1 NMR Spectroscopy As a Robust Tool for the Analysis of Lipids In

NMR Spectroscopy as a Robust Tool for the Analysis of Lipids in Fish Oil Supplements and Coffee

Thesis

Presented in Partial Fulfillment of the Requirements for the Degree Master of Science in

the Graduate School of The Ohio State University

Kathryn Maxine Williamson

Graduate Program in Food Science and Technology

The Ohio State University

2018

Thesis Committee

Emmanuel Hatzakis, Advisor

Yael Vodovotz

Luis Rodriguez-Saona

Copyrighted by

Kathryn Maxine Williamson

2018

Abstract

Fish Oil

The western diet is poor in n-3 fatty acids, therefore the consumption of fish oil supplements is recommended to increase the intake of these essential nutrients. The objective of this work is to demonstrate the qualitative and quantitative analysis of encapsulated fish oil supplements using high-resolution 1H and 13C NMR spectroscopy utilizing two different NMR instruments; a 500 MHz and an 850 MHz instrument. Both proton (1H) and carbon (13C) NMR spectra can be used for the quantitative determination of the major constituents of fish oil supplements. Quantification of the lipids in fish oil supplements is achieved through integration of the appropriate NMR signals in the relevant 1D spectra. Results obtained by 1H and 13C NMR are in good agreement with each other, despite the difference in resolution and sensitivity between the two nuclei and the two instruments. 1H NMR offers a more rapid analysis compared to 13C NMR, as the spectrum can be recorded in less than 1 min, in contrast to 13C NMR analysis, which lasts from 10 min to one hour. The 13C NMR spectrum, however, is much more informative. It can provide quantitative data for a greater number of individual fatty acids and can be used for determining the positional distribution of fatty acids on the glycerol backbone.

Both nuclei can provide quantitative information in just one experiment without the need

ii of purification or separation steps. The strength of the magnetic field mostly affects the

1H NMR spectra due to its lower resolution with respect to 13C NMR, however, even lower cost NMR instruments can be efficiently applied as a standard method by the food industry and quality control laboratories.

Coffee oil

Approximately 15% of the mass of an Arabica coffee bean, Coffea Arabica, commonly known as Arabica coffee, consists of lipids. This lipid fraction has the potential for a number of applications in the food, cosmetic and pharmaceutical industries, and contains several compounds that may act as biomarkers for classification and evaluation of Arabica coffee. The objective of this study is to employ multinuclear and multidimensional NMR spectroscopy as a rapid and reliable method for the quantitative analysis and evaluation of the non-polar, including unsaponifiable, fraction of Coffea Arabica. Our results suggest that NMR can be a valuable tool for the determination of many compounds in coffee oil and can be used for quantifying the impact of the coffee roasting process. Green and roasted coffee beans, as well as brewed coffee and spent coffee grounds, were analyzed for their lipid components. A number of gradient-selected two-dimensional NMR techniques were applied for a systematic two- dimensional analysis of the various components in coffee oil, including FA, terpenes, oxidation and hydrolysis products, caffeine and sterols. Quantification was achieved by integration of the appropriate diagnostic signals in the NMR spectra using 2,6-Di-tert- butyl-4-methylphenol (BHT) as an internal standard (IS), as well as the PULCON method, which offers several advantages compared to IS. Bland-Altman analysis showed iii that PULCON and IS approaches are in a good agreement. Overall, it was found that the major fatty acids in coffee oil are linoleic, oleic, linolenic and saturated fatty acids.

Targeted analysis showed that, with the exception of linolenic acid, only minor changes occur in the fatty acid profile during roasting. A statistically significant increase occurs in the secondary oxidation product 2,4-hepta-dienal and free fatty acids after roasting.

Additionally, 1,3-diacylglycerides significantly decrease with roasting due to their instability to hydrolysis. Untargeted analyses, namely PCA and OPLS-DA, revealed differences between green and roasted samples. MRI indicated significant morphological changes in coffee beans due to roasting, which may be responsible for these compositional variations. Finally, lipids extracted from spent coffee grounds can be successfully epoxidized and are therefore promising precursors for the production of bioplastics.

Dedication

This thesis is dedicated to my newfound family in Guatemala. My research would not be possible without their hard work and the hard work of millions of other farmers around the world dedicated to growing coffee.

Acknowledgments

This thesis would not have been possible without the dedicated work of my graduate committee members Dr. Emmanuel Hatzakis, Dr. Yael Vodovotz, and Dr. Luiz

Rodriguez-Saona for guiding me in coursework, research direction, and thesis compilation. I would like to express my extreme gratitude to Dr. Hatzakis who instilled a love of NMR in me and has supported me through my entire graduate experience. This research would not be possible without his expertise in NMR, chemistry, and lipid analysis and I have learned more than I could have imagined thanks to him. I would also like to thank Dr. Vodovotz for introducing me to research as a Freshman at Ohio State and influencing my decision to attend graduate school. Additionally, Dr. Rodriguez has provided me with a majority of my background in food analysis and I would not be the student I am today without his guidance.

Additionally, I would like to thank Chunhua Yuan, who helped me run every experiment on the NMR spectrometers and provided great assistance in data acquisition and analysis. MRI analysis would not have been possible without the assistance of Jiadi

Xu at the MRI facility at the Kennedy Krieger Institute at John’s Hopkins University.

Special thanks to Sravanti Paluri for providing MRI expertise and for organizing the research collaboration with the Kennedy Krieger Institute. I would like to thank the

Peterson lab, specifically Sichaya Sittipod for providing me with coffee samples, as well as Geoff Dubrow for allows offering research guidance.

Finally, I would like to thank the Hatzakis lab and my undergraduate students

Morgan Vasas and Connie Yuan for their assistance in the lab and willingness to learn.

Thank you to the Rodriguez lab, Jimenez lab, Kopec lab, and Vodovotz lab for always being willing to lend lab supplies or equipment as the Hatzakis lab got started. Finally, I would like to thank my parents for their continued support, as well as my best friends,

Meredith Myers and Kenzi Hannum, for their support with schoolwork and their willingness to watch many NMR presentations.

vii

Vita

November 18th, 1993……………………………….. Born- Camden, NJ, USA

2012………………………………………………… High School Degree- Lenape High

School, Medford, NJ

2016………………………………………………… Bachelor’s degree in Food Science and

Technology with Research Distinction at

The Ohio State University

2016-2018…………………………………………… Graduate Research Associate at The

Ohio State University. Masters of Science in

Food Science and Technology utilizing

NMR spectroscopy for the analysis of fish

oil and coffee oil.

Publications

1) Dais, P.; Plessel, R.; Williamson, K.; Hatzakis, E. Complete 1H and 13C NMR

assignment and 31P NMR determination of pentacyclic triterpenic acids.

Analytical Methods, 2017, 9, 949-957.

2) Williamson, K.; Hatzakis, E. NMR Spectroscopy as a robust tool for the rapid

evaluation of the lipid profile of fish oil supplements. Journal of Visualized

Experiments, 2017, 123, e55547. viii

Field of Study Major Field: Food Science and Technology

Table of Contents

Abstract ...... ii Fish Oil...... ii Coffee oil ...... iii Dedication ...... v Acknowledgments...... vi Vita ...... viii Table of Contents ...... x List of Tables ...... xiii List of Figures ...... xiv List of Abbreviations ...... xvi Chapter 1. Introduction ...... 1 Nuclear Magnetic Resonance spectroscopy ...... 1 NMR-based metabolomics...... 4 Quantitation...... 5 NMR in food ...... 7 Fish oil supplements ...... 8 Analysis...... 10 Coffee ...... 12 Coffee oil ...... 14 Lipids in green and roasted coffee beans ...... 16 Coffee lipid analysis ...... 18 Chapter 2. Statement of the Problem ...... 20 Chapter 3. Objectives ...... 22 Fish Oil...... 22 Coffee Oil...... 22 Chapter 4. Materials and Methods ...... 23 Chapter 5. NMR Spectroscopy as a Robust Tool for the Rapid Evaluation of the Lipid Profile of Fish Oil Supplements ...... 24 Short Abstract ...... 24 x

Long Abstract...... 24 Introduction ...... 25 Protocol ...... 27 1. NMR Sample Preparation ...... 27 2. NMR Instrument preparation ...... 28 Representative Results ...... 39 1H NMR analysis ...... 40 13C NMR analysis ...... 43 Figure Legends...... 48 Discussion ...... 55 Modifications and strategies for troubleshooting ...... 55 Limitations of the technique ...... 57 Significance with respect to existing methods ...... 60 Future applications ...... 61 Critical steps within the protocol ...... 62 Acknowledgments...... 63 Disclosures ...... 64 Chapter 6. Coffee paper ...... 65 Introduction ...... 65 Material and Methods ...... 72 Coffee samples ...... 72 Chemicals ...... 73 Measurement of water activities and moisture content ...... 73 Synthesis of coffee oil epoxides ...... 73 Sample preparation for NMR experiments ...... 74 NMR experiments ...... 75 One-dimensional (1D) NMR spectra ...... 76 Two-dimensional (2D) NMR experiments ...... 77 MRI ...... 79 Spectral Data Processing and Multivariate Data Analysis ...... 80 Univariate data analysis ...... 80 PULCON-based quantification ...... 80

Results and Discussion ...... 81 Compound identification-Spectral assignment ...... 81 Fatty acid composition ...... 82 Diterpenes, sterols and caffeine ...... 84 Oxidation-hydrolysis products and phospholipids...... 87 Quantitative NMR-PULCON method ...... 92 Effect of roasting on coffee lipids composition ...... 100 Physical transformation in the coffee bean due to roasting ...... 105 Coffee lipids following brewing ...... 107 Supplemental Material ...... 112 Chapter 7. Conclusions ...... 114 Bibliography ...... 116

xii

List of Tables

1 Table 1. H-NMR chemical shifts of fish oil lipids in CDCl3 solution ...... 51 13 Table 2. C -NMR chemical shifts of fish oil lipids in CDCl3 solution ...... 52 Table 3. Diagnostic signals in the 1H and 13C NMR spectra that can be used for the quantitative determination of various components in coffee oil...... 95 Table 4. LO, 1,3-DG, Cafestol, Campesterol, Cafestol and Kaheol (μmoles/g) determined by NMR Spectroscopy in green coffee using IS and PULCON methods ...... 97 Table 5. Linear regression data of PULCON values with internal standard values...... 98 Table 6. Paired sample t-test values comparing the aldehyde doublet, aldehyde region, and 1,3-DG region in 1H and FFA in 13C values in 18 green and roasted coffee samples, normalized to total intensity (ɑ 0.05) ...... 103

xiii

List of Figures

Figure 1. PULCON schematic and equation. (2018) ...... 7 Figure 2. Chemical structures of DHA and EPA...... 9 Figure 3: Chemical composition of Green Coffea Arabica and Coffea Canephora. (Farrah, 2010) ...... 13 Figure 4. The 1H NMR analysis. 850.23 (A) and 500.20 MHz (B) 1H-NMR spectrum of a fish oil supplement in CDCl3 solution...... 48 Figure 5. The 13C NMR analysis. 213.81 (A) and 125.77 MHz (B) 13C-NMR spectrum of a fish oil supplement in CDCl3 solution in the carbonyl carbon region...... 49 Figure 6. Fish oil oxidation. The 1H NMR spectrum of oxidized fish oil depends on the oxidation conditions...... 50 Figure 7. Comparison between the 13C NMR spectra acquired using the standard broadband decoupling (A) and the inverse gated decoupling (B) pulse sequences...... 54 Figure 8. 850 MHz (A) 1H-NMR spectrum and (B) 213 MHz 13C-NMR spectrum of coffee oil in the olefinic region of a coffee oil sample in CDCl3:DMSO-d6 solution. OL, oleic acid; LO, linoleic acid; LN, linolenic acid; K, Kahweol; C, Cafestol; Cf, Caffeine...... 82 1 1 Figure 9. 850 MHz H- H TOCSY spectrum of green coffee oil in CDCl3:DMSO-d6 solution, showing the connectivity between protons in the same spin system...... 83 Figure 10. 850 MHz 1H–13C HSQC-DEPT spectrum of roasted coffee oil in CDCl3:DMSO-d6 solution, showing the one bond connectivity between protons and carbons...... 85 Figure 11. 850 MHz band selective constant time 1H–13C HMBC spectrum of roasted coffee oil in CDCl3:DMSO-d6 solution, showing the long range connectivity between the carbonyl carbons and the glycerol backbone protons and the CHα, CHβ protons...... 90 Figure 12. 850 MHz 1H-31P HMBC spectrum of phospholipids isolated from roasted + coffee grounds, in CDCl3/CD3OD/D2O-EDTA-K (400:95:5 v/v/v)...... 92 Figure 13. Comparison of the 1H NMR spectra of a coffee oil sample with different concentrations of Cr(acac)3. 0 mM (A), 0.6 mM (B), 1.2 mM (C), 2.4 mM (D), 4.8 mM...... 93 Figure 14. Difference (bias) plots obtained from Bland Altman analysis of green coffee oil samples analyzed by IS and PULCON methods for (A) LO, (B) 1,3-DG, (C) Campesterol, and (D) Cafestol...... 100 Figure 15. The effect of roasting on the oxidation status of coffee oil. Comparisons between coffee oil extracted from green beans (A) and roasted beans (B)...... 102

xiv

Figure 16. PCA (A) and OPLS-DA (B) scores plots of green (green DP) coffee and roasted (brown DP) coffee samples...... 104 Figure 17. Proton density image of a green (A) and roasted (B) Arabica coffee bean. . 105 1 Figure 18. Expansions of a 700 MHz H NMR spectrum in CDCl3 showing the regions of methyl protons of lipids (A), aromatic compounds and oxygenated compounds with double bonds (B), aldehydes (C), peroxides/ enols/ FFA (D)...... 108 Figure 19. Epoxidation process as monitored through 1H NMR spectroscopy: pure coffee oil (A), partially epoxidized coffee oil (B), and fully epoxidized coffee oil (C)...... 111 Figure 20. Resolution enhancement of a 1H spectrum by applying a window function and vurce fitting...... 112 Figure 21. PULCON equation for quantification...... 112 Figure 22. Permutation tests for OPLS-DA...... 112 Figure 23. Lipids as appear in the 1H NMR spectrum of a water coffee extract ...... 113

List of Abbreviations

NMR Nuclear Magnetic Resonance

GC Gas Chromatography

MS Mass Spectrometry cGC Capillary Gas Chromatography

EIS-MS Electrospray Ionization- Mass Spectrometry

FT-IR Fourier Transformation Infrared Spectroscopy

FDA Food and Drug Administration

TG Triacylglycerol or Triglyceride

FA Fatty acids

FFA Free fatty acids

DAG/DG Diglyceride

EE Ethyl Ester

PLs Phospholipids n-3 Omega-3 n-6 Omega-6

CDCl3 Chloroform-d (deuterated chloroform) uL microliter mL milliliter mg milligram

ODS Office of Dietary Supplements

xvi

FDA Food and Drug Administration

NIH National Institute of Health

AOACI Association of Analytical Chemists (International)

DSLD Dietary Supplement Label Database

PCA Principle Component Analysis

OPLS-DA Orthogonal Projections to Latent Structures Discriminant Analysis

ERETIC Electronic Reference To access In vivo Concentrations

PULCON Pulse length–based Concentration determination

SCAA Specialty Coffee Association of America

DMSO-d6 Dimethyl Sulfoxide- d6

xvii

Chapter 1. Introduction

Nuclear Magnetic Resonance spectroscopy

Nuclear Magnetic Resonance (NMR) spectroscopy is a powerful and robust analytical method that exploits the magnetic properties of certain nuclei such as 1H, 13C,

21P and 19F. NMR can be applied to liquid and/or solid materials and it is mainly used to measure the concentration of various molecules in a sample, study the interaction between them, and elucidate the structure of organic compounds and biomolecules.

NMR-active nuclei possess nuclear spin associated with a nuclear magnetic moment, μ, and thus generate a tiny magnetic field. When a sample with NMR-active nuclei is placed into a strong magnetic field, such the ones generated by the NMR instruments, the nuclear magnetic moment, μ, of the nuclei interacts with the external field B0 generated by the NMR instrument and aligns in two spin states, +1/2 and -1/2, associated with two energy states. There is an excess of nuclei in the +1/2 energy state, and the difference in energy between these two spin states, which is B0 dependent, is what is measured. A radio frequency is pulsed through the sample and the energy emitted by the sample is measured. The most common NMR observations are the chemical shift (δ), which is expressed in ppm and provides information about the chemical environment of a nuclei; the scalar coupling (J), which is measured in Hz and is responsible for the signal splitting

1 as a result of the interaction between neighboring nuclei; and the signal area, which is proportional to the number of nuclei giving rise to the signal and thus can be used for identification purposes and under certain conditions for quantitative measurements.

The most common NMR experiment utilizes is the nuclei of a hydrogen (1H isotope) atom and is called a proton NMR experiment (1H NMR). Because of the high natural abundance of 1H, as well as the high gyromagnetic ratio and the short relaxation times of proton nuclei, 1H NMR experiments are characterized by high sensitivity and extremely fast experimental times as short as 2 to 3 minutes. However, because of the narrow spectral width of the 1H NMR spectra (~ 15 ppm) and the presence of scalar coupling, they often suffer from spectral crowding depending on the molecular complexity of the sample.

Carbon NMR experiments (13C NMR), rely on the presence of the 13C isotope, present in nature at only about 1.1%. This low abundance causes 13C NMR to be low in sensitivity in most cases. In addition, carbon nuclei have longer relaxation times compared to protons, causing experimental times of carbon experiments to be as long as one to two hours, or even longer. The most significant advantage of 13 C NMR experiments is its high spectral resolution resulting from the large spectral width of the

13C NMR spectrum (>220 ppm) and the appearance of singlets due to the application of proton decoupling which creates a single peak for each carbon in the sample and allows for rapid determination. When running carbon analysis for quantitative purposes, it is important that an appropriate pulse sequence be selected. For quantifications using 13C

NMR, an inverse gated decoupling experiment that employs broadband proton

2 decoupling only during the acquisition period is important to ensure there is no polarization transfer from 1H to 13C via the nuclear Overhauser effect (NOE). Fully decoupled NMR experiments can be used for quantitation, but it is important to keep in mind that there are different NOE factors and relaxations times among carbons with different multiplicities, so integral comparison between methyl, methylene, methane, and carbonyl carbons should be avoided. The fully decoupled method does allow for reliable comparison among carbons of similar multiplicity and similar chemical environments/relaxation times.

The sensitivity and the experimental time of an NMR experiment depends on several factors, such as sample concentration, the MW of the compounds to be analyzed and the type of nuclei that will be used for analysis (eg 1H vs 13C). In addition to the nature of the samples there are also factors related to the NMR hardware that will affect the final NMR results and experimental time. NMR instruments range in their magnetic frequency from 200 MHz all the way up to 1.2 GHz. Generally, the higher the magnetic field strength, the more sensitive the instrument. However, variations in instrumentation, such as the presence of cryoprobes, the type of coil (inverse vs. observe) and the abundance of NMR-active nuclei can also affect the obtained result. Cryogenically cooled probes, or cryoprobes, can enhance detection and sensitivity of a liquid sample up to 3-4 times its original detection, allowing smaller amounts of sample to be analyzed and faster acquisition (Kovacs et al.). The two most common types of NMR probes are inverse probes, where the coil for proton detection is closest to the sample, and is surrounded by an X-nuclei coil (eg carbon, phosphorus, nitrogen, etc). An observe probe

3 has the opposite makeup, with the X coil closest to the sample, followed by the proton.

This affects the sensitivity of the carbon and proton experiments, and the choice of probe depends on the nuclei to be analyzed.

The major drawback of NMR is its low sensitivity compared to other analytical techniques such as chromatography and mass spectrometry although recent advances in instrumentation and technology have increased the sensitivity up to the nanomolar range.

Food consists of great amounts of H and C, and thus NMR active nuclei, are present in measurable quantities in most foods. 1H NMR is usually the method of choice in food related studies, however, 13C and 31P NMR have also been used with great success, especially in cases that spectral resolution is an issue. In complex food matrices, there can be spectral crowding which makes it difficult to identify and quantify specific compounds. In this case, the application of 2D NMR spectroscopy and sample extraction and purification methods may help mitigate this challenge. An additional drawback of

NMR is its high cost due to the expensive instrumentation and software required for sample analysis. However, the long-life of NMR spectrometers and the diverse applications of NMR lower the cost of the analysis in the long term.

NMR-based metabolomics

Metabolomics is an emerging field of science with a wide range of applications in a variety of research disciplines, including food science. The two most common approaches for metabolomics are targeted analysis, focused on the identification and quantification of individual components, and untargeted analysis which relies on the comparison of various spectral patterns, using multivariate statistical analysis techniques, 4 such as Principal component analysis (PCA) and Projections to Latent Structures

Discriminant Analysis (OPLS-DA), without quantifying specific biomarkers. Despite the complementary information that targeted and non-targeted approaches can deliver when used together, they are often applied separately due to challenges in sample preparation and instrumental limitations. NMR is a popular analytical technique for metabolomics because of its reproducibility, its non-destructive nature, and the fact that it can be employed for both targeted and untargeted metabolomics simultaneously. It is inherently quantitative, and various methods exist for the quantification of analytes without contamination of the sample.

Most metabolomics and foodomics studies involving NMR deal with 1H -NMR spectroscopy due to the higher sensitivity of proton nuclei. Other nuclei such as 13C are not used as often due to the low sensitivity of carbon nuclei. In addition, for targeted analysis, where the full relaxation of magnetization is required, the long relaxation times of carbon dramatically increase the experimental time and cost and render the analysis challenging. However, as discussed earlier, 13C NMR spectra are characterized by excellent resolution due to the large spectral width and the elimination of scalar coupling thanks to decoupling pulse parameters.

Quantitation

There are various methods that allow for quantitation using NMR spectroscopy.

The most commonly used quantitation technique utilizes an internal standard of known concentration that is soluble in the sample and has resonance signals that do not overlap with the resonance signals of interest (Pauli 2001; Pauli et al. 2005). The integral of this 5 known standard signal is compared to the signal of unknown concentrations for their quantification.

Identification and selection of an appropriate internal standard for a sample, especially a complex sample such as food, is extremely important and can be challenging.

If the reference signal of the internal standard overlaps with unknown analyte peaks, the measurement will be inaccurate or impossible. Additional challenges of this method include that it contaminates the sample and prevents further analysis using additional analytical methods such as Gas chromatography (GC) or sensory analysis. It is also extremely important to know the exact concentration of the reference signal, so precise sample preparation is necessary.

A number of alternative quantification methods exist that mitigate some of these challenges. One such method is ERETIC (Electronic Reference To access In vivo

Concentrations), which electronically generates an artificial reference signal using a free channel on the NMR spectrometer. This artificial signal is calibrated against absolute concentrations of unknown peaks (Akoka et al. 1999; Billault et al. 2002). This method requires additional hardware and alterations in the spectrometer set up that are not available for all instrumentation, as well as a known standard compound to reference the artificial signal from (Bharti and Roy 2012).

Another quantitative NMR approach is the ERETIC2 tool, based on PULCON

(Pulse length–based Concentration determination). PULCON (Figure 1) employs an external standard which has one spectra, that is then correlated to the absolute intensity of a spectra with unknown analyte concentrations. Because the concentration of the

“external standard” is known, the unknown sample concentrations can be obtained based on the principle of reciprocity (Hoult and Richards 2011; Tyburn and Coutant 2015). The

PULCON equation can be viewed in Figure 2. Although PULCON allows the quantification of various compounds without the contamination of the sample with an internal standard, it is still rarely used, most likely due to the lack of comprehensive studies related to its analytical efficiency. Here we performed the first systematic study for evaluating the reliability of PULCON, using ERETIC2 tool as part of our targeted- untargeted analysis platform.

Figure 1. PULCON schematic and equation. (2018) NMR in food

NMR’s very first use in food dates back to 1957, when a low-resolution NMR instrument was used to measure moisture in foods (Spyros and Dais 2013). Since then,

NMR methodologies have been further developed for studying food’s chemical makeup,

7 but much more research is needed to further develop and fully establish this approach as a rapid and reliable analytical tool.

NMR spectroscopy has several advantages for analyzing foods: it is a nondestructive and quantitative technique, it requires minimal to no sample preparation, experimental times are fast compared to many analytical methods, and it is characterized by excellent accuracy and reproducibility. Additionally, NMR spectroscopy is an environmentally friendly methodology because it utilizes extremely small amounts of solvents (no more than 600-800 uL in most cases).

Fish oil supplements

Fish oil is composed by several major and minor compounds such as TGs, PLs, and DGs and is an excellent source of omega-3 (n-3) fatty acids, which play a vital role in human health. Fish oil is consumed in two forms: through eating a whole fish rich in oil, or through consumption of fish oil supplements. A large review of the current research published in 2009 on fish oil’s health benefits examined the literature comparing the benefits of consuming fish oil supplements to the whole fish. The authors concluded that while fish oil supplements have been linked to the same health effects of fish consumption, there is a possibility that other constituents in fish work synergistically with health promoting n-3 fatty acids. It is, therefore, recommended to eat whole fish over consuming fish oil supplements because fish may contain added nutrients. However, those who do not consume fish regularly should take fish oil supplements to increase their intake of n-3 fatty acids (He 2009).

Fish oil supplements are an extremely common supplement and sold in the vitamin section of stores around the world. In a National Institute of Health (NIH) survey, it was reported that nearly 50% of adults in the United States consume dietary supplements. Additionally, fish oil was reported as the most popularly consumed

“natural” supplement used by adults in the United States (National Health Interview

Survey 2012). These supplements tout health claims such as “Support Heart & circulatory health” and “supports heart, brain, skin, eye, & joint health.” One of the major reasons fish oil gets so much attention for its health benefits is its high content of essential n-3 fatty acids, eicosapentaenoic acid (EPA) (C20:5 n-3) and docosahexaenoic acid (DHA) (C22:6 n-3), shown in Figure 2. Several animal studies and clinical trials indicate that these two fatty acids have been linked with numerous health effects, such as heart disorders, inflammatory diseases and diabetes.

Figure 2. Chemical structures of DHA and EPA.

Fish oil is often obtained from fish such as salmon, sardines, mackerel, and anchovies. It has been found that the fish oil obtained from different sources can differ in composition from different fatty acid compositions to different types of lipids (TG, EE,

Phospholipids) which in turn may affect the bioavailability and metabolic pathways of

FA like EPA and DHA (Wakil et al. 2010; Ghasemi Fard et al. 2014). Increase in n-3 fatty acid through consumption of fish oil has been shown to mediate certain diseases when consumed in large enough doses. It has been linked with decreased inflammation in patients with inflammatory disease, such as rheumatoid arthritis, and can help protect against certain inflammatory diseases in mice models (Prickett et al. 1983; Robinson et al. 1986; Kremer et al. 1995). Additionally, increased consumption of fish oil has been linked to increased platelet concentrations leading to milder bruising tendency and lowered atherosclerotic vascular disease (Goodnight et al. 1981). In a systematic review assessing clinical data of fourteen randomized clinical trials, it was found that there is evidence suggesting that consumption of EPA and DHA from fish oil leads to significant reduction in total mortality, reduction in coronary heart disease death, and reduction sudden death (Harper and Jacobson 2005). The Western diet is relatively low in n-3 fatty acids and higher in n-6 fatty acids, thus the consumption of fish oil supplements can help improve the n-6/n-3 balance in consumer’s nutrition (Simopoulos).

Analysis

Despite the health benefits of fish oil and increase in fish oil supplement consumption, questions remain about the safety, authenticity, and quality of some of these products because they are not regulated by the Food and Drug Administration

(FDA). The rapid and accurate compositional analysis of fish oil supplements is essential to properly evaluate the quality of these commercial products and ensure consumer safety.

In 2008, the Office of Dietary Supplements (ODS) at the NIH and the FDA partnered with the Association of Official Analytical Chemists International (AOACI) to improve methods for fish oil supplement analysis. Omega-3 fatty acids were one of the components on the AOACI list to develop validation methods (Dwyer et al. 2008). While this project resulted in the Dietary Supplement Label Database (DSLD) from the NIH in

2013, there is still a need for more data and rapid analytical techniques (National Institute of Health).

Currently, the AOAC method for analyzing encapsulated fish oil supplements is

Gas Chromatograph with Flame Ioniazation Decetor (GC-FID). This is the most common methodology for the assessment of fish oil supplements, and Infrared Spectroscopy (IR) is also commonly used. Both of these methods are highly sensitive, but they have several drawbacks. GC analysis is time consuming (4-8 h) because separation and derivatization of individual compounds is required and lipid oxidation may occur during the analysis

(Sacchi et al. 1993; Igarashi et al. 2000; Guillén and Ruiz 2003). While IR spectroscopy can be quantitative, a prediction model is required to be constructed using partial least squares regression (PLSR), although there are exceptions in which IR bands can be attributed to a single compound (Plans et al. 2015). PLSR requires the analysis of a large number of samples, which increases the time of the analysis (Jian•hua 2014). For this reason, there is an increasing interest in the development of new analytical methodologies 11 that allow accurate and fast analysis of a large number of fish oil samples in a high throughput way.

NMR has been used in an array of applications for the study of lipids. In fish oil, specifically, it has been proven to identify and quantify a number of polyunsaturated fatty acids, including n-3 fatty acids, as well as lipids in forms other than triglycerides (Sacchi et al. 1993; Igarashi et al. 2000; Guillén and Ruiz 2003). The protocols and conditions for sample experiments are not well defined, however, and because this method is still being tested and compared to more widely accepted analytical techniques, there is no general protocol for the analysis.

Coffee

A coffee bean begins as a coffee cherry on a coffee tree, where it then undergoes processing to remove the pulp and reveal the coffee seed, commonly referred to as the coffee bean. This bean, which can be Coffea arabica species, commonly called Arabica coffee, or Coffea canephora species, commonly called Robusta coffee, is a green color prior to roasting and the two varieties composition can be summarized in Figure 2. The green coffee bean is traded on the market, and is then roasted by coffee shops or industry.

The roasting of a coffee bean greatly changes the morphology and the biochemical composition of the beans.

Figure 3: Chemical composition of Green Coffea Arabica and Coffea Canephora. (Farrah, 2010) There are many changes that occur in a coffee bean during roasting. The most noticeable change is the Maillard browning and caramelization, which generate a majority of roasted coffee’s characteristic taste and aroma, as well as the dark brown color compounds that are characteristic of the beverage (Bradbury 2001; Schenker and

Rothgeb 2017). Additionally, intense dehydration from 10-12% moisture down to 1-2% moisture occurs in the bean during roasting (Farah 2012). The bean becomes more porous after roasting and can increase in size to as much as double its original size (Schenker et al. 2000; Schenker and Rothgeb 2017). Finally, after roasting there is oil liberation and

13 migration from the inside of the coffee bean toward the surface, and the measurable oil content can increase in the coffee bean (Budryn et al. 2012; Schenker and Rothgeb 2017).

This oil migration can be observed as an oily surface on the bean that is not observed on the green coffee bean.

Coffee oil

The lipid fraction of Arabica and Robusta coffee beans differs, with an Arabica coffee bean consisting of approximately 15% lipids on a wet basis, compared to a

Robusta bean with about 10% lipids (Calligaris et al. 2009). The lipid makeup of coffee consists of about 75% TG, with oleic, linoleic, and linolenic acids being the most important for determining coffee freshness (Folstar 1985; Speer and Kölling-Speer

2006b). This lipid fraction has many applications in the food, pharmaceutical, and cosmetic industry.

Coffee oil contains a number of nutritionally beneficial components, namely cafestol and kahweol, two coffee-specific diterpenes. These diterpenes have antioxidant properties that have been found to have hepatoprotective effects on certain liver damage, and chemoprotective effects on peroxide-induced oxidative stress and DNA damage (Lee and Jeong 2007; Lee et al. 2007). A meta-analysis of data related to coffee consumption and decrease in colorectal cancer also found a lower risk of colorectal cancer associated with substantial consumption of coffee, however the results were inconclusive and required more studies to confirm the observation (Giovannucci 1998).

In addition to the antioxidant properties, coffee oil has also been found to have cosmetic applications. Green coffee oil (GCO) has the ability to absorb UVB radiation, 14 and has been tested as a successful sun protectant (Grollier and Plessis 1983; Aleriana et al. 2010). GCO has also been found to increase skin hydration, help to repair skin cells, and have anti-aging properties (Del Carmen Velazquez Pereda et al. 2009).

The lipids in coffee also play an integral role in determining coffee’s shelf life. In both green and roasted coffee, lipid oxidation occurs and can have a negative effect on the overall sensory perception of the final coffee beverage (Toci et al. 2013; Rendón et al.

2014). Roasted coffee has an extremely low water activity, and therefore lipid oxidation is expected to be a major shelf-life limiting factor for coffee (Farah 2012). Stored, green coffee can become oxidized and have a flat, woody taste. These off-notes that occur arise mainly from undesirable changes in the lipid fraction of the bean (Selmar et al. 2008).

Holscher and Steinhart discovered that a large number of low-odor threshold molecules in green coffee possess a carbonyl group and are degradation products generated during autoxidation of lipids that contribute to final roasted coffee flavor (Holscher and Steinhart

1995). Additionally, maillard reactions occurring during green coffee storage decrease the amount of volatile precursors necessary during roasting and may influence the final cup quality (Speer and Kölling-Speer 2006b).

In current work that is not yet published but is based on research presented by our collaborators at Texas A&M, Echeverria- Beirute, it was found that there is likely a link between the lipid fraction of the green coffee bean and stress during growing. Coffee beans that are more stressed due to the presence of coffee leaf rust, extreme sun exposure, or an abundance of coffee fruit, likely have a different lipid profile compared to coffee plants that do not experience such stress (Echeverria-Beirute et al. 2017)

Lipids in green and roasted coffee beans

Despite the importance of lipids in coffee chemistry and the high value of coffee oil, there is conflicting literature regarding the change in lipids during roasting of coffee.

It is important to understand the changes in coffee oil during roasting to understand if there are changes in their health benefits following roasting, as well as for authentication purposes. Upon researching the literature regarding the changes in coffee lipids during roasting, I found many different results. In many studies that exist on the coffee lipids, increase in FFA is used as an indicator of oxidation. In one study analyzing the lipid fraction of coffee using Attenuated Total Reflectance- Fourier Transform Infrared (ATR-

FTIR) Spectroscopy and Peroxide Value (PV), they found that there was not a significant difference in the oxidative status of green and roasted coffee lipids when using FTIR, but they did find an increase in FFA in roasted coffee and significantly higher PV in roasted coffee compared to green (Raba et al. 2015). However, the authors claimed that overall, there was no difference in oxidation between green and roasted lipids and attributed this oxidative stability to the lipid-soluble colored maillard products that are formed during roasting and have antioxidant properties.

In another article analyzing coffee lipids using cGC by Martin et. al., the authors report that there is no significant difference in the fatty acid content of the lipids before and after roasting, however they could discriminate between green and roasted coffee samples using Linear Descriptive Analysis (LDA) (Martín et al. 2001). Another article analyzing lipids using GC-FID found a difference in the amount of tocopherols present in the bean but no difference in FA profile, with larger amounts of tocopherols present in 16 roasted beans attributed to the liberation of tocopherols during roasting (González et al.

2001).

Finally, one article analyzing the effect of roasting conditions, such as time, temperature, and moisture, found that PV increases during roasting at certain temperatures, but actually decreases at most temperatures likely from the transformation of peroxides into products with lower molecular mass which are not measured when measuring PV (Budryn et al. 2012). They additionally found a decrease in conjugated dienes and trienes as a measure of oxidation after roasting, and only a small influence of roasting on the composition of fatty acids. These authors concluded that coffee oil is relatively thermally stable during roasting, with slight changes in oxidation depending on roasting conditions.

In a broad literature review of coffee lipids by Speer and Kolling-Speer, the main fatty acids, diterpenes, tocopherols, and sterols are discussed. They report on the extensive body of literature regarding changes in nutritionally important diterpenes, stating that new diterpenes, dehydrocafestol and dehydrokahweol, form during roasting.

Speer and Kolling-Speer conclude, however, that the literature review of fatty acid changes during roasting is incomplete and they could not discuss the influence of roasting on changes in the lipid fraction of coffee (Speer and Kölling-Speer 2006b).

Overall, it can be concluded that some authors have found differences in fatty acid composition or free fatty acids in coffee lipids after roasting, however, some authors have found differing results. Differences in the thermal behavior of coffee oil between green and roasted has been reported in terms of thermogravimetry and differential thermal

17 analysis, however differential scanning calorimetry did not differ between green and roasted (Kobelnilk et al. 2014).

It is known, however, that during storage of roasted coffee, lipid oxidation occurs,

FFA in coffee increase, and oxidation is an important shelf limiting parameter (Ortolá et al. 1998; Toci et al. 2013; Rendón et al. 2014). While storage has an effect on oxidation, it is unclear if oxidation products increase with roasting, which would be expected due to the high temperature of coffee roasting. If oxidation does not occur to some degree during roasting, there are likely antioxidants protecting the lipids during roasting. We believe NMR is a powerful tool to analyze the total lipid profile of coffee oil before and after roasting.

Coffee lipid analysis

Coffee oil has been analyzed using a variety of analytical methods including High

Performance Liquid Chromatography (HPLC), Gas Chromatography (GC), Capillary gas chromatography (cGC), Electrospray ionization-mass spectrometry (EIS-MS), and Fourier

Transform Infrared Spectroscopy (FT-IR) (Valdenebro et al.; Carrera et al. 1998; González et al. 2001; Martín et al. 2001). Despite the benefits of these methods, some of them require lengthy extraction techniques, derivatization, and complex sample preparation, or may not be capable of providing quantitative data or information on the molecular level. For example, the GC analysis of fatty acids requires separation and derivatization of individual compounds and lipid oxidation may occur during the analysis, which can cause further inaccuracies (Sacchi et al. 1993; Igarashi et al. 2000; Guillén and Ruiz 2003). Similarly, the current procedure for the sterolic profile determination in coffee is very time consuming

18 and it involves several steps, including lipid saponification and chromatographic separations with TLC and GC.

Nuclear Magnetic Resonance (NMR) Spectroscopy is a robust method that can rapidly analyze the lipid profile of coffee on the molecular level without requiring separation and/or purification steps. The analysis is conducted in one snapshot, which provides a significant advantage compared to traditional analysis that focuses on one specific class of compounds. The major drawback of NMR is its low sensitivity compared to other analytical techniques, such as chromatography and mass spectrometry, however

NMR’s nondestructive nature, as well as its rapid, robust nature offers a great non-targeted approach when combined with multivariate statistical analysis.

So far, NMR has been used once before to evaluate green coffee lipids of beans from different origins and species (arabica and robusta) (D’Amelio et al. 2013). In this study, researchers analyzed pure coffee oil without dilution, which allows the fast analysis of several lipid components, although there is a reduction in resolution.

Additionally, NMR has been used to study the diterpenes cafestol, kahweol, and 16-O- methocafestol which have important nutritional properties and can be used for coffee authentication (Scharnhop and Winterhalter 2009). NMR has been used to measure the polar part of coffee and its ability to indicate changes in roasting, between Arabica and

Robust, and to analyze green coffee according to variety and origin (Ciampa et al. 2010;

Wei et al. 2012b, a; Schievano et al. 2014; Monakhova et al. 2015b).

Chapter 2. Statement of the Problem

The composition of fish oil supplements from various origins has been studied using several different analytical techniques, however most of them have their own limitations. NMR has been shown to offer a novel approach to analyze these lipids rapidly with minimal sample preparation and high reproducibility. However, there is currently not a sound protocol published for analyzing fish oil using NMR. Therefore, I published a protocol in the Journal of Visualized Experiments (JoVE), titled “NMR

Spectroscopy as a Robust Tool for the Rapid Evaluation of the Lipid Profile of Fish Oil

Supplements.” This protocol discusses the sample preparation of the NMR sample, the use of 1H and 13C NMR experiments, as well as the difference between different NMR hardware, such as inverse and observe NMR probes and the analytical capabilities of 500 and 850 MHz instruments. This peer-reviewed publication was a training experience for me to use NMR, and I became acquainted with NMR sample preparation, experimentation, and data analysis. Using the skills I learned in analyzing fish oil supplements by NMR and in publishing my first journal article, I was able to apply my knowledge to study the lipid fraction of coffee.

Although coffee lipids accounts for about 15% of an Arabica coffee bean and have a vast number of applications as a valuable material and as a biomarker, they are not well studied compared to the polar part of coffee. Additionally, there is a conflicting body of literature regarding the oxidative status and compositional changes of coffee lipids upon roasting and storage. This is likely due to the lack of efficient analytical tools that

20 allow the rapid and reliable evaluation of coffee oil at the molecular level. NMR is a promising analytical tool for this type of analysis, however, there is currently only one study in literature for the analysis of coffee lipids using NMR spectroscopy (D’Amelio et al. 2013). This publication only analyzed the lipids of green coffee oil and authors used a concentrated oil sample with a small amount of deuterated solvent (DMSO- d6) in the

NMR instrument, which caused peak broadening and low spectral resolution. Because

NMR requires minimal sample preparation and no separation or derivatization steps, it has the potential to be a powerful tool to capture any changes that may occur in coffee lipids during roasting. Therefore, NMR methodologies and analyses need to be created to understand its ability to measure the changes, if any, in coffee lipids. In addition, NMR may allow the analysis of lipids in spent coffee grounds, which currently are a waste product with limited applications. This coffee waste has an environmental cost but its rich source of lipids give it the potential to be a precursor for development of novel, value added products.

Chapter 3. Objectives

Fish Oil

1) Develop a sound protocol for the identification and quantitative analysis of

various fish oil components, such as fatty acids sterols and other lipids in

encapsulated supplements.

2) Compare the effects of different NMR spectrometers (inverse 500 MHz, and

observe 850 MHz) on qualitative 1H and 13C NMR spectroscopy and discuss

opportunities and limitations.

Coffee Oil

1) Develop a method for analysis of lipids in green and roasted coffee, as well as a

coffee beverage and spent coffee grounds.

2) Investigate the impact of roasting on the lipid profile of coffee.

3) Perform a systematic comparison between the traditional internal standard method

with the PULCON-based quantitation.

4) Evaluate spent coffee grounds for their lipid content and for their ability to act as

precursors for bioplastics.

Chapter 4. Materials and Methods

Please visit Chapter 5: Protocol and Chapter 6: Materials and Methods for complete materials and methods for each objective.

Chapter 5. NMR Spectroscopy as a Robust Tool for the Rapid Evaluation of the Lipid

Profile of Fish Oil Supplements

Short Abstract

Here, high-resolution 1H and 13C Nuclear Magnetic Resonance (NMR) spectroscopy was used as a rapid and reliable tool for quantitative and qualitative analysis of encapsulated fish oil supplements.

Long Abstract

24 can provide quantitative data for a greater number of individual fatty acids and can be used for determining the positional distribution of fatty acids on the glycerol backbone.

Both nuclei can provide quantitative information in just one experiment without the need of purification or separation steps. The strength of the magnetic field mostly affects the

Introduction

The consumption of n-3 fatty acids in the diet has proven to be beneficial against several conditions such as heart disorders(Simopoulos; Goodnight et al. 1981; Harper and

Jacobson 2005), inflammatory diseases and diabetes(Malasanos and Stacpoole; Kremer et al. 1995). The Western diet is considered poor in n-3 fatty acids and thus the consumption of fish oil supplements is recommended to improve the n-6/n-3 balance in consumer’s nutrition(Simopoulos). Despite the recent increase in fish oil supplement consumption, questions remain about the safety, authenticity, and quality of some of these products. The rapid and accurate compositional analysis of fish oil supplements is essential to properly evaluate the quality of these commercial products and ensure consumer safety.

The most common methodologies for the assessment of fish oil supplements are gas chromatography (GC) and Infrared Spectroscopy (IR). While these are highly sensitive methods, they suffer from several drawbacks (Han et al. 2011). GC analysis is time consuming (4-8 h) because separation and derivatization of individual compounds is 25 required and lipid oxidation may occur during the analysis(Sacchi et al. 1993; Igarashi et al. 2000; Guillén and Ruiz 2003). While IR spectroscopy can be quantitative, a prediction model is required to be constructed using partial least squares regression (PLSR), although there are exceptions in which IR bands can be attributed to a single compound

(Plans et al. 2015). PLSR requires the analysis of a large number of samples, which increases the time of the analysis (Jian•hua 2014). For this reason, there is an increasing interest in the development of new analytical methodologies that allow accurate and fast analysis of a large number of fish oil samples. Organizations such as the Office of

Dietary Supplements (ODS) at the National Institutes of Health (NIH) and the Food and

Drug Administration (FDA) have collaborated with the Association of Official Analytical

Chemists (AOAC) to develop these new methods (Millen et al. 2004; Dwyer et al. 2008).

One of the most promising analytical methods for the screening and the evaluation of multi-component matrices, such as dietary supplements, is Nuclear Magnetic Resonance

(NMR) spectroscopy(Frank et al. 2014). NMR spectroscopy has several advantages: it is a non-destructive and quantitative technique, it requires minimal to no sample preparation, and it is characterized by excellent accuracy and reproducibility. In addition,

NMR spectroscopy is an environmentally friendly methodology because it utilizes only small amounts of solvents. The main drawback of NMR spectroscopy is its relatively low sensitivity compared to other analytical methods, however, recent technological advances in instrumentation such as stronger magnetic fields, cryogenic probes of various diameters, advanced data processing, and versatile pulse sequences and techniques have increased the sensitivity up to the nM range. While NMR instrumentation is high cost, the

26 long-life of NMR spectrometers and the many applications of NMR lower the cost of the analysis in the long run. This detailed video protocol is intended to help new practitioners in the field avoid pitfalls associated with 1H and 13C NMR spectroscopic analysis of fish oil supplements.

Protocol

1. NMR Sample Preparation

Note: Caution, please consult all relevant material safety data sheets (MSDS) before use.

Deuterated chloroform (CDCl3) used in sample preparation is toxic. Please use all the appropriate safety practices when performing sample preparation including the use of a fume hood and personal protective equipment (safety glasses, gloves, lab coat, full length pants, closed-toe shoes).

1.1) Preparation of 1H and 13C samples

1.1.1) Extract 120 μL (~ 110 mg) of fish oil from a dietary capsule using a syringe and place it in a 4 mL glass vial. Record the weight of the fish oil.

1.1.2) Sample dissolution

1.1.2.1) Dissolve approximately 120 μL of fish oil in 500 μL of CDCl3 containing 0.01% of Tetramethylsilane (TMS) which is used as a reference for the 1H and 13C chemical shifts.

Note: TMS is used only for chemical shift calibration (see step numbers 2.2.1.2.7 and

2.2.2.2.7), not for quantification (see step numbers 2.2.1.3 and 2.2.2.3) purposes.

1.1.2.2) Prepare a 2,6-Di-tert-butyl-4-methylphenol (BHT) stock solution, if quantification expressed in mg/g is desired, by dissolving approximately 220 mg of BHT and 15 mg of

Chromium (III) acetylacetonate (Cr(acac)3) in 20 mL of CDCl3 containing 0.01% of TMS.

Use 500 μL of the stock solution to dissolve 100 mg (± 10 mg) of fish oil.

1.1.3) After dissolving the oil (this takes a few seconds), transfer all of the solution directly into a high quality 5-mm NMR tube and attach a cap. Analyze the samples within 24 h after preparing the samples.

2. NMR Instrument preparation

Note: Caution, beware that the presence of strong magnetic fields produced by NMR instruments can affect medical devices and implants such as pacemakers and surgical prostheses, as well as electronic items such as credit cards, watches, etc. Additional caution is required when the analysis is performed using non-shielding magnets. Two

NMR instruments were used for the acquisition of 1H and 13C NMR spectra; a spectrometer operating at 850.23 MHz and 213.81 MHz for 1H and 13C nuclei, respectively, equipped with a triple resonance helium-cooled inverse (TCI) 5 mm probe and a spectrometer operating at 500.20 MHz and 125.77 MHz for 1H and 13C nuclei, respectively, equipped with a broad band observed (BBO) nitrogen-cooled 5 mm probe.

All experiments were performed at 25 ± 0.1 ºC and the spectra were processed by a standard NMR data analysis acquisition and processing software package (see table of materials).

2.1) Preparation for acquiring the NMR spectra

Note: 1H and 13C NMR spectra can be acquired consequently without removing the sample from the instrument.

2.1.1) Insert the NMR tube into a spinner turbine (see table of materials). 28

2.1.2) Place the spinner and the tube on the top of a graded depth gauge and gently push the top of the tube until its bottom part touches the bottom of the gauge.

2.1.3) Place NMR sample in an open spot of the SampleCase. Note the slot number the sample is placed in.

2.1.4) To load the sample in the NMR, return to the control computer and type 'sx #', where # is the slot in the SampleCase holding your sample.

2.1.5) Wait for the deuterium signal of CDCl3 to appear on the lock window screen. If it does not automatically appear, type “lockdisp”. As soon as the deuterium signal is visible, type “lock” on the command line and select “CDCl3” from the solvent’s list in order to lock the sample using the CDCl3 deuterium resonance.

Note: Deuterium signal may not appear if previous user used a different solvent. User should wait for the indicator that the sample is down, then lock.

2.1.6) Type “bsmsdisp” in the command line to ensure spinning is not active. If the

“SPIN” button is green, click it to deactivate spinning.

2.1.7) Type the “new” command to create a new data set. Enter a name for the data set in the “NAME” tab and the experiment number in the “EXPNO” tab. Use number “1” in the

“PROCNO” tab. In the “Experiment” tab, hit “Select” and choose the “PROTON” parameter file. Write the title of the experiment in the “TITLE” tab. Click “OK.”

2.1.8) Type “getprosol” in the command line to obtain the standard parameters for the current NMR probe and solvent.

2.1.9) Repeat step 2.1.7 for 13C, selecting the “C13IG” pulse sequence in the

“Experiment” tab for the 1D 13C inverse gated decoupled experiment.

2.1.10) Type “getprosol” in the command line to obtain the standard parameters for the current NMR probe and solvent.

2.1.11) Type the command “atma” to perform automatic tuning and matching of the probe for both carbon and proton nuclei.

2.1.12) Perform one-dimensional gradient shimming to achieve a highly homogeneous magnetic field, and thus optimum line shape for the NMR signals.

2.1.12.1) Use the standard automatic procedure for 1D shimming, simply by sequentially executing the commands “qu topshim 1dfast ss”, “qu topshim tuneb ss,” and “qu topshim report” on the command line.

2.2) Parameter optimization

2.2.1) 90° pulse calibration

2.2.1.1) Create a new data set for 1H (see steps 2.1.7 and 2.1.8).

2.2.1.2) Type the command “paropt” on the command line to start the automation program for calibrating the 90° pulse. Select pulse duration, p1, as the parameter to be modified.

2.2.1.3) Start with “2” μs as the initial value of p1, enter “2” μs increments and perform

“16” experiments.

2.2.1.4) Create a new data set for 13C (see step 2.1.9) and repeat the process for 13C nuclei

(see steps 2.2.1.2 and 2.2.1.3).

16 1 2.2.2) T1 measurement measured by the null method for H

Note: The null method uses the inversion recovery pulse sequence, consisting of a 180° pulse follow by a delay (tau), to allow relaxation along the z axis and a final 90° pulse which creates the observable transverse magnetization.

2.2.2.1) Create a new data set for 1H (see steps 2.1.7 and 2.1.8).

2.2.2.2) Type “pulprog t1ir1d” to change the pulse sequence to the inversion-recovery experiment.

2.2.2.3) Type the following commands on the command line to set up the spectral width in ppm, the center of the RF transmitter, the number of scans the number of dummy scans and the number of data points “sw 8”, “o1p 3.8”, “ns 2”, “ds 2” and “td 64K”.

2.2.2.4) Type “p1 (value)” and enter the duration values for 90° pulse as determined by the pulse calibration (see step 2.2.1) and type “p2 (value)” for the 180° pulse (the duration value for the 180° pulse is the 90° pulse duration multiplied by two).

2.2.2.5) Set the recycle delay to a very large value, such as 10 s by typing “d1 10”.

2.2.2.6) Set tau to a short value, such as 10 ms, by typing “d7 10ms” in the command line.

2.2.2.7) Set the receiver gain (RG) to an appropriate value using the command “rga” for automatic calculation of RG.

2.2.2.8) Run a spectrum by typing the command “zg”.

2.2.2.9) Execute Fourier-transformation by typing “efp” in the command line.

2.2.2.10) Perform automatic phase correction by typing the command “apk” in the command line. If additional phase adjustments are required to further improve the

31 spectrum, click on the “Process tab,” then click on the “Adjust Phase” icon to enter the phase correction mode.

2.2.2.10.1) Use the zero-order (0) and first-order (1) phase correction icons by dragging the mouse until all the signals are in negative absorption mode. Apply and save the phase correction values by clicking the “Return and Save” button to exit the phase correction mode.

2.2.2.11) Increase the tau until all peaks are either positive or nulled by repeating steps

2.2.2.6-2.2.2.9. To determine the T1 value, simply divide the tau value where the peak is nulled with ln2.

16 13 2.2.3) T1 measurement measured by the null method for C

2.2.3.1) Create a new data set for 13C (see step 2.1.9)

2.2.3.2) Type “pulprog t1irpg” to change the pulse sequence to the inversion-recovery experiment for carbon nuclei.

2.2.3.3) Type the following commands on the command line to set up the spectral width in ppm, the center of the RF transmitter, the number of scans, the number of dummy scans and the number of data points: “sw 200”, “o1p 98”, “ns 8”, “ds 2”and “td 64K”.

2.2.3.4) Type “p1 (value)” and enter the duration values for 90° pulse as determined by the pulse calibration (see step 2.2.1) and type “p2 (value)” for the 180° pulse (the duration value is the 90° pulse duration multiplied by two).

2.2.3.5) Set the recycle delay to a very large value, such as 100 s by typing “d1 100”.

2.2.3.6) Set tau to a short value, such as 100 ms by typing “d7 100ms” in the command line.

2.2.1.7) Set the receiver gain (RG) to an appropriate value using the command “rga” for automatic calculation of RG.

2.2.3.8) Run a spectrum by typing the command “zg”.

2.2.3.9) Execute Fourier-transformation by typing “efp” in the command line.

2.2.3.10) Perform Automatic phase correction by typing the command “apk” in the command line. If additional phase adjustments are required to further improve the spectrum, click on the “Adjust Phase” icon and the phase correction icons for zero-order

(0) and first-order phase (1) correction.

2.2.3.10.1) While clicking on the zero-order and first-order phase correction icons, drag the mouse until all the signals are in negative absorption mode. Apply and save the phase correction values by clicking the “Return and Save” button to exit the phase correction mode.

2.2.3.11) Increase the tau until all peaks are either positive or nulled by repeating steps

2.2.3.6-2.2.3.9. To determine the T1 value, simply divide the tau value where the peak is nulled with ln2.

2.3) One-dimensional (1D) NMR Spectra

2.3.1) 1H-NMR spectra

2.3.1.1) Acquisition of the NMR data

2.3.1.1.1) Go to the 1H data set created in step 2.1.7 and use the standard “pulse-acquire” pulse sequence, “zg”, by typing “pulprog zg” in the command line.

2.3.1.1.2) Type the following commands on the command line to set up the spectral width in ppm, the center of the RF transmitter, the number of scans, the number of dummy

33 scans, the number of data points and the pulse duration for a 90° pulse angle: “sw 8”,

“o1p 3.8”, “ns 2”, “ds 2”, “td 64K” and “p1 (as determined by pulse calibration)” (see step 2.2.1).

Note: 32K data points can be used for the 500 MHz instrument.

2.3.1.1.3) Set a relaxation delay of 7 s for the 500 MHz instrument or 9 s for the 850

MHz instrument by typing “d1 7s” or “d1 9s”, respectively, in the command line.

2.3.1.1.4) Set the receiver gain (RG) to an appropriate value using the command “rga” for automatic calculation of RG.

2.3.1.1.5) Type “digmod baseopt” to acquire a spectrum with improved baseline.

2.3.1.1.6) Start the acquisition by typing the pulse-acquire command “zg” in the command line.

2.3.1.2) Processing of the NMR data

2.3.1.2.1) Type “si 64K” in the command line to apply zero-filling and set the size of the real spectrum to 64K.

2.3.1.2.2) Set the line broadening parameter to 0.3 Hz by typing “lb 0.3” in the command line to apply a weighting function (exponential decay) with a line broadening factor of

0.3 Hz prior to Fourier transform.

2.3.1.2.3) Execute Fourier-transformation by typing “efp” in the command line.

2.3.1.2.4) Perform Automatic phase correction by typing the command “apk” in the command line. If additional phase adjustments are required to further improve the spectrum, click on the “Process tab,” then click on the “Adjust Phase” icon and the phase correction icons for zero-order (0) and first-order (1) phase correction.

2.3.1.2.4.1) While clicking on the zero-order and first-order phase correction icons, drag the mouse until all of the signals are in positive absorption mode. Apply and save the phase correction values by clicking the “Return and Save” button to exit the phase correction mode.

2.3.1.2.5) Apply a polynomial fourth-order function for base-line correction upon integration by typing the command “abs n”.

Note: This ensures a flat spectral baseline with a minimum intensity.

2.3.1.2.6) Report chemical shifts in ppm from TMS (δ = 0). Click on the calibration

(“Calib. Axis”) icon, and place the cursor with the red line on top of the TMS NMR signal (peak closest to 0). Left click and type in “0”.

2.3.1.3) NMR data analysis

2.3.1.3.1) Integrate the spectral region from δ 1.1 to δ 0.6 as well as the peaks at δ 4.98, δ

5.05 and δ 5.81 using the “Integrate” icon (under the “Process” tab) and the highlight

(“Define new Region”) icon. Left click and drag through the integrals.

Note: If there is need to focus on a region, click on the highlight icon to deactivate and left click and drag the mouse to zoom in on the region. To adjust the threshold intensity, use the middle mouse button if needed. Click on the highlight icon again to make the integration function active, then move to the next peak.

2.3.1.3.1.1) Normalize the sum of the above integrals to 100 by right clicking on the integral value that appears under the signal and select “Normalize sum of integrals”.

Input the value “100” in the box and click the “Return and Save” to exit the integration mode.

2.3.1.3.2) When using BHT as an internal standard, integrate the peak at δ 6.98 and set the integral equal to the mmoles of BHT per 0.5 mL of the stock solution.

2.3.1.3.3) Integrate the peaks of interest (see step 2.3.1.3.1) extending 10 Hz from each side of the peak, when possible.

2.3.1.3.4) Proceed to perform 13C-NMR spectra acquisition and processing in a similar manner.

2.3.2) 13C-NMR spectra

2.3.2.1) Acquisition of the NMR data

2.3.2.1.1) Go to the 13C data set and use the inverse gated decoupled pulse sequence,

“zgig” by typing “pulprog zgig” in the command line.

Note: To run a carbon experiment with the standard broadband decoupled pulse sequence, type “pulprog zgpg” in the command line.

2.3.2.1.2) Type the following commands on the command line to set up the spectral width in ppm, the center of the RF transmitter, the number of scans, the number of dummy scans, the number of data points and the pulse duration for a 90° pulse angle: “sw 200”,

“o1p 95”, “ns 16” “ds 2”, “td 64K” and “p1 (as determined by pulse calibration)” (see step 2.2.1.4).

2.3.2.1.3) Set a relaxation delay of 35 s for the 500 MHz instrument or 45s for the 850

MHz instrument by typing “d1 35s” or "d1 45s”, respectively, in the command line.

When using BHT, relaxation delay should be 50s in the 500 MHz instrument and 60s in the 850 MHz instrument.

2.3.2.1.4) Set the receiver gain (RG) to an appropriate value using the command “rga” for automatic calculation of RG.

2.3.2.1.5) Type “digmod baseopt” in the command line to acquire a spectrum with improved baseline.

2.3.2.1.6) Start the acquisition by typing the pulse-acquire command “zg” in the command line.

2.3.2.2) Processing of the NMR data

2.3.2.2.1) Type “si 64K” in the command line to apply zero-filling and set the size of the real spectrum to 64K.

2.3.2.2.2) Set the line broadening parameter to 1.0 Hz by typing “lb 1.0” in the command line to apply a weighting function (exponential decay) with a line broadening factor of

1.0 Hz prior to Fourier transform.

2.3.2.2.3) Execute Fourier-transformation by typing “efp” in the command line.

2.3.2.2.4) Perform Automatic phase correction by typing the command “apk” in the command line. If additional phase adjustments are required to further improve the spectrum, click on the “Process tab,” then click on the “Adjust Phase” icon and the phase correction icons for zero-order (0) and first-order phase (1) correction.

2.3.2.2.4.1) While clicking on the zero-order and first-order phase correction icons, drag the mouse until all of the signals are in positive absorption mode. Apply and save the phase correction values by clicking the “Return and Save” button to exit the phase correction mode.

Note: For carbon spectra recorded on Larmor frequency of 214 MHz (the 850 MHz instrument) the correction of the frequency dependent errors (first-order) may be challenging and time consuming for less experienced users because of the large off- resonance effects of the 90° pulse.

2.3.2.2.5) Apply a polynomial fourth-order function for base-line correction upon integration by typing the command “abs n” in the command line.

2.3.2.2.6) Report chemical shifts in ppm from TMS (δ = 0). Click on the calibration

(“Calib. Axis”) icon, and place the cursor with the red line on top of the NMR signal to be referenced. Left click and type in “0”.

2.3.2.3) NMR data analysis

2.3.2.3.1) Integrate the spectral region from δ 175 to δ 171 using the “Integrate” icon

(under the “Process” tab) and the highlight (“Define new Region”) icon. Left click and drag through the integrals.

Note: If there is need to focus on a region, click on the highlight icon to deactivate and left click and drag the mouse to zoom in on the region. Click on the highlight icon again to make the integration function active, then move to the next peak.

2.3.2.3.1.1) Set the integral to 100 by doing a right click on the integral value that appears under the signal and select “Calibrate Current Integral”. Input the value “100” in the box and click the “Return and save” to exit the integration mode.

2.3.2.3.2) When using BHT as an internal standard, integrate the peak at δ 151.45 and set the integral equal to the mmoles of BHT per 0.5 mL of the stock solution.

2.3.2.3.3) Integrate the peaks of interest extending 5 Hz from each side of the peak (see step 2.3.2.3.1).

Representative Results

1H and 13C NMR spectra were collected for commercially available fish oil supplements using two NMR instruments; an 850 MHz and a 500 MHz spectrometer. These spectra can be used for the quantitative determination of components of fish oil, such as docosahexaenoic acid (DHA) and eicosapentaenoic acid (EPA), as well other compounds such as n-1 acyl chains and nutritionally important index such as the n-6/n-3 ratio. The quantification can be performed even without the use of an internal standard, however, the quantitative results must be expressed as relative molar percentages. When the data need to be expressed in absolute values (mg/g), an internal standard is required. The results obtained by NMR are highly reproducible with relative standard deviations (RSD) ranging from 0.3% to 2% for 13C NMR analysis and from 0.5% to 2.5 % for 1H NMR analysis, depending on the lipid. The slightly higher RSD for 1H NMR is often observed because proton spectra tend to be overcrowded, which affects the accuracy of the analysis, especially for resonances that have a lower signal to noise ratio (S/N). A very good agreement was found between the 850 MHz and the 500 MHz instrument with

RSDs ranging from 1% to 4%. Relatively high RSDs (up to 8%) were observed when comparing results obtained by 1H and 13C, especially for compounds that appear in lower concentrations such as n-1 acyl chains. NMR spectroscopy has been previously validated as a tool for lipid analysis, including the determination of some fish oil components.

Results showed that it is in good agreement with traditional methods, such as GC17,18. 39

1H NMR analysis

Figure 1 compares the 1H NMR spectra acquired on (A) an 850 MHz and (B) a 500 MHz instrument. The 850 MHz spectrum is characterized by higher resolution, however the major components of fish oil including DHA, EPA, and n-6/n-3 ratio can also be determined from the 500 MHz spectrum. The 1H-NMR signals of fish oil fatty acids that can be used for quantitation purposes are shown in Table 1, whereas the complete NMR assignment of the 1H NMR spectrum of fish oil can be found elsewhere19.

1H NMR gave reliable data for the quantification of the total amount of n-3, n-6, DHA, trans fatty acids, n-1 acyl chains, and saturated fatty acids (SFA). For the 1H NMR analysis, the use of appropriate relationships is required because most of the signals belong to groups of protons that are common to different fatty acids and lipids. For that reason, in most cases the concentration of fatty acids in fish oil can be determined only by combination of various 1H NMR signals, incorporated in the appropriate relationships.

In addition, these equations contain arithmetic coefficients that normalize the different number of protons associated with each group. When an internal standard is used the following equation should be considered: C = I/IIS × NIS/N × A × MW/m (1), where C is the concentration of the analyte in mg/g of fish oil, I is the integral of a resonance that is uniquely attributed to the lipid of interest, IIS is the area of a proton signal that belongs uniquely to the internal standard, N is the number of protons of the functional group that is analyzed, NIS is the number of protons of the internal standard that are used for the analysis, A is the mmoles of internal standard, MW is the molecular weight of the fatty acid (expressed in methylesters), and m is the amount of fish oil expressed in g. 40

Example 1, DHA: The proportion of DHA is determined by the equation CDHA = ¾

IDHA/S, where IDHA is the integral of the signal at δ 2.39 which belongs to the Hα and Hβ protons of DHA, and S is the sum of integrals of the methyl protons of SFA, n-6, n-9, n-

3, trans fatty acids plus the integrals of the peaks of n-1 acyl chains at δ 4.98, δ 5.05 and

δ 5.81. The integral IDHA is normalized by multiplying by 3/4 because it corresponds to four protons, whereas the integral S corresponds to three protons. 1H NMR is not capable to giving information about positional distribution of fatty acids on the glycerol backbone and thus can only be used for the quantification of the total amount of fatty acids. The 1H

NMR analysis of an encapsulated fish oil supplement showed that it consists of 10.5 % of

DHA. The concentration of DHA in the same sample using BHT was found to be 105.23 mg/g. These values are very close to the values obtained with 13C NMR (see example 2 for 13C analysis).

Example 2, n-1 acyl chains: The concentration of n-1 acyl chains is given by the relationship Cn-1 = 3In-1/S, where In-1 is the integral of the signal at δ 5.818. This signal corresponds to one proton and thus needs to be normalized by multiplying by three.

When using BHT, n-1 acyl chains are determined by the equation Cn-1 = 2In-1/IBHT. The results cannot be expressed in mg/g because the MW of n-1 acyl chains is unknown.

Example 3, n-6/n-3 ratio: This important index can be calculated from the ratio of the normalized intensities of the resonance at δ 2.77, which corresponds to the bis-allylic protons of n-6 acyl chains (two protons) over the triplet at δ 0.97 that belongs to n-3 fatty acids and corresponds to three protons. The relationship is Cn-6/ Cn-3 = 3/2 IA/IB, where IA and IB are the integrals of the signals at δ 2.77 and δ 0.97, respectively. n-6 fatty acids are

41 determined from the relationship Cn-6 = 3/2In-6/S, where In-6 is integral of the bis-allylic protons at δ 2.77.

Example 4, trans fatty acids: Trans fatty acids can be calculated from the equation

Ctrans= Itrans/S, where Itrans is the integral of the signal at δ 0.91. The present sample contained 3.07% of trans fatty acids, as determined by 1H NMR using the 850 MHz instrument. The same sample analyzed in a 500 MHz instrument was found to contain

3.03% of trans fatty acids.

Example 5, saturated fatty acids (SFA): The concentration of SFA can be calculated from the equation CSFA = S – Cn-3 – Cn-6 – Cn-9 – Cn-1 - Ctrans. n-9 fatty acids (mainly oleic acid) can be quantified according to the equation Cn-9 = (3/4Q - 3/2In-6)/S, where Q is the integral of the allylic protons of n-6 and n-9 at δ 2.01. The amount of SFA in a commercially available fish oil sample was found to be 36.1%. The same sample analyzed with 13C NMR was found to contain 33.8% SFA. SFA represent a group of various FA (eg stearic and palmitic) with different MW and thus their concentration if fish oil cannot be expressed in mg/g.

Example 6, Total sterols: The amount of total sterols (free and esterified) can be determined by the signal of the methyl protons at carbon 18 which appears at δ 0.68, using the equation C = Iste/S. The molar ratio of total sterols in a commercially available fish oil sample was found to be 0.32%. BHT can also be used for the determination of the absolute concentration of sterols. The main sterols in fish oil are cholesterol and vitamin

D (or its precursor 7-dehydrocholesterol) and are often added in the supplements. These compounds have a very similar MW. Therefore, results can be expressed in mg/g and are

42 calculated according to the equation C = 2/3 ISTE/IIS A × MWSTE/m, where MWSTE is the molecular mass (386) of cholesterol, which constitutes the majority of the sterolic fraction in fish oil20. The amount of sterols in the same sample using BHT was 3.8 mg/g of fish oil. The individual determination of cholesterol (δ 0.680) and 7- dehydrocholesterol (δ 0.678) is feasible on an 850 MHz instrument after the application of a window function for resolution enhancement.

13C NMR analysis

Figure 2 illustrates the 13C NMR spectra acquired on (A) an 850 MHz and (B) a 500

MHz instrument in the carbonyl carbon area. The two spectra are very similar and can provide the same amount of information. The 13C NMR spectrum can be successfully used for the analysis of additional fatty acids such as stearidonic (SDA) and eicosatetraenoic (ETA) acids, however more scans are required for samples in which these acids are in lower concentrations. The 13C spectra are characterized by high resolution because of the large spectral width and the application of broadband decoupling, which eliminates the effect of scalar coupling and produces singlets. For this reason, there is limited overlapping even when using a 500 MHz instrument.

The 13C NMR spectrum is much more informative compared to the 1H NMR spectrum and can provide more comprehensive quantitative data because less signal overlapping is observed (Figures 1 and 2). The most useful spectral region of the 13C spectrum is the carbonyl carbon region because it provides quantitative information for a large number of fatty acids as well as for their positional distribution on the glycerol skeleton19,21,22. The methyl group area from δ 14.5 to δ 13.5 can be used for the quick determination of the 43 total amount of n-3, n-6, n-9 and saturated fatty acids (SFA), as well as trans fatty acids.

However, in the 500 MHz NMR spectrometer, there is a partial overlapping of the n-6 and n-9 saturated fatty acids (SFA). The application of a window function for resolution enhancement may solve this problem although the 850 MHz instrument is still considered a more reliable option. The olefinic region of the carbon spectrum can be used for the total amount of n-3 and n-1 acyl chains as well as for determination of individual fatty acids such as DHA, EPA, Arachidonic acid (AA), Linolenic (Ln) n-3, and oleic acid (OL)

(see Table 2). 13C NMR can be also applied for the characterization of fish oil from other sources, such as supplements rich in ethyl esters (EE) using the carbon signals at δ 14.31

(methyl) and δ 60.20 (methylene).

For carbon analysis, fatty acids can be determined by dividing the integral of the appropriate aliphatic, olefinic, and carbonyl signals with the total integral of all acyl chains, according to the general relationship C = I/S (2), where C is the concentration of the analyte in mole (%), I is the integral of a resonance that is uniquely attributed to the lipid of interest, and S is the total integral of signal(s) that represents the total lipid content of the sample. The total integral S of acyl chains can be determined by integrating the region from δ 175 to δ 171 and is set to 100.

Quantification of fatty acids in mg/g of fish oil is performed using an internal standard on the basis of the following relationship: C = I/IIS × A× MW/m (3), where C is the concentration of the analyte in mg/g of fish oil, I is the integral of a resonance that is uniquely attributed to the lipid of interest, IIS is the area of a carbon signal that belongs uniquely to the internal standard, A is the mmoles of internal standard, MW is the

44 molecular weight of the compound of interest (for fatty acids expressed in methylesters), and m is the amount of fish oil in g. The 13C-NMR signals of fish oil fatty acids that can be used for quantitation purposes are shown in Table 2, whereas the complete NMR assignment of the 13C NMR spectrum can be found elsewhere19.

Example 1, EPA at sn-2 position: The amount (%) of EPA on the sn-2 position is calculated by dividing the integral of the signal at δ 172.56 by S. The amount of EPA at the sn-2 position in a commercially available sample was found to be 3.4% using the 850

MHz instrument. Using the same spectrometer and BHT as an internal standard, the amount of EPA at the sn-2 position expressed in mg/g of fish oil is 29.73 mg/g. The same sample analyzed in a 500 MHz instrument was found to contain 3.6% or 31.39 mg/g of

EPA in the sn-2 position. Similar results can be obtained when calculating the relative molecular ratios of EPA at sn-2 using a fully decoupled spectrum. This is because the carbonyl carbon of EPA is affected by proton decoupling to the same negligible degree as the other carbonyl carbons, which are used as reference. However, large deviations are observed when using BHT, because the carbon of BHT at δ 151.45, which is used for quantification, receive a different NOE enhancement compared to the carbonyl carbons of fatty acids. For that reason, the fully decoupled spectrum should be avoided when using internal standards or integrating carbons with different multiplicities.

Example 2, total amount of DHA: The total amount (%) of DHA is simply calculated by adding the amounts of DHA in sn-1,3 and sn-2 position as determined by the NMR signals at δ 172.48 and δ 172.08, respectively. The same sample analyzed with 1H NMR

(see example 1 of 1H analysis) was found to contain 10.3% DHA according to 13C NMR

45 analysis. The amount of DHA can also be expressed in mg/g by using an internal standard and Equation 3. The total amount of DHA was 103.25 mg/g.

Example 3, total amount of SDA: The total amount (%) of SDA is determined by adding the integrals of the signals at δ 172.99 and δ 172.60 which belong to the carbonyl carbons of SDA on position sn-1,3 and sn-2, respectively, then dividing the sum by S.

The sample analyzed was found to contain 3.93% SDA or 34.54 mg/g.

Example 4, n-3 Ln: n-3 Ln (%) can be determined by dividing the integral of the signal at δ 131.85 with the integral S. The molar ratio of n-3 Ln in the analyzed fish oil sample was 0.7%. The absolute concentration using BHT was calculated as 5.5 mg/g.

Example 5, trans fatty acids: The molar ratio of trans fatty acids is determined by dividing the integral of the signal at δ 13.80 with S. The analysis of the same sample that was analyzed with 1H NMR and was found to be 3.07% of trans FA, was also analyzed with 13C NMR and its trans fatty acid content was found to be 3.42%. The 13C NMR analysis of the same sample on a 500 MHz instrument showed a 3.64% content of trans fatty acids. The amount of trans FA in mmoles/g of fish oil can be determined using BHT as an internal standard and the equation C = I/IIS × A/m, however results cannot be expressed in mg/g because the peak at δ 13.80 corresponds to various trans fatty acids, mainly trans DHA and trans EPA, with different MW.

Example 6, EE: The concentration of EE in a fish oil sample is calculated by dividing the integral of the spectral area from δ 60.50 to δ 60.00, which corresponds to the methylene carbons of the EE of various fatty acids, with S. The analysis of an EE fish oil sample showed that it consisted of 100% EE. It should be noted that in EE samples, EPA

46 can be calculated either by the carbonyl peak at δ 173.60 or by the methylene EE carbon at δ 60.20, whereas DHA can be calculated using the signal at δ 60.31 and/or the signal at

δ 173.09.

A complete list of the diagnostic signals that can be used for quantification purposes with

13C and 1H NMR analysis can be found in Tables 1 and 2, respectively, whereas a detailed description of the equations that can be used for this analysis can be found elsewhere19.

NMR can additionally be applied for the assessment of the oxidation status of fish oil supplements. Figure 3 compares the 1H NMR spectra of a fish oil sample under two oxidation conditions; exposure to heating and exposure to ultraviolet (UV) light. Lipid oxidation is a complicated process, and the composition of oxidation products depends on the conditions of oxidation. The main oxidation products are hydroperoxides (δ 8.0–8.8), conjugated dienes hydroperoxides (δ 5.4–6.7), and aldehydes (δ 9.0- 10).

Figure Legends

Figure 4. The 1H NMR analysis. 850.23 (A) and 500.20 MHz (B) 1H-NMR spectrum of a fish oil supplement in CDCl3 solution. The NMR signals of EPA and DHA that can be used for their determination are shown. The peak at δ 0.97 can be used for the determination of the total amount of n-3 fatty acids. The envelop at δ 1.39-1.20 is cropped, as it belongs to the methylene protons of all fatty chains and cannot be used for any identification or quantification purposes. The 1H NMR spectrum is characterized by a narrower spectral width (SW) compared to the 13C NMR spectrum and thus by lower spectral resolution.

Figure 5. The 13C NMR analysis. 213.81 (A) and 125.77 MHz (B) 13C-NMR spectrum of a fish oil supplement in CDCl3 solution in the carbonyl carbon region. The NMR signals of EPA and DHA on sn-1,3 and sn-2 position are shown. These signals can be used for the quantitative determination of EPA and DHA. Although the spectra recorded at 213.81 MHz are characterized by a higher resolution and sensitivity, the 125.77 MHz spectra can also be used for the determination of the major compounds. The application of decoupling in the 13C NMR experiment eliminates the effect of scalar coupling between the carbon and hydrogen nuclei and thus the signals appear as singlets making the analysis easier compared to the 1H NMR spectrum.

FO: Exposure to heating Conjugated diene hydroperoxides Aldehydes

10.0 9.5 9.0 8.5 8.0 7.5 7.0 6.5 ppm FO: Exposure to UV Conjugated diene hydroperoxides light Aldehydes OOH

10.0 9.5 9.0 8.5 8.0 7.5 7.0 6.5 ppm

FO: No oxidation stress

10.0 9.5 9.0 8.5 8.0 7.5 7.0 6.5 ppm

Figure 6. Fish oil oxidation. The 1H NMR spectrum of oxidized fish oil depends on the oxidation conditions. The resonances attributed to hydroperoxides (δ 8.0–8.8), conjugated dienes hydroperoxides (δ 5.4–6.7), and aldehydes are shown.

1 Table 1. H-NMR chemical shifts of fish oil lipids in CDCl3 solution

δ ppm Proton Compound

0.677 CH3 (18) Cholestarol

7- 0.678 CH (18) 3 dehydrocholesterol

CH CH (t), J = 7.27 n-9, SFA acyl 0.88 2 3 ω1, ω2 Hz chains CH CH (t), J = 7.08 0.883 2 3 ω1, ω2 n-6 acyl chains Hz CH CH (t), J = 7.65 0.911 2 3 ω1, ω2 Trans acyl chains Hz CH CH (t), J = 7.63 0.973 2 3 ω1, ω2 n-3 acyl chains Hz

1.25 CH2CH3 (t), J = 7.20 Hz Ethyl esters OCOCH2CH (t) J = 1.697 2 Hα, Ηβ EPA acyl chain Hz

2.391 OCOCH2CH2 (t) DHA acyl chain

2.772 CH=CHCH2CH=CH n-6 acyl chains

2.81 CH=CHCH2CH=CH n-3 acyl chains Glycerol of 1- 3.593 3’a-CH OCO 2 MAG Glycerol of 1,2- 3.722 3’a, 3’b-CH OCO (br) 2 DAG Glycerol of 1,3- 4.073 2’-CHOH (br) DAG

4.121 CH2CH3 multiplet Ethyl esters Glycerol of 1,3- 4.173 1’b, 3’b-CH OCO (dd) 2 DAG Glycerol of 1,2- 4.238 1’a-CH OCO (dd) 2 DAG Glycerol of 1,2- 4.329 1’b-CH OCO (dd) 2 DAG

4.989 -CH=CH2 cis (dd) n-1 acyl chains

5.052 -CH=CH2 trans (dd) n-1 acyl chains Glycerol of 1,2- 5.082 2’-CHOCO DAG 5.268 2’-CHOCO Glycerol of TG

5.436 CH=CHCH2CH=CH2 n-1 acyl chains

5.818 -CH=CH2 n-1 acyl chains

The assignment of the 1H NMR spectrum. The 1H-NMR chemical shifts of fish oil fatty acid signals that can be used for quantification purposes in CDCl3 solution are presented. The chemical shifts are measured in ppm and provide information about the chemical environment of the nuclei.

13 Table 2. C -NMR chemical shifts of fish oil lipids in CDCl3 solution

δ ppm Carbon 173.24 C1 SFA (sn-1,3) 172.21 C1 OL, LO (sn-1,3) 173.16 C1 ETA (sn-1,3) 173.13 C1 DPA (sn-1,3) 173.03 C1 SDA (sn-1,3) 172.97 C1 EPA (sn-1,3) 172.73 C1 ETA (sn-2) 172.69 C1 DPA (sn-2) 172.61 C1 SDA (sn-2) 172.56 C1 EPA (sn-2) 172.48 C1 DHA (sn-1,3) 172.08 C1 DHA (sn-2) 136.8 Cω1, n-1 131.85 Cω3 LN 130.37 C15 AA 130.11 C9 LN 130.06 C13 LO 129.54 C5 DHA sn-2 129.47 C5 DHA sn-1,3 128.94 C5 EPA 128.76 C6 EPA 128.45 C17 n-3 127.71 n-3 127.53 C4 DHA sn-2 127.5 C4 DHA sn-1,3 126.86 Cω4, all n-3 114.71 Cω2, n-1

60.08 DHA, Ethyl esters 59.96 EPA, Ethyl esters 59.95- Other FA, Ethyl 59.85 esters 33.48 C2 EPA sn-2 33.32 C2 EPA sn-1,3 31.44 C3 n-1 27.05 Allylic n-6 26.49 C4 EPA sn-1,3 26.47 C4 EPA sn-2 24.6 C3 EPA 24.48 C3 SDA sn-1,3 24.44 C3 SDA sn-2 14.27 Cω1, all n-3 14.13 Cω1, SFA 14.11 Cω1, OL 14.07 Cω1, LO 13.8 Cω1, trans FA

The assignment of the 13C NMR spectrum. The 13C-NMR chemical shifts of fish oil fatty acid signals that can be used for quantitation purposes in CDCl3 solution are presented.

Figure 7. Comparison between the 13C NMR spectra acquired using the standard broadband decoupling (A) and the inverse gated decoupling (B) pulse sequences. Supplemental Figure S1: Comparison between the 13C NMR spectra acquired using the standard broadband decoupling (A) and the inverse gated decoupling (B) pulse sequences. The spectra were recorded for the same sample with the same number of

54 scans, processed with the same processing parameters and are shown with the same scale factor.

Discussion

Modifications and strategies for troubleshooting

Spectral quality. The linewidth of the NMR signal and thus the resolution of the NMR spectrum is highly dependent on shimming, which is a process for the optimization of the homogeneity of the magnetic field. For routine analysis, 1D shimming is adequate and a

3D shimming is not required, given that it is performed by NMR personnel on a regular basis. If this is not the case, a 3D shimming must be performed prior to analysis using a sample containing 0.6 mL of H2O:D2O (90:10). To achieve a better and faster shimming, the sample needs to be centered in the excitation/detection region of the radio frequency

(RF) coil, using the graded depth gauge, before it is placed in the magnet bore. Another factor that affects shimming is spinning the sample at a spin rate of 10-20 Hz. Although spinning the sample at this spin rate improves the radial shims (X, Y, XY, XZ, YZ, X2-

Y2, etc), it is generally not recommended in order to avoid the appearance of spinning side bands of the first or higher order. However, when working on instruments that operate in Larmor frequencies lower than 400 MHz, spinning is recommended for 1D

NMR experiments.

Resolution, as well as sensitivity, are affected by the receiver gain (rg) value. Low values of the receiver gain reduce the sensitivity, whereas values higher than appropriate cause overflow of the analog-to-digital converter (ADC). ADC overflow results in

55 nonsymmetrical line-shapes and the signals cannot be used for quantitative purposes because the first points of the free induction decay (FID) may be lost. In most cases, the command “rga” calculates an appropriate rg value. However, in some cases, the rg value calculated by the software is higher than the ideal value and there is a distortion in the

Lorentzian shape of the NMR signal. In such a case, the user should manually input a smaller rg value by typing “rg (value)” in the command line. A typical RG value for the samples that analyzed with this protocol is 8.

Often, when using cryogenically cooled NMR probes with a high Quality-factor (Q- factor), a large delay (dead time) >200us between the last pulse and the detection period is required to avoid artifacts such as a hump around the transmitter’s frequency and a rolling in the spectrum’s baseline. However, such a long delay causes a large negative first-order phase error, which can also introduce a baseline rolling and large dips around the base of the strong signals. In these cases, a z-restored spin-echo pulse sequence can be used to produce NMR spectra with significantly improved baselines, although a small sensitivity reduction may occur23.

Phospholipids. In addition to the analysis of fish oil samples rich in triglycerides and ethyl esters, NMR can be used for the analysis of fish oil samples rich in phospholipids

(PLs). However, special care is required for such samples because PLs form aggregates, which can cause a significant reduction in the spectral resolution and sensitivity. For the analysis of these samples, a solvent mixture of deuterated chloroform:methanol

(CDCl3:CD3OD) in a ratio of 70:30 is required for obtaining spectra of high quality.

Internal standard. BHT was selected as an internal standard in this study because it is a highly symmetric molecule with simple 1H and 13C NMR spectra and none of its peaks overlap with those of fish oil constituents. BHT has a signal in the 1H NMR spectrum, which appears as a singlet at δ 6.97 and belongs to the two equivalent aromatic protons

(para- position in respect to the OH group) and a signal at δ 151.45 in the 13C NMR spectrum which belongs to the aromatic quaternary carbon bearing the -OH group. Both of these signals have no overlap with any of the constituents in fish oil, and thus can be used for quantification purposes. Other compounds such as 1,2,4,5-Tetrachloro-3- nitrobenzene (TCNB) or ethylene chloride can also be used as alternative internal standards, however, they are characterized by longer T1 values.

Limitations of the technique

The quantification of various fatty acids and lipids in fish oil supplements is achieved through integration of the appropriate diagnostic NMR signals in the 1D spectra. Such signals should belong only to a specific sample component and must have no interference with signals from other compounds. This may be an issue for 1H NMR analysis since the

1H NMR spectrum is characterized by low resolution due to the short range of chemical shifts. In addition, the presence of scalar coupling (J) produces multiplets and makes the analysis more complicated. For example, ethylesters (EE) can be quantified using 1H

NMR by the characteristic triplet (J = 7.20 Hz) of the methyl group at δ 1.25 and the multiplet at δ 4.12, which belongs to the methylene protons of the ester group. However, when using NMR instruments operating in Larmor frequencies lower than 850 MHz, the analysis of EE using 1H NMR should be avoided because of the partial overlapping of the 57 peak at δ 4.12 with the peak at δ 4.14 of TGs, and the overlapping of the signal at δ 1.25 with the broad signal of the aliphatic methylene protons at δ 1.23–1.35. Large deviations were also observed between the 1H and 13C analysis of EPA in some samples, 13C NMR was closer to the labeled composition provided by the manufacturer. This is probably due to the overlapping of the signal at δ 1.69, which is used for EPA analysis, with signals of other compounds that appear in some types of fish oil supplements. Additional errors in quantifications can arise when using an internal standard due to the uncertain purity of the internal standard and from errors in weighing.

The compositional analysis can be expressed in relative molar concentrations without the use of an internal standard. If results need to be expressed in absolute concentrations, for example as milligram of fatty acid per gram of oil (mg/g), the use of an internal standard is required. However, in cases where the NMR signal of interest belongs to multiple compounds with different molecular weights, the results cannot be expressed as mg/g even when using an internal standard. In addition, the use of internal standard usually increases the length of the analysis because the most common internal standards, such as

BHT, are small molecules with high molecular symmetry, which results in long relaxation times. Since the repetition time (delay between pulses + acquisition time) is set according to the longest relaxation time T1 in the sample, the use of an internal standard will increase the duration of the experiments as longer delays between pulses are required. This is an especially important factor for 13C NMR analysis because of the exceptionally long T1 relaxation time of carbon nuclei. The addition of a paramagnetic compound such as Cr(acac)3 can efficiently reduce the T1 relaxation time. The

58 recommended concentration of Cr(acac)3 is 0.75 mg/ml of solution. Higher concentrations of Cr(acac)3 may be considered for further reduction of T1, however, caution is required in order to avoid decreases in the S/N due to the line broadening.

Although the 13C NMR is characterized by a much higher spectral resolution compared to

1H, the sensitivity of the 13C NMR experiment is significantly lower because of the low natural abundance (1.1 %) and the low gyromagnetic ratio (67.26 106 rad s-1 T-1) of 13C

13 nuclei. In addition, the long T1 relaxation times of C increase the length of the analysis.

This may be an issue when the available oil for analysis is limited, because an increased number of scans should be used to achieve a reasonable signal to noise ratio.

Limitations in the sensitivity and the resolution of the NMR spectra prevent the analysis of many minor compounds in fish oil that can be analyzed with other techniques such as

GC. For example, 1Η NMR is unable to separate individual sterols or fatty acids (eg palmitic and stearic) whereas 13C is not able to determine compounds that appear in very low concentration in fish oil such as dodecanoic and myristic acid, which overlap with the signals of all saturated fatty acids at δ 173.24 and δ 172.82. Although increasing the amount of sample that is analyzed makes the analysis of some minor compounds feasible, caution is required for very concentrated samples, because of their increased viscosity.

Very viscous solutions containing more than 150 mg of oil should be avoided because there is a decrease in S/N due to the line broadening caused by the reduced spin-spin T2 relaxation times. In addition, longer delays between pulses are required because of the longer T1 and there are several issues in shimming and thus in resolution.

All the compounds analyzed in fish oil with NMR can be quantified simultaneously in one snapshot without using any separation or purification steps. The NMR analysis is rapid as the 1H spectrum can be recorded in less than one minute, whereas the 13C NMR acquisition lasts 10 min. It should be noted, however, that there are a few factors that affect the data acquisition time. Specifically, for 13C NMR, the 10 min run time can only be achieved without the use of internal standards, and with the use of cryogenically cooled probes, in which the RF coil and the preamplifier are cooled and thus the thermal noise is minimized. A 10-15 fold increase in the experimental time should be expected for 13C NMR analysis when room temperature (conventional) probes are used.

Significance with respect to existing methods

NMR spectroscopy proved to be a powerful tool for qualitative and quantitative determination of the composition of fish oil supplements, and because of its rapidness it has the potential to be applied for the high throughput screening of a vast number of fish oil samples. NMR spectroscopy is by definition a quantitative methodology since the signal area is directly proportional to the number of nuclei that cause the signal. While acutely toxic chemicals are required to prepare NMR samples, this method is environmentally friendly because such small amounts of this chemicals (eg CDCl3) are used as opposed to other methods that require large amounts of solvent to elute samples.

In addition, NMR has several advantages compared to other analytical methods. No calibration with standards is required prior the analysis, and a minimal sample preparation without any separation and purification steps is usually adopted, which renders NMR a very fast analytical tool. Additionally, 13C NMR is the best available 60 methodology for determining the positional distribution of various fatty acids on the glycerol skeleton. While the enzymatic hydrolysis has been used as an alternative is not always reliable24. This is of specific importance because there is a significant interest in studying the regiospecificity of various fatty acids in foods, as it has been found that this affects their function in human diet25,26.

Future applications

Despite the agreement between NMR analysis and the products’ label, as well as the fact that there are some studies showing agreement between GC and NMR, we believe that more rigorous and comprehensive intra-laboratory studies are required to examine the agreement between NMR and traditional methodologies for the analysis of fish oil constituents using a larger number of samples, fish oil products of different origins, and certified standard solutions.

Another important future application of NMR in fish oil analysis will be the determination of the oxidation products. In addition to the determination of the major compounds in fish oil, several primary and secondary oxidation products in fish oil, such as aldehydes and peroxides, are present. 1H NMR can be potentially applied for the evaluation of the oxidation status in fish oil supplements, under different oxidation conditions, as shown in Figure 3. The biggest challenge in this analysis will be the NMR assignment and the identification of individual oxidation products. Advances in sensitivity of the NMR hardware will also allow the identification of individual sterols using 13C NMR. NMR spectroscopy can also be applied for the analysis of fish tissue as a

61 whole even without any extraction by using High Resolution Magic Angle Spinning (HR-

MAS) NMR.

Critical steps within the protocol

Two of the most critical steps that affect the accuracy of quantitative NMR spectra involve the selection of a 90° pulse and the use of a delay between pulses ≥ 5×T1. The pulse angle is proportional to pulse width which is a calibrated NMR parameter that depends on the instrumentation and the sample. A 90° pulse is essential for the complete conversion of longitudinal (z) magnetization to the observable transverse (xy) magnetization. It is important to note that before pulse calibration, the NMR probed needs to be well tuned and matched. This will optimize the transfer of the RF power to the sample and thus maximize S/N and ensure effective decoupling. The probe tuning is mostly affected by the dielectric constant of the sample, so if there are differences in the concentration between samples, repeat the tuning process for each one. The 1D 13C NMR experiment involves both 13C and 1H channels so automatic tuning and matching is necessary for both nuclei.

A delay between pulses longer than 5×T1 ensures the complete recovery of the net magnetization to its initial value. If all the resonances in the spectrum have not completely relaxed before each pulse, the signal is partially suppressed and this leads to inaccuracies in the integration. T1 value is a critical factor that affects the length of the experiment and it depends on the magnetic field strength as well as the viscosity of the sample. Given that the viscosity between samples is similar, T1 relaxation times should be determined for each instrument only in the beginning of the analysis session. 62

Another important feature of the fish oil analysis with 13C NMR is the selection of the appropriate pulse sequence. The most reliable method for quantitative 13C analysis is the inverse gated decoupling experiment, where broadband proton decoupling is applied only during the acquisition period and thus there is no polarization transfer from 1H to 13C via the nuclear Overhauser effect (NOE). However, while the fully decoupled NMR experiment can be used for quantitative purposes, caution is required when using this experiment because there are different NOE factors among carbons with different multiplicities and therefore integral comparison between methyl, methylene, methane and carbonyl carbons must be avoided. Despite this, when only carbons of similar multiplicity and chemical environment are considered in the analysis, the fully decoupled method is reliable. One example of this is carbonyl carbons of fatty acids which have been found to have no significant differences in the NOE factors after decoupling27. In addition, for carbons bearing protons, the fully decoupled experiment provides higher sensitivity due to the NOE contributions on the NMR signal intensity. A comparison between spectra acquired with the two pulse sequences is shown in figure S1.

Acknowledgments

This work was supported by the Foods for Health Discovery Theme at The Ohio State

University and the Department of Food Science and Technology at The Ohio State

University. The authors would like to thank the NMR facility at The Ohio State

University and the NMR facility at Penn State University.

Disclosures

The authors have nothing to disclose.

Chapter 6. Analysis of coffee lipids using NMR spectroscopy

Introduction

Coffee is among the most traded and consumed commodities in the world. The two most commercially available coffee species are Coffea Arabica and Coffea Canephora, commonly referred to as Arabica and Robusta coffee, respectively, with Arabica considered to have the highest quality and commercial value. The mass of a green Arabica coffee bean is composed of about 15-17% lipids, while a green Robusta bean is composed of about 10% lipids (Calligaris et al. 2009). These lipids, collectively called coffee oil, are mainly found in the endosperm (Crisafulli et al.) of the green coffee bean and composed of nearly 75% triacylglycerols (TG), with oleic (OL), linoleic (LO), and linolenic (LN) acids being the most important for determining coffee freshness (Folstar 1985; Speer and

Kölling-Speer 2006b) (Wilson and Illy P.E. 1997; Speer and Kölling-Speer 2006b;

Calligaris et al. 2009). Partially esterified compounds, such as diglycerides can be also found. Unlike many oils of plant or animal origins, coffee oil also contains a unique unsaponifiable fraction which accounts for nearly 24% of the lipids (Calligaris et al. 2009).

This unsaponifiable fraction is rich in diterpenes, sterols, tocopherols, and waxes

(Calligaris et al. 2009; D’Amelio et al. 2013).

Because of its fatty acid composition and its unique content of bioactive compounds, coffee oil has many applications as a health promoting, natural product. Two coffee-specific diterpenes, cafestol and kahweol, are particularly important for health as they have been shown to induce the degradation of toxic substances, protect against

65 aflatoxin (Cavin et al. 2001), and to have positive effects against cancer(Cavin et al. 2003).

Additionally, they have antioxidant, anti-inflammatory, and hepatoprotective properties in humans (Cavin et al. 2002; Lee and Jeong 2007; Lee et al. 2007), whereas they are used as biomarkers for coffee processing and coffee authentication (Speer et al. 1992; Speer and

Kurt 2001; Scharnhop and Winterhalter 2009). Due to its peculiar taste and flavoring, coffee oil is used as a basic ingredient for the production of many food products, such as candies, cakes, and beverages, where it is added to improve the sensory characteristic but also increase the health promoting effects of foods (Frascareli et al. 2012; Raba et al. 2015).

In addition to its nutritional properties, coffee oil has the potential in an array of applications in the cosmetic and pharmaceutical industries. It can be used as a sun protectant because of its ability to block UV-B radiation, as well as help with the formation of connective tissue (Del Carmen Velazquez Pereda et al. 2009) and help with skin dryness and serious dermatological conditions, such as eczema and psoriasis (Raba et al. 2015;

Wagemaker et al. 2015). Another diverse application of coffee oil is using the oil obtained from the wasted spent coffee grounds (SCG) following coffee beverage extraction as a source for biodiesel production (Oliveira et al. 2008; Maressa Aparecida de Oliveira et al.

2014; Caetano et al. 2014; Park et al. 2016). Due to its large number of current and potential applications in the food, cosmetic, pharmaceutical and bioenergy industry, coffee oil is valued at a higher price compared to coffee itself.

In the coffee bean, the lipids have the potential to act as indicators of coffee quality, biomarkers for classification purposes, and as indicators regarding the function and the health status of the coffee plant. For example, the lipid composition of coffee

66 bean is affected by the stress level of the coffee plant as related to external factors, such as breeding and environmental factors (Echeverria-Beirute et al. 2017). In addition, coffee oil sterols have been found to be an effective tool for the classification of coffee varieties, and other coffee lipid components play an important role in determining shelf life and sensory perception of the final coffee beverage (Carrera et al. 1998; Toci et al.

2013; Rendón et al. 2014). These lipid components either pre-exist in the green coffee bean or they are formed during roasting.

Roasting is the most important processing step of a coffee bean, which turns it into a porous, brown coffee bean. The roasting process induces dramatic changes in the chemical properties of the coffee, and may be the most important factor in the development of desirable and complex flavors in a final cup of a coffee beverage (Illy and Viani 2005).

Most of the changes in biochemical composition due to roasting are related to Maillard reaction products and carbohydrate degradation (Calligaris et al. 2009; Wei et al. 2012a).

Surprisingly, there are not as many studies involved with the impact of roasting in the lipid fraction. While there is a huge impact of roasting on the polar compounds in coffee, such as sugars and acids, some studies indicate that coffee lipids and the oxidation and/or hydrolysis products associated with lipids are not affected by the high roasting temperatures (Anesei et al. 2000; Vila et al. 2005; Kobelnilk et al. 2014). However, other studies have found significant differences in certain lipid components during roasting, although the total lipid profile was not examined (Martín et al. 2001).

During roasting, several lipid compounds such as primary and secondary oxidation products that do not exist in the green coffee bean are produced, and have been found to

67 contribute to the final taste and aroma (Budryn et al. 2012). Holscher and Steinhart discovered that a large number of low-odor threshold molecules in green coffee possess a carbonyl group and are degradation products generated during autoxidation of lipids that contribute to final roasted coffee flavor (Holscher and Steinhart 1995). Overall, the literature is conflicting regarding the compositional changes in lipids during roasting.

Controversial data also exist about the effect of storage in the lipids and oxidation/hydrolysis status of coffee (Vila et al. 2005; Toci et al. 2013), although there is strong evidence that it is during storage that these lipids are oxidized and hydrolized, and lipid oxidation and hydrolysis products can cause rancid, off-flavors attributed to a stale cup of coffee (Rendón et al. 2014).

Generating knowledge regarding the impact of roasting, as well as other factors, such as storage and water activity in the lipids composition is of specific importance given the significance of these compounds and their oxidation and hydrolysis products on the quality and the commercial value of coffee (Toci et al. 2013). Oxidation and hydrolysis products from lipid oxidation may be used as indicators for quality control purposes and assessment of important factors such as freshness and storage. In addition, several compounds that have been found to impact the sensory perception of a coffee beverage are lipophilic and thus become a part of the lipid fraction (Calligaris et al. 2009). Further, given the great potential of coffee oil as a value-added material, it is important to understand the impact of various factors in its quality and composition.

The lack of consensus regarding coffee lipid changes during roasting due to few studies reported in coffee lipid analysis may be partially explained by the lack of efficient

68 analytical tools that allow the rapid and reliable evaluation of coffee oil. Coffee oil has been analyzed using a variety of analytical methods including High Performance Liquid

Chromatography (HPLC), Gas Chromatography (GC), Capillary gas chromatography

(cGC), Electrospray ionization-mass spectrometry (EIS-MS), and Fourier Transform

Infrared Spectroscopy (FT-IR) (Valdenebro et al.; Carrera et al. 1998; González et al. 2001;

Martín et al. 2001). Despite the benefits of these methods, some of them require lengthy extraction techniques, derivatization, and complex sample preparation, or may not be capable of providing quantitative data or information on the molecular level. For example, the GC analysis of fatty acids requires separation and derivatization of individual compounds, and lipid oxidation may occur during the analysis, which can cause further inaccuracies (Sacchi et al. 1993; Igarashi et al. 2000; Guillén and Ruiz 2003). Similarly, the current procedures for the determination of terpenes and the sterolic profile in coffee are very time consuming and involves several steps, including lipid saponification and chromatographic separations with TLC and GC.

Nuclear Magnetic Resonance (NMR) Spectroscopy is a robust method that can rapidly analyze the lipid profile of coffee at the molecular level without requiring separation and/or purification steps. The analysis is conducted in one snapshot, which provides a significant advantage compared to traditional analysis that focuses on one class of compounds at a time. The major drawback of NMR is its low sensitivity compared to other analytical techniques, such as chromatography and mass spectrometry, however

NMR’s nondestructive, rapid, and robust nature offers a great non-targeted approach when combined with multivariate statistical analysis.

NMR has been used in the past to analyze the polar coffee components, especially for classification purposes, to differentiate Arabica and Robusta coffee, to analyze green coffee according to variety and origin, and to measure changes that occur during roasting

(Ciampa et al. 2010; Wei et al. 2012b, a; Schievano et al. 2014; Monakhova et al. 2015b).

So far, NMR has been used once before by D’Amelio et al. to evaluate green coffee lipids of coffee beans from different origins and species (arabica and robusta) (D’Amelio et al.

2013). In that study, green coffee oil was analyzed without dilution, which allows for fast analysis of several lipid components, but causes a reduction in resolution because of the high sample concentration. In addition, Monakova et al. applied 1H NMR spectroscopy to determine the presence of Arabica and Robusta species by screening lipophilic extracts

(Monakhova et al. 2015b).

Here we developed an NMR method for the qualitative and quantitative determination of various coffee lipid components in green and roasted coffee beans, as well as spent coffee grounds and coffee beverage. Further, we combined NMR with a hybrid targeted-untargeted multinuclear metabolomic analysis to investigate the impact of roasting and other factors, such as storage and water activity in the lipid profile of coffee.

Most metabolomics and foodomics studies involving NMR deal with 1H-NMR spectroscopy due to the higher sensitivity of proton nuclei. Other nuclei such as 13C are not often used due to low sensitivity of carbon nuclei, and for targeted analysis where the full relaxation of magnetization is required, the long relaxation times of carbon experiments dramatically increase the experimental time and cost, rendering the analysis difficult.

However, 13C NMR spectra are characterized by excellent resolution due to the large

70 spectral width and the elimination of scalar coupling due to decoupling. Therefore, we developed an analytical approach for simultaneous targeted and untargeted analysis using a single sample preparation utilizing both 1H and 13C nuclei.

Targeted analysis was initially performed using an internal standard which is required when quantitative NMR results need to be expressed in absolute concentrations.

The selection of a good internal standard can be a difficult task, especially for complex mixtures as there is an increasing chance for signal overlapping. Additionally, using an internal standard can contaminate the sample ad may prevent further analysis and sample recovery, which is often necessary for biological and clinical studies, as well as sensory analysis studies. An alternative approach is (Pulse Length–based Concentration determination) PULCON which is based on the use of an external standard in a separate

NMR tube and with a predefined concentration. This method was first introduced by Wider et al and it is now commercially available as ERETIC2 (Electronic REference To access

In vivo Concentrations 2) tool (Wider and Dreier 2006; Tyburn and Coutant 2015).

PULCON/ERETIC2 allows for quantification of various compounds without the contamination of the sample with an internal standard, which may prevent further analysis using other techniques. Although this method offers an excellent and convenient tool for mixture analysis, only a few studies utilizing this technique have been conducted so far, mostly in the field of pharmaceutical analysis (Akoka et al. 1999; Billault et al. 2002;

Monakhovaa et al. 2014) whereas as only one study involves application in food analysis for the determination of polar compounds (Monakhova et al. 2013, 2015a; Ackermann et al. 2017). In any case no systematic study to evaluate its performance has been conducted.

In addition, no application of PULCON in lipids has been reported so far. Magnetic

Resonance Imaging (MRI) was also utilized to study the structural changes in coffee bean as related to coffee oil sweating and lipid migration as a result of roasting. To our knowledge, MRI has been used in one previous application to view the water distribution of a green coffee bean (Schmidt et al. 1996) and this is the first application of MRI for the study of roasted coffee lipids.

The overall goal of this study is to employ multinuclear and multidimensional 1H and 13C NMR spectroscopy as a reliable tool for the metabolomic analysis of lipids in coffee before and after roasting, as well as after coffee beverage extraction. Our method offers an excellent tool for the high-resolution analysis of coffee lipids, which can be used for the assessment of coffee oil. This protocol can be also applied to several other complex matrices, not just coffee oil. In addition, we present the first MRI application for the analysis of water and lipids in a roasted coffee bean, as well as a novel application of coffee lipids obtained from spent coffee grounds, a waste product, with a potential for the development of biopolymers.

Material and Methods

Coffee samples

Eighteen green coffee beans and the corresponding roasted samples, all processed under the same roasting conditions determined by the Specialty Coffee Association of

America (SCAA), were used in this study. All samples were harvested in the 2015-2016 harvest year and stored at -40°C under the same conditions.

Chemicals

Chloroform d1, Dimethyl sulfoxide-d6, D2O, CD3OD, were purchased from

Cambridge Isotope Laboratories (Tewksbury, MA). Cr(acac)3, 3,5-Di-tert-butyl-4-

+ hydroxytoluene (BHT), hydrogen peroxide, formic acid, NaCl, NaHCO3, EDTA-K , chloroform, petroleum ether, ethyl acetate, and ethanol were obtained from Fisher

Chemical (Waltham, MA, USA).

Measurement of water activities and moisture content

Water activities for green coffee samples were measured using an Aqua Lab water activity meter (Meter, Pullman, WA, USA). Moisture content was measured using a vacuum oven (Isotemp Vaccum Oven, Fisher Scientific, Waltham, MA, USA) by weighing the samples before oven drying at 200°C under vacuum, and after drying for 24 hours.

Difference in weight was attributed to moisture and divided by the total weight of the sample to calculate moisture content.

Synthesis of coffee oil epoxides

160 g of dried coffee waste was placed in a supercritical fluid extraction system (Waters

Inc., Milford, MA, USA) with 1L sample capacity. Coffee oil was extracted using 100%

CO2 as extraction medium (70°C, 300 bar, 100g/min). Coffee oil was epoxidized at a molar ratio of 1.0:1.0:5.6. (coffee oil: formic acid: hydrogen peroxide). First, 2.5g of the extracted oil was mixed with hydrogen peroxide (1.794 g, 30 %) in a round bottom flask and stirred at 400 rpm to form a homogenous mixture. Next, 0.5 ml of formic acid (0.154 g, 85 %) was added drop-wise. The mixture was stirred at 50°C for 48 hours. Epoxidized oil was taken from the reaction mixture and transferred into a separatory funnel where it was mixed

73 with 8.3 ml of ethyl acetate. After shaking the funnel the mixture was left to relax until the separation of the two layers. The water (lower) layer was removed and the organic layer was shaken with 12.5 mL brine (saturated salt solution). The mixture was left to relax until the two layers separated, and the water layer was removed. 12.5 ml of a NaHCO3 1M solution was added and shaken. The water phase was discarded and the organic phase was further washed with 4.5 ml brine. Finally, the organic phase was collected and dried using

3.5 g of sodium sulfate (anhydrous), filtered (#4 Whatman), and the solvent was removed under reduced pressure evaporation (Buchi , Flawil, CH).

Sample preparation for NMR experiments

Coffee oil from green and roasted beans: 2 g of coffee were ground using a L’equip mini coffee grinder & seed mill (Nutrimill, Salt Lake City, UT) for 30 seconds. Ground samples were mechanically shaken with 10 ml of chloroform for 5 min at room temperature. The extracts were filtered using filter paper (#4 Whatman) and the solvent was removed under reduced pressure evaporation. The oil was dissolved in 600 uL of a stock solution (20 mL) composed of CDCl3/DMSO d6 in 5:1 volume ratio containing 10 mg of chromium acetylacetonate, Cr(acac)3 (28.6 uM), and 110 mg of BHT (500 uM), protected from moisture with 5A molecular sieves. The sample was inserted directly into a 5 mm NMR tube for acquiring one- and two-dimensional 1H and 13C NMR spectra.

Coffee oil extracted from coffee beverage: A coffee beverage of the roasted beans was prepared at 5% coffee: water. 20 mL of brewed coffee was shaken with 20 mL of CHCl3.

The mixture was centrifuged at 2700 rpm for 20 minutes at 20°C (Eppendorf Centrifuge

5810R, Hamburg, DE). The bottom CHCl3 layer was collected in a rotary evaporator flask.

The remaining coffee was extracted a second time with an additional 15 mL of CHCl3 under the same centrifugal parameters. The solvent from the two extractions was combined and removed under reduced pressure evaporation. The remaining oil was dissolved in 600 uL of the same CDCl3:DMSO (5:1) mixture as mentioned above and placed into a 5 mm

NMR tube.

Phospholipids extraction from ground coffee: 5 g of ground coffee were extracted by 2

× 20 mL of ethanol/water (85:15 v/v). The ethanolic extracts were washed with 10 mL of petroleum ether, and the phospholipids were obtained upon removal of the solvent under vacuum. The residual oil that contained the phospholipids was re-dissolved in 0.5 ml of

+ CDCl3/CD3OD/D2O-EDTA-K (400:95:5 v/v/v).

Coffee oil and epoxides from coffee waste-reaction monitoring: 2.5 ml of sample was taken after 5 hours and after 10 hours of reaction time and was treated as described in section 2.4. After the removal of ethyl acetate the epoxidized oil was dissolved in 0.5 ml of CDCl3. The solution was then transferred into a 5mm NMR tube.

NMR experiments

1H and 13C NMR experiments were conducted on a Brüker Avance III spectrometer

(Bruker, Ettlingen, DE) operating at 700.13 MHz and 176.04 MHz for 1H and 13C nuclei, respectively, equipped with a TXO helium-cooled 5mm probe, whereas 31P NMR experiments were carried out on a Brüker Avance III spectrometer operating at 323.89

MHz for 31P nuclei, equipped with a TCI helium-cooled 5mm probe. Certain 2D NMR experiments were also performed on a spectrometer operating at 850.23 MHz and 213.81

MHz for 1H and 13C nuclei, respectively, equipped with a triple resonance helium-cooled

75 inverse (TCI) 5 mm probe. All experiments were performed at 25 ± 0.1 ºC and the spectra were processed by the Topspin software package provided by Brüker Biospin.

One-dimensional (1D) NMR spectra

1H-NMR spectra were recorded using the following acquisition parameters: 16 scans and 4 dummy scans, 64K data points, 90º pulse angle (10.5 μs), relaxation delay 10 s to ensure quantitative results, spectral width 13 ppm. A polynomial fourth-order function was applied for base-line correction in order to achieve accurate quantitative measurements upon integration of signals of interest. The spectra were acquired without spinning the

NMR tube in order to avoid artifacts, such as spinning side bands of the first or higher order. Chemical shifts are reported in ppm from TMS (δ = 0).

13C-NMR spectra were obtained with proton decoupling using the inverse gated decoupling version of a z-restored spin echo method(Xia et al. 2008) to minimize NOE effects and produce improved baselines. Repetition delays between consecutive 90º pulses equal to five times the longitudinal relaxation times measured by the null method. 13C NMR spectra were recorded using a modified version (inverse gated decoupling) of the z-restored spin-echo pulse sequence (Xia et al. 2008), with spectral width of 200 ppm, using 64K data points, a 90º excitation pulse (9.8 μs); acquisition time 0.9 s and relaxation delay of 60 s in order to avoid signal saturation. 128 scans were collected and spectra zero-filled to 128K.

For all FIDs, line broadening of 1 Hz was applied prior to Fourier transform. Chemical shifts are reported in ppm from CDCl3 (δ = 77).

31P NMR experiments were obtained as follows: 90° pulse width, 15 μs; sweep width of 40 ppm; relaxation delay, 2 s; memory size, 32K (zero-filled to 64K). Line broadening

76 of 3 Hz was applied, and drift correction was performed prior to Fourier transform.

Polynomial fourth-order baseline correction was performed before integration. For each spectrum 2K transients were acquired.

Two-dimensional (2D) NMR experiments

Experimental details and pertinent references for most of the 2D pulse sequences used in this study can be found in Reference (Berger & Broun, 2004).

Gradient selected 1H-1H Correlation Spectroscopy (H-H-gCOSY) experiments were performed in the magnitude mode using 8 dummy scans, 8 scans and 256 increments.

Spectral width of 13 ppm in both dimensions, 2K data points) in F2 dimension and a relaxation delay of 1.5 s. The spectra were zero-filled to a final size of 2K × 1K prior to

Fourier transformation.

1H-1H DQF (Double Quantum Filtered) COSY experiments were performed using 8 dummy scans, 32 scans and 256 increments. Spectral width of 13 ppm in both dimensions,

2K data points in F2 dimension and a relaxation delay of 1.5 s were used. The spectra were zero-filled to a final size of 2K × 1K prior to Fourier transformation.

1H–1H Total Correlation Homonuclear Spectroscopy (H–H-TOCSY). These spectra were acquired in the phase sensitive mode with TPPI, using the DISPI2 pulse sequence for spin lock. Typically, 16 dummy scans, 32 scans and 512 increments were collected, with

SW of 13 ppm in both dimensions, 2K data points in F2 dimension, spin-lock time of 60 ms, and a relaxation delay of 2.0 s. The data points in the second dimension were increased to 2 K real data points by linear prediction, and the spectra were zero-filled to a final size

77 of 4K × 4K prior to Fourier transformation. A sine-bell squared window function was used in both dimensions.

The Gradient selected 1H-13C heteronuclear multiple bond correlation (gHMBC) experiment was performed using a low-pass J-filter (3.4 ms) and delays of 65 and 36 ms to observe long-range C–H couplings optimized for 3 and 7 Hz with 256 increments and

32 transients of 4096 data points. The relaxation delay was 1.5 s. Zero-filling to a 4K × 1K matrix and π/2-shifted sine square bell multiplication was performed prior to Fourier transform.

The hybrid HSQC-TOCSY experiment consists of the initial basic gradient enhanced

HSQC sequence, followed by a phase-sensitive TOCSY transfer step with the echo- antiecho method using the DISPI2 pulse train for the spin lock. The experiment was conducted with 1K × 256 complex points and a spectral width of 13 ppm for 1H and 220 ppm for 13C. 64 transients were collected for each point with 32 dummy scans. The mixing time 60 ms and the relaxation delay was 2.0 s. the spectra were zero filled to 2K × 2K and processed with Qsine-square bell in both dimension.

The combined experiment Gradient selected 1H-13C multiplicity-edited heteronuclear single quantum coherence (HSQC-DEPT or edited HSQC) was performed with 512 × 512 complex points and a spectral width of 180 ppm for 13C (F1) and 10 ppm for 1H (F2), 512 increments, 32 dummy scans and 64 scans for each increment according to the echo- antiecho procedure, relaxation delay of 2 s; delays 3.6 ms (1/2 J) for multiplicity selection, and 1.8 ms (1/4 J) for sensitivity improvement were used. Carbon decoupling during proton acquisition was achieved by applying the GARP pulse train. Gradient strengths were 20

78 and 5 G/cm. The data were multiplied in the 1H time domain with a qsine weighting function. The 13C time domain was doubled by forward linear prediction prior to a cosine window function.

The band-selective constant Heteronuclear Multiple Bond Correlation (HMBC) experiments14,15 were acquired using the standard shmbcctetgpl2nd Bruker pulse sequence with a SW of 15 ppm in 1H dimension and 10 ppm in 13C dimension, with 2K × 128 data points. Selective excitation of 13C was achieved using a 180º Gaussian pulse. Pulsed field gradients were applied as half-sine shaped pulses. All spectra were zero-filled to 2K × 1K of data points for 1H and 13C dimensions respectively and are presented in magnitude- mode.

The Gradient selected 1H-31P heteronuclear multiple bond correlation (gHMBC) experiment was performed with 128 increments and 256 scans of 1K data points. The relaxation delay was 2.0 s. The evolution delay for long-range proton-phosphorus couplings was set to 65ms associated with JH-P of about 15 Hz. Zero-filling to a 1K × 1K matrix and π/2-shifted sine square bell multiplication was performed prior to Fourier transform.

MRI

All MRI experiments were performed on a Bruker Ascend 750 MHz Micro imaging system

(Bruker, Ettlingen, DE). The system had field strength of 17.6T and was equipped with 89 mm wide bore (WB) and standard bore (SB) systems. A rapid acquisition with relaxation enhancement Rapid Acquisition with Relaxation Enhancement (RARE) sequence was used to acquire proton density (PD) images o with the following parameters: Repetition time

(TR) 1 s, echo time (TE) 3.8 ms, slice thinkness (SI) 0.50/0.50 mm, field of view (FOV)

1.70/1.20 cm and matrix size 196/128(a).

Spectral Data Processing and Multivariate Data Analysis

For 1H NMR untargeted analysis the spectral regions δ 0.30−12.50 was integrated into regions with equal width of 0.01 ppm using the AMIX software package (V3.9, Bruker-

Biospin). For 13C NMR based metabolomic analysis the regions δ 5−200 and a bin width equal to 0.01 ppm were used. The regions where BHT, water, CDCl3 and DMSO d6 signals appear were discarded. Each bucketed region was then normalized to the total sum of the spectral integrals to compensate for the overall concentration differences prior to statistical data analysis. Multivariate data analysis was carried out with SIMCA-P+ software (version

14.1, Umetrics, Sweden). Principal component analysis (PCA) and orthogonal projection to latent structures with discriminant analysis (OPLS-DA) were conducted on the NMR data. The OPLS-DA models were validated using a 7-fold cross validation method, and the quality of the model was described by the parameters R2Y and Q2 values.

Univariate data analysis

Bland Altman and paired t-tests (two-tailed, α levels = 0.05), were performed using IBM

SPSS Statistics package version 24 (NY, USA).

PULCON-based quantification

PULCON analysis was performed using the equation as described previously (Wider and

Dreier 2006; Monakhovaa et al. 2014)

퐴푢푛푘푇푢푛푘휃푢푛푘푛푟푒푓 퐶푈푁퐾 = 푘퐶푟푒푓 퐴푟푒푓푇푟푒푓휃푟푒푓푛푢푛푘

0.6 ml of the BHT stock solution was transferred into a 5 mm NMR tube and NMR experiments were acquired under quantitative conditions, as described in section 2.6.1.

All experiments for external standard (PULCON) samples and coffee oil samples were processed using the same parameters. Comparisons were made only between samples that were run in the NMR instrument/probe.

Results and Discussion

Compound identification-Spectral assignment

The high-resolution NMR analysis of coffee oil offers a rapid and robust method for the compositional analysis of its major and minor compounds. These compounds can be used for targeted and untargeted metabolomics approaches for quality control and classification purposes, as well as aid in understanding compositional changes in coffee due to processing. Both 1H and 13C NMR spectra contain diagnostic signals that can be attributed to specific compounds and thus can be used for quantification. Because the reliability of the qualitative and quantitative analysis depends on the correct NMR assignment of the 1D spectra, we performed the first high resolution systematic 2D NMR analysis that provides the unambiguous assignment of 1H and 13C NMR signals of coffee oil constituents.

Figure 8A and Figure 8B display the high resolution 1H spectrum and the

olefinic region of the 13C NMR spectrum of coffee oil obtained from a roasted coffee

bean, acquired in CDCl3:DMSO d6 (5:1) solution. The use of a solvent mixture that

contains DMSO-d6 is essential for the determination of compounds bearing labile protons

such as hydroperoxides that bear OOH protons in the 1H NMR spectra. This solvent

mixture slows down the proton exchange and has been found to yield spectra with

excellent resolution (Skiera et al. 2012). The main classes of compounds that can be

determined with 1H and 13C NMR are fatty acids (FA), esterified and partially hydrolyzed

lipids, terpenes, sterols, caffeine and oxidation products.

A B

Figure 8. 850 MHz (A) 1H-NMR spectrum and (B) 213 MHz 13C-NMR spectrum of coffee oil in the olefinic region of a coffee oil sample in CDCl3:DMSO-d6 solution. OL, oleic acid; LO, linoleic acid; LN, linolenic acid; K, Kahweol; C, Cafestol; Cf, Caffeine. Fatty acid composition

The total amount of unsaturated and saturated fatty acids (SFA), except linolenic

(LN), can be determined by the signal that appears at δ 0.88 in the 1H NMR spectrum and

belongs to the protons of the terminal methyl group of FA (Figure 8). For most oils with

82 a plant or animal origin, these methyl protons appear as triplets. However, in the case of coffee oil they appear as a pseudo-quartet that arises from the overlapping between the methyl protons of oleic (OL) acid and those of linoleic (LO) acid, which appears in increased amounts in coffee oil. Acquiring the spectrum in instruments that operate in higher Larmor frequencies (eg 850 MHz) reveals two distinct triplets (Figure 20). In addition, the TOCSY spectrum shows correlations of the methyl protons with methylene and allylic protons in two different chemical shifts, δ 0.875 for the methyl group of OL and SFA and δ 0.887 for LO. The TOCSY spectrum of a green coffee oil sample is shown in Figure 9.

1 1 Figure 9. 850 MHz H- H TOCSY spectrum of green coffee oil in CDCl3:DMSO-d6 solution, showing the connectivity between protons in the same spin system. The most convenient way for the determination of individual fatty acids such as

OL and LO fatty acids using 1H NMR spectrum is the use of the distinct signals of their 83 allylic protons, at δ 2.00 and δ 2.04, respectively. For the determination of LN, the signal of bis-allylic protons at δ 2.79 is recommended instead of the signal of allylic protons due to overlapping of the allylic protons of Ln with the methine proton H14 of Cafestol as found by the HSQC-TOCSY spectrum.

The 13C NMR spectrum can also be very informative for the analysis of fatty acids that appear in coffee oil. For the determination of various lipids of plant or animal origin, the most useful area in the 13C NMR spectrum is the carbonyl carbon region.

However, in the case of coffee oil, because of its unique FA composition which causes overlapping between signals, as well as the presence of FA in the form of diglycerides

(DGs) and free fatty acids (FFA) that require multiple integrations, the most informative area is the one where the olefinic carbons appear. More specifically, LO can be determined from carbon C9 at δ 129.25, OL from C9 at δ 128.73 (sn-1,3) and δ 128.70

(sn-2) and LN from carbon C15 at δ 126.23 or carbon C16 at δ 131.89 (Zamora et al.

2001).

Diterpenes, sterols and caffeine

NMR offers an efficient, alternative approach for the analysis of Cafestol (C) and

Kahweol (K) in coffee oil which is less time consuming than traditional chromatographic separation. The two diterpenes are structurally very similar which can cause challenges in their separation required for chromatographic analysis (Carrera et al. 1998). The signal which is more convenient for the determination of Cafestol appears at δ 7.22 and belongs to proton H19, whereas Kahweol can be easily determined by its signal at δ 7.27, attributed to proton H19. The corresponding C19 carbons of Cafestol at δ 139.57 and

Kahweol at δ 140.20 can be easily determined by the HSQC-DEPT spectrum Figure 10 and may also be used for the determination of these compounds. Kahweol has an extra

AB system with two olefinic protons H2 and H1, which appear as doublets at δ 6.21 and

δ 5.89 with a J3 of 9.86 Hz and can also be used for its quantification.

Figure 10. 850 MHz 1H–13C HSQC-DEPT spectrum of roasted coffee oil in CDCl3:DMSO-d6 solution, showing the one bond connectivity between protons and carbons. Cafestol and Kawheol may also appear in the form of esters of fatty acids. The signals at δ 172.5 and δ 172.46 may be attributed to the carboxyl groups of these compounds, because they don’t show any correlation peak in the HMBC spectra with glyceridic protons, however, further evidence is required to support this hypothesis.

The main sterols that appear in coffee are β-Sitosterol, Stigmatasterol and

Campesterol (Guercia et al. 2016), which belong to the group of 4-desmethylsterols. 4- desmethylsterol consumption has been shown to exhibit antitumor, antidiabetic and antifungal properties(Abdul et al. 2016), as well as cause decrease the blood cholesterol levels(Ostlund et al. 2002; de Jong et al. 2003). These three sterols represent more than

90% of the sterolic fraction of coffee oil and can be reliably determined by 1H NMR

(Hatzakis et al. 2010b). β-Sitosterol and Stigmatasterol both resonate at δ 0.68 because of the close isochronicity of their methyl groups at position 18, while the methyl protons at

C18 of Campesterol have a signal at δ 0.7 which can be used for its quantification. In instruments operating at Larmor frequencies higher than 700 MHz, β-Sitosterol and

Stigmatasterol have distinct signals at δ 0.679 and δ 0.682, which allow for their individual determination. The signal at δ 0.54 is accredited to the C18 methyl group of

Δ7-stigmastenol and Δ7-avenasterol which are known to appear in coffee oil and resonate in that chemical shift (Wilson et al. 1996; Speer and Kölling-Speer 2006a). 13C NMR can also be used for the analysis of β-Sitosterol, Stigmatasterol and Campesterol from the signals of C18 at δ 10.95, δ 11.08 and δ 11.14, respectively, however the analysis requires an additional number of scans and thus increased experimental time.

Caffeine in coffee oil can be easily quantified using the 1H NMR signals of the aromatic proton H2 at δ 7.66 and the methyl protons H10, H14 and H12, which all appear as singlets at δ 3.98 3.55 and δ 3.36, respectively. Quantifications can also be performed using the signals of carbons C2, C10 and C12 that resonate at δ 141.09, 32.61, 26.84, respectively, as determined by the HSQC-DEPT experiment (Figure 10). The carbon

C14 at δ 28.71 cannot be used for quantification purposes because of its overlapping with

C4 carbons of fatty acids. NMR can be used as a fast screening tool for the determination of caffeine in coffee oil, which is very important considering the variety of different applications of this material in food, as well as in the pharmaceutical and cosmetic industries.

Oxidation-hydrolysis products and phospholipids

Several oxidation and hydrolysis products can be determined in coffee oil using

NMR spectroscopy. The lipid fraction of both green and roasted coffee beans contains primary and secondary oxidation products, such as hydroperoxides and aldehydes (Bosco et al. 1999) as a result of the oxidation of unsaturated fatty acids. These compounds play an important role in the sensory perception of a final coffee beverage and can potentially be used as indicators of the oxidative changes that occur in coffee during roasting or storage.

The main secondary oxidation product in coffee oil has a signal at δ 10.12 (dd, J =

7.01 Hz), which is characteristic for aldehydic protons. The application of a window function for resolution enhancement reveals a four-bond coupling of 1.65 Hz. This proton has a cross peak in the HSQC-DEPT spectrum with a methine carbon at δ 191.83 and belongs to the same spin system with the olefinic protons at δ 5.94 (bd, J=10.10 Hz) as found by the HSQC-TOCSY spectrum. In addition, there are correlation peaks in HSQC-

TOCSY between the protons at δ 5.94 and δ 10.12 with the protons at δ 6.44 (1H, ddd,

J=15.19 Hz, 10.67 Hz, 3.75 Hz) and δ 5.64 (1H, dt, J=15.19 Hz, J=7.01). Two allylic protons at δ 2.11 and three methylic protons at δ 0.93 were also found to belong to the same molecule, however we were not able to identify the corresponding carbons because they overlap with those of fatty acids. The multiplicity pattern of the signals indicates the presence of an α,β-unsaturated aldehyde. A close inspection of spectral properties, such as chemical shifts, J coupling values, and relative integral ratios led us to attribute these assignments to 2,4-hepta-dienal.

The other main signal in the aldehyde region appears as a broad singlet at δ 9.68.

This can be assigned to a quaternary aldehyde due to its multiplicity and the absence of any correlation peak in the HSQC-TOCSY spectrum. Utilization of signal integrals and coupling constants helped us to associate the singlet at δ 9.68 with the signals at δ 7.16 and δ 6.70, belonging to two aromatic protons that form an AB system and have a cross peak in the TOCSY spectrum, and the aliphatic protons at δ 4.55. This chemical assignment is attributed to 5-hydroxymethyl-2-furaldehyde, which has been found in coffee previously (Bosco et al. 1999). These compounds may have an impact on the taste and aroma of the coffee beverage, however, they also have known genotoxicity and cytotoxicity (Esterbauer et al. 1991; Zarkovic 2003; Guillén and Goicoechea 2008)

In addition to the main secondary oxidation products mentioned above, the NMR spectrum of coffee oil indicates the presence of several minor aldehydes, saturated and unsaturated, as well as the presence of primary oxidation products such as peroxides, which are formed during the autoxidation of unsaturated fatty acids in coffee oil. In addition to fatty acids, other compounds that appear in significant quantities in coffee oil and contain double bonds in their structures, such Cafestol and Kawheol, are also expected to be subject of autoxidation and display signals in this area. The signals of

OOH protons of hydroperoxides usually appear between δ 8.9–8.3 as broad signals

(Guillen and Goicoechea 2009; Skiera et al. 2012) when CDCl3 is used as a solvent. The use of CDCl3:DMSO-d6 (5:1) significantly improves this resolution as OOH protons appear as sharp singlets between δ 12.5–10.5. The integration of this area of the spectrum

88 can be used for the fast determination of the peroxide value of coffee oil, as has been applied previously for the analysis of olive oil (Skiera et al. 2012).

The main hydrolysis products in coffee oil are 1,3-Diglycerides (1,3-DG), 1,2-

Diglycerides (1,2-DG) and FFA. Some small amounts of monoglycerides (MG) may also appear in some samples. The band selective constant time HMBC spectrum optimized for three-bond C–H coupling, which provides increased spectral resolution in the indirect dimension (F1), compared to the regular HMBC, revealed the chemical shift assignments of the carbonyl carbon signals of fatty acids in the form of 1,3-DG and the signals Hα protons of FFA, as shown in Figure 10. The total amount of FFA can be determined by the triplet at δ 2.24 (J=6.44 Hz) in the 1H NMR spectrum, which belongs to the CHα protons of FFA and has a correlation peak in the band-selective constant time HMBC spectrum. The carboxylic carbons at δ 174.77 and δ 174.74 can also be used for FFA determination. Because of the ration between the integrals of the two signals, the signal at

δ 174.77 may be attributed to SFA/OL FFA chains and the signals at δ 174.74 may belongs to LO/LN FFA chains.

Figure 11. 850 MHz band selective constant time 1H–13C HMBC spectrum of roasted coffee oil in CDCl3:DMSO-d6 solution, showing the long range connectivity between the carbonyl carbons and the glycerol backbone protons and the CHα, CHβ protons.

1,3-DGs can be determined by the distinct signals of 1’b, 3’b-CH2OCO protons in

1 the H NMR spectrum at δ 4.21 (dd, J=11.30 Hz). 1’a, 3’a-CH2OCO at δ 4.27 and 2’-

CHOH at δ 4.13, cannot be used for its determination due to the overlapping with the glyceridic protons of TGs. Carbons C1’ and C3’ of 1,3-DGs give one signal at δ 67.40 which can be used for quantitative purposes. However, the signal of C2’ of 1,3-DGs is not appropriate for quantitative measurements because it appears at δ 67.94 and overlaps with the C2’ signal of TG, as revealed by the HSQC-DEPT spectrum. 1,2-DGs appear only in small amounts in coffee and can be determined by their signals at δ 5.07 (2’a-

CHOCO, m), and δ 3.65 (3’a, 3’b-CH2OCO; br). The signals at δ 4.34 (1’b-CH2OCO; dd,

J=11.97 Hz) and δ 4.14 (1’α-CH2OCO; dd, J=11.97 Hz) cannot be used for quantification purposes because of overlapping. The carbons C1’, C2’ and C3’ resonate at δ 61.68, δ

71.16, and δ 59.53 and can all be used for quantification.

Phopsholipids have been found in coffee and the current methodology for their determination is based on Folch method followed by HPLC analysis (Zhou et al. 2013).

Because of the small amounts of phospholipids that appear in coffee, the 1H and 13C

NMR signals of phospholipids usually overlap with the signals of the major lipids in coffee and thus cannot be used for their determination. 31P NMR is a great alternative for their analysis, however, PLs form aggregates and also electrostatic complexes with paramagnetic ions, and this has a negative impact in the resolution and sensitivity of 31P

NMR spectra. The first problem can be confronted using a solvent mixture of

+ CDCl3/CD3OD/D2O-EDTA-K (400:95:5 v/v/v) and the latter by washing the phospholipids solution with EDTA-cation salts. Figure 12 shows the 1H-31P-HMBC spectrum of phospholipids extracted from coffee oil. The chemical shifts correspond to

Phosphatidylcholine (PC), Lyso- Phosphatidylcholine (LPC) and Phosphatidylinositol

(PI), however, unknown signals still remain and require further investigation.

Figure 12. 850 MHz 1H-31P HMBC spectrum of phospholipids isolated from roasted coffee + grounds, in CDCl3/CD3OD/D2O-EDTA-K (400:95:5 v/v/v). Quantitative NMR-PULCON method

1H and 13C NMR has been used previously for the quantification of lipids

(Alexandri et al., 2017; Williamson & Hatzakis 2017; Dais et al. 2015). For both nuclei, accurate quantification involves the use of a 90° pulse and a delay pulse ≥ 5×T1 to ensure the complete recovery of the net magnetization. Slow relaxation when utilizing carbon nuclei for the analysis is a large issue, as it increases the delay required to ensure full recovery of magnetization between scans. We reduced the T1 relaxation times of the mixture compounds using a relaxation agent, chromium (III) acetylacetenoate

1 13 (Cr(acac)3). This rendered the analysis rapid for both H and C, and allowed the use of an integrated targeted-untargeted metabolomics approach using both nuclei in one sample preparation. The optimum Cr(acac)3 conditions for coffee lipids analysis were examined and a concentration of 1.4 mM of Cr(acac)3 was found to be ideal for obtaining fast quantitative results while preserving sufficient spectral resolution. Higher concentrations 92

13 may be used for C analysis, which was found to be surprisingly tolerant to Cr(acac)3

1 addition. However, high concentrations of Cr(acac)3 have a negative impact on the H

NMR spectrum, which contains important information, such as signal multiplicities, shown in Figure 13. Although these conditions can be applied for the quantification of lipids from other sources, special care is required when applied to compounds with different chemical structures as these molecules may be characterized by different T1 relaxation times. In addition, the role of the matrix and existence of other paramagnetic cations in the solution should be taken into account.

Figure 13. Comparison of the 1H NMR spectra of a coffee oil sample with different concentrations of Cr(acac)3. 0 mM (A), 0.6 mM (B), 1.2 mM (C), 2.4 mM (D), 4.8 mM. Carbon NMR analysis requires the elimination of NOE contributions. To do this, inverse gated decoupling with broadband proton decoupling applied only during the acquisition period is usually the pulse sequence used. In modern instruments, especially 93 the ones that operate in high Larmor frequencies and are equipped with cryoprobes, carbon NMR analysis suffers from baseline and phasing issues. For that reason, we performed 13C NMR experiments using the z-restored spin-echo pulse sequence which produces spectra with smoother baselines compared to standard 13C {1H} experiments

(Xia et al. 2008). This z-stored spin-echo sequence minimizes first-order phase errors and baseline humps that occur as a result of the high Q-factor of modern NMR probes, especially cryoprobes. To our knowledge, this is the first application of this sequence in both lipid analysis and quantitative analysis in general. Here, we used an inverse gated decoupling version of the z-restored spin-echo pulse sequence, which is highly recommended for coffee lipid analysis, and lipid analysis in general, when using instruments operating in Larmor frequencies higher than 600 MHz. In addition, the use of adiabatic pulses allows the uniform excitation of all frequencies, even in cases of wide bandwidths which are becoming more common as high field instruments are more easily available. Offset effects can cause significant errors in analysis and create limitations in signal integrations, as only signals that appear in similar spectral regions can be compared (D’Amelio et al. 2013). The use of adiabatic pulses, one for inversion

(Crp80,0,5,20,1) and one for refocusing (Crp80comp.4), ensures the elimination of any off-resonance issue and the equal excitation of all frequencies.

Table 3 summarizes the compounds that can be determined quantitatively by 1H and 13C NMR from a specific diagnostic signal using BHT as an internal standard. 1H and

13C NMR were found to be in a good agreement, confirming the quantitative nature of the z-stored pulse sequence, although a systematic deviation was found between 1H and 13C

NMR for the calculation of Ln, which can be attributed to the low S/N in the carbon spectrum.

Table 3. Diagnostic signals in the 1H and 13C NMR spectra that can be used for the quantitative determination of various components in coffee oil.

Diagnostic Number of Diagnostic Number of signal 1H NMR proton atoms signal 13C carbons atoms Compound (δ); contribute to NMR (δ); contribute to signal notation the signal signal notation the signal LN 2.79; (A) 4 126.23; (A) 1 LO 2.04; (B) 4 129.25; (B) 1 OL 2.00; (C) 4 128.83; (C) 1 SFA 0.88; (D) 3 13.28; (D) 1 Cafestol 7.22; (E) 1 139.57; (E) 1 Kawheol 7.27; (F) 1 140.20 (F) 1 Caffeine 3.98; (G) 3 32.60; (F) 1 Campesterol 0.70; (H) 3 - - β-Sitosterol 0.679; (I) 3 - - Stigmatasterol 0.682; (J) 3 - - Δ7-stigmastenol 0.54; (K) 3 - - and Δ7-avenasterol Hydroperoxides 12.5–10.5 (L) 1 - - 2,4-hepta-dienal 10.12 (M) 1 191.83; (K) 1 5-hydroxymethyl- 9.68 (N) 1 122.08 (N) 1 2-furaldehyde 1,3-DGs 4.21 (O) 2 67.40 (O) 2 1,2-DGs 5.07 (P) 1 61.68 (P) 1 174.77-174.74 FFA 2.24 (Q) 2 1 (Q)

Next, we performed a systematic comparative study to examine the agreement between PULCON and the regular internal standard approach, using the Bland-Altman methodology. To our knowledge, this is the first application of PULCON for lipid quantification and the first attempt to systematically investigate the efficiency of

PULCON in small molecules using a robust statistical approach. The two most critical experimental parameters that need to be considered for minimizing the quantification errors are the pulse length and the receiver gain value. Quantification with PULCON seems to be sensitive to pulse miscalibration, and even small variations (eg 1 us) in the 95 pulse length can cause significant deviation in the calculated concentration of the analyte.

This is not surprising, as the concentration to be determined has a direct dependence on the pulse length used for the acquisition of the reference and the unknown spectra as shown in Equation 1 (Figure 21). For that reason, the probe should be tuned and matched for all samples as the efficiency of the delivery of the RF power to the sample for a given pulse length value depends on tuning and matching. The signal area also has a dependence on the receiver gain (RG) value, although our results indicated that the integral areas of a given sample was constant for a relatively wide range of RG values.

To validate the reliability of the PULCON method we quantified several components of the lipid fraction of green coffee, namely OL, LO, LN, SFA, K, C,

Sterols, 1,3-DG, 1,2-DG, FFA, using the traditional internal standard method as well as

PULCON. Representative quantitative data for 1H NMR and 13C NMR analysis are shown in

Table 4.

Table 4. LO, 1,3-DG, Cafestol, Campesterol, Cafestol and Kaheol (μmoles/g) determined by NMR Spectroscopy in green coffee using IS and PULCON methods

Sample LO 1,3-Diglyceride Campesterol Cafestol Kahweol BHT PULCON BHT PULCON BHT PULCON BHT PULCON BHT PULCON 1 62.25 62.41 8.39 8.49 0.43 0.43 5.90 5.96 11.93 14.66 2 96.20 97.80 11.88 12.07 0.64 0.65 8.76 8.89 11.82 18.67 3 126.50 126.91 13.03 13.04 0.68 0.68 11.90 11.90 12.65 20.15 4 77.54 83.40 11.58 13.33 0.43 0.47 9.78 9.14 9.68 17.64 5 109.92 105.88 13.97 13.48 0.73 0.70 10.94 10.57 11.30 18.77 6 60.97 63.36 11.69 12.23 0.55 0.58 6.44 6.75 15.27 25.50 7 90.81 91.89 10.60 10.75 0.73 0.73 11.36 11.51 11.78 19.64 8 92.93 89.94 11.79 11.41 0.75 0.73 11.32 10.96 9.25 15.64 9 86.06 84.93 9.59 9.41 0.51 0.50 9.80 9.65 7.23 11.17 10 99.16 99.55 11.86 11.90 0.81 0.80 6.96 6.99 14.62 23.01 11 74.84 76.15 8.95 9.10 0.33 0.33 6.98 7.10 5.02 8.64 12 82.01 80.32 8.79 8.60 0.44 0.43 6.60 6.46 8.39 13.92 13 97.84 95.69 11.77 11.51 0.57 0.56 10.46 10.24 11.52 17.80 14 74.84 75.33 10.53 10.60 0.63 0.62 12.34 12.43 7.81 12.90 15 62.28 67.82 10.01 9.94 0.48 0.48 8.20 8.14 6.55 11.22 16 73.69 71.02 9.64 9.27 0.41 0.40 9.70 9.32 7.81 12.35 17 72.94 72.05 9.73 9.61 0.55 0.55 9.40 9.29 8.02 12.81 18 74.05 75.17 9.10 9.58 0.45 0.46 8.36 8.50 6.11 10.46 Mean 84.16 84.42 10.72 10.80 0.56 0.56 9.18 9.10 9.82 15.83 Std. 16.99 16.12 1.52 1.59 0.13 0.13 1.96 1.89 2.85 4.48

Based on the data in these tables, we generated linear regression equations of the type:

(Analyte, PULCON)= a+b (Analyte, IS)

to evaluate the correlation between the data obtained with PULCON and those obtained

with IS. Table 5 contains the linear regression data of the dependent variable (PULCON)

97 with the independent variable (IS). All of the linear regression models had R2 values above 0.9, showing that PULCON and the IS method are well correlated. Because linear regression only correlates the two methods, but does not assess the existence of any agreement between them, we performed a Bland-Altman analysis to compare PULCON with IS. Bland-Altman is an alternative approach to linear regression that is used for evaluating the agreement between two methods and allows the assessment of how the measurements differ systematically from zero (bias) and how much the difference varies

(error).

Table 5. Linear regression data of PULCON values with internal standard values.

Compound R2 value Equation Linoleic 0.9788 0.9388x+5.4166 1,3-Diglycerides 0.9064 0.9976x+0.1061 Campesterol 0.9874 0.9586x+0.0229 Cafestol 0.9578 0.9578x+0.3103

The Bland-Altman plots for LO, 1,3-DG, Campesterol, and Cafestol are shown in

Figure 14. Prior to Bland-Altman analysis, a two-tailed t test was performed to the difference of the measurements obtained by the two methods, PULCON and internal standard, to ensure that they are not significantly different from 0. Additionally, a 2-way

ANOVA was performed to confirm that the differences observed among sample values was not a result of the measurement or a result of coffee growing region. One sample’s measurement consistently produced an outlier in each bland altman analysis. This systematic error is likely due to a slight difference in the samples matrix, causing the sample to have a different p1 value and incorrect quantitation to be achieved. In all cases

98 there was no significant difference between the measurements and Bland-Altman analysis was conducted. As seen in Figure 14, nearly all measurements are located inside the limits of agreement and there is no strong relationship between the difference and the average. In addition, there is no systematic error because the 95% confidence interval

(mean2SD) includes zero. Additionally, the mean differences of the two methods only slightly differ from zero, with all points equally distributed below and above the mean difference line as linear regression analysis of the Bland-Altman data showed. These data indicate that the two quantification approaches are in a good agreement and thus

PULCON can be used for the quantification of lipids in coffee and lipids in general. A similar analysis for 13C NMR showed that PULCON is not limited to 1H NMR analysis but can be extended to other nuclei. It is important to note that PULCON showed reliability and high accuracy for the analysis of various classes of compounds that have different chemical structures and appear in a wide range of concentrations in the sample.

PULCON can be applied to 1H NMR, as well as to 13C NMR, regardless of the carbon multiplicity (eg methyl, methylene, methine). This is important as it shows the general applicability of the PULCON method.

A B

C D

Figure 14. Difference (bias) plots obtained from Bland Altman analysis of green coffee oil samples analyzed by IS and PULCON methods for (A) LO, (B) 1,3-DG, (C) Campesterol, and (D) Cafestol.

Effect of roasting on coffee lipids composition

Roasting can cause compositional changes in coffee lipids. It is expected that

these changes will result in the disappearance of certain functional groups and the

appearance of new ones. Further, it is expected that these changes are detectable via

NMR spectroscopy. Targeted (IS or PULCON) and untargeted NMR metabolomics

analysis can be applied to investigate the impact of roasting on coffee lipids. Although 100 there are spectral similarities between green and roasted coffee beans, a closer inspection of the spectra reveals remarkable qualitative and quantitative differences. Figure 15 shows the expanded spectral regions from δ 12.5-8 and δ 8-5.6 for the coffee lipids of a green bean and the lipids extracted from the same sample after roasting. It can be observed that roasting has a significant impact on the oxidation product profile as it causes a degradation of unsaturated hydroperoxides which are converted to aldehydes.

These aldehydes have been associated with the characteristic aroma of roasted coffee

(Calligaris et al. 2009). In general, there appear to be more diverse oxidation products in roasted coffee compared to green. When comparing the entire aldehyde region, δ 10.5-

9.0, although there was a difference between green and roasted samples, it was not significant (p= 0.091, Table 6). A 2-way ANOVA also indicated that there was no significant influence of growing region on the aldehyde regions. However, the variations of the aldehyde doublet at δ 10.12 confirm that there are differences in oxidation products between green and roasted samples. The concentration of this aldehyde at δ 10.12 was significantly different in green and roasted coffee (p= 0.037, Table 6), as found by a 2- way ANOVA. Again, it was found that there was no significant influence of growing region on the aldehyde doublet. This indicates that lipid oxidation does occur to some degree during roasting.

101

Figure 15. The effect of roasting on the oxidation status of coffee oil. Comparisons between coffee oil extracted from green beans (A) and roasted beans (B). Roasting also seems to accelerate the hydrolysis of TG in coffee oil. When comparing the quantity of FFA between green and roasted coffee, a significant difference was observed (p< 0.001, Table 6). A 2-way ANOVA also found a significant influence of coffee origin on the amount of FFA. This finding is in agreement with a few studies that found increases in FFA after roasting coffee (Speer and Kölling-Speer 2006b; Raba et al. 2015). 1,3-DGs were found in lower concentrations in roasted coffee compared to green coffee and this difference was statistically significant. This is somewhat unexpected if one assumes that hydrolysis of TGs takes place during roasting, resulting in an increase in DGs. This finding indicates that TGs are more resistant to hydrolysis compared to DGs, which may be further hydrolized to produce FFA and MGs. This difference in reactivity of TGs and DGs explains the higher concentrations of FFA and the lower concentrations of DGs and is in agreement with previous studies (Lykidis et al.

102

1995; Hatzakis et al. 2010a). Additionally, a 2-way ANOVA found significant differences among samples based on growing region, as well. It was also found that there was a significant decrease in linolenic fatty acid with roasting, which may be attributed to oxidation, which alters the chemical structure of linolenic acid measured by 1H NMR.

This is in agreement with a previous study (Martín et al. 2001). While it appears that roasting is influencing the amount of specific lipids in a coffee bean, more research is required to understand additional influences on these fatty acids such as growing region.

Table 6. Paired sample t-test values comparing the aldehyde doublet, aldehyde region, and 1,3-DG region in 1H and FFA in 13C values in 18 green and roasted coffee samples, normalized to total intensity (ɑ 0.05)

Mean Peak Area Spectral Area Sample p-value (x103) 2,4-hepta-dienal Green 0.281±0.170 0.037 (δ 10.12 1H NMR) Roasted 0.398±0.70 Total aldehydes Green 1.15±0.530 0.091 (δ 10.5-9.0 1H NMR) Roasted 1.69±0.986 1,3-DG Green 25.1±2.57 <0.001 (δ 4.24-4.19, 1H NMR) Roasted 23.05±2.14 FFA (δ 175.5 13C Green 2.17±0.730 <0.001 NMR) Roasted 2.90±0.865 Green 0.830±0.0722 LN <0.001 Roasted 0.735±0.0680

To further understand the impact of roasting in the coffee bean metabolome, we performed PCA analysis on the normalized NMR data. The PCA plot helps to better visualize compositional similarities and dissimilarities between the samples, as well as detect outliers and abnormal samples. Although PCA is an unsupervised method and includes the total variance between samples, there is still a good separation between the green and roasted samples. The PCA plot obtained by the covariance matrix is shown in

103

Figure 16A. The first principal component explains 56.2% and the second principal component, 16.5% of the total variance. The corresponding variable loadings plot indicates that chemical compounds that appear in spectral areas where oxidation products appear act as important biochemical descriptors for group classification. A comparative

OPLS-DA was also performed for the two groups and the cross-validated scores plots showed a clear separation between them, indicating the significant differences between their global biochemical profiles. R2Y and Q2, the most important quality indicators for

OPLS-DA (Figure 16B), were 0.986 and 0.902 respectively, indicating the reliable distinction obtained by the model. The model was further validated using CV-ANOVA

(Table 7) and permutation tests (Figure 22). Compared to the green coffee samples, roasting resulted in significant alterations in the biochemical composition of coffee lipids in primary and secondary oxidation products, hydrolysis products, and fatty acid chains.

Figure 16. PCA (A) and OPLS-DA (B) scores plots of green (green DP) coffee and roasted (brown DP) coffee samples. PCA: PC1:57.0 %, PC2:16.5%.; OPLS-DA: Q2: 0.902, R2Y: 0.986, CV ANOVA, p=3.64 x 10-8.

104

Physical transformation in the coffee bean due to roasting

A B

Figure 17. Proton density image of a green (A) and roasted (B) Arabica coffee bean. Both targeted and untargeted analysis revealed significant differences in the composition and structure of several major coffee lipids and other minor compounds during roasting. In addition to reasons described in previous sections, another reason is because roasting causes important structural changes, as indicated by MRI. A rapid acquisition with relaxation enhancement RARE sequence was used to acquire proton density (PD) images of green and roasted Arabica coffee beans. PD MRI with the same acquisition parameters allowed for the comparison of PD distribution in green (Figure

17A) and roasted (dry) (Figure 17B) coffee beans. In Figure 17, the mobility of protons is visible in grayscale with the brighter pixels indicating the presence of highly mobile protons, and the darker areas representing either a decrease in mobility or absence of protons (void). It appears that the distribution of mobile protons in the green bean image

105

(Figure 17A) is denser and more homogeneous compared to the distribution in the roasted bean (Figure 17B). The moisture content of the green bean was measured using a vacuum oven to be 7.0% moisture, compared to the lower 1.9% moisture in its roasted counterpart. With this knowledge, we hypothesize that the protons in Figure 17A represent water and lipids, while the protons in Figure 17B represent mostly mobile lipid protons. As expected, Image Figure 17A appears brighter than Figure 17B because the roasting process leads to 5.1% loss of moisture.

It is known that lipids exist mostly in the endosperm of a green coffee bean and that during bean expansion caused by roasting, the oleosomes are disrupted, and lipids migrate to the surface of the bean (Schenker and Rothgeb 2017). This increase in bean volume or expansion is visually apparent while comparing the small bean size in Figure

17A to the larger size of the roasted bean in Figure 17B. Therefore, it can be concluded that the bright edge around the bean in Figure 17B represents the mobile lipids that migrated from the endosperm to the surface. This representative result is in agreement with the results obtained using cryo-scanning electron microscopy micrographs

(Schenker et al. 2000). Furthermore, due to the extremely short T2 of the coffee bean, and the acquisition parameters used to facilitate comparison of the beans pre- and post- roasting, the lipids and water protons in the green bean were not distinguishable. Similar to our work, this work presents a proton density image of a coffee bean of 35% moisture

(Schmidt et al. 1996) which, due to its moisture content, might be a green bean. To our knowledge, we present the first MR image of a roasted coffee bean which may be a promising tool to investigate oil distribution during roasting of coffee.

106

Coffee lipids following brewing

Lipids in coffee beverage: Small amounts of lipids can be found in the water extracts of coffee, however, they form aggregates with short T2 relaxation times and thus they appear as a broad peak at δ 0.92, which belongs to the terminal methyl protons of fatty acids (Figure 23). For that reason, they cannot be studied without extraction from the beverage using a non-polar solvent, such as chloroform. The dominant NMR signals in the chloroform extraction of a coffee beverage belong to caffeine. However, this extract is also rich in various lipid or lipid-related components, as shown in Figure 18.

The spectral area from δ 13.0-10.0 (Figure 18D) is rich in enols, lipid peroxides and carboxylic acids, whereas several aldehydes appear in the δ 10.0-9.0 region (Figure

18C). The spectra contain several signals attributed to aromatic compounds and oxygenated compounds with double bonds as can be seen in the area from δ 9.0 to δ 5.90

(Figure 18B). Finally, the typical signals of methylene and methyl protons of FA at δ

1.26 and δ 0.88, as well as small amounts of sterols at δ 0.68, also appear (Figure 18A).

Although this first NMR screening was able to indicate the various classes of compounds, only a small amount of information was provided about individual compounds. Further analysis using 2D NMR spectroscopy and/or other analytical methods, such as GC, may be required to obtain information for individual substances.

107

1 Figure 18. Expansions of a 700 MHz H NMR spectrum in CDCl3 showing the regions of methyl protons of lipids (A), aromatic compounds and oxygenated compounds with double bonds (B), aldehydes (C), peroxides/ enols/ FFA (D). Epoxidation of coffee waste lipids: Lipids, such as triglycerides and fatty acids have been previously used for the production of biopolymers (Montero de Espinosa and

Meier 2011; Plaza et al. 2017). One possible pathway for creating bioplastics from raw plant material lipids is epoxidation for the creation of epoxy resins. TGs are in general neutral molecules with low reactivity and thus the introduction of an easily polymerizable and more reactive functional group, such as an alcohol or an epoxide, is essential for increasing their applicability. These epoxy resins consist of at least two epoxide groups, and through a series of chemical reactions, polymers with strong mechanical properties and high temperature and chemical resistance can be produced. After the production of a coffee beverage in which coffee grounds are extracted using water, a polar solvent, a large amount of the coffee lipids remain in the spent coffee grounds (SCG). This coffee 108 waste was found to be a rich source of triglycerides with an excellent quality a relatively large unsaturated profile. Therefore, SCG utilization for the production of resins and biopolymers is an attractive approach for achieving a low cost, renewable application.

Because the oil from spent coffee grounds contains significant amounts of oxidation products that may affect the reaction and the stability of the products, it was initially extracted using supercritical CO2, which is a very non-polar extraction medium and thus will not extract semi-polar compounds such as lipid peroxides and aldehydes.

This technique also allows the avoidance of organic solvents and is an effective and green replacement to Sohxlet extraction. The quality and chemical composition of the oil extracted by spent coffee grounds were monitored using 1H NMR spectroscopy and it was similar to the coffee oil extracted by roasted beans. Coffee lipids displayed a remarkable stability to hydrolysis during storage. This is probably because spent coffee grounds are produced by roasted coffee where the hydrolytic enzymes have been deactivated during the heat treatment. The extracted oil was then epoxidized using H2O2 and formic acid and the reaction was monitored by NMR. In the epoxidation reaction,

H2O2 initially reacts with formic acid to form peroxy-formic acid and water. In a next step, an epoxy group is added on the double bonds of the unsaturated fatty acid chains.

Coffee oil is free of epoxides before the reaction, whereas two possible isomers are possible after reaction initiation, as shown if Figure 19C. This is supported by the fact that four distinct signals of the epoxidized oil are observed at δ 3.30-2.90. A good indication for the yield and the status of the reaction are the signals intensities of olefinic protons at δ 5.42-5.29, which are reduced as the reaction takes place. Small amounts of

109 monoepoxides are still in the mixture even after the completion of the reaction, as found by the signals at δ 2.92 and the olefinic signals at δ 5.57-5.39, which belong to olefinic protons located next to an epoxide group. Signals that were observed between δ 8.5-7.9 belong to peroxides and are considered side-products of the reaction, however they were only formed in very minor concentrations and they don’t affect the overall yield of the process. It is also worthy to note that the chemical shift of the methyl groups of LO and

OL are not affected by the epoxidation, but the methyl group of LN is affected because there is a double bond in n-3 position which is epoxidized. The reaction scheme and reaction monitoring using NMR is shown in Figure 19. Quantitative determination of epoxides can be easily performed using the PULCON method, described in a previous section.

110

Figure 19. Epoxidation process as monitored through 1H NMR spectroscopy: pure coffee oil (A), partially epoxidized coffee oil (B), and fully epoxidized coffee oil (C).

111

Supplemental Material

Figure 20. Resolution enhancement of a 1H spectrum by applying a window function and vurce fitting.

Figure 21. PULCON equation for quantification

Figure 22. Permutation tests for OPLS-DA.

112

Figure 23. Lipids as appear in the 1H NMR spectrum of a water coffee extract Table 7. CV-Anova for OPLS-DA analysis

M7 SS DF MS F p SD Total corr. 35 35 1 1 Regression 31.5622 13 2.42786 15.5368 3.64464e-008 1.55816 Residual 3.43783 22 0.156265 0.395304

113

Chapter 7. Conclusions

Overall, it can be concluded that NMR spectroscopy is an effective tool for the analysis and evaluation of lipids in fish oil supplements and coffee oil. The 1H NMR experiment provides an abundance of information and is a rapid technique for the

1 determination of many compounds in fish oil, due to the sensitivity and the short T1 of H nuclei. These compounds include DHA, EPA, as well as total amount of n-3, n-6, and n-9 fatty acids. However, the small spectral width of the 1H NMR spectra (~15ppm), as well as the presence of scalar coupling, causes a loss in resolution and makes the determination of minor compounds challenging. 13C NMR can be more informative, despite longer experimental times due to longer T1 and the lower sensitivity of carbon nuclei. More individual fatty acids can be identified in a 13C NMR spectrum, and their positional distribution on the glycerol skeleton can also be determined. While there is a low natural abundance of 13C nuclei, the protocol here uses enough sample to provide a large amount of information from 13C nuclei. Following the published protocol will allow for the qualitative and quantitative analysis of several components in fish oil supplements. 1H experiments in conjunction with 13C experiments can provide the most complete information about the lipid profile of the sample.

Coffee oil was found to be rich in linoleic and saturated fatty acids, as well as oleic and linolenic fatty acids. Additionally, minor compounds such as sterols and diterpenes, as well as oxidation and hydrolysis products, can be identified and quantified in the 1H and 13C spectrum of coffee oil. The 2D 1H-1H TOCSY spectrum confirmed the

114 assignment of the proton spectrum and the selective 1H-13C HMBC in the carbonyl carbon region allowed for assignment of protons and carbons on the glycerol skeleton and fatty acid chains. The PULCON method for quantitation was in good agreement with the traditional internal standard (IS) quantification method and can be applied in future studies to prevent sample contamination and other issues related to the addition of an IS in the NMR sample. NMR spectroscopy coupled with univariate and multivariate statistical analysis was able to detect significant differences in the amount of aldehydes,

1,3-DG, and FFA, as well as linolenic acid following roasting. Additionally, MRI was proven as an effective tool for tracking lipid migration in a coffee bean following roasting. Finally, spent coffee grounds have the potential to be a promising source for bioplastics. This is because a majority of lipids remain in the coffee grounds following beverage extraction with water. The lipid fraction of coffee is highly unsaturated, making it a good source for the production of epoxides which are known precursor for bioplastics.

115

Bibliography

Abdul QA, Choi RJ, Jung HA, Choi JS (2016) Health benefit of fucosterol from marine algae: a review. J Sci Food Agric 96:1856–1866. doi: 10.1002/jsfa.7489 Ackermann SM, Dolsophon K, Monakhova YB, et al (2017) Automated Multicomponent Analysis of Soft Drinks Using 1D 1H and 2D 1H-1H J-resolved NMR Spectroscopy. Food Anal Methods 10:827–836. doi: 10.1007/s12161-016-0643-y Akoka S, Barantin L, Trierweiler M (1999) Concentration Measurement by Proton NMR Using the ERETIC Method. doi: 10.1021/AC981422I Aleriana T, Wagemaker L, Regina C, et al (2010) Sun protection factor, content and composition of lipid fraction of green coffee beans. Ind Crop Prod 33:469–473. doi: 10.1016/j.indcrop.2010.10.026 Alexandri E, Ahmed R, Siddiqui H, et al (2017) High Resolution NMR Spectroscopy as a Structural and Analytical Tool for Unsaturated Lipids in Solution. Molecules 22:1663. doi: 10.3390/molecules22101663 Anesei M, De Pillp T, Massinf R, Lericp CR (2000) OXIDATIVE STABILITY OF THE LIPID FRACTION IN ROASTED COFFEE STABILITÀ DELLA FRAZIONE LIPIDICA IN CAFFÈ TOSTATO. J Food Sci n 4:12–2000 Bharti SK, Roy R (2012) Quantitative 1H NMR spectroscopy. TrAC Trends Anal Chem 35:5–26. doi: 10.1016/J.TRAC.2012.02.007 Billault I, Richard Robins, Akoka S (2002) Determination of Deuterium Isotope Ratios by Quantitative 2H NMR Spectroscopy: the ERETIC Method As a Generic Reference Signal. doi: 10.1021/AC025867P Bosco M, Toffanin R, Palo D De, et al (1999) High-resolution1H NMR investigation of coffee. J Sci Food Agric 79:869–878. doi: 10.1002/(SICI)1097- 0010(19990501)79:6<869::AID-JSFA302>3.0.CO;2-6 Bradbury AGW (2001) Coffee: recent developments. Blackwell Science Ltd Budryn G, Nebesny E, Żyżelewicz D, et al (2012) Influence of roasting conditions on fatty acids and oxidative changes of Robusta coffee oil. Eur J Lipid Sci Technol 114:1052–1061. doi: 10.1002/ejlt.201100324 Caetano NS, Silva VFM, Melo AC, et al (2014) Spent coffee grounds for biodiesel production and other applications. Clean Technol Environ Policy 16:1423–1430. doi: 10.1007/s10098-014-0773-0 Calligaris S, Munari M, Arrighetti G, Barba L (2009) Insights into the physicochemical properties of coffee oil. Eur J Lipid Sci Technol 111:1270–1277. doi: 10.1002/ejlt.200900042 Carrera F, Leo Ân-Camacho M, Pablos F, Gonza Âlez AG (1998) Authentication of 116

green coffee varieties according to their sterolic profile. Anal Chim Acta 370:131– 139 Cavin C, Bezencon C, Guignard G, Schilter B (2003) Coffee diterpenes prevent benzo[a]pyrene genotoxicity in rat and human culture systems. Biochem Biophys Res Commun 306:488–495. doi: 10.1016/S0006-291X(03)00994-X Cavin C, Holzhaeuser D, Scharf G, et al (2002) Cafestol and kahweol, two coffee specific diterpenes with anticarcinogenic activity. Food Chem Toxicol 40:1155–1163. doi: 10.1016/S0278-6915(02)00029-7 Cavin C, Mace K, Offord E, Scilter B (2001) Protective effects of coffee diterpenes against aflatoxin B1-induced genotoxicity: mechanisms in rat and human cells. Food Chem Toxicol 39:549–556 Ciampa A, Renzi G, Taglienti A, et al (2010) Studies on coffee roasting process by means of nuclear magnetic resonance spectroscopy. J Food Qual 33:199–211. doi: 10.1111/j.1745-4557.2010.00306.x Crisafulli P, Navarini L, Silizio F, et al Ultrastructural Characterization of Oil Bodies in Different Coffea Species TBO Toluidine blue O staining solution ATcell Total area of cell Acell Area of content cell Aob Area of oil bodies Acw Area of cell wall. doi: 10.1007/s12042-013-9132-2 D’Amelio N, De Angelis E, Navarini L, et al (2013) Green coffee oil analysis by high- resolution nuclear magnetic resonance spectroscopy. Talanta 110:118–127. doi: 10.1016/j.talanta.2013.02.024 Dais P, Misiak M, Hatzakis E (2015) Analysis of marine dietary supplements using NMR spectroscopy. Anal Methods 7:5226–5238. doi: 10.1039/C5AY00794A Dais P, Plessel R, Williamson K, Hatzakis E (2017) Complete 1H and 13C NMR assignment and 31 P NMR determination of pentacyclic triterpenic acids. Anal Methods. doi: 10.1039/C6AY02565J de Jong A, Plat J, Mensink RP (2003) Metabolic effects of plant sterols and stanols (Review). J Nutr Biochem 14:362–369. doi: 10.1016/S0955-2863(03)00002-0 Del Carmen Velazquez Pereda M, de Campos Dieamant G, Eberlin S, et al (2009) Effect of green Coffea arabica L. seed oil on extracellular matrix components and water- channel expression in in vitro and ex vivo human skin models. J Cosmet Dermatol 8:56–62. doi: 10.1111/j.1473-2165.2009.00425.x Dwyer JT, Frances Picciano M, Betz JM, et al (2008) Progress in developing analytical and label-based dietary supplement databases at the NIH Office of Dietary Supplements. J Food Compos Anal 21:S83–S93. doi: 10.1016/j.jfca.2007.07.010 Echeverria-Beirute F, Murray SC, Klein P, et al (2017) Rust and Thinning Management Effect on Cup Quality and Plant Performance for Two Cultivars of Coffea arabica L. J Agric Food Chem acs.jafc.7b03180. doi: 10.1021/acs.jafc.7b03180 Esterbauer H, Schaur RJ, Zollner H (1991) Chemistry and biochemistry of 4- hydroxynonenal, malonaldehyde and related aldehydes. Free Radic Biol Med 11:81– 128. doi: 10.1016/0891-5849(91)90192-6 Farah A (2012) Coffee: Emerging Health Effects and Disease Prevention, 1st edn. Blackwell Publishing Ltd. Folstar P (1985) Lipids. In: Coffee Chemistry. Springer Netherlands, Dordrecht, pp 203– 117

222 Frank O, Kreissl JK, Daschner A, Hofmann T (2014) Accurate Determination of Reference Materials and Natural Isolates by Means of Quantitative 1H NMR Spectroscopy. J Agric Food Chem 62:2506–2515. doi: 10.1021/jf405529b Frascareli EC, Silva VM, Tonon RV, Hubinger MD (2012) Effect of process conditions on the microencapsulation of coffee oil by spray drying. Food Bioprod Process 90:413–424 Ghasemi Fard S, Turchini GM, Sinclair AJ (2014) Bioavailability of fish oil vs krill oil – Influence of gender. J Nutr Intermed Metab 1:8. doi: 10.1016/j.jnim.2014.10.019 Giovannucci E (1998) Meta-analysis of coffee consumption and risk of colorectal cancer. Am J Epidemiol 147:1043–1052 González AG, Pablos F, Martín MJ, et al (2001) HPLC analysis of tocopherols and triglycerides in coffee and their use as authentication parameters. Food Chem 73:93– 101. doi: 10.1016/S0308-8146(00)00282-X Goodnight SH, Harris WS, Connor WE (1981) The Effects of Dietary w3 Fatty Acids on Platelet Composition and Function in Man: A Prospective, Controlled Study. Blood 58: Grollier J, Plessis S (1983) Use of coffee bean oil as a sun filter Guercia E, Berti F, Navarini L, et al (2016) Isolation and characterization of major diterpenes from C. canephora roasted coffee oil. Tetrahedron: Asymmetry 27:649– 656. doi: 10.1016/J.TETASY.2016.06.008 Guillen MD, Goicoechea E (2009) Oxidation of corn oil at room temperature: Primary and secondary oxidation products and determination of their concentration in the oil liquid matrix from 1H nuclear magnetic resonance data. Food Chem 116:183–192. doi: 10.1016/j.foodchem.2009.02.029 Guillén MD, Goicoechea E (2008) Toxic Oxygenated α,β-Unsaturated Aldehydes and their Study in Foods: A Review. Crit Rev Food Sci Nutr 48:119–136. doi: 10.1080/10408390601177613 Guillén MD, Ruiz A (2003) 1H nuclear magnetic resonance as a fast tool for determining the composition of acyl chains in acylglycerol mixtures. Eur J Lipid Sci Technol 105:502–507. doi: 10.1002/ejlt.200300799 Han Y, Wen Q, Chen Z, Li P (2011) Review of Methods Used for Microalgal Lipid- Content Analysis. Energy Procedia 12:944–950. doi: 10.1016/J.EGYPRO.2011.10.124 Harper CR, Jacobson TA (2005) Usefulness of Omega-3 Fatty Acids and the Prevention of Coronary Heart Disease. Am J Cardiol 96:1521–1529. doi: 10.1016/j.amjcard.2005.07.071 Hatzakis E, Agiomyrgianaki A, Dais P (2010a) Detection and Quantification of Free Glycerol in Virgin Olive Oil by 31P-NMR Spectroscopy. J Am Oil Chem Soc 87:29–34. doi: 10.1007/s11746-009-1466-4 Hatzakis E, Dagounakis G, Agiomyrgianaki A, Dais P (2010b) A facile NMR method for the quantification of total, free and esterified sterols in virgin olive oil. Food Chem 122:346–352. doi: 10.1016/j.foodchem.2010.02.043 He K (2009) Fish, Long-Chain Omega-3 Polyunsaturated Fatty Acids and Prevention of 118

Cardiovascular Disease—Eat Fish or Take Fish Oil Supplement? Prog Cardiovasc Dis 52:95–114. doi: 10.1016/J.PCAD.2009.06.003 Holscher W, Steinhart H (1995) Aroma compounds in green coffee. Food Flavours Gener Anal Process Influ 785 803 Hoult DI, Richards RE (2011) The signal-to-noise ratio of the nuclear magnetic resonance experiment. J Magn Reson 213:329–343. doi: 10.1016/j.jmr.2011.09.018 Igarashi T, Aursand M, Hirata Y, et al (2000) Nondestructive quantitative determination of docosahexaenoic acid and n−3 fatty acids in fish oils by high-resolution 1H nuclear magnetic resonance spectroscopy. J Am Oil Chem Soc 77:737–748. doi: 10.1007/s11746-000-0119-0 Illy A, Viani R (2005) Espresso Coffee: The Science of Quality, Second. San Diego Jian•hua C (2014) Near-infrared Spectrum Detection of Fish Oil DHA Content Based on Empirical Mode Decomposition and Independent Component Analysis. J Food Nutr Res 2:62–68. doi: 10.12691/jfnr-2-2-1 Kobelnilk M, Fontanari GG, Cassimiro DL, et al (2014) Thermal behavior of coffee oil ( Robusta and Arabica species). J Therm Anal Calorim 115:2045–2052 Kovacs H, Moskau D, Spraul M Cryogenically cooled probes—a leap in NMR technology. doi: 10.1016/j.pnmrs.2005.03.001 Kremer JM, Lawrence DA, Petrillo GF, et al (1995) Effects of high-dose fish oil on rheumatoid arthritis after stopping nonsteroidal antiinflammatory drugs clinical and immune correlates. Arthritis Rheum 38:1107–1114. doi: 10.1002/art.1780380813 Lee KJ, Choi JH, Jeong HG (2007) Hepatoprotective and antioxidant effects of the coffee diterpenes kahweol and cafestol on carbon tetrachloride-induced liver damage in mice. Food Chem Toxicol 45:2118–25. doi: 10.1016/j.fct.2007.05.010 Lee KJ, Jeong HG (2007) Protective effects of kahweol and cafestol against hydrogen peroxide-induced oxidative stress and DNA damage. Toxicol Lett 173:80–7. doi: 10.1016/j.toxlet.2007.06.008 Lykidis A, Mougios V, Arzoglou P (1995) Kinetics of the two-step hydrolysis of triacylglycerol by pancreatic lipases. Eur J Biochem 230:892–8 Malasanos TH, Stacpoole PW Biological Effects of co-3 Fatty Acids in Diabetes Mellitus Maressa Aparecida de Oliveira P, Henrique de Almeida R, Albertina de Oliveira N, et al (2014) Enrichment of diterpenes in green coffee oil using supercritical fluid extraction – Characterization and comparison with green coffee oil from pressing. J Supercrit Fluids 95:137–145. doi: 10.1016/j.supflu.2014.08.016 Martín MJ, Pablos F, González AG, et al (2001) Fatty acid profiles as discriminant parameters for coffee varieties differentiation. Talanta 54:291–297. doi: 10.1016/S0039-9140(00)00647-0 Millen A, Dodd K, Subar A (2004) Use of vitamin, mineral, nonvitamin, and nonmineral supplements in the United States: The 1987, 1992, and 2000 National Health Interview Survey results. American Dietetic Association Monakhova Y, Kuballa T, Lobell-Behrends S, et al (2013) Standardless 1H NMR determination of pharmacologically active substances in dietary supplements and medicines that have been illegally traded over the internet. John Wiley & Sons Monakhova YB, Lachenmeier DW, Kuballa T, Mushtakova SP (2015a) Standardless 119

multicomponent qNMR analysis of compounds with overlapped resonances based on the combination of ICA and PULCON. Magn Reson Chem 53:821–828. doi: 10.1002/mrc.4275 Monakhova YB, Ruge W, Kuballa T, et al (2015b) Rapid approach to identify the presence of Arabica and Robusta species in coffee using 1H NMR spectroscopy. Food Chem 182:178–184. doi: 10.1016/j.foodchem.2015.02.132 Monakhovaa Y, Kohl-Himmelseher M, Kuballa T, Lachenmeier D (2014) Findings in Pharmaceutical and Biomedical Analysis Reported from N.G. Chernyshevsky Saratov State University (Determination of the purity of pharmaceutical reference materials by H-1 NMR using the standardless PULCON methodology). Biotech Week 347–347 Montero de Espinosa L, Meier MAR (2011) Plant oils: The perfect renewable resource for polymer science?! Eur Polym J 47:837–852. doi: 10.1016/J.EURPOLYMJ.2010.11.020 National Health Interview Survey (2012) Use of Complementary Health Approaches in the U.S. In: Natl. Inst. Heal. https://nccih.nih.gov/research/statistics/NHIS/2012/natural-products/omega3. Accessed 26 Feb 2018 National Institute of Health Dietary Supplement Label Database (DSLD). https://ods.od.nih.gov/Research/Dietary_Supplement_Label_Database.aspx. Accessed 27 Feb 2018 Oliveira LS, Franca AS, Camargos RRS, Ferraz VP (2008) Coffee oil as a potential feedstock for biodiesel production. Bioresour Technol 99:3244–3250. doi: 10.1016/j.biortech.2007.05.074 Ortolá MD, Gutiérrez CL, Chiralt A, Fito P (1998) Kinetic study of lipid oxidation in roasted coffee. Rev Agaroquimica y Tecnol Aliment 4:67–73. doi: 10.1177/108201329800400109 Ostlund RE, Racette SB, Okeke A, Stenson WF (2002) Phytosterols that are naturally present in commercial corn oil significantly reduce cholesterol absorption in humans. Am J Clin Nutr 75:1000–1004. doi: 10.1093/ajcn/75.6.1000 Park J, Kim B, Lee JW (2016) In-situ transesterification of wet spent coffee grounds for sustainable biodiesel production. doi: 10.1016/j.biortech.2016.09.001 Pauli GF (2001) qNMR - a versatile concept for the validation of natural product reference compounds. Phytochem Anal 12:28–42. doi: 10.1002/1099- 1565(200101/02)12:1<28::AID-PCA549>3.0.CO;2-D Pauli GF, Jaki BU, Lankin DC (2005) Quantitative 1H NMR: Development and Potential of a Method for Natural Products Analysis. J Nat Prod 68:133–149. doi: 10.1021/np0497301 Plans M, Wenstrup MJ, Rodriguez-Saona LE (2015) Application of Infrared Spectroscopy for Characterization of Dietary Omega-3 Oil Supplements. J Am Oil Chem Soc 92:957–966. doi: 10.1007/s11746-015-2666-8 Plaza DD, Strobel V, Heer PKK, et al (2017) Direct valorisation of waste cocoa butter triglycerides via catalytic epoxidation, ring-opening and polymerisation. J Chem Technol Biotechnol 92:2254–2266. doi: 10.1002/jctb.5292 120

Prickett JD, Robinson DR, Steinberg AD (1983) Effects of dietary enrichment with eicosapentaenoic acid upon autoimmune nephritis in female NZBXNZW/F1 MICE. Arthritis Rheum 26:133–139. doi: 10.1002/art.1780260203 Raba DN, Poiana M-A, Borozan AB, et al (2015) Investigation on Crude and High- Temperature Heated Coffee Oil by ATR-FTIR Spectroscopy along with Antioxidant and Antimicrobial Properties. PLoS One 10:e0138080. doi: 10.1371/journal.pone.0138080 Rendón MY, De T, Salva JG, Bragagnolo N (2014) Impact of chemical changes on the sensory characteristics of coffee beans during storage. Food Chem 147:279–286. doi: 10.1016/j.foodchem.2013.09.123 Robinson DR, Prickett JD, Makoul GT, et al (1986) Dietary fish oil reduces progression of established renal disease in (NZB × NZW)F1 mice and delays renal disease in BXSB and MRL/1 strains. Arthritis Rheum 29:539–546. doi: 10.1002/art.1780290412 Sacchi R, Medina I, Aubourg SP, et al (1993) Proton Nuclear Magnetic Resonance Rapid and Structure-Specific Determination of Polyunsaturated Fatty Acids in Fish Lipids. J Am Oil Chem Soc 70:225–228 Scharnhop H, Winterhalter P (2009) Isolation of coffee diterpenes by means of high- speed countercurrent chromatography. J Food Compos Anal 22:233–237. doi: 10.1016/j.jfca.2008.10.018 Schenker S, Handschin S, Frey B, et al (2000) Pore Structure of Coffee Beans Affected by Roasting Conditions. J Food Sci 65:452–457. doi: 10.1111/j.1365- 2621.2000.tb16026.x Schenker S, Rothgeb T (2017) The Roast—Creating the Beans’ Signature. In: The Craft and Science of Coffee. Elsevier, pp 245–271 Schievano E, Finotello C, De Angelis E, et al (2014) Rapid Authentication of Coffee Blends and Quantification of 16-O-Methylcafestol in Roasted Coffee Beans by Nuclear Magnetic Resonance. J Agric Food Chem 62:12309–12314. doi: 10.1021/jf505013d Schmidt SJ, Sun X, Litchfield JB, Eads TM (1996) Applications of magnetic resonance imaging in food science. Crit Rev Food Sci Nutr 36:357–385. doi: 10.1080/10408399609527730 Selmar D, Bytof G, Knopp SE (2008) The storage of green coffee (Coffea arabica): Decrease of viability and changes of potential aroma precursors. Ann Bot 101:31– 38. doi: 10.1093/aob/mcm277 Simopoulos AP Dossier: Polyunsaturated fatty acids in biology and diseases The importance of the ratio of omega-6/omega-3 essential fatty acids Skiera C, Panagiotis S, Kuballa T, et al (2012) 1H-NMR Spectroscopy as a New Tool in the Assessment of the Oxidative State in Edible Oils. Am Oil Chem Soc 89:1383– 1391. doi: 10.1007/s11746-012-2051-9 Speer K, Kölling-Speer I (2006b) The lipid fraction of the coffee bean. Braz J Plant Physiol 18:201–216 Speer K, Kölling-Speer I (2006a) The lipid fraction of the coffee bean. Braz J Plant Physiol 18:201–216 121

Speer K, Kurt A (2001) Effects of steam treatment on diterpenes. Proc 19th Int Sci Colloq Coffee Speer K, Tewis R, Montag A (1992) 16-o-methylcafestol: a quality indicator for coffee. Proc 14th Int Sci Colloq Coffee 237–244 Spyros A, Dais P (2013) NMR spectroscopy in food analysis. Royal Society of Chemistry Toci AT, Neto VJMF, Torres AG, Farah A (2013) Changesss in triacylglycerols and free fatty acids composition during storage of roasted coffee. LWT - Food Sci Technol 50:581–590. doi: 10.1016/j.lwt.2012.08.007 Tyburn J-M, Coutant J (2015) TopSpin ERETIC 2. 1–34 Valdenebro MS, León-Camacho M, Pablos F, et al Determination of the arabica/robusta composition of roasted coffee according to their sterolic content Vila MA, Andueza S, Paz de Pena M, Cid C (2005) Fatty Acid Evolution During the Storage of Ground, Roasted Coffees. J Am Oil Chem Soc 82:639–646 Wagemaker TAL, Rijo P, Rodrigues LM, et al (2015) Integrated approach in the assessment of skin compatibility of cosmetic formulations with green coffee oil. Int J Cosmet Sci 37:506–510. doi: 10.1111/ics.12225 Wakil A, Mir M, Mellor DD, et al (2010) The bioavailability of eicosapentaenoic acid from reconstituted triglyceride fish oil is higher than that obtained from the triglyceride and monoglyceride forms. Asia Pac J Clin Nutr 19:499–505 Wei F, Furihata K, Koda M, et al (2012b) 13C NMR-based metabolomics for the classification of green coffee beans according to variety and origin. J Agric Food Chem 60:10118–10125. doi: 10.1021/jf3033057 Wei F, Furihata K, Koda M, et al (2012a) Roasting Process of Coffee Beans as Studied by Nuclear Magnetic Resonance: Time Course of Changes in Composition. J Agric Food Chem 60:1005–1012. doi: 10.1021/jf205315r Wider G, Dreier L (2006) Measuring Protein Concentrations by NMR Spectroscopy. doi: 10.1021/ja055336t Williamson K, Emmanuel H (2017) NMR spectroscopy as a robust tool for the rapid evaluation of the lipid profile of fish oil supplements. J Vis Exp. doi: doi:10.3791/55547 Wilson AJ, Illy P.E. (1997) Some preliminary investigations of oil biosynthesis in the coffee fruit and its subsequent re-distribution within green and roasted beans. In: 17th International Scientific Colloquium on Coffee. ASIC - Association for science and information on coffee, Paris Wilson WK, Sumpter RM, Warren JJ, et al (1996) Analysis of unsaturated C27 sterols by nuclear magnetic resonance spectroscopy. J Lipid Res 37:1529–55 Xia Y, Moran S, Nikonowicz EP, Gao X (2008) Z-restored spin-echo 13 C 1D spectrum of straight baseline free of hump, dip and roll. doi: 10.1002/mrc.2195 Zamora R, Alba V, Hidalgo FJ (2001) Use of high-resolution 13C nuclear magnetic resonance spectroscopy for the screening of virgin olive oils. J Am Oil Chem Soc 78:89–94. doi: 10.1007/s11746-001-0225-z Zarkovic N (2003) 4-Hydroxynonenal as a bioactive marker of pathophysiological processes. Mol Aspects Med 24:281–291. doi: 10.1016/S0098-2997(03)00023-2 122

Zhou L, Khalil A, Bindler F, et al (2013) Effect of heat treatment on the content of individual phospholipids in coffee beans. doi: 10.1016/j.foodchem.2013.06.056 (2018) Application of PULCON to purity assay of Acetaminophen. In: JEOL. https://www.jeol.co.jp/en/applications/detail/1229.html. Accessed 7 Mar 2018

123