure of Liu and colleagues [18] with modification. Briefly, a Thermo Scientific LTQ Orbitrap XL™ hybrid ion trap-orbitrap mass spectrometer equipped with an electrospray ionization probe (Thermo Fisher Scientific, San Jose, CA, USA) was applied to both positive and negative mode compound detection, which coupled with the same HPLC and column as we used in the targeted metabolite analyses. The major parameters are as listed: heater temperature, 120 °C; sheath gas, 30; auxiliary gas, 10; sweep gas, 3; spray voltage, 3.6 kV for positive mode, and 2.5 kV for negative mode. Capillary temperature was set to 320 °C, and S-lens was 55 V. A full scan ranged from 50 to 800 (m/z). The resolution was set to 60,000. Automated gain control was set to 3 × 104 in full scan, 1 × 104 in SIM, 1 × 104 in MSn, 3000 in zoom for ion trap. For Fourier transformations, we utilized 3 × 106 in full scan, 1 × 105 in SIM, and 1 × 105 in MSn [19]. Customized mass calibration was performed before every batch of samples.

Data analysis

The Quanbrowser module of Xcalibur 4.0 was used to manually process targeted metabolite profiling data from the HPLC-MS/MS experiments. Acquired peak intensities were normalized to corresponding optical density values of the bacterial cultures. MetaboAnalyst 4.0 was used for statistical analyses (http://www.metaboanalyst.ca/). Peak intensities were subjected to a log transform and auto-scaling (mean-centered and divided by the standard deviation of each variable) to achieve a normal distribution. ANOVA module, a principal component analysis module, and heatmap module were used for data analyses and visualizations. Compound discoverer 2.1 (Thermo Scientific) was used to process untargeted metabolite data. The data were filtered as follows: precursor mass range, 50-1000 Da; mass tolerance, 10 ppm; minimum peak intensity, 5000; retention time tolerance, 1 min. Metabolic features fell between RT 0-11 min.

5.3 Results The combination metabolomics workflow developed in this study is shown in Figure 5.1. Polar metabolites were extracted (Figure 5.1, step 1) from the bacterial cultures. Liquid

96 chromatography was then coupled with a TSQ triple quadrupole MS (Step 2) or a LTQ linear ion trap-Orbitrap MS (step 3) for targeted or untargeted metabolite analyses, respectively. For the targeted metabolite study, a rapid polarity switch mode was applied to increase the metabolite coverage. The LTQ does not allow for the rapid polarity switch between positive and negative mode, so two injections are normally required for coverage of both detection polarities. A 20 min chromatography run was employed. The targeted metabolomics analyses were conducted as steps 4 and 6 (Figure 5.1). Targeted peaks were extracted from chromatography, and all the peaks were manually inspected by Quanbrowser module of Xcalibur version 4.0 (Thermo Fisher Scientific). The detailed list of metabolites detected in this study and their HPLC–MS/MS experimental conditions are reported in supporting information (Table S5.1). Normalized data were subjected to MetaboAnalyst 4.0, and statistical analyses were applied. Step 5 and 7 showed untargeted metabolite analysis, which was performed with no pre-existing knowledge of metabolites to be measured [18]. Figure 5.2a shows the general steps of data acquisition, targeted metabolites were extracted, as identified by their chromatography peaks, and statistical analyses were based on peak intensities where filtering process of “extra” metabolites was applied to obtain untargeted metabolic features. Targeted metabolites were identified using their precursor ions and fragmentations in specified collision energies (determined by analyzing chemical standards first), while accurate masses were used for identifying untargeted metabolites. Multiple data filtering steps were applied for untargeted metabolic data analysis (Figure 5.2b). The multiple criteria, including the peak intensity (integrated peak area), coefficient variation (CV) within biological replicates, the retention time of extracted peak and multiple database search result, were used. The accurate HR mass spectra were annotated by chemical formula and database matching at a 10 ppm mass tolerance. The putative identity of a compound was then generated. Detailed putative compound identities are reported in Table S5.2 (an example of putatively-identified compound can be seen in Fig. 5.3.).

Figure 5.1 Overview of polar metabolites platform. Biological samples were separated and analyzed by either LC-QQQ-MS (2) or LC-Orbi-MS (3) for targeted and untargeted metabolites analysis, respectively. The targeted study was applied to MetaboAnalyst 4.0 online for manual integrated data analyses (4), while Compound discoverer 2.1 was used for untargeted method studies (5). Statistical analyses of targeted and untargeted data (7 and 8) were performed to identify Lactobacillus species via metabolic profiling.

We next compared the overlapping metabolites/metabolic features detected by the two established platforms. This approach was achieved by applying the same samples under the same chromatography conditions for two platforms. A total of 44 metabolites from targeted metabolomics workflow exhibited detectable counterparts from the untargeted workflow with

98 mass differences less than 0.5 Da and correlations greater than 0.5 in the four bacterial samples. One representative compound, leucine/isoleucine, is shown as an example in Figure 5.3. The isomer pair is presented because our chromatography system could not separate the isomers. The targeted compound showed a sharp peak at a retention time of 1.94 min with fragments of m/z = 44.18 and m/z = 86.18, which confirms the identity of leucine/isoleucine. The same sample was scanned in the untargeted platform, and the compound molecular weight of 131.0954 g/mol corresponded to a sharp peak at a retention time of 1.92 min (Figure 5.3b); the detected molecular weight showed a mass accuracy within 5 ppm compared to the exact mass of leucine/isoleucine. The full scan MS spectrum (positive mode with isotopic patterning) is shown as an insert in Figure 5.3.

Figure 5.2 Triple quadrupole MS and Orbitrap MS data analyses workflow. (a) Workflow for targeted and untargeted metabolomics study and (b) specific workflow for untargeted metabolic feature peaks filtering. Values in brackets are number of metabolic features from untargeted platform after applying each filtering step.

99 The quantitative abilities of the two platforms were compared and shown in Figure 5.4. In our study, the dynamic range of the instrument was demonstrated by detecting a series of diluted, pooled quality control (QC) samples. Overall, 151 metabolites from targeted platform and 142 metabolic features from untargeted platform exhibited good correlations between sample concentrations and peak intensities with R2> 0.9. Figure 5.4a shows the selected ion chromatogram (SIC) of leucine/isoleucine in the positive mode of the targeted metabolite study. The compound was chosen because it exhibited the highest peak intensity in untargeted analyses. As the concentration of sample increased, the relative peak intensity increased following the same trend with R2 > 0.98 (the Y-axis is the relative peak abundance, the absolute peak intensities are given next to each of these peaks). Figure 5.4b shows that the LC-Orbi-MS used for the untargeted workflow yielded a similar intensity vs concentration relationship for leucine/isoleucine compared to the triple quadrupole analysis and a R2 > 0.99 within the dynamic range. Figures 5.5a and 5.5c demonstrated the MS intensity range across metabolites/metabolic features detected from either the targeted or untargeted platforms. Figure 5.5b and 5.5d demonstrate a correlation between CV (%) and MS intensity within biological replicates (N = 5 per group). All compounds detected in the targeted platform and the metabolic features after critical filtering but before database annotation from the untargeted platform were used in the analysis. Targeted metabolites (177/234) and 86/130 untargeted metabolic features detected from Lactobacillus acidophilus showed an average CV < 10% within five biological replicates.

100 a

Exact mass=131.0946 MW=131.0954

Figure 5.3 A representative spectral comparison of the same compound between targeted and untargeted platforms. (a) Leucine/isoleucine detected from the targeted metabolomics approach, the insert showed the top two SRM transitions. (b) The compound with MW = 131.09542 g/mol from untargeted metabolomics approach, which showed identical RT and a mass accuracy within 5 ppm with leucine/isoleucine. The insert showed high-resolution mass spectrum of m/z 132.1028 (M+H) and its isotopic peaks. In this study, a total of 20 targeted metabolites were matched with HRMS data with a mass accuracy <10 ppm. The detailed MS spectrum is shown in Figure S5.1.

101

The platforms were then applied to study the metabolic profiles in four anaerobic bacterial strains, Lactobacillus acidophilus, Lactobacillus fermentum, Lactobacillus reuteri, and Lactobacillus delbrueckii. Each bacterial strain was cultured in an identical growth environment to avoid any variation in metabolism due to environmental factors, such as nutrition, oxygen level, and temperature. Compounds (234) were detected from the targeted platform, while 130 putatively- identified metabolites were obtained after the critical filtering process demonstrated in Figure 5.2b. Metabolites (20) from both platforms were identified as the same, while the rest were unique in each study. Quantile normalization was applied to the metabolite intensity (Figure 5.6). All the bacterial strains showed similarities in the peak intensity range within the targeted platform (Figure 5.6a) and untargeted platform (Figure 5.6b). The result illustrates the of performance stability of both instruments when analyzing bacterial samples.

Figure 5.4 Analytical performance of both MS-Metabolomics workflows. (a) Selected ion chromatogram (SIC) of leucine/isoleucine, an increasing concentration of the sample was used. (b) SIC of leucine/isoleucine from the untargeted panel (most abundant), Figure legends 1 to 7 indicate the decreasing folds of sample concentrations used by a factor of 2 where 1 is the highest concentration.

102 Figure 5.5 Overview of data quality. The log-transformed intensity vs. number of untargeted (a) and targeted (c) metabolites from Lactobacillus acidophilus, n=5. MS intensity vs. coefficient variation (CV) were evaluated and the untargeted platform result showed in (b) where targeted result showed in (d).

Figure 5.6 MS intensity distribution of four anaerobic bacterial strains from targeted (a) and untargeted (b) metabolomics workflow.

103 Figure 5.7 PLS-DA analysis of targeted and untargeted metabolomics study. (a) Targeted study and (b) untargeted study.

Figure 5.8 Heat map of targeted and untargeted metabolomics study. (a) Targeted study and (b) untargeted study.

104 We also conducted partial least squares discriminant analyses (PLS-DA) (Figure 5.7) to the normalized metabolites data extracted from the bacterial strains to analyze the differences in the metabolic profile among bacterial strains. As seen in Figure 5.7, both targeted platform (Figure 5.7a) and untargeted platform (Figure 5.7b) showed four major clusters of samples generated by the Lactobacillus strains, which indicates the success in differentiating Lactobacillus strains by their unique metabolic profiles. The ovals of different colors indicate the 95% confidence region, and there were 182/234 and 117/130 metabolites from the targeted and untargeted platform, respectively. These compounds were significantly-different in comparison among the four Lactobacillus species by applying one-way analysis of variance (ANOVA) test. Twenty metabolites that were shared by these two platforms were significantly-different; specifically phenethylamine, nicotinamide, proline, valine, leucine/isoleucine, aspartic acid, serine, asparagine, glucosamine, tyrosine, tryptophan, phenylalanine, ornithine, lysine, glutamine, desthiobiotin, citrulline, cytosine, creatinine, and adenine. Figure 5.8 demonstrates the metabolic profile (top 50 based on ANOVA) heatmap of Lactobacillus species. The horizontal axis represents the group information, while the vertical axis represents individual metabolites. The colors indicate the expression levels of each metabolites, as dark red means highest abundance and dark blue indicates the lowest abundance of the metabolite in the sample. The heat map analysis (top 50) reveals that the four Lactobacillus species generally clustered in their own groups.

5.4 Discussion Previous studies have shown the impact of Lactobacillus strains on the metabolism of food in the human intestine [20]. Others also proved that probiotics, such as Lactobacillus fermentum, contribute to ammonia and protein metabolism as well as altering the gut microbiota community [21]. In order to assess the efficacy of Lactobacillus in humans, a thorough understanding of Lactobacillus species is required as each species is unique [22]. Because the metabolome of an organism can be uniquely expressed to fit the needs of that organism, metabolomics approaches can be useful tools in the identification and differentiation of different Lactobacillus strains [23]. Techniques, such as ultra-performance liquid chromatography (UPLC), LC-MS, high- resolution magic angle spinning-nuclear magnetic resonance spectroscopy (HRMAS-NMR) coupled with complex, statistical data analyses, allows for the detection of multiple metabolites [21]. The widespread adoption of LC-QQQ-MS for quantitative analyses in metabolomics research

105 is due to its high specificity, which is realized by using the SRM mode [24]. While powerful, limitations of triple quadrupole based, targeted metabolomics analysis, such as low resolution and a limited number of compounds in targeted panel, also exist. High-resolution analysis, such as analysis using LC-Orbi-MS instrument, is a complementary approach that can be used to analyze compounds that do not fragment favorably in LC-QQQ-MS [24]. Additionally, the Orbitrap MS achieves many of the requirements that are needed for quantitative analyses, such as excellent mass accuracies, good dynamic ranges, decent sensitivities, and high selectivities [25]. In our study, both a triple quadrupole mass spectrometer platform and an orbitrap mass spectrometer platform were tested for their robustness in performance. Some of the identical compounds were detected by both platforms at similar signal levels, while each of the platforms provided a substantial amount metabolite/metabolic feature information that is unique to each platform. This result demonstrates the complementary nature of compound detection and identification with the two techniques. Also, on both platforms, increasing peak intensities were observed with increasing sample concentrations for a majority of the detected metabolites/metabolic features. Additionally, the CVs of compounds that showed sufficient peak intensities are low among multiple biological replicates, which demonstrated great reproducibility for both targeted and untargeted platforms. Distinct metabolic patterns were exhibited by these four Lactobacillus strains based on their metabolic profiles from both platforms. The PLS-DA and heatmap data indicated that the 4 Lactobacillus species displayed unique metabolic fingerprints when they were cultured in identical media, which is likely due to their differential expression levels of metabolic enzymes, as reported by Kralj and Tieking [26, 27]. Twenty metabolites, which contributed to the identification of four bacterial strains, were detected by both platforms. However, not all of the 20 metabolites were present at the same levels in all Lactobacillus strains. However, on the other hand, the remaining 110 putatively-identified metabolites from the untargeted platform could be used to provide insight to the results from the targeted method. With appropriate validation studies, the combination approach has the potential to expand the existing targeted metabolite detection panel in the future.

5.5 Conclusions In summary, our study used both targeted and untargeted approaches to analyze polar metabolites from anaerobic bacteria to systematically differentiate four Lactobacillus species based on their metabolic profile. The identical peaks detected from both instruments show the

106 precision of peak detection in both LC-MS systems, and the good linear relationships of each platform provided confidence in using both instruments for quantitative analyses. The applications of these platforms on four Lactobacillus strains showed the great potential of combination metabolomics approaches in rapidly differentiating closely-related bacterial strains. We will continue to work on both expanding these methods to many other bacterial strains and validating their utility in more complicated bacterial growth conditions in our future studies.

ASSOCIATED CONTENT Supporting information This material is available in Appendix C. Acknowledgments The authors thank Miami University for financial support for this project (startup fund to JZ). We thank Dr. Theresa Ramelot for her help in linear ion trap-Orbitrap mass spectrometer experiments.

107 Reference

[1] Song, Y.L., Kato, N., Liu, C.-X., Matsumiya, Y., Kato, H., and Watanabe, K., Rapid identification of 11 human intestinal Lactobacillus species by multiplex PCR assays using group- and species-specific primers derived from the 16S–23S rRNA intergenic spacer region and its flanking 23S rRNA. FEMS Micro Lett, 2000. 187(2): p. 167-173. [2] Isolauri, E., Rautanen, T., Juntunen, M., Sillanaukee, P., and Koivula, T., A human Lactobacillus strain (Lactobacillus casei sp strain GG) promotes recovery from acute diarrhea in children. Pediatrics, 1991. 88(1): p. 90-97. [3] Goldin, B.R., Swenson, L., Dwyer, J., Sexton, M., and Gorbach, S.L., Effect of Diet and Lactobacillus acidophilus Supplements on Human Fecal Bacterial Enzymes23. JNCI: J Natl Cancer I, 1980. 64(2): p. 255-261. [4] Olivares, M., Díaz-Ropero, M.P., Sierra, S., Lara-Villoslada, F., Fonollá, J., Navas, M., Rodríguez, J.M., and Xaus, J., Oral intake of Lactobacillus fermentum CECT5716 enhances the effects of influenza vaccination. Nutrition, 2007. 23(3): p. 254-260. [5] Krasse, P., Carlsson, B., Dahl, C., Paulsson, A., Nilsson, A., and Sinkiewicz, G., Decreased gum bleeding and reduced gingivitis by the probiotic Lactobacillus reuteri. Swed Dent J, 2006. 30(2): p. 55-60. [6] Picchietti, S., Fausto, A.M., Randelli, E., Carnevali, O., Taddei, A.R., Buonocore, F., Scapigliati, G., and Abelli, L., Early treatment with Lactobacillus delbrueckii strain induces an increase in intestinal T-cells and granulocytes and modulates immune-related genes of larval Dicentrarchus labrax (L.). Fish Shellfish Immun, 2009. 26(3): p. 368-376. [7] Sneath, P.H., Mair, N.S., Sharpe, M.E., and Holt, J.G., Bergey's manual of systematic bacteriology. Volume 2. 1986: Williams & Wilkins. [8] Song, Y.-L., Kato, N., Matsumiya, Y., Liu, C.-X., Kato, H., and Watanabe, K., Identification of and hydrogen peroxide production by fecal and vaginal lactobacilli isolated from Japanese women and newborn infants. J Clin Microbiol, 1999. 37(9): p. 3062-3064. [9] Ampe, F., Omar, N.B., and Guyot, J.P., Culture‐independent quantification of physiologically‐active microbial groups in fermented foods using rRNA‐targeted

108 oligonucleotide probes: application to pozol, a Mexican lactic acid fermented maize dough. J Appl Microbiol, 1999. 87(1): p. 131-140. [10] Lindon, J.C., Holmes, E., and Nicholson, J.K., Metabonomics Techniques and Applications to Pharmaceutical Research & Development. Pharm Res, 2006. 23(6): p. 1075-1088. [11] Dettmer, K., Aronov, P.A., and Hammock, B.D., Mass spectrometry‐based metabolomics. Mass Spectrom Rev, 2007. 26(1): p. 51-78. [12] Sauer, S. and Kliem, M., Mass spectrometry tools for the classification and identification of bacteria. Natr Rev Micro, 2010. 8: p. 74. [13] Hugues, H., Reza, S.H., Olaf, S., Maciej, B., B., N.S., and Bertrand, R., Comparison between a high‐resolution single‐stage Orbitrap and a triple quadrupole mass spectrometer for quantitative analyses of drugs. Rapid Commun Mass Sp, 2012. 26(5): p. 499-509. [14] Majumdar, T.K., Vedananda, S., and Tse, F.L., Rapid analysis of MMI270B, an inhibitor of matrix metalloproteases in human plasma by liquid chromatography–tandem mass spectrometry: matrix interference in patient samples. Biomed Chromatogr, 2004. 18(2): p. 77-85. [15] Xu, M., Zhong, F., and Zhu, J., Evaluating metabolic response to light exposure in Lactobacillus species via targeted metabolic profiling. J Microbiol Methods, 2017. 133: p. 14-19. [16] Schelli, K., Rutowski, J., Roubidoux, J., Zhu, J., Staphylococcus aureus methicillin resistance detected by HPLC-MS/MS targeted metabolic profiling. J Chromatogr B Analyt Technol Biomed Life Sci, 2017. 1047: p. 124-130. [17] Yang, K.D., M. L. Zhu, J., Metabolomics Study Reveals Enhanced Inhibition and Metabolic Dysregulation in Escherichia coli Induced by Lactobacillus acidophilus- Fermented Black Tea Extract. J Agric Food Chem, 2018. 66(6): p. 1386-1393. [18] Liu, X., Ser, Z., and Locasale, J.W., Development and quantitative evaluation of a high- resolution metabolomics technology. Anal Chem, 2014. 86(4): p. 2175-84. [19] Ugurlu, A., Karahasan Yagci, A., Ulusoy, S., Aksu, B., and Bosgelmez-Tinaz, G., Phenolic compounds affect production of pyocyanin, swarming motility and biofilm formation of Pseudomonas aeruginosa. Asian Pac J Trop Biomed, 2016. 6(8): p. 698-701.

109 [20] Rabot, S., Rafter, J., Rijkers, G.T., Watzl, B., and Antoine, J.-M., Guidance for Substantiating the Evidence for Beneficial Effects of Probiotics: Impact of Probiotics on Digestive System Metabolism–3. J Nutrition, 2010. 140(3): p. 677S-689S. [21] Mozzi, F., Ortiz, M.E., Bleckwedel, J., De Vuyst, L., and Pescuma, M., Metabolomics as a tool for the comprehensive understanding of fermented and functional foods with lactic acid bacteria. Food Res Int, 2013. 54(1): p. 1152-1161. [22] Corcionivoschi, N., Drinceanu, D., Stef, L., Luca, I., and Julean, C., Probiotics- identification and ways of action. Innov Rom Food Biotech, 2010. 6: p. 1. [23] Dauchy, R.T., Dauchy, E.M., Tirrell, R.P., Hill, C.R., Davidson, L.K., Greene, M.W., Tirrell, P.C., Wu, J., Sauer, L.A., and Blask, D.E., Dark-phase light contamination disrupts circadian rhythms in plasma measures of endocrine physiology and metabolism in rats. Comparat Med, 2010. 60(5): p. 348-356. [24] Bateman, K.P., Kellmann, M., Muenster, H., Papp, R., and Taylor, L., Quantitative– Qualitative Data Acquisition Using a Benchtop Orbitrap Mass Spectrometer. ACS Mass Spectr 2009. 20(8): p. 1441-1450. [25] Makarov, A., Electrostatic axially harmonic orbital trapping: a high-performance technique of mass analysis. Anal Chem, 2000. 72(6): p. 1156-1162. [26] Krajl, S., van Geel-Schutten, G., Van Der Maarel, M., and Dijkhuizen, L., Efficient screening methods for glucosyltransferase genes in Lactobacillus strains. Biocatal Biotransfor, 2003. 21(4-5): p. 181-187. [27] Tieking, M., Korakli, M., Ehrmann, M.A., Gänzle, M.G., and Vogel, R.F., In situ production of exopolysaccharides during sourdough fermentation by cereal and intestinal isolates of lactic acid bacteria. Appl Environ Microb, 2003. 69(2): p. 945-952.

110

CHAPTER 6 Analysis of Barrel-Aged Kentucky Bourbon Whiskey by Ultrahigh Resolution Mass Spectrometry

Kundi Yang1, Arpad Somogyi2, Caitlyn Thomas1, Huan Zhang1, Zishuo Cheng1, Shenyuan Xu1, Callie Miller1, Devin Spivey1, Colin Blake3, Clay Smith3, David Dafoe3, Neil D. Danielson1 and Michael W. Crowder1*

1Department of Chemistry & Biochemistry, Miami University, Oxford, OH, 45056 2Campus Chemical Instrumentation Center, Mass Spectrometry and Proteomics Facility, The Ohio State University, Columbus, OH, 43210 3Moonshine University and Grease Monkey Distillery, Louisville, KY

Contributions to the chapter. Arpad Somogyi helped with the FT-ICR MS analysis and the LC/MS experiments were completed with the help of Caitlyn Thomas, Huan Zhang, Zishuo Cheng, Callie Miller and Devin Spivey. The bourbon samples were made and provided by Colin Blake, Clay Smith and David Dafoe. The data analyses were performed by Kundi Yang with the help of Shenyuan Xu. This chapter was written by Kundi Yang and edited by Neil D. Danielson and Michael W. Crowder.

This work appeared in Food Analytical Methods (2020), 1-11. Reprinted (adapted) with permission from the Springer Nature, Copyrights 2020.

111 Abstract In an effort to characterize differently-aged bourbons and to determine whether bourbons could be “fingerprinted” by their chemical compositions, we used Fourier-transform ion cyclotron resonance mass spectrometry (FT-ICR MS) and liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) to characterize 2 distinct sets of bourbon samples. The first set of bourbons were prepared using the same mash bill but were aged-differently (unaged (0 years), 2 years, 4 years, and 6 years) in oak barrels. The results showed an increase in the number of chemical compounds present as the bourbon ages. Most of the large changes in chemical composition occur in the first two years of aging. We also analyzed single barrel bourbons, which were produced identically but maturated in different, adjacent barrels, and the results suggested that significant differences exist among these samples. These results suggest that “fingerprinting” of different bourbons for authentication purposes may be complicated and that careful analyses, coupled with more comprehensive identification of chemical compounds in bourbons, are needed.

Keywords: Fourier-transform ion cyclotron resonance (FT-ICR), Orbitrap, mass spectrometry, bourbon whiskey, barrel aging.

112 6.1 Introduction Whiskey is one of the largest classes of spirits sold in the world, and it is produced in Scotland (major form is called Scotch whisky), Ireland (major form is called Irish whiskey), Japan (Japanese Whisky), India, Canada (Canadian whisky), the United States (Bourbon Whiskey), and in many other countries [1]. One of the largest producers of whisky in the world is Scotland, and Scotch has been analyzed with sensory analyses [2] and many analytical techniques, such as liquid chromatography (LC) and gas chromatography (GC) coupled with flame ionization detection (FID) or mass spectrometry (MS) [3-6]. Recently, Kew et al. utilized Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR-MS) to analyze 85 different Scotch samples and identified thousands of compounds in these complex samples [7]. Roullier-Gall et al. subsequently utilized FT-ICR-MS and LC-MS/MS to analyze 150 different whisky samples from 49 different distilleries and reported, not surprisingly, that the history of the wood used in the barrel had an impact on the chemical composition of the resulting barrel-aged spirit [8]. Elaborate statistical analyses of the resulting spectra were reported; however, conclusions were complicated by the differences between the Scotch whisky samples, which were caused by different aging barrels, different peats/malts, and different styles. Bourbon, America’s whiskey as decreed by Congress in 1964, must be produced in the United States, made from a mash bill that is at least 51% corn, aged in new, charred oak containers, distilled to no more than 80% alcohol by volume (ABV), entered into the oak barrel at no more than 62.5% ABV, and bottled at 40% ABV or more [9-13]. Despite relatively strict guidelines to produce this spirit, there are many different types of bourbons with different flavors, colors, and %ABV. Differences in bourbons are due to different mash bills (in addition to the required at least 51 percent corn, there are variable barley, malts, wheat, and rye compositions), different yeasts, different proofs of the hearts collected off of the stills and of the distillates when entered into the barrel, different aging times and locations of the barrels in the rick houses, and if the bourbon is conditioned in alternative barrels after the oak barrels [13]. Despite the large number of different bourbons, there have been far fewer analytical studies done on bourbons as compared to Scotch whiskies. LC-MS [14] and GC-olfactometry [15] have been used to analyze bourbons. GC-MS [16] appears to be the industry-standard, although there are more recent application notes describing the use of LC-MS with data analyses for bourbon profiling [3, 8].

113 In this work, we report the use of FT-ICR-MS and LC-ultrahigh resolution mass spectrometry (LC-Orbitrap MS) to characterize bourbon samples, which all had the same mash bill and that were aged 0 years (unaged), 2 years, 4 years, and 6 years in oak barrels. This work is complementary to studies recently published by Heinz and Elkins, who reported GC-MS studies on volatile compounds in unaged and aged whiskies [15], and Collins et al. who used UHPLC- QTOF MS to examine a number of whiskey samples [3]. Based on these previously published studies, we hypothesized that FT-ICR-MS and LC-ultrahigh resolution mass spectrometric techniques are sensitive enough to detect hundreds of chemical compounds in different bourbon samples, allowing us to probe the chemistry occurring during barrel aging. Our analyses suggest that the chemical compounds that appear in the bourbon samples as they age vary, in part, on different barrels. Therefore, we also characterized three 8 year aged bourbons, all from the same mash bill, but stored in different barrels that were adjacent to each other in the rick house. Our results strongly suggest that use of high-resolution mass spectrometry to “fingerprint” bourbons for authenticity and for signs of adulteration will be very challenging, given the large differences in contributions from the barrels.

6.2 Materials and Methods Whiskey samples Four bourbon whiskey samples were provided by Moonshine University: (1) unaged bourbon (0 year), (2) bourbon that had been aged 2 years in an oak barrel (2 year), (3) whiskey that had been aged 4 years in an oak barrel (4 year), and (4) whiskey that had been aged 6 years in an oak barrel (6 year). In addition, three bourbon samples that had been aged for 8 years in different barrels in the same rick house were also provided by Moonshine University. Information about the samples are provided in Table S1. Samples were directly collected from barrels and stored in 100 mL amber vials at room temperature prior to analysis. LC-MS grade methanol and water were purchased from Sigma Aldrich (Sigma, Saint Louis, MO, USA), and 30 μL of each bourbon sample were diluted 1:10 in methanol:water (50:50) immediately prior to direct infusion into the ESI-MS source. The dilution ratio was optimized to minimize potential carryover effects.

FT-ICR-MS Metabolome Profiling

114 Ultrahigh resolution mass spectra were acquired on a Bruker SolariX Fourier transform ion cyclotron resonance mass spectrometer (FT-ICR-MS) (Bruker Daltronics GmbH, Bremen, Germany) equipped with a 15 Tesla superconducting magnet (Magnex Scientific Inc., Yarnton, GB) and an Apollo II ESI source (Bruker Daltonics GmbH, Bremen, Germany) operated in the negative ionization mode. Negative ionization has already been shown to be the preferred ionization mode in fingerprinting wines by FT-ICR-MS [16]. Each whiskey sample (1 mL) was collected and centrifuged (10,000 rpm, 5 min). Each supernatant (100 µL) was diluted with 900 µL methanol and then direct injected into the FT-ICR instrument. Samples were introduced into the electrospray source at a flow rate of 120 μL/h. FT-ICR-MS was externally calibrated by using an Agilent calibration mixture (Agilent Technologies, Santa Clara, CA, USA) diluted ten-fold with acetonitrile:water 1:1. Further internal calibration was performed for each sample, yielding mass accuracies lower than 0.1 ppm in routine day-to-day measurements.

Untargeted LC-HRMS method for metabolic profiling The untargeted metabolite analysis was performed following the procedure from Roullier et al. with modification [8]. Briefly, a Thermo Scientific LTQ Orbitrap XLTM hybrid ion trap- orbitrap mass spectrometer equipped with an electrospray ionization probe (Thermo Fisher Scientific, San Jose, CA, USA) was applied to negative mode compound detection, which was coupled with a Phenomenex® Kinetex C18 reverse phase column (column dimensions: 100 × 2.1 mm; particle size: 2.6 µm). Under optimized conditions, the column oven was thermostated to 40 oC. Eluent A consisted of 10 percent acetonitrile (ACN) in water and Eluent B of 100 percent ACN, both containing 0.1 percent formic acid. Flow rate was set to 0.25 mL/min. The major parameters for MS scanning were: sheath gas, 30; auxiliary gas, 10; sweep gas, 3; and spray voltage, 2.5 kV for the negative mode. Capillary temperature was set to 275 °C, and S-lens was 55 V. A full scan ranging from 50 to 1000 (m/z) was used. The resolution was set to 60,000. Automated gain control was set to 3×104 in full scan, 1×104 in SIM, 1×104 in MSn, and 3000 in zoom for ion trap, 3×106 in full scan, 1×105 in SIM, and 1 × 105 in MSn for Fourier transform. Customized mass calibration was performed before every batch of sample analysis. Each whiskey sample (1 mL) was collected and centrifuged (10,000 rpm, 5 min). Each supernatant (100 µL) was diluted with 900 µL methanol and then load in the auto-sampler of the HPLC system.

115 Data analysis Spectra were calibrated and peak picked in Data Analysis 4.4 (Bruker Daltonics GmbH, Bremen, Germany). Spectra were peak picked with a signal to noise ratio (S/N) threshold of 4 and a minimum absolute intensity of 2 × 106, a value based on visual inspection of the data. Assignment was set with the elemental limits C0-100 H0-200 O1-20 N1-20 S0-1 with a maximum error threshold of 1 ppm and a minimum of 15 species per class. Only singly-charged species were observed and assigned. MetaboAnalyst 4.0 was used for statistical analysis (http://www.metaboanalyst.ca/). Relative peak intensities were subjected to a log transform and auto-scaling (mean- centered and divided by the standard deviation of each variable) to achieve a normal distribution. ANOVA module, a principal component analysis module, and heatmap module were used for data analysis and visualization by MetaboAnalyst 4.0. Compound discoverer 2.1 (Thermo Scientific) was used to process untargeted metabolite data. The data were filtered as follows: precursor mass range, 50–1000 Da; mass tolerance, 5 ppm; minimum peak intensity, 10,000; retention time tolerance, 1 min. Group coefficient variance (CV) was set at 25 percent in any sample group. Metabolic features fall between RT 0–35 min was displayed at the chromatogram and MS spectrum. All detected molecular weights from LC-MS were matched with ChemSpider database by their full mass with a mass tolerance of 10 ppm and m/z Cloud database by their MS2 patterns to give the potential compound IDs.

6.3 Results and Discussion FT-ICR MS analysis of different aged bourbon samples In order to explore the nature and the chemical composition of each different-aged bourbon sample, all samples were first analyzed by direct infusion on a 15T FT-ICR MS instrument. Negative ion mode was chosen since it has been reported to provide a higher number of resolved ion signals in the m/z 100-1000 mass range than the positive ion mode [17]. About 50 singly- charged mass signals within the mass range of m/z 200-800 with a signal to noise ratio of greater than or equal to 4 (S/N >4) were detected in the unaged (0 year) bourbon sample (Figure 6.1A – top spectrum), and over 240 similar signals were detected in the 2, 4, and 6 year bourbon samples (Figure 6.1A – bottom three spectra, respectively). It is not surprising that there were many more compounds in the bourbon samples that had been aged in a barrel since it is well known that compounds from the barrel are released into the spirit as it ages [8, 18]. Visual comparison of the

116 four spectra in Figure 6.1A shows a number of clear differences: (1) the overall number of peaks in the spectra increased from unaged < 2 year < 4 year < 6 year samples, (2) the peaks at m/z 255.2330, 318.9798, and 362.9696 are larger in the spectrum of the unaged (0 year) bourbon than in the spectra of the aged bourbons (see Table S6.2 for a list of other peaks observed in higher amounts in the unaged sample), and (3) new, relatively large peaks (at m/z 300.9990, 487.1821, and 517.3171) emerged in the spectra of the aged spirits but were absent in the spectrum of the unaged bourbon. Our database analyses predict that the peak at m/z 255.2330 has a mass that is consistent with 2-methylpentadecanoate (C16H31O2), which is a fatty acid most likely from yeast [19], but we were unable to unambiguously assign the peaks at m/z 318.9798 and m/z 362.9696. Similarly, the compounds corresponding to the m/z 300.9990 and m/z 487.1821 are tentatively assigned as ellagic acid and 4-allyl-2-methoxyphenyl-6-O-b-D-glucopyranosyl-b-D- glucopyranoside, respectively; the peak at m/z 517.3171 peak could not be unambiguously assigned but has a mass consistent with a chemical formula of C30H45O7. Ellagic acid is one of the predominant polyphenols in barrel-aged wines [20], and the compound corresponding to the m/z of 487.1821 is likely a substituted disaccharide that originated from the charred barrel.

Figure 6.1 FT-ICR-MS spectra of bourbon samples aged (top to bottom) from 0, 2, 4, and 6 years in oak barrels. (A) Spectra from m/z 200–800 (note that the y-axis scale is larger for the spectrum of the unaged bourbon than the other three spectra so that the peaks at m/z 318.980 and 362.970 would be on scale). (B) Enlargements of the areas showing the nominal masses of m/z 401, with the molecular formula assignments.

Following a similar approach as previously reported [8], we expanded the FT-ICR-MS spectra into single m/z regions (in this case, for quasi isobaric ions with a nominal mass of m/z

117 401), which allowed for a more complete visual comparison of the bourbon samples in this region of the spectrum (Figure 6.1B). Peaks at m/z 401.0514, 401.0878, 401.1089, 401.1242, and 401.1606 increased in intensities for the aged bourbons, as compared to the unaged sample, and have masses consistent with chemical formulas of shoyuflavone B (C19H14O10), 3-(4-acetoxy-1- hydroxy-2-butanyl)-4,5,7-trihydroxy-9,10-dioxo-9,10-dihydro-2-anthracenolate (C20H17O9), 4- hydroxy-5-(3',5'-dihydroxyphenyl)-valeric acid-O-glucuronide (C17H22O11), nobiletin (C21H22O8), and ethyl 3-[4-(cyclopentyloxy)phenyl]-3-[3-hydroxy-6-(hydroxymethyl)-4-oxo-4H-pyran-2- yl]propanoate (C22H27O7), respectively. Shoyuflavone B has been found in certain spices and in fermented soy sauce and probably was produced in fermentation [21]. Nobiletin is a flavonoid that is found in citrus peels; it is likely that this compound, or its parent, originated from the barrel [22]. While none of the other compounds have been reported to be found in barrel-aged whiskies, they all have hydroxylated ring systems, reminiscent of polyphenols, and it very likely that they originated from the barrel. Peaks at m/z 401.2180 and 401.2545 decreased in intensity for the aged bourbons and have m/z values consistent with 4-acetoxy-2-{2-[(3,4-diethyl-3-octanyl)oxy]-2-oxoethyl}-2-hydroxy-4- oxobutanoate and tris(3-methylbutyl) citrate, respectively. Both compounds have backbone structures reminiscent of citrate and were most likely formed during fermentation. Predicted molecular formula are given for all of the peaks in Figure 6.1B, and analyses of other expanded FT-ICR-MS spectra are in progress. While there are subtle differences in the spectra for the 2, 4, and 6 year aged bourbons (note undetectable ellagic acid in the unaged sample, a dramatic increase in the 2 year old sample, and only a 20 percent increase from the 2 year old sample to the 4 and 6 year samples (Figure S6.1)), we are unable to draw unambiguous conclusions about these differences since all of the bourbon samples were produced at different times and/or were aged in different barrels. It is clear that most of the new compounds in aged bourbon, at least with molecular masses between m/z 200 – 800, are present after 2 years of aging (Figure 6.1A and 6.1B). The FT-ICR-MS spectra of the 6 year sample showed over 2,700 peaks with a signal to noise ratios greater than S/N > 4. Using a method previously described [8], we were able to assign molecular formula with mass tolerances of less than 1 ppm (using only C, H, O, N, and S atoms) that were consistent with the masses of m/z 401 of these peaks, noting that some of these peaks are also found in spectra of the other samples. With these “empirical” formula from each of the samples, we were able to calculate H/C ratios and O/C ratios for all compounds in each sample. The H/C ratio provides an estimate of the degree of saturation (double/triple bonds and ring

118 systems) of the compound; the lower the H/C ratio, the more double bonds/ring systems in the compound. On the other hand, the O/C ratio provides an estimate of the degree of oxygenation of the compound; the larger the O/C ratio, the more oxygenation of the compound. Plots of H/C ratios vs O/C ratios are called van Krevelen (VK) plots, and these plots were used to monitor changes in compounds during the aging process (Figure 6.2A) [8]. Different regions of the plot suggest different classes of chemicals [8]. For example, compounds with H/C ratios of 1.7 to 2.3 and O/C ratios of 0.05 to 0.25 are usually consistent with fatty acids. Compounds with H/C ratios of 1.6 to 2.4 and O/C ratios of 0.7 to 1.0 correspond to carbohydrates, and compounds with H/C ratios of 0.6 to 1.1 and O/C of 0.5 to 0.8 are consistent with tannins. For the VK plots in Figure 6.2A, all chemical formula were divided into four groups based on their formula compositions: compounds with only C, H, and O (CHO) are in blue, CHOS in green, CHON in orange, and CHONS in red (Table 6.1). In agreement with previous results [8, 23, 24], there are considerably more compounds with compositions of CHO and CHON than compounds containing sulfur (CHOS or CHONS) (Table S6.3 and Figure 6.2B).

119 0-Year 2-Year A 2.5 2.5

2 2

1.5 1.5 H/C H/C 1 1

0.5 0.5

0 0 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 O/C O/C 4-Year 6-Year 2.5 2.5

2 2

1.5 1.5 H/C H/C 1 1

0.5 0.5

0 0 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 O/C O/C B

Figure 6.2 FT-ICR MS analyses. (A) Van Krevelen diagrams (H/C vs. O/C atomic ratio) of the unaged (0 year) and 2-, 4-, and 6-year aged bourbons. Each dot in the plot represents a chemical composition that has been detected in the experiment and the color represents the elemental formula composition of each class of compounds containing the denoted elements. (B) Frequency histogram of CHO, CHOS, CHON, and CHONS elemental compositions. The vertical axis represents the number of compounds detected. Color represents the elemental compositions for each class of compounds.

120 Table 6.1 Representative compounds that have been detected from LC-MS, which showed significant different abundances among aged bourbon whiskey samples.

Calculated Detected Delta mass Annotation Formula Sensory descriptor mass mass (ppm) Mild, plastic, woody, tonka, 182.0579 182.0582 1.81 Syringaldehyde C9H10O4 sweet

130.0994 130.0982 -9.30 Isoamyl acetate C7H14O2 Sweet, fruity, banana

Ethyl hexanoate/Octanoic Sweet, fruity, pineapple/Fatty, 144.1150 144.1153 1.53 C8H16O2 acid waxy Ethyl caprylate/Decanoic 172.1463 172.1468 2.67 C10H20O2 -/unpleasant, rancid, sour acid

200.1776 200.1780 1.90 Ethyl Caprate C12H24O2 -

476.1530 476.1601 14.85 Vanillin Lactoside C20H28O13 -

152.0473 152.0462 -7.23 Vanillin C8H8O3 Vanilla, sweet, chocolate, creamy

228.2089 228.2092 0.96 Ethyl laurate C14H28O2 -

256.2402 256.2405 0.94 Ethyl myristate C16H32O2 -

308.2715 308.2747 10.25 Ethyl linoleate C20H36O2 Mild, fatty, fruity

Inspection of the VK plots in Figure 6.2A reveals several interesting pieces of information. The VK plot for the unaged sample showed relatively few compounds, consistent with spectra in Figure 6.1A, and a clustering of the compounds with H/C ratios of 2 and O/C ratios of less than 0.4. Many of these compounds have H/C and O/C ratios consistent with them being fatty acids or esters, which are most likely from fermentation (yeast) [7]. VK plots of the barrel-aged bourbon samples reveal a larger number of compounds and a far greater chemical diversity of compounds (Figure 6.2B and Table S6.3). The H/C and O/C ratios clearly suggest the presence of lignins, tannins, flavanols, and carbohydrates in all of the barrel-aged samples. While it is clear that most of the barrel-associated chemicals are released by the 2-year point in aging, there are subtle differences in the plots as the spirit ages further. These changes include other compounds being released from the barrels and compounds reacting with oxygen and/or each other during the aging process [25]. As mentioned above, caution will need to be exercised as we compare the V/K plots of the 2, 4, and 6 year samples in this study because some differences may be due to the different barrels (and production dates) of these samples. As we continue to assign the points in the VK

121 plots to compounds, we may better be able to understand the extraction of barrel compounds and the chemistry that occurs in the barrel during aging. In the future, we plan to obtain and analyze bourbon samples that had one origin (aged in a single barrel and collected from the same barrel over six consecutive years).

Figure 6.3 HPLC-MS/MS (Orbitrap) analysis of different aged bourbons. (A) Total ion chromatographs of the 0, 2, 4, and 6-year bourbon samples (top to bottom). (B) Descriptive statistics view plot. An increase trend of signal level with the increase of maturation time. (C) PCA approach differentiates the 0 (red), 2 (green), 4 (dark blue) and 6 (light blue)-year whiskey based on their metabolic profiles. PCA score plot (left) demonstrates the clear separation of four clusters of samples. Loading plot of the first two principal components (right). Representative compounds that were found higher in concentration in the unaged bourbon than in the aged samples were labeled A – L and their molar masses plus peak intensities are indicated in Table S6.2.

122 HPLC-MS analysis of different aged bourbon samples While FT-ICR-MS is one of the best analytical techniques to detect the presence of low abundance compounds in complex samples, an HPLC-MS approach allows for quantitation of the compounds in the samples with high-throughput [26]. The total ion chromatograms (TIC) (after a C18 column) show significant differences between the unaged bourbon and the barrel-aged bourbon samples (Figure 6.3A), particularly in the region showing compounds with retention times of 0 to 14 minutes. Data from these chromatograms were filtered by using the Compound Discoverer 2.1 software package (see Materials and Methods for filtering criteria), and masses corresponding to each peak were provided by the MS. After data filtering, 271 out of initial 3,126 m/z features were selected and statistically-analyzed. Initially, a box-and-whisker plot was used to compare the areas of all of the peaks corresponding to chemical compounds in the filtered list (Figure 6.3B). A box-and-whisker plot displays the data for a variable (in this case peak area) as a rectangular box with a set of whiskers at each end [27, 28]. The horizontal line through the rectangle represents the median value of the variable in the data set. The lower portion of the rectangle represents the data points that fall within the third quartile, and the upper portion represents the data points that fall within the second quartile. The bottom whisker represents the fourth quartile and the top whisker represents the first quartile. The circles that fall outside the fence whiskers are outliers. Consistent with the total ion chromatograms (Figure 6.3A) and the FT-ICR-MS data (Figure 6.1), the unaged bourbon sample showed the lowest levels of signals, while all three of the barrel-aged bourbon samples have higher total peak areas than the unaged sample, indicating, not surprisingly, more compounds in the aged samples (Figure 6.3B). Moreover, the total peak areas gradually increased (see median lines in Figure 6.3B) as the bourbon samples aged, with the highest value of the median peak area in the 6 year bourbon sample. These results suggest an increase in the concentrations of compounds in the spirit as it ages, which is consistent with new compounds entering the spirit from the barrel as it ages. As mentioned above though, we cannot completely discount the possibility that some of these observed differences between the barrel-aged samples are due to the different production details and different barrels. In an effort to evaluate whether the differently-aged bourbons had unique metabolic profiles, we applied principle component analysis (PCA) (Figure 6.3C, score plot on the left and loading plot on the right) to the normalized metabolites. As shown in the PCA score plot, four

123 major clusters of samples were observed, which indicates the unaged, 2, 4, and 6 year bourbon samples exhibit unique metabolic profiles (at 95 percent confidence). Extended statistical analysis was performed to generate a loading plot to identify potential chemical markers for the discrimination between unaged and aged bourbons. The distances between the symbols (compounds) and the mean center reflect the contribution of the variables in discrimination of the four samples on the PCA components. Representative compounds far away from the center were labeled as potential chemical markers for the discrimination of unaged and aged bourbons, and their detailed retention times and molecular weights (MW) can be found in Table S6.2. The concentrations of compounds corresponding to peaks distributed along PC1 were higher in the unaged group than in the other samples, for example, compounds with MW of 521.9151, 385.9407, and 243.9508. Since the metabolite concentrations for the unaged sample were so different than those of the aged bourbon samples, we re-analyzed the data using only those from the barrel-aged bourbons. As seen in Figure S6.2A, a PCA score plot shows clear separation of the metabolic profiles of the differently-aged bourbon samples, respectively. Figure S6.2B shows the loading plot that corresponds to the PCA score plot, which was used to identify the potential chemical markers that discriminate the three differently-aged bourbons. The concentrations of compounds corresponding to peaks distributed along PC2 were higher in the 4 year and 6 year samples. For example, compounds with MW 194.0427 and 208.0737 were found more concentrated in the 4 year aged sample than in the others. The compound with molecular weight of 194.0427 was assigned as galacturonic acid, which is an oxidized form of galactose and is the main component of pectin, and compound with MW 208.0737 was assigned as sinapaldehyde, which is a plant metabolite. Similarly, compounds with MW 170.0215 and 434.0474 were found highest in concentration in the 4 year aged bourbon. Gallic acid (MW 170.0215) and quercetin (MW 434.0474) are phenolic compounds that were most likely introduced from the barrel [8, 29]. On the other hand, concentrations of compounds corresponding to peaks distributed along PC1 were higher in the 2 year aged bourbon than in the other samples, and two representative compounds were found to have MW 500.3118 and 502.3277. Based on ANOVA analysis, 209 of the 271 metabolites from the untargeted metabolomic study were significantly-different when the three barrel-aged bourbon samples were compared. The corresponding heat map of these samples showed 271 metabolites with substantially-different concentrations (top 75 compounds are shown in Figure S6.2C). About

124 85 percent metabolites (64/75) in the analysis exhibited increased concentrations as the age of the bourbon increased. By using the masses from the Orbitrap MS and analysis software (see Materials and Methods), 10 representative metabolites were assigned to potential compounds (Table 6.1 and Figure S6.3), as a proof of concept. These compounds, such as syringaldehyde, vanillin lactoside and vanillin, are commonly found in whiskies [15]. Box-and-whisker plots were generated for these compounds to show how the concentrations changed from the 2, 4, and 6 year samples (Figure S6.3). While some of the differences can be explained by the different barrels used for the 2, 4, and 6 year samples, the overall trend is consistent with metabolic changes occurring even up to 6 years.

Different chemical compositions of single barrel bourbons One of the most popular marketing strategies for bourbon is single barrel products. Single barrel bourbons are bottled directly from individual barrels rather than bottled from containers that hold bourbon blended from many different barrels. Rather than having the uniformity and consistency of the blended bourbons, the single barrel varieties are thought to have distinctive, yet subtle, differences in flavor profiles [10]. The analytical approaches described above to compare different aged bourbons showed unique chemical compositions of the differentially-aged bourbons. In an effort to determine whether different chemical compositions could be detected in single barrel bourbons, we analyzed 3 different 8-year old bourbons, each made with the same mash bill and distilled at the same time, but aged in different barrels that were next to each other in the rick house. These samples were chosen to remove variabilities due to different mash bills, different yeasts, different distillation cuts, or different aging conditions, and therefore, allow for differences to be explained solely on the different barrels. FT-ICR MS were obtained on the three single barrel samples, and the resulting data were analyzed as above to yield VK plots. As seen in Figure 6.4, VK plots showed similar, but not identical, chemical compositions in bourbons from barrels 1 and 3 with similar amounts of fatty acids (points with H/C ratios between 1.7-2.3 and O/C ratios between 0.05-0.25), carbohydrates (points with H/C ratios between 1.6-2.4 and O/C ratios between 0.7-1.0), and lignins (points with H/C ratios between 0.6-1.1 and O/C ratios between 0.5-0.8) [8]. Bourbon from barrel 2, on the other hand, has very few points in the region with H/C ratios between 0 – 1 and O/C ratios between

125 0 – 0.5; this region, as mentioned above, corresponds to region denoting lignins (see circled area in Figure 6.4) [8]. In all three plots, there is a central region at H/C ratio of 1.0 and an O/C ratio of 0.5, from which there appears to be points that form lines that emanate from this central region, and this central region corresponds to aromatic molecules, lignins, and tannins [8]. All of the bourbons in these barrels clearly have compounds that correspond to barrel aging; however, these data clearly show very different chemical compositions of these single barrel aged bourbons.

Figure 6.4 Van Krevelen diagrams (H/C vs. O/C atomic ratio) of the three 8-year bourbons that aged parallel in three different barrels. Each dot in the plot represents a chemical composition that has been detected in the experiment and the color represents the elemental formula composition of each compound class. The dashed ovals highlighted the main differences among these three samples, which may indicate the chemicals that contribute to the “single barrel” phenomenon.

To further characterize the single barrel bourbons, we conducted the LC-MS based untargeted metabolic approach on these samples as we reported above on the differentially-aged bourbons. An untargeted metabolomic analysis and application of a one-way analysis of variance (ANOVA) test revealed that 104 out of 249 metabolites are statistically-different (p<0.05). Revealed by the Figure 6.5A, PCA score plots showed three major clusters, which indicates the single barrel bourbon samples exhibit unique metabolic profiles (95 percent confidence), and a loading plot corresponding to the PCA score plot is shown in Figure S6.4. Six potential compounds (2-furoic acid, glutaric anhydride, dimethyl itoconate, gallic acid, monobutyrin, and 1,4- dihydroxy-2-napthoic acid) were chosen to examine their relative concentrations in the single

126 barrel bourbons and their contributions to the discrimination. Box and whiskers plots of the corresponding data show significantly different amounts of these compounds in the single barrel bourbons (Figure 6.5C). Figure 6.5B shows the heat map of the single barrel samples (top 75 metabolites that are significantly-different). Visually, the heat map analysis reveals that the three bourbon samples generally clustered in their own groups, and the heat map clearly shows significant differences in the metabolites detected in these samples. As we continue our efforts to identify more of the compounds in bourbon, we will be in position to understand better how different barrels, mash bills, yeast, distillation practices, and aging practices correlate with different chemical compositions and possibly to different flavor profiles.

127

A B

C 2-Furoic acid Glutaric anhydride Dimethyl itaconate

Gallic acid Monobutyrin 1,4-Dihydroxy-2-naphthoic acid

Figure 6.5 LC-MS/MS analyses on ‘single barrel’ bourbons. (A) PCA approach differentiates the barrel 1 (red), barrel 2 (green), and barrel 3 (dark blue) whiskey based on their metabolic profiles. Ovals show 95% confidence level. (B) Heatmap (top 75) presentation of metabolic profiles from three “single barrel” whiskey samples. (C) Box plots of six compounds showing significant differences barrels.

128 6.4 Conclusions In this study, we combined ultrahigh resolution mass spectrometry (FT-ICR-MS and LC- MS/MS) and statistical analyses to analyze two different sets of bourbons. In the first set of samples, we examined 4 differently-aged samples, and not surprisingly, our data clearly show an increase in the number of chemical compounds present in the samples as the bourbon ages. In fact, most of the large changes in chemical composition occurred in the first two years of aging (compare data from unaged to 2-year old bourbon), and it is clear that most of the “new” compounds from the barrel have entered during this first aging point. This conclusion is consistent with the colors of the samples, with the unaged sample being clear, and the 2, 4, and 6 year samples all being the same typical brown typically observed with barrel-aged bourbons. The FT-ICR-MS data also show that the O/C ratios of many compounds increase as the sample ages, suggesting that compounds are reacting with oxygen during the aging process. We have started to identify compounds in the samples using the data from the FT-ICR-MS, which offers ultrahigh mass resolution and mass accuracy [7]. We also examined these same samples with LC-Orbitrap MS. While with somewhat lower mass resolution and mass measurement accuracy than FT-ICR MS, our LC-Orbitrap MS affords high-throughput analyses and the ability to quantitate chemical compounds in our samples [30]. Similar to a published study [3], we were able to identify specific chemical compounds that are normally found in bourbons, and our data clearly show that the chemical composition of the samples changes as the bourbon ages. Also, like the published study [3], our data clearly show that unaged and 2, 4, and 6 year aged samples can be differentiated. It is important to note that the previous study and our study suffer from a similar sample issue. In both studies, samples that were produced at different times and under different conditions were analyzed. Some of the observed differences are likely explained by these different conditions as well as by the different barrels used. It is clear that these techniques are sufficiently sensitive to identify many chemical compounds in the samples, even those at low concentrations. We will need to continue our efforts to assign peaks in both techniques to actual chemical compounds in order to better understand the complex chemistries that occur during barrel aging. We also analyzed three 8 year bourbon samples, which were produced at the same time, using the same mash bill, yeast, and distillation conditions but were aged in different adjacent barrels. FT-ICR-MS and HPLC-Orbitrap MS analyses clearly show distinct differences between

129 these three samples. This result suggests that “fingerprinting” different bourbons, for authentication purposes for example, may be complicated because researchers will need to determine whether observed chemical differences are due to the different types of bourbon or to the variability in barrels. As with the studies on the barrel aged samples, we will continue to analyze the data to assign peaks to actual chemical compounds. After these analyses, we will be in better position to determine which chemical compounds are “diagnostic” with specific bourbons, even though their relative concentrations may be affected by different barrels. The use of ambient MS methods [31] could likely also be an effective future approach for the analysis of different whiskey samples.

ASSOCIATED CONTENT Supporting information This material is available in Appendix D.

Acknowledgements The 15 T Bruker SolariXR FT-ICR instrument was supported by NIH Award Number Grant S10 OD018507.

Funding Information This research was supported by a Miami University Committee on Faculty Research grant (to MWC).

Conflict of interest The authors claim no conflict of interest.

Ethical Approval This article does not contain any studies with human participants or animals performed by any of the authors.

Informed Consent Informed consent was obtained from all individual participants included in this study.

130 References [1] Russell, I., C. Bamforth, and G. Stewart, Whisky: technology, production and marketing. 2014: Elsevier. [2] Murray, J., Jim Murray's Whisky Bible 2018. 2017: Dram Good Books. [3] Collins, T.S., weigenbaum, J. Z. and Ebeler, S.E., Profiling of nonvolatiles in whiskeys using ultra high pressure liquid chromatography quadrupole time-of-flight mass spectrometry (UHPLC–QTOF MS). Food Chem, 2014. 163: p. 186-196. [4] Teodoro, J.A.R., Pereira, H.V., Sena, M.M., Piccin, E., Zacca, J.J. and Augusti, R., Paper spray mass spectrometry and chemometric tools for a fast and reliable identification of counterfeit blended Scottish whiskies. Food Chem, 2017. 237: p. 1058-1064. [5] Pryde, J., Conner, J., Jack, F., Lancaster, M., Meek, L., Owen, C., Paterson, R., Steele, G., Strang, F. and Woods, J., Sensory and Chemical Analysis of ‘Shackleton's’ Mackinlay Scotch Whisky. J I Brewing, 2011. 117(2): p. 156-165. [6] Wiśniewska, P., Dymerski, T., Wardencki, W. and Namieśnik, J., Chemical composition analysis and authentication of whisky. J Sci Food Agr, 2015. 95(11): p. 2159-2166. [7] Kew, W., Goodall, I., Clarke, D. and Uhrín, D., Chemical diversity and complexity of scotch whisky as revealed by high-resolution mass spectrometry. J Am Soc Mass Spectr, 2017. 28(1): p. 200-213. [8] Roullier-Gall, C., Signoret, J., Hemmler, D., Witting, M.A., Kanawati, B., Schäfer, B., Gougeon, R.D. and Schmitt-Kopplin, P., Usage of FT-ICR-MS Metabolomics for characterizing the chemical signatures of barrel-aged whisky. Front Chem, 2018. 6: p. 29. [9] Rogers, A., Proof: the science of booze. 2014: Houghton Mifflin Harcourt. [10] Minnick, F., Bourbon: The Rise, Fall, and Rebirth of an American Whiskey. 2016: Voyageur Press (MN). [11] Veach, M.R., Kentucky Bourbon Whiskey: An American Heritage. 2013: University Press of Kentucky. [12] Mitenbuler, R., Bourbon Empire: The Past and Future of America's Whiskey. 2016: Penguin. [13] Minnick, F., Bourbon Curious: A Tasting Guide for the Savvy Drinker with Tasting Notes for Dozens of New Bourbons. 2019: Harvard Common Press.

131 [14] Glabasnia, A. and T. Hofmann, Sensory-directed identification of taste-active ellagitannins in American (Quercus alba L.) and European oak wood (Quercus robur L.) and quantitative analysis in bourbon whiskey and oak-matured red wines. J Agr Food Chem, 2006. 54(9): p. 3380-3390. [15] Heinz, H.A. and J.T. Elkins, Comparison of unaged and barrel aged whiskies from the same Mash Bill using gas chromatography/mass spectrometry. J Brewing Distilling, 2019. [16] Roullier-Gall, C., Witting, M., Gougeon, R.D. and Schmitt-Kopplin, P., High precision mass measurements for wine metabolomics. Front Chem, 2014. 2(102). [17] Roullier-Gall, C., Witting, M., Tziotis, D., Ruf, A., Gougeon, R.D. and Schmitt-Kopplin, P., Integrating analytical resolutions in non-targeted wine metabolomics. Tetrahedron, 2015. 71(20): p. 2983-2990. [18] Gollihue, J., Richmond, M., Wheatley, H., Pook, V.G., Nair, M., Kagan, I.A. and DeBolt, S., Liberation of recalcitrant cell wall sugars from oak barrels into bourbon whiskey during aging. Sci Rep, 2018. 8(1): p. 15899. [19] Suomalainen, H. and A. Keränen, The fatty acid composition of baker's and brewer's yeast. Chem Phys Lipids, 1968. 2(3): p. 296-315. [20] Yoshioka, S., Terashita, T., Yoshizumi, H. and Shirasaka, N., Inhibitory effects of whisky polyphenols on melanogenesis in mouse B16 melanoma cells. Biosci Biotech Biochem, 2011: p. 1110242700-1110242700. [21] Kataoka, S., Functional effects of Japanese style fermented soy sauce (shoyu) and its components. J Biosci Bioeng, 2005. 100(3): p. 227-234. [22] Crozier, A., Jaganath, I.B., and Clifford, M.N., Dietary phenolics: chemistry, bioavailability and effects on health. Nat Prod Rep, 2009. 26(8): p. 1001-1043. [23] Ochiai, N., K. Sasamoto, and K. MacNamara, Characterization of sulfur compounds in whisky by full evaporation dynamic headspace and selectable one-dimensional/two- dimensional retention time locked gas chromatography–mass spectrometry with simultaneous element-specific detection. J Chromatogr A, 2012. 1270: p. 296-304. [24] Sha, S., Chen, S., Qian, M., Wang, C. and Xu, Y., Characterization of the typical potent odorants in Chinese roasted sesame-like flavor type liquor by headspace solid phase microextraction–aroma extract dilution analysis, with special emphasis on sulfur- containing odorants. J Agr Food Chem, 2016. 65(1): p. 123-131.

132 [25] Spedding, G., 80 Years of Rapid Maturation Studies—Why Are We There Yet? Part 1 of 3 Part 1: Key Analytics and Solvent Chemistry. [26] Ghaste, M., R. Mistrik, and V. Shulaev, Applications of Fourier transform ion cyclotron resonance (FT-ICR) and orbitrap based high resolution mass spectrometry in metabolomics and lipidomics. Int J Mol Sci, 2016. 17(6): p. 816. [27] Chambers, J.M., Cleveland, W.S., Kleiner, B. and Tukey, P.A., Graphical Methods for Data Analysis. Wadsworth Int'l. Group, Belmont, CA, 1983. [28] Tukey, J.W., Box-and-whisker plots. St Class Dat Anal 1977: p. 39-43. [29] Bukovsky-Reyes, S.E., Lowe, L.E., Brandon, W.M. and Owens, J.E., Measurement of antioxidants in distilled spirits by a silver nanoparticle assay. J I Brewing, 2018. 124(3): p. 291-299. [30] Wang, X., Liu, Y., Su, Y., Yang, J., Bian, K., Wang, Z. and He, L.M., High-throughput screening and confirmation of 22 banned veterinary drugs in feedstuffs using LC-MS/MS and high-resolution orbitrap mass spectrometry. J Agr Food Chem, 2014. 62(2): p. 516- 527. [31] Black, C., O.P. Chevallier, and C.T. Elliott, The current and potential applications of Ambient Mass Spectrometry in detecting food fraud. Trac-Trend Anal Chem, 2016. 82: p. 268-278.

133 CHAPTER 7 Conclusions MS-based metabolomics is a powerful tool to evaluate the complex samples and systems and how external stimuli can alter these systems. In the dissertation, we utilized several different MS-based approaches to evaluate such samples and systems in an effort to address some long- standing questions. In Chapter 2, we examined how four common nutritional components of growth media affected the human gut microbiome. The effects of these components in the culture medium to the metabolome of human gut microbiome were investigated by using a targeted metabolic profiling approach. Four nutritional components, certain inorganic salts, mucin, short chain fatty acid, and bile salts were tested because the concentrations of these nutrients were varied in previous studies and there were reports that these components could modulate the gut microbial populations and functions.[1] Our results indicate that the concentration of inorganic salts in media is the major factor that needs to be considered, while mucin is the second most important component that affects the gut microbiome profile. Subsequent analyses revealed the highly-impacted biological pathways and the metabolites that were affected most by the different media components. These results allowed us to propose which biosynthetic pathways in the host are regulated by gut bacteria. This preliminary study serves as a first attempt to evaluate individual nutritional components in their contribution to gut microbial metabolic functions and provide a LC-MS based, targeted metabolomics approach for researchers to optimize in vitro cultures for the gut microbiome. Future studies will need to completed to evaluate other components of the culture media so that an accurate human gut model can be used to evaluate how food and drugs affect the gut. In Chapter 3, we performed a comprehensive metabolic study to evaluate the ability of Lactobacillus acidophilus (LA) to ferment black tea extract (BTE) and to measure the enhanced uptake of phenolic compounds by Escherichia coli when incubated with fermented BTE. Targeted LC-MS results showed an increase in the uptake of total phenolic compounds by E. coli after the BTE was fermented by LA, and this increased uptake lead to increased antimicrobial activity against E. coli by the fermented BTE. Scanning electron microscopic results revealed the mechanism of antimicrobial ability is due to an increased oxidative stress on the bacterial membranes, which was supported by metabolic profiling studies on the E. coli cultures. Sensitive and specific detection of phenolic compounds and metabolites enabled a molecular level

134 understanding of the antimicrobial activity of fermented BTE. This research will be continued in the future to further elucidate the antimicrobial activity of fermented BTE (and other fermented foods) on other bacterial systems in order to improve human health and to understand the complicated synergism of bacteria and the human gut. In Chapter 4, multi-omics approaches, including targeted metabolomics and microbiome analyses, were used to evaluate the ultrafine particle (UFP)-related effects on the gut microbial population and its function. Genomic results obtained from 16s rRNA sequencing techniques predicted the possible functional impacts of ultrafine particles on the mouse gut microbiome, and our metabolomics analyses validated many of the functional predictions. Great differences were observed between male and female mice after exposure to B0 and B20 UFPs, and female mice appear to exhibit a higher risk of metabolic dysregulation. This study serves as a preliminary attempt to better understand the potential sex-specific health effects of UFPs exposure in humans in the future. These future studies should be used to address emissions when these fuels are burned so that human health can be improved. A hybrid platform that demonstrated the capability of targeted and untargeted metabolomics in differentiating closely-related bacterial strains/species was established by incorporating a LC- QQQ targeted metabolomics approach with an LC-Orbitrap untargeted metabolomics approach. Four closely-related Lactobacillus species, Lactobacillus acidophilus, Lactobacillus fermentum, Lactobacillus reuteri, and Lactobacillus delbrueckii were differentiated successfully. Both the LC- QQQ platform and LC-Orbitrap platform showed robustness in performance. Many identical compounds were detected by both platforms at similar signal levels, but each of the platform showed its complementary capability in compound detection and identification by covering a substantial amount metabolite/metabolism information that is unique to each approach. This study is a great example to show the potential of combinatory metabolomics approaches in rapidly differentiating closely-related human health-related bacterial strains. Future works will focus on expanding this hybrid method to many other bacterial strains and validating their utility in more complicated, bacterial environments. It is interesting to consider whether this approach could be applied to viral communities in an effort to quickly and accurately identify which viruses are present in different communities. Through the use of two ultrahigh resolution mass spectrometric systems, which include a Fourier-transform ion cyclotron resonance mass spectrometry (FT-ICR MS) and liquid

135 chromatography coupled with tandem mass spectrometry (LC-MS/MS), four Kentucky bourbon whiskies, which were matured in barrels differently (new made distillate, 2, 4, and 6-year aged bourbons), were analyzed for chemical compositions. A great alteration of the chemical composition occurred in the first two years of barrel aging and was readily detected by the MS systems. Further aging resulted in more subtle changes in the chemical composition of the bourbons. FT-ICR MS showed, not surprisingly, that oxidation occurred during the maturation process, especially in the first two years. The LC-Orbitrap MS system identified specific chemical compounds that are normally found in bourbons, and we used them as chemical markers for quantitative analyses. This combinatory approach is sufficiently-sensitive to identify many chemical compounds in the samples, even those at low concentrations. Bourbons that produced identically but matured in different barrels for 8 years were also found to be chemically-distinct; this result confirms the uniqueness of “single barrel” bourbons, which is a significant marketing tool used by the bourbon industry. Bourbon contains a complex mixture of volatile compounds, generated by yeast during fermentation and isolated during distillation, and non-volatile compounds, which are introduced by the charred barrels.[2-3] This complex mixture is perfect for analysis by GC- and LC-coupled MS techniques. Future work involves both targeted and untargeted approaches. In targeted studies, LC- and GC-MS techniques will be used to quantitate the amounts of known chemicals in bourbon in an effort to “fingerprint” bourbons based on the producer and hopefully the mash bill. Untargeted studies will involve identifying more chemical compounds in an effort to better understand the different flavor profiles and possibly identify any toxic compounds that are produced during barrel aging. Finally, results from all of these techniques could be used to better understand the complicated chemistry that occurs during barrel aging. Mass spectrometry is an important technique that has been used in biology, chemistry, geology, physics, and other sciences. While mature, new coupled MS approaches are introduced frequently and as more complicated questions need to be addressed. This dissertation was application-based, in which we applied MS-based techniques to answer specific questions in food chemistry and in human health. As with most research, our studies and results led to more questions and other possible avenues to explore. It is clear that MS-based techniques will continue to provide great insight into the chemical composition of complex samples.

136 Reference

[1] Li, L., Zhang, X., Ning, Z., Mayne, J., Moore, J.I., Butcher, J., Chiang, C.-K., Mack, D., Stintzi, A., and Figeys, D., Evaluating in Vitro Culture Medium of Gut Microbiome with Orthogonal Experimental Design and a Metaproteomics Approach. J Proteome Res, 2017. 17(1): p. 154- 163. [2] Stupak, M., Goodall, I., Tomaniova, M., Pulkrabova, J. and Hajslova, J., A novel approach to assess the quality and authenticity of Scotch Whisky based on gas chromatography coupled to high resolution mass spectrometry. Anal Chim Acta, 2018. 1042, pp.60-70.

[3] Roullier-Gall, C., Signoret, J., Hemmler, D., Witting, M.A., Kanawati, B., Schäfer, B., Gougeon, R.D. and Schmitt-Kopplin, P., Usage of FT-ICR-MS Metabolomics for characterizing the chemical signatures of barrel-aged whisky. Front Chem, 2018. 6: p. 29.

137

APPENDIX A Supporting information for Chapter 2

a b c 5-Hydroxymethyluracil Indole-3-acetate

Figure S2.1 Comparison between high and low concentration of inorganic salts. (a) PCA plot for group comparison between L-4 and H-4. (b) and (c) representative metabolites as biomarkers of oxidative stress.

138 Figure S2.2 Heatmap of (a) low concentration inorganic salt groups and (b) high concentration inorganic salt groups.

139 a b

Figure S2.3 (a) PCA plot showed L-3 group clearly different from L-4 group. (b) and (c) are 5- hydroxymethyluracil and indole-3-acetaldehyde showed lower concentrations in mucin treated groups.

140

Figure S2.4 Bile salts and SCFAs effect without mucin treatment in low concentration inorganic salts medium.

141 Table S2.1 Detailed list of the overview of metabolic profiles of all studied samples with the various nutritional enrichments.

Retention time Precursor Product 1 Product 2 Compound Polarity (min) (m/z) (m/z) (m/z) Phenylpyruvate Negative 2.51 163.0398 91.0554 4-Hydroxybenzoate (p-Salicylic acid) Negative 2.41 137.0225 93.0341 Oxaloacetic acid Negative 1.94 131.0058 59.015 87.0083 Methylmalonate Negative 2.23 117.0189 73.0296 Succinic acid Negative 1.61 117.0266 73.03 99.0086 Pyruvate Negative 1.68 87.016 43.0185 86.556 2,3-dihydroxybenzoic acid Negative 2.93 153.0169 109.0287 Alpha-ketoglutaric acid Negative 1.7 145.096 101.04 Uracil Negative 1.98 111.0272 41.9984 111.0192 Indole-3-acetate Negative 2.72 174.0529 130.064 3-dehydroshikimate Negative 3 171.0328 109.0318 127.0424 Malate Negative 2.16 133.0215 71.0138 115.0026 Myo-inositol Negative 3.37 179.0547 59.0142 161.0436 Glucose Negative 2.36 179.0633 59.0142 71.0139 (GMP) guanosine 5'-monophosphate Negative 4.9 362.049 78.959 210.999 Allantoin Negative 4.66 157.0344 41.9988 97.0034 (ADP) adenosine 3'5'-diphosphate Negative 3.5 426.022 134.046 328.044 3-phosphoglyceric acid Negative 5.5 184.989 78.959 96.969 Fructose-6-phosphate Negative 4.38 259.0297 79.143 96.9689 UDP-N-acetylglucosamine Negative 5.41 606.0758 282.0388 384.9857 (CMP) cytidine monophosphate Negative 3.75 322.043 78.959 96.968 Phenylacetic acid Negative 1.34 135.061 91.155 3-Hydroxybenzoate Negative 1.46 137.052 93.111 2-Methylmaleate/itaconate Negative 1.47 129.043 41.222 85.111 Lipoamide Positive 2.64 206.0659 161.0436 189.0369 (S)-Nicotine Positive 4.61 163.1271 84.0803 106.0643 Caffeine Positive 3.3 195.0875 110.0715 138.0659 Deoxyadenosine Positive 2.28 252.1081 136.0606 Deoxyguanosine Positive 2.37 268.0967 110.0343 135.0294 Urocanate Positive 2.13 139.0504 93.0451 121.0393 Nicotinamide Positive 4.77 123.0558 80.0449 Tyramine Positive 2.1 138.094 77.038 121.065 5-hydroxymethyluracil Positive 3.03 143.0437 82.0289 125.0339 Riboflavin Positive 1.62 377.1448 172.0873 243.0887 Trigonelline Positive 2.51 138.0552 94.0654 Adenine Positive 1.76 136.0544 94.0404 119.0346

142 Adenosine Positive 2.51 268.1038 136.0611 Hypoxyxanthine Positive 2.49 137.0457 110.0348 119.0351 4-Aminobenzoate Positive 2.59 138.0567 77.0934 94.0655 Dopamine Positive 2.31 154.086 137.0595 Cytidine Positive 2.85 244.0924 112.0502 Noradrenaline Positive 2.47 170.0842 135.0459 152.0731 Pterin Positive 2.39 164.0557 119.0345 147.0296 Creatinine Positive 1.81 114.0699 44.1134 Guanosine Positive 5.33 284.983 152.0558 Pyridoxine Positive 2.46 170.0807 152.0703 Xanthine Positive 2.54 153.0405 55.03 110.0348 Purine Positive 1.84 121.1509 94.0401 5-aminolevulinic acid Positive 2.74 132.0649 86.06 114.0543 Cytosine Positive 1.96 112.0432 69.0454 95.0242 Histamine Positive 5.1 112.089 68.0521 95.0623 Sorbate Positive 2.83 113.0601 67.0555 95.05 Guanine Positive 2.6 152.0494 108.0198 135.0299 Uridine Positive 2.59 245.0695 113.0349 Homocysteine Positive 3.03 136.0421 56.0506 90.038 Guanidinoacetate Positive 2.13 118.0611 72.056 76.396 Tryptamine Positive 2.31 161.1074 144.0781 dGMP Positive 3.38 348.0705 152.0565 5-aminopentanoate Positive 2.13 118.0838 55.0552 101.0602 Thiamine Positive 3.36 265.1123 122.0714 144.0476 Carnosine Positive 2.95 227.1132 156.0757 210.0866 Acetylcholine Positive 3.1 146.1178 43.0182 87.0448 4-guanidinobutanoate Positive 3.16 146.0916 86.0607 87.0446 Ethanolamine Positive 2.41 62.0609 44.0509 Glucosamine-6-phosphate Positive 4.44 260.0516 84.0448 98.0549 Serotonin Positive 3.79 177.102 160.0752 Citrulline Positive 2.96 176.102 113.071 159.076 Glutathione Positive 4.1 308.0934 162.0235 179.0501 Melatonin Positive 3.76 233.1281 174.0908 3-Hydroxyanthranilate Positive 2.6 154.0496 136.0386 Octopamine Positive 2.21 154.0869 91.0545 136.0756 Theophylline Positive 2.1 181.0716 96.0556 124.0499 NAD Positive 4.72 664.116 428.04 524.057 N-Acetyl-ornithine Positive 5.47 175.1075 115.0864 158.0804 3-ureidopropionate Positive 5.63 133.0597 90.0545 115.0491 Ornithine Positive 4.53 133.097 70.066 116.071 4-aminobutanoate Positive 3.13 104.0708 87.0445

143 Cadaverine Positive 6.01 103.1231 86.0969 S-Adenosyl-methionine Positive 5.88 399.1478 136.0604 250.0918 Putrescine Positive 6.32 89.1082 72.0816 Glutathione disulfide Positive 6.31 613.1603 355.0733 484.1146 Tryptophan Positive 1.86 205.152 146.06 187.968 Phenylalanine Positive 1.91 166.152 103.111 120.111 Leucine/Isoleucine Positive 1.98 132.152 44.183 86.183 Aspartic acid Positive 1.92 134.243 74.058 88.111 Methionine Positive 1.99 150.122 104.111 133.054 Asparagine Positive 1.93 133.152 74.024 87.111 Tyrosine Positive 2.12 182.152 136.183 165.196 Valine Positive 2.15 118.213 55.165 72.183 Proline Positive 2.26 116.152 70.183 Alanine Positive 2.43 90.152 44.222 45.222 Glycine Positive 2.78 76.122 30.333 47.222 Glutamic acid Positive 3.04 148.061 102.183 130.05 Histidine Positive 3.65 156.122 95.111 110.111 Arginine Positive 4.36 175.152 60.056 70.111 Lysine Positive 4.5 147.152 84.183 130.165 1-methyl-6,7-dihydroxy-1,2,3,4- Positive 1.92 180.1 145.17 163.12 tetrahydroisoqui Adipic acid Positive 1.35 147.078 129.111 Ribitol Positive 2.04 153.078 135.054 Dethiobiotin Positive 1.56 215.14 179.17 197.11 Biotin Positive 1.5 245.1 227.1 Mannitol Positive 2.2 183.078 165.054 Mono-methyl-glutarate Positive 1.36 147.087 129.111 3-Hydroxyphenylacetate Positive 1.46 153.06 125.22 4-Hydroxy-3-methoxyphenylglycol Positive 1.46 455.2 437.155 Salicylate Positive 1.61 139.07 121.054 Salicylamide Positive 1.45 138.078 121.054 N,N-dimethyl-1,4-phenylenediamine Positive 1.7 137.11 122.12 Phenethylamine Positive 1.65 122.1 105.17 2-Methylglutaric acid Positive 1.39 147.078 129.054 Pyridoxal Positive 1.78 168.07 150.12 N-Acetylserotonin Positive 1.52 219.11 160.18 202.16 Trans-cinnamaldehyde Positive 1.13 133.096 115.111 2,6-Dihydroxypyridine Positive 1.71 112.096 94.054 Ethyl-3-indoleacetate Positive 1.34 204.1 130.14 Methyl indole-3-acetate Positive 1.4 190.09 130.16 172.12 2',4'-Dihydroxyacetophenone Positive 1.47 153.07 135.054

144 Homovanillate Positive 1.48 183.07 137.04 3-Methyladenine Positive 2.24 150.08 133.12 N-Acetyl-L-leucine Positive 1.54 174.11 128.21 156.12 Indole-3-acetaldehyde Positive 1.93 160.08 118.11 142.07 Pyruvic aldehyde Positive 1.33 73.183 55.165

145 Table S2.2 Metabolites that have fold change larger than 2 in terms of their abundant in inorganic salts comparison.

Metabolites Fold Change p-value

5-aminopentanoate 6.37E-08 4.01E-14 Lipoamide 1.27E+07 1.79E-13 Acetylcholine 2.44E-07 7.76E-13 3-hydroxybenzoate 3.25E+06 3.17E-14 Dopamine 1.74E+06 3.82E-13 Mannitol 1.39E+06 3.63E-13 Thiamine 9.42E-07 1.21E-13 Dethiobiotin 9.28E+05 5.20E-13 Glycine 1.08E-06 1.34E-11 1-methyl-6,7-dihydroxy-1,2,3,4-tetrahydroisoqui 8.62E+05 6.39E-14 dGMP 1.27E-06 3.57E-11 N-acetylserotonin 1.35E-06 3.23E-12 GMP 7.20E+05 1.35E-09 Glucosamine-6-phosphate 6.31E+05 1.02E-12 Methyl_indole-3-acetate 1.71E-06 3.41E-15 Cinnamaldehyde 1.77E-06 4.68E-13 Theophylline 2.14E-06 9.03E-13 S-nicotine 4.28E+05 4.53E-12 Aspartic acid 2.36E-06 8.62E-13 Cytosine 2.85E-06 2.23E-13 2,6-dihydroxypyridine 4.02E-06 6.78E-13 5-hydroxymethyluracil 4.94E-06 5.18E-14 Arginine 5.09E-06 3.63E-12 Riboflavin 5.61E-06 2.36E-11 Alpha-ketoglutaric acid 1.56E+05 2.79E-13 4-hydroxy-3-methoxyphenylglycol 6.49E-06 5.96E-14 Adipic acid 6.85E-06 1.27E-12 Mono-methyl glutarate 6.87E-06 9.55E-10 2-methylglutaric acid 7.35E-06 2.90E-12 Deoxyguanosine 7.36E-06 6.54E-13 Pyruvic_aldehyde 7.55E-06 1.76E-12 Ethyl_3-indoleacetate 9.54E-06 3.16E-11 N-acetyl-ornithine 9.69E-06 4.71E-12 3-dehydroshikimate 1.21E-05 9.93E-13 Uridine 1.26E-05 1.88E-11 Pyruvate 1.29E-05 8.72E-11

146 2-methylmaleate/itaconate 1.57E-05 1.01E-10 Phenylpyruvate 5.52E+04 1.74E-13 Guanosine 1.98E-05 6.54E-13 Nicotinamide 2.33E-05 1.10E-11 3-phosphoglyceric acid 2.40E-05 1.34E-13 Ethanolamine 2.74E-05 3.23E-11 Cytidine 8.82E-03 2.94E-02 Salicylate 6.26E+01 2.70E-08 Urocanate 5.81E+01 2.35E-08 F6P 5.24E+01 1.07E-05 Hypoxyxanthine 4.63E+01 1.19E-09 Lysine 4.35E+01 7.21E-10 Methylmalonate 4.07E+01 1.12E-07 Succinic acid 3.68E+01 1.20E-07 Tryptophan 3.64E+01 1.44E-09 Sorbate 3.00E-02 3.45E-02 Adenosine 3.46E-02 1.25E-09 Putrescine 3.65E-02 1.52E-06 Uracil 2.67E+01 1.20E-02 Creatinine 2.53E+01 1.06E-07 Cadaverine 4.63E-02 1.88E-06 Myo-inositol 2.11E+01 8.24E-07 Indole-3-acetate 2.11E+01 6.83E-06 4-hydroxybenzoate 2.05E+01 3.88E-07 Malate 1.56E+01 1.45E-04 Glucose 1.29E+01 5.53E-06 Adenine 1.14E+01 5.86E-03 3-ureidopropionate 1.14E+01 1.20E-06 Ornithine 1.08E+01 2.32E-06 Guanine 1.02E+01 2.18E-06 Tyramine 1.01E-01 4.94E-08 Salicylamide 1.02E-01 1.92E-08 CMP 8.98E+00 3.22E-05 Homovanillate 1.33E-01 1.43E-02 Tyrosine 7.46E+00 1.56E-07 Deoxyadenosine 1.88E-01 2.33E-03 Histamine 5.23E+00 4.34E-07 4-aminobenzoate 1.92E-01 2.74E-06 S-adenosyl-methionine 4.89E+00 2.45E-05 Allantoin 4.80E+00 4.81E-05

147 Biotin 4.11E+00 1.78E-04 Xanthine 4.04E+00 1.19E-09 Glutathione disulfide 3.70E+00 2.26E-02 Pterin 3.44E+00 2.15E-06 Serotonin 3.41E+00 4.99E-04 Histidine 3.32E+00 5.13E-05 Glutathione 3.32E+00 2.26E-02 Citrulline 3.30E+00 1.34E-03 Ribitol 2.95E+00 1.98E-06 3-hydroxyphenylacetate 2.92E+00 6.22E-05 3-hydroxyanthranilate 2.83E+00 3.30E-05 Octopamine 2.81E+00 1.36E-05 Carnosine 3.60E-01 3.31E-05 N-acetyl-l-leucine 2.77E+00 5.13E-05 Pyridoxine 2.71E+00 1.14E-05 2',4'-dihydroxyacetophenone 2.68E+00 1.54E-05 Noradrenaline 2.51E+00 1.37E-05 Glutamic acid 2.08E+00 8.27E-05

148

Table S2.3 Metabolites that have fold change larger than 2 in terms of their abundant in mucin comparison.

Metabolites Fold Change p-value

Uracil 2.25E-06 1.20E-02 GMP 2.48E+04 9.41E-09 ADP 9.58E+03 1.05E-09 Histidine 8.19E-02 8.06E-05 Glycine 8.53E-02 9.82E-06 N-acetyl-l-leucine 7.26E+00 7.49E-07 Arginine 4.97E+00 3.08E-06 dGMP 2.16E-01 4.77E-03 Tyrosine 2.23E-01 3.36E-06 N-acetyl-ornithine 4.37E+00 4.67E-06 Glutamic_acid 2.72E-01 4.10E-05 Homovanillate 2.81E-01 9.69E-06 Melatonin 3.10E-01 1.81E-05 Glucose 3.13E-01 3.59E-03 CMP 2.66E+00 7.38E-03 Theophylline 3.86E-01 2.52E-05 Tryptamine 2.58E+00 3.93E-05 2-methylmaleate/itaconate 3.88E-01 3.16E-04 Acetylcholine 4.05E-01 3.75E-05 4-guanidinobutanoate 4.05E-01 9.32E-06 2,3-dihydroxybenzoic acid 4.19E-01 3.01E-03 Lysine 2.30E+00 2.28E-06 Pyridoxine 4.35E-01 3.59E-05 Noradrenaline 4.37E-01 3.07E-05 Myo-inositol 4.46E-01 6.28E-03 2',4'-dihydroxyacetophenone 4.51E-01 1.59E-04 4-hydroxybenzoate 4.57E-01 1.49E-03 Biotin 4.58E-01 1.38E-03 Allantoin 4.62E-01 4.10E-03 Ribitol 4.67E-01 4.53E-05 Uridine 4.67E-01 1.86E-03 3-hydroxyphenylacetate 4.72E-01 5.91E-04 4-hydroxy-3-methoxyphenylglycol 4.78E-01 9.45E-05 NAD 2.08E+00 8.06E-05

149 Hypoxyxanthine 4.82E-01 1.42E-03 5-aminopentanoate 4.88E-01 1.02E-05 Adenosine 4.96E-01 2.28E-04 5-aminolevulinic acid 4.96E-01 1.14E-05 2-methylglutaric acid 2.01E+00 2.08E-03 Leucine/isolucine 5.00E-01 1.98E-05

150 APPENDIX B Supporting information for Chapter 4

Figure S4.1 PLS-DA plot of B0 vs B20 in (a) female obese mice and (b) male obese mice.

Figure S4.2 Intestinal and plasma pyridoxal levels detected from female obese mice. (a) cecum data and (b) plasma data. The box and whisker plots summarize the normalized values.

151

Figure S4.3 Plasma Myo-inositol detected from male obese mice. The box and whisker plots summarize the normalized values.

152

Table S4.1 Mice cecum microbial metabolic pathways identified in Figure 4.5.

Labels Pathway Identity a Taurine and hypotaurine metabolism b Alanine, aspartate and glutamate metabolism c Glycolysis or Gluconeogenesis d Pyruvate metabolism e Arginine and proline metabolism f Butanoate metabolism g Amino sugar and nucleotide sugar metabolism h Propanoate metabolism i Cysteine and methionine metabolism j Glycerophospholipid metabolism k Pentose phosphate pathway l Inositol phosphate metabolism m Biotin metabolism n Cysteine and methionine metabolism o Purine metabolism p Aminoacyl-tRNA biosynthesis q Glyoxylate and dicarboxylate metabolism r Citrate cycle (TCA cycle)

153

Table S4.2 Host metabolic pathways (using plasma metabolites) identified in pathway analysis (Figure 4.7).

Labels Pathway identity a Phenylalanine, tyrosine and tryptophan biosynthesis b D-Glutamine and D-glutamate metabolism c Alanine, aspartate and glutamate metabolism d Phenylalanine metabolism e Histidine metabolism f Vitamin B6 metabolism g Glutathione metabolism h Pyrimidine metabolism i Arginine and proline metabolism j Thiamine metabolism k Methane metabolism l Riboflavin metabolism m Glycolysis or Gluconeogenesis n Amino sugar and nucleotide sugar metabolism o Starch and sucrose metabolism p Pentose phosphate pathway

154

APPENDIX C Supporting information for Chapter 5

Figure S5.1 Zoomed mass spectrums of leucine/isoleucine in targeted and untargeted platform. S5.1a, SRM transitions of leucine/isoleucine showed fragments at m/z=44.18 and m/z=86.18. S5.1b, HR mass spectrum of compound m/z=132.1028. The lavender bar means the labeled centroid matches the monoisotopic mass of the expected compound ion, green bar means the labeled centroid matches the delta mass and the relative intensity of the theoretical isotope pattern within the specified tolerances and red means the expected centroid for this m/z value is missing or its intensity does not fall within the tolerance range for the theoretical isotope pattern.

155

Table S5.1 Targeted compounds detected by LC-QQQ-MS.

Precursor Product 1 Product 2 Collision Collision Compound Polarity (m/z) (m/z) (m/z) Energy 1 (V) Energy 2 (V) Caffeine Positive 195 110 138 20 20 4-Hydroxybenzoate Positive 139 104 121 21 10 Deoxyadenosine Positive 252 136 10 Oxaloacetic acid Negative 131 59 87 10 10 Thiamine Positive 266 122 123 16 10 Methylmalonate Negative 117 55 73 25 11 Succinic acid Negative 117 73 99 10 10 Lauroylcarnitine Positive 344 85 285 23 16 Riboflavin Positive 377 198 243 36 22 Trigonelline Positive 138 92 120 19 10 Adenine Negative 134 92 107 18 Adenosine Positive 268 136 10 Hypoxyxanthine Negative 135 92 133 17 20 Dopamine Positive 154 137 10 Cytidine Positive 244 95 112 41 16 Pterin Negative 162 119 129 17 11 Creatinine Positive 114 44 86 17 11 Guanosine Positive 285 152 10 N-acetyl-D-glucosamine Positive 222 126 138 10 10 Pyridoxine Positive 170 134 152 21 12 Xanthine Positive 153 99 121 10 13 Noradrenaline Positive 170 93 152 26 10 Purine Positive 121 67 94 28 21 Tryptamine Positive 161 117 144 26 10 5-aminolevulinic acid Positive 132 90 114 10 11 Cytosine Positive 112 69 95 10 10 Uracil Negative 111 42 111 10 10 Histamine Positive 112 68 95 20 11 Fumarate Negative 115 27 71 10 10 Indole-3-acetate Positive 176 130 145 16 14 Sorbate Negative 111 67 93 10 11 Carnosine Positive 227 193 209 26 10 Maleic acid Negative 115 71 97 10 10 Acetylcholine Positive 147 87 88 14 10 Guanidinoacetate Positive 118 72 100 16 14 Guanine Positive 152 108 135 10 10 5-aminopentanoate Positive 118 55 101 16 10 4-guanidinobutanoate Positive 146 86 87 16 13 2, 3-Dihydroxybenzoic Negative 153 109 10 acid Pyruvate Negative 87 43 87 10 10 Uridine Positive 245 113 117 10 10 Theophylline Positive 181 124 163 19 10 (S)-dihydroorotate Negative 157 42 113 26 10 (cyclic CMP) cytidine 2'3'- Positive 306 112 178 10 10 cyclic monophosphate (cyclic GMP)guanosine 3', Negative 344 108 150 20 20 5'-cyclic monophosphate Glucose Negative 179 59 71 10 10 Ethanolamine Positive 62 44 46 12 34 Phosphoenolpyruvate Positive 169 123 151 10 40 Melatonin Positive 233 159 174 27 14 Pentanoate Positive 103 43 85 14 10

156

Serotonin Positive 177 115 160 29 10 Acetyl-CoA Positive 810 303 428 10 20 Citrulline Positive 176 113 159 10 10 3-Hydroxyanthranilate Positive 154 122 136 10 10 (cyclic AMP) adenosine 3'5'-cyclic Positive 330 136 312 20 20 monophosphate (DAMP) deoxyadenosine Negative 330 134 195 20 20 monophosphate Glyceraldehyde-3- Negative 169 79 97 10 10 phosphate Lipoamide Positive 206 161 189 14 10 Octopamine Positive 154 113 136 10 10 Urocanate Positive 139 91 111 17 10 AMP Positive 348 69 119 40 40 Glutathione Positive 308 162 179 15 11 FAD Negative 784 181 346 20 20 4-Aminobenzoate Positive 138 94 120 13 15 Glucosamine-6- Positive 260 84 98 10 20 phosphate Malate Negative 133 71 115 10 10 Tyramine Positive 138 77 121 27 10 (DCMP) 2'-deoxycytidine Positive 308 95 112 20 20 5'-monophosphate 5-hydroxymethyluracil Positive 143 97 125 10 10 Myo-inositol Positive 181 133 163 22 10 Allantoin Negative 157 97 114 17 15 (GMP) guanosine 5'- Negative 362 79 211 20 20 monophosphate (CMP) cytidine Negative 322 79 97 20 20 monophosphate NAD Positive 664 428 524 20 20 Nicotinamide Positive 123 78 80 23 20 (DGDP)2-deoxyguanosine Negative 426 159 408 20 20 5'-diphosphate (ADP) adenosine 3'5'- Negative 426 134 328 10 10 diphosphate 3-phosphoglyceric acid Negative 185 79 97 10 10 Homocysteine Positive 136 90 91 11 19 UDP-glucose Negative 565 323 40 N-Acetyl-ornithine Positive 175 115 158 10 10 4-aminobutanoate Positive 104 86 87 11 10 Cadaverine Positive 103 69 86 15 10 Ornithine Positive 133 70 116 10 10 3-ureidopropionate Positive 133 90 115 10 10 Stachyose Positive 667 417 16 S-Adenosyl-methionine Positive 399 136 250 10 10 Putrescine Positive 89 61 71 10 10 NADPH Positive 746 136 150 40 40 Glutathione disulfide Positive 613 355 484 10 10 (CMP) cytidine Negative 322 79 97 20 20 monophosphate Succinate semialdehyde Negative 101 57 69 10 10 Orotate Negative 155 42 111 23 10 N-acetyl-L-alanine Negative 130 88 12 Adenosine 2'3'-cyclic Negative 266 150 176 20 21 monophosphate (S)-dihydroorotate Negative 157 42 113 26 10

157

N-acetyl-DL-methionine Negative 190 142 148 15 12 Ascorbate Negative 175 87 115 19 10 Guanosine 3',5'-cyclic Negative 344 133 150 34 23 monophosphate Glycolate Negative 75 31 47 10 10 Xanthosine Negative 283 151 165 20 22 3-nitro-L-tyrosin Negative 225 163 181 10 10 Urate Negative 167 124 147 15 10 Sarcosine Negative 88 42 45 13 10 Thymidine-5'-diphospho- Negative 563 241 321 30 22 alpha-D-glucose Xanthosine 5'- Negative 363 151 211 27 18 monophosphate 3'-CMP Negative 322 110 211 26 18 (S,S)-tartaric acid Negative 149 73 87 17 11 L-cystathionine Negative 221 120 134 12 11 2'-deoxyguanosine 5'- Negative 506 408 426 23 24 triphosphate Inosine Positive 269 181 253 46 26 Tetrahydrofolate Positive 446 342 430 16 11 Homocystine Positive 269 181 253 46 25 Theobromine Positive 181 138 163 17 16 Citramalate Positive 149 93 121 18 16 Deoxyribose Positive 135 94 107 10 10 Indoxyl sulfate Positive 134 93 106 10 15 Isopentyl pyrophosphate Positive 298 206 282 55 10 Thymidine Positive 243 117 127 11 10 Mevalolactone Positive 131 69 113 10 10 Methylguanidine Positive 74 33 42 13 10 Adenosine Positive 268 119 136 46 19 1-methylnicotinamide Positive 138 93 95 21 20 Taurine Positive 126 81 108 16 10 2, 4-dihydroxypteridine Positive 165 120 148 21 15 6'-hydroxynicotinate Positive 140 94 122 22 18 Aniline-2-sulfonic acid Positive 174 92 156 17 10 (2R, 3R)-(-)-2, 3- Positive 91 50 65 10 20 butanediol Betaine Positive 118 58 59 23 19 2, 4- dihydroxypyrimidine-5- Positive 157 116 139 10 12 carboxylic acid Deoxycytidine Positive 228 95 112 35 10 Kynurenine Positive 209 179 193 22 19 Paraxanthine Positive 181 109 124 10 21 Trigonelline Positive 138 92 120 19 10 Ethyl 3-ureidopropinoate Positive 161 92 120 13 10 Hypotaurine Positive 110 82 92 12 10 3-aminoisobutanoate Positive 104 30 86 13 10 D-(+)-glucosamine Positive 180 139 163 12 10 1, 3-diaminopropane Positive 75 43 57 10 10 3-hydroxybutanoic acid Positive 105 73 87 10 10 Thiourea Positive 77 43 60 36 18 Uridine-5- Positive 325 185 307 26 12 monophosphate Ethanolamine Positive 62 44 46 12 34 N-formylglycine Positive 104 45 63 17 10 S-hexyl-glutathione Positive 392 246 263 14 11 3-methylhistamine Positive 126 96 109 21 16

158

Maleamate Positive 116 72 88 13 13 Maleimide Positive 98 57 70 10 14 Creatinine Positive 114 44 86 17 11 SN-glycero-3- Positive 259 105 125 15 27 phosphocholine Cytidine 2'3'- Negative 322 79 97 20 20 cyclicmonophosphate Choline Positive 105 51 77 27 10 Mesoxalate Positive 119 91 101 16 10 2-amino-2- Positive 104 45 58 10 11 methylpropanoate Ophthalmic acid Positive 290 161 215 12 10 Creatine Positive 132 86 114 10 10 Homoserine Positive 120 56 74 17 10 Deoxycarnitine Positive 146 60 87 14 16 Glyceraldehyde Positive 91 61 73 10 10 Hydroxypyruvate Positive 105 63 64 10 10 Inosine 5'-triphosphate Positive 509 263 345 25 10 Galactitol Positive 183 129 165 10 10 Pyridoxamine Positive 169 134 152 22 2-hydroxybutyric acid Positive 105 23 64 10 Epinephrine Positive 184 166 11 Glucosaminate Positive 196 132 178 14 10 Inosine 5'- Positive 349 137 14 monophosphate Nepsilon, nepsilon- Positive 189 84 130 21 14 trimethyllysine Agmatine sulfate Positive 131 89 113 10 10 N1-acetylspermine Positive 245 129 171 14 14 D-Psicose Positive 181 149 163 10 10 Mannose Positive 181 163 10 Biotin Positive 245 227 10 3-Hydroxyphenylacetate Positive 153 125 11 Phenylacetic acid Negative 135 91 11 3-Hydroxybenzoate Negative 137 93 13 Diacetyl Negative 85 41 14 2- Negative 129 41 85 15 10 Methylmaleate/Itaconate 4-Hydroxy-3- Positive 455 437 10 methoxyphenylglycol Myo-inositol Positive 181 133 163 22 10 Trans-cinnamaldehyde Positive 133 115 11 3-(4- Negative 181 163 10 Hydroxyphenyl)lactate Salicylamide Positive 138 121 121 15 11 Phenethylamine Positive 122 105 10 N,N-Dimethyl-1,4- Positive 137 122 10 phenylenediamine 3-Methyl-2-oxindole Positive 148 130 18 Pyruvic aldehyde Positive 73 55 10 1-Methyl-6,7-dihydroxy- 1,2,3,4- Positive 180 145 163 10 10 tetrahydroisoquinoline D-Gulonic acid gama- Negative 177 89 12 lactone Dethiobiotin Positive 215 179 197 10 10 Pyridoxal Positive 168 150 11 2,6-Dihydroxypyridine Positive 112 94 16

159

Suberic acid Positive 175 157 10 3-Methyl-2-oxovaleric Positive 131 78 30 acid Resorcinol monoacetate Positive 153 111 10 Indole-3-acetaldehyde Positive 160 118 142 10 10 3-Methylbutanal Negative 85 41 10 5-Hydroxyindoleacetate Positive 192 146 15 Ribitol Positive 153 135 10 3-Methoxy-4- Negative 197 137 23 hydroxymandelate 4-Quinolinecarboxylic Positive 174 146 10 acid Ethyl 3-indoleacetate Positive 204 130 10 Methyl indole-3-acetate Positive 190 130 172 10 10 N-acetylserotonin Positive 219 160 202 10 10 Homovanillate Positive 183 137 10 3-Methyladenine Positive 150 133 10 N-acetyl-L-leucine Positive 174 128 156 10 10 Azelaic acid Positive 189 171 10 2-Methylglutaric acid Positive 147 129 10 Adipic_acid Positive 147 129 10 Mono-methyl glutarate Positive 147 129 10 2',4'- Positive 153 135 15 dihydroxyacetophenone Homogentisate Positive 169 123 12 Mannitol Positive 183 165 10 Tryptophan Positive 205 146 188 10 10 Phenylalanine Positive 166 103 120 10 10 Leucine/Isoleucine Positive 132 44 86 10 10 Aspartic acid Positive 134 74 88 10 10 Methionine Positive 150 104 133 10 10 Asparagine Positive 133 74 87 10 10 Tyrosine Positive 182 136 165 10 10 Cysteine Positive 122 59 76 10 10 Valine Positive 118 55 72 10 10 Proline Positive 116 70 10 Alanine Positive 90 44 45 10 10 Glycine Positive 76 30 47 10 10 Threonine Positive 120 56 74 10 10 Serine Positive 106 42 60 10 10 Glutamine Positive 147 84 130 10 10 Glutamic acid Positive 148 102 130 10 10 Histidine Positive 156 95 110 10 10 Arginine Positive 175 60 70 10 10 Lysine Positive 147 84 130 10 10

160

Table S5.2 Untargeted compounds detected by LC-Orbi-MS. The compounds were matched with Chemspider database.

Name Formula Molecular Weight

Zalcitabine C9 H13 N3 O3 211.0964 Welkstoff C9 H15 N3 O7 277.0917 Vigabatrin C6 H11 N O2 129.0796 UDP-GlcNAc C17 H27 N3 O17 P2 607.0866 Trimethadione C6 H9 N O3 143.0591 Threonylserine C7 H14 N2 O5 206.091 Threonylglutamine C9 H17 N3 O5 247.1178 Tetraacetylethylenediamine C10 H16 N2 O4 228.1119 Terbuthylazine C9 H16 Cl N5 229.1072 Styrene C8 H8 104.0632 S-Propyl hexanethioate C9 H18 O S 174.1065 Sinomenine C19 H23 N O4 329.1603 S-Adenosyl-L-methionine C15 H22 N6 O5 S 398.1379 Risperidone C23 H27 F N4 O2 410.2159 Risedronic acid C7 H11 N O7 P2 283.0025 Riboprine C15 H21 N5 O4 335.1591 Resolvin D2 C22 H32 O5 376.2261 Propanidid C18 H27 N O5 337.1867 Propafenone C21 H27 N O3 341.1961 Prometon C10 H19 N5 O 225.158 Pregabalin C8 H17 N O2 159.1267 Pivagabine C9 H17 N O3 187.1217 Pilocarpine C11 H16 N2 O2 208.1219 Phenethylamine C8 H11 N 121.0897 Paeoniflorin C23 H28 O11 480.1665 Oxiracetam C6 H10 N2 O3 158.0695 Oxazolidinone C3 H5 N O2 87.03229 Oxagrelate C14 H16 N2 O4 276.1117 Norhaman C11 H8 N2 168.0695 Nicotinamide C6 H6 N2 O 122.0486 N1,N12-Diacetylspermine C14 H30 N4 O2 286.2375 N6-[(2R)-3,4-Dihydro-2H-pyrrol-2- C11 H19 N3 O3 241.1433 ylcarbonyl]-L-lysine N2-(1-Carboxyethyl)-N5- C9 H18 N4 O4 246.1332 (diaminomethylene)ornithine N-(4,6-Diamino-5- C5 H7 N5 O 153.0657 pyrimidinyl)formamide N-(1-Hydroxycyclopropyl)glutamine C8 H14 N2 O4 202.0961 N(1)-acetylspermidine C9 H21 N3 O 187.1687

161

Methyl (2E)-4-{[(5alpha,6beta)-17- (cyclopropylmethyl)-3,14-dihydroxy-4,5- C25 H30 N2 O6 454.2086 epoxymorphinan-6-yl]amino}-4-oxo-2- butenoate Methionylserine C8 H16 N2 O4 S 236.084 meprobamate C9 H18 N2 O4 218.1276 L-Pyroglutamic acid C5 H7 N O3 129.0433 L-Proline C5 H9 N O2 115.064 Leu-Val C11 H22 N2 O3 230.1633 Leu-pro C11 H20 N2 O3 228.148 L-Alanyl-L-glutamine C8 H15 N3 O4 217.1072 L-(+)-Valine C5 H11 N O2 117.0795 L-(+)-Leucine C6 H13 N O2 131.0951 L-(+)-Aspartic acid C4 H7 N O4 133.0381 L-(-)-Serine C3 H7 N O3 105.043 L-(-)-Asparagine C4 H8 N2 O3 132.0541 Isopimpinellin C13 H10 O5 246.0516 Isoleucylasparagine C10 H19 N3 O4 245.1389 Ipazine C10 H18 Cl N5 243.1228 Hexahydro-2-oxo-1h-thieno(3,4- C10 H16 N2 O3 S 244.0889 d)imidazole-4-Pentanoic acid Glucosaminate C6 H13 N O6 195.0749 gamma-Glutamylleucine C11 H20 N2 O5 260.1381 gamma-Glu-gln C10 H17 N3 O6 275.1124 gamma-Glu-Ala C8 H14 N2 O5 218.0912 Ethyl (4beta,8alpha,9R)-6'- C23 H28 N2 O4 396.2019 methoxycinchonan-9-yl Carbonate Epirizole C11 H14 N4 O2 234.1121 Dulcin C9 H12 N2 O2 180.0908 D-PANTOTHENIC ACID C9 H17 N O5 219.1112 DL-TYROSINE C9 H11 N O3 181.0745 DL-Tryptophan C11 H12 N2 O2 204.0906 DL-Phenylalanine C9 H11 N O2 165.0798 DL-Ornithine C5 H12 N2 O2 132.0904 DL-Lysine C6 H14 N2 O2 146.1061 DL-Glutamine C5 H10 N2 O3 146.0702 DL-Glutamic acid C5 H9 N O4 147.0537 DL-dethiobiotin C10 H18 N2 O3 214.1324 DL-Citrulline C6 H13 N3 O3 175.0961 Difeterol C25 H29 N O2 375.2167 Dexamethasone C22 H29 F O5 392.2025 Daminozide C6 H12 N2 O3 160.0855 Cytosine C4 H5 N3 O 111.0436 Creatinine C4 H7 N3 O 113.0594 CDP-glycerol C12 H21 N3 O13 P2 477.0559 Caroverine C22 H27 N3 O2 365.2071

162

Butopyronoxyl C12 H18 O4 226.1186 Butabarbital C10 H16 N2 O3 212.117 beta-D-Fructofuranosyl 4-O-dodecanoyl- C24 H44 O12 524.2822 alpha-D-glucopyranoside beta-D-Ethyl glucuronide C8 H14 O7 222.0754 Benserazide C10 H15 N3 O5 257.1036 Atraton C9 H17 N5 O 211.1426 Asp-gln C9 H15 N3 O6 261.0966 Arg-pro C11 H21 N5 O3 271.165 Aminolevulinic acid C5 H9 N O3 131.059 Alanyltryptophan C14 H17 N3 O3 275.1278 Afegostat C6 H13 N O3 147.0903 Adenine C5 H5 N5 135.0551 Aceglutamide C7 H12 N2 O4 188.0806 6-(6-Amino-9H-purin-9-yl)tetrahydro- 4H-furo[3,2-d][1,3,2]dioxaphosphinine- C10 H12 N5 O6 P 329.0539 2,7-diol 2-oxide 5-hydroxy-2-oxo-4-ureido-2,5-dihydro- C5 H6 N4 O5 202.0334 1H-imidazole-5-carboxylic acid 5-Amino-1-(5'- C8 H14 N3 O7 P 295.0594 phosphofuranoribosyl)imidazole methyl [9-(3-carbamoyl-3,4-dihydroxy- 16-methoxy-1-methyl-6,7- didehydroaspidospermidin-15-yl)-5- ethyl-5-hydroxy-1,4,5,6,7,8,9,10- C18 H20 N O3 P S 361.0904 octahydro-2H-3,7- methanoazacycloundecino[5,4-b]indol-9- yl]acetate 4-Aminophenol C6 H7 N O 109.0524 Prometryn C12 H18 N4 O2 250.1436 3-Methylsulfolene C5 H8 O2 S 132.0251 3,6,9,12-tetraoxatridecan-1-ol C9 H20 O5 208.1315 3,6,9,12,15,18-Hexaoxaicosane-1,20-diol C14 H30 O8 326.1947 3,4-Dihydroxy-7-methoxy-2,2,6- trimethyl-2,3,4,6-tetrahydro-5H- C16 H19 N O5 305.1236 pyrano[3,2-c]quinolin-5-one

3-(3,4-Dihydroxyphenyl)-2,7,8- C18 H10 O8 354.0405 trihydroxydibenzo[b,d]furan-1,4-dione

2-Acetamido-2-deoxy-D-glucitol C8 H17 N O6 223.1064 2-{[(3alpha,5beta,6alpha,7alpha)-3,6,7- Trihydroxy-24-oxocholan-24- C26 H45 N O7 S 515.2949 yl]amino}ethanesulfonic acid 2,5-Diethyl-3,6-dimethylpyrazine C10 H16 N2 164.1322 13S-hydroxyoctadecadienoic acid C18 H32 O3 296.2353 (E)-p-coumaric acid C9 H8 O3 164.048

163

(3Z,5E,9E,17E,19E)-1,7,15-Trihydroxy- 4,6,10,16,22,28-hexamethyl-13-[(4E)- 6,7,9,11-tetrahydroxy-4,10-dimethyl-4- C48 H76 O12 844.534 tridecen-2-yl]-12,29- dioxabicyclononacosa-3,5,9,17,19- pentaene-8,11,23-trione (2S)-5-Carbamimidamido-2-(2-oxo-1- C9 H16 N4 O3 228.123 azetidinyl)pentanoic acid (2S)-2-Piperazinecarboxamide C5 H11 N3 O 129.0913

(1R,2S,3S,6R,7R,11R,14R,16S)-7- Isopropyl-2,6-dimethyl-19-oxa-17- C22 H33 N O2 343.2479 azapentacyclononadec-9-en-8-one

(10E,12Z)-9-Hydroperoxy-10,12- C18 H32 O4 312.2306 octadecadienoic acid

164

APPENDIX D Supporting information for Chapter 6 Table S6.1 Characteristics of bourbon samples used in this study. Table includes number of samples (n), percent alcohol by volume (%ABV), pH values (mean and standard deviation), and colors of sample.

0-year 2-year 4-year 6-year Barrel-1 Barrel-2 Barrel-3 n= 3 3 3 3 3 3 3 pH (Mean) 5.09 4.17 4.08 4.18 3.99 4.00 3.98 pH (STDEV) 5.07E-02 9.57E-03 5.77E-03 1.00E-02 1.29E-02 2.06E-02 1.91E-02 ABV (%) 40 40 40 40 40 40 40 Color Clear Light brown Brown Brown Brown Brown Brown

165

Table S6.2 Twelve compounds initially abundant in the unaged spirit but are at much lower levels in the 6-year aged bourbon.

# MW RT(min) Intensity in 0-year Intensity in 2-year Intensity in 4-year Intensity in 6-year

A 175.9639 0.76 8.64E+04 7.79E+04 6.86E+04 5.37E+04

B 147.9688 0.75 1.62E+04 1.55E+04 1.43E+04 1.10E+04 C 435.8775 0.75 9.75E+03 1.12E+04 8.38E+03 4.36E+03

D 515.8985 0.80 1.45E+04 - - -

E 521.9151 0.82 2.79E+04 - - - F 589.9022 0.81 1.75E+04 - - -

G 379.9243 0.79 1.93E+04 1.78E+03 1.00E+03 2.27E+02

H 385.9407 0.81 2.45E+04 5.84E+02 2.20E+02 4.95E+01 I 243.9508 0.78 2.93E+04 1.63E+04 1.25E+04 8.45E+03

J 181.9807 0.81 2.04E+04 1.26E+04 1.07E+04 6.48E+03

K 317.9546 0.81 6.14E+04 6.22E+03 3.13E+03 5.64E+02 L 249.9676 0.81 5.03E+04 1.49E+04 1.02E+04 3.79E+03

166

Table S6.3 Number of elemental formulae detected from different bourbon samples by FT-ICR MS.

0-year 2-year 4-year 6-year Elemental Number of Number of Number of Number of Formula Percentage Percentage Percentage Percentage annotation annotation annotation annotation CHO 29 50.0% 149 59.6% 163 52.1% 153 54.8% CHOS 6 10.3% 9 3.6% 6 1.9% 10 3.6% CHON 21 36.2% 90 36.0% 144 46.0% 115 41.2%

CHONS 2 3.4% 2 0.8% 0 0.0% 1 0.4%

167

Figure S6.1 An example of ion with m/z of 300.99989 was observed in maturated samples but not in non-matured whiskey sample and was identified as the deprotonated ellagic acid (C14H5O8).

168

A B 500.3118 C 502.3277

170.0215

434.0474

208.0737

194.0427

Figure S6.2 Metabolic patterns of 2, 4 and 6-year bourbons. (A) PCA score plot differentiates the 2 (red), 4 (green) and 6 (dark blue)- year whiskey based on their metabolic profiles. Ovals show 95% confidence level. (B) Loading plot of the first two principal components. Representative compounds that discriminate the groups were labeled by their m/z. (C) Heat map (top 75) presentation of metabolic profiles from 2, 4 and 6-year whiskey.

169

Figure S6.3 Box plots of ten characteristic compounds showing significant differences from bourbon whiskies that aged differently (2, 4 and 6 years). All these compounds were annotated to their potential IDs based on their accurate m/z. Detailed information can be seen in Table 6.1.

170

Gallic acid

Dimethyl itaconate 2-Furoic acid

Glutaric anhydride Monobutyrin

1,4-Dihydroxy-2-naphthoic acid

Figure S6.4 Loading plot of the first two principal components. Representative compounds that discriminate the groups were labeled by their potential IDs.

171

172