Profiling Lipids for Authentication of High Value Ingredients by Mid-Infrared Spectroscopy Combined with Multivariate Analysis

THESIS

Presented in Partial Fulfillment of the Requirements for the Degree Master of Science in the Graduate School of The Ohio State University

Natalie E. Maurer

Graduate Program in Food Science and Technology

The Ohio State University

2012

Master's Examination Committee:

Dr. Luis Rodriguez-Saona, Advisor

Dr. Sheryl Barringer

Dr. John Litchfield

Copyrighted by

Natalie E. Maurer

2012

Abstract

Over the past decade, there has been an increase in demand for rapid techniques for authentication and detection of adulterants in food ingredients. Economic adulteration and counterfeiting of global food and consumer products is expected to cost the industry

$10 to $15 billion per year. Verifying authenticity and detection of adulterants is something that can happen anywhere in the food supply chain, from the farmers to consumers and the industry to the manufacturer. The development of sensitive and robust screening tool(s) for assuring the quality of incoming raw materials would supplement the assurances provided by food manufacturer vendor auditing programs. The objective of this study was to develop a rapid and accurate method for the characterization and authentication of high valued ingredients. A temperature-controlled ZnSe ATR mid- infrared benchtop and diamond ATR mid-infrared portable handheld spectrometers were used to characterize sacha inchi oil and evaluate its oxidative stability compared to commercial oils. Soft independent model of class analogy (SIMCA) and partial least squares regression (PLSR) analyzed the spectral data. Fatty acid profiles showed that sacha inchi oil (44% linolenic acid) had similar levels of PUFA as flax oils. PLSR showed good correlation coefficients (R2>0.9) between reference tests and spectra from infrared devices, allowing for rapid determination of fatty acid composition and predicting oxidative stability. Oils formed distinct clusters allowing the evaluation of commercial sacha inchi oils from Peruvian markets and showed some prevalence of ii adulteration. Determining oil adulteration and quality parameters using the ATR-MIR portable handheld spectrometer allowed for portability and ease-of-use, making it a great alternative to traditional testing methods. Forty different cocoa butter samples encompassing an acceptable range of compositional variability for the chocolate industry were included. Cocoa butters were characterized for their melt characteristics (DSC

Hardness), triacylglycerol content and fatty acid composition (GC-FAME). SIMCA and

PLSR were used for classification and quantification analysis of cocoa butters. SIMCA classified all cocoa butters in distinct clusters in a 3-dimensional space but no sample clustering patterns were associated with melt characteristics. Spectral differences responsible for the separation of classes were attributed to stretching vibrations of the ester (–C=O) linkage (1660-1720 cm-1). PLSR models showed correlation coefficients >

0.93 and prediction errors (SECV) of 1.5 units for melt characteristics, 0.2-0.3% and 0.4-

0.8% for major fatty acids and triacylglycerols, respectively. ATR-IR spectroscopy combined with pattern recognition analysis provides robust models for characterization and determination of cocoa butter composition. Overall, FT-IR has proven to be a fast, reliable and highly reproducible method for the characterization and authentication of high valued ingredients.

iii

Dedicated to Joel.

Thank you for always supporting me in following my dreams.

Acknowledgments

I would like to thank my advisor, Dr. Rodriquez-Saona for allowing me to work with him and for giving me the confidence I need in my abilities as a scientist. Thank you to Dr. Barringer and Dr. Litchfield for serving as committee members.

Thank you to the members of my lab group, especially Emily Birkel and Ting

Wang, who have always shown me great support and encouragement.

Vita

June 2006 ...... Oak Hills High School

May 2010 ...... B.S. Chemistry, Miami University

2010 to present ...... Graduate Research Associate, Department

of Food Science and Technology,

The Ohio State University

Publications

Characterization and Authentication of a Novel Vegetable Source of Omega-3 Fatty

Acids, Sacha Inchi (Plukenetia volubilis L.) Oil

Rapid Assessment of Quality Parameters in Cocoa Butter using FT-IR Spectroscopy and

Multivariate Analysis

Fields of Study

Major Field: Food Science and Technology

Table of Contents

Abstract ...... ii

Dedication...... iv

Acknowledgments...... v

Vita ...... vi

List of Tables ...... ixx

List of Figures ...... x

Chapter 1: Literature Review ...... 1

1.1 Infrared Spectroscopy ...... 1

1.2 Risk of Chemical Hazards ...... 11

1.3 Vibrational Spectroscopy for Detection of Chemical Contaminants ...... 17

1.4 References ...... 23

Chapter 2: Characterization and Authentication of a Novel Vegetable Source of Omega-3

Fatty Acids, Sacha Inchi (Plukenetia volubilis L.) Oil ...... 28

2.1 Abstract ...... 29

2.2 Introduction ...... 30

vii

2.3 Materials and Methods ...... 32

2.4 Results and Discussion ...... 37

2.5 Conclusion ...... 53

2.6 Acknowledgement ...... 53

2.7 References ...... 54

Chapter 3: Rapid Assessment of Quality Parameters in Cocoa Butter using ATR-IR

Spectroscopy and Multivariate Analysis ...... 57

3.1 Abstract ...... 58

3.2 Introduction ...... 59

3.3 Materials and Methods ...... 62

3.4 Results and Discussion ...... 67

3.5 Conclusion ...... 76

3.6 References ...... 78

List of References ...... 80

viii

List of Tables

Table 2.2. PLSR model statistical analysis for determining fatty acid composition in oils

(corn, high oleic sunflower, flax and sacha inchi) using a benchtop and hendheld infrered system ...... 45

Table 2.3. Peroxide value (PV), free fatty acid (FFA), and fatty acid composition for the different oil samples studied during a 20 day oxidative stability test...... 49

Table 2.4. Statistical analyses for the PLSR models developed to determine the PV and

FFA of oils (corn, high oleic sunflower, flax and sacha inchi) during their oxidation process using a mid-IR ATR benchtop and mid-IR ATR portable handheld spectrometers...... 51

Table 3.1. Statistical data for the hardness, triacylglycerol (TAG) and fatty acid composition levels using the 40 commercial cocoa butter samples...... 71

Table 3.2. Summary of statistical multivariate analysis for determining hardness, triacylglycerol (TAG) and fatty acid composition levels...... 74

List of Figures

Figure 1.1. General setup of an infrared spectrometer and design of a Michelson

Interferometer ...... 3

Figure 1.2. Infrared band assignments for a typical lipid spectrum...... 4

Figure 1.3. TruDefender FT handheld portable infrared spectrometer ...... 11

Figure 1.4. spectra of melamine and infant formula collected using ATR method ...... 19

Figure 2.1. FT-IR mid-infrared spectra of high oleic sunflower, canola, corn, sunflower, cottonseed, olive, sacha inchi and flax oil using a temperature-controlled single-bounce

ZnSe ATR mid-IR...... 37

Figure 2.2. Soft independent modeling of class analogy (SIMCA) 3D projection plots of second derivative transformed spectral data collected by a mid-IR ATR benchtop (A) and mid-IR ATR portable handheld (C) spectrometers. SIMCA discrimination plots for the mid-IR ATR benchtop (B) and mid-IR ATR portable handheld (D) spectrometers ...... 41

Figure 2.3. SIMCA prediction plots for authentication of sacha inchi oils using the mid-

IR ATR benchtop (A) and mid-IR ATR portable handheld (B) spectrometers ...... 43

Figure 2.4. Partial least squares regression (PLSR) of free fatty acids using the mid-IR

ATR benchtop (A- linolenic acid, C- linoleic acid, E- oleic acid) and the mid-IR ATR portable handheld spectrometers (B- linolenic acid, D- linoleic acid, F- oleic acid) ...... 46

Figure 2.5. Peroxide value (PV) of corn, high oleic sunflower, flax and sacha inchi oils versus the storage time (in days) in a 65 oC oven ...... 48

Figure 2.6. Soft independent modeling of class analogy (SIMCA) 3D projection plots of second derivative transformed spectral data collected by a mid-IR ATR benchtop (A) and mid-IR ATR portable handheld (B) spectrometers for corn, high oleic sunflower, flax and sacha inchi oils during their oxidation process. For SIMCA projection plots, boundaries marked around the sample clustered represent a 95% confidence interval for each class. If the residual variance of a sample exceeds the upper limit of the boundary for the modeled classes in the data set, it is not assigned to any of the classes; either it is an outlier, or it comes from a class not represented in the data set ...... 52

Figure 3.1. FT-IR mid-infrared spectra and band assignments of cocoa butter using a temperature-controlled single-bounce ZnSe ATR mid-IR ...... 68

Figure 3.2. Soft independent modeling of class analogy (SIMCA) 3D projection plots of second derivative transformed spectral data collected by ATR-MIR spectrometer for all

40 cocoa butter samples (A) and the cocoa butter samples disregarding the softest cocoa butter (DSC 58) (B)...... 69

Figure 3.3. SIMCA discrimination plots for the ATR-MIR spectrometer for all 40 cocoa butter samples (A) and the cocoa butter samples disregarding the softest cocoa butter

(DSC 58) (B) ...... 72

Figure 3.4. Partial least squares regression (PLSR) of hardness by DSC (A), stearic acid fatty acid composition (B) and SOS triacylglycerol composition (C) using selected sample from the 40 cocoa butters ...... 75

Chapter 1: Literature Review

1.1 Infrared Spectroscopy

Spectroscopy is the study of the interaction between light and matter and explores the production, measurement and interpretation of spectra arising from this interaction.

Differences in spectroscopic methods such as sample, type of interaction to be monitored and the electromagnetic region used, allows for several types of spectroscopic methods to be used to solve analytical problems. Spectroscopic methods are widely used for quantitative and qualitative analysis (Penner, 1998). The most common spectroscopic methods are based on the absorption or emission of radiation in the ultraviolet (UV), visible (Vis), infrared (IR) and radio (nuclear magnetic resonance) frequency ranges.

Infrared radiation is electromagnetic energy with wavelengths (λ) longer than visible light but shorter than microwaves (Wehling, 1998). Infrared radiation is divided into three parts: far-IR (40-400 cm-1), mid-IR (400-4,000 cm-1) and near-IR (4,000-14,000 cm-

1) (Guillen & Cabo, 1997). A molecule can absorb IR radiation if it vibrates in such a way that its charge distribution changes during the vibration. The most important vibrations that cause a change in dipole moment are stretching and bending (scissoring) motions. The vibrational energy is directly proportional to the strength of the bond. The vibrating functional group can absorb radiant energy to move from the lowest vibrational state (υ=0) to the first excited state (υ=1). The frequency of radiation that will cause the energy jump is equal to the initial frequency of the vibration of the bond and this 1 frequency is called the fundamental absorption. Molecules that absorb radiation and move to a higher excited state (υ=2 or 3) are referred to as overtones. (Wehling, 1998).

Every molecule has slightly different vibrational modes; therefore, the infrared spectrum of a given molecule is unique and can be used to identify that molecule (Griffiths & de

Haseth, 2007).

1.1.1 Mid-Infrared Spectroscopy

Mid-infrared spectroscopy is a great qualitative analysis tool because of its capability to elucidate the chemical functional groups of molecules. Chemical changes in food components whether from processing or even storing, may potentially be monitored by comparing differences in mid-IR spectra (Li-Chan, Griffiths, & Chalmers, 2010a). This region is very interesting to use in the study of organic compounds because the intensity of the absorption bands due to the vibrations of certain functional groups (Guillen &

Cabo, 1997) are proportional to the concentration. This makes mid-IR spectroscopy ideal for qualitative and quantitative analysis of foods (Li-Chan, Griffiths, & Chalmers,

2010a).

Fourier Transform infrared Spectroscopy (FT-IR) has been gaining acceptance as a good analytical tool for the last three decades (Ismail, van de Voort, & Sedman, 1997). In

Fourier transform (FT) instruments the sample is irradiated with a continuous band of radiation and all wavelengths arrive at the detector simultaneously (Guillen & Cabo,

1997). The radiation source is commonly a Globar which is resistively heated silicon carbide (Griffiths & de Haseth, 2007).

Figure 1.1 General setup of an infrared spectrometer and design of a Michelson Interferometer (Baeten and Dardenne 2002)

Figure 1.1 shows the general setup of an infrared spectrometer. In an FT-IR instrument, the monochromator of a conventional infrared spectrometer is replaced with a Michelson interferometer which splits the infrared beam using a semi-permeable beamsplitter

(Günzler & Gremlich, 2002). The two partial beams are then reflected on a fixed and moveable mirror back to the beamsplitter where they recombine and are brought to interfere (Wehling, 1998; Ismail, van de Voort, & Sedman, 1997). A shift in the moveable mirror changes the optical pathlength in this interferometer and the intensity of the radiation reaching the detector varies as a function of this optical path difference. The intensity of the signal from the detector, as a function of the change in optical pathlength corrected by a constant component, is called an interferogram (Ismail, van de Voort, &

Sedman, 1997). This interferogram showing intensity versus pathlength is converted by a 3 mathematical algorithm called a Fourier transform, into an IR spectrum giving absorbance versus frequency (Wehling, 1998). Figure 1.2 shows a mid-infrared spectrum for a typical lipid.

Figure 1.2. Infrared band assignments for a typical lipid spectrum (van de Voort and others 2001).

FT-IR spectroscopy has significant advantages over dispersive instruments (Guillen &

Cabo, 1997). FT-IR is a rapid, non-destructive method that requires no reagents. The multiplexing property allows all frequencies to be detected simultaneously allowing for a higher energy throughput and a superior wavelength resolution (Guillen & Cabo, 1997).

The wear on the equipment is low and a reduction in scan time is possible without the loss of resolution (Griffiths & de Haseth, 2007). The signal to noise ratio of a spectrum is

4 significantly increased by repetitive scanning (co-adding) and FT-IR allows for cleaner spectra (Ismail, van de Voort, & Sedman, 1997). FT-IR also allows for high wavelength calibration accuracy, improved light throughput and a more rapid speed of analysis

(Juanéda, Ledoux, & Sébédio, 2007). The use of automated operations and personal computing capabilities allows for easier operations and data processing (Juanéda,

Ledoux, & Sébédio, 2007).

FT-IR is a reliable and well recognized fingerprinting method. One of the strengths of IR spectroscopy is its capability to obtain spectra from a wide range of solids, liquids and gases. Traditionally, IR spectrometers analyze samples by transmitting the infrared radiation directly through the sample (PerkinElmer, 2005). An interesting technique for surface analysis of most liquids and solids has been developed (Guillen & Cabo, 1997) and combats the most challenging aspects of infrared analysis; sample preparation and spectral reproducibility (PerkinElmer, 2005). Attenuated total reflectance (ATR) is a versatile, nondestructive technique for obtaining infrared spectrum of the surface of a material (Günzler & Gremlich, 2002). ATR has grown into one of the most widely used techniques in infrared spectroscopy. The technique requires little or no sample preparation and consistent results can be obtained with little training required (Griffiths & de Haseth, 2007). To obtain an ATR spectrum the infrared radiation from the interferometer is passed into a high-refraction-index crystal and is internally reflected, generally several times (Guillen & Cabo, 1997). The light interacts with the sample which is in contact with this high-refraction-index crystal. The sample then interacts with the evanescent wave at the surface of the crystal formed by the internal reflectance,

5 resulting in the absorption of radiation by the sample at each point of reflection (Günzler

& Gremlich, 2002; Ismail, van de Voort, & Sedman, 1997). The evanescent wave is attenuated by the absorption of radiation and the measurement of the attenuation by the detector yields an interferogram which if Fourier transformed to yield an absorption spectrum (Ismail, van de Voort, & Sedman, 1997).

There are several common crystal materials for ATR such as zinc selenide (ZnSe), germanium or diamond. ZnSe is a fairly low priced ATR crystal and is ideal for analyzing liquids (PerkinElmer, 2005). The use of an ATR crystal has two major requirements. First, the sample must be in direct contact with the crystal to allow interaction with the evanescent wave. Second, the refractive index of the crystal must be significantly greater than the sample or else internal reflectance will not occur (the light will be transmitted instead) (PerkinElmer, 2005).

The use of mid-IR spectroscopy can be used on all types of food systems from lipids to meat to fruit and vegetables and even chemical contaminants in the food system. What makes mid-IR spectroscopy so valuable is that it can be used to identify specific functional groups present in an unknown substance. The relative intensities of the bands are also related to the concentration of that chemical group in the sample which makes mid-IR spectroscopy successful in making quantitative measurements (Wehling, 1998).

For oils, the characteristic mid-IR absorption bands in the fingerprint region provides direct information about the ratio between saturated and monounsaturated fatty acids in the sample (Li-Chan, Griffiths, & Chalmers, 2010b). Information like this can be useful

6 in not only identifying and characterizing unknown samples but it can help authenticate samples and find adulterants.

Overall, mid-IR spectroscopy is a very powerful analytical technique that allows rapid and accurate characterization of food sample. It is especially useful in identifying specific functional groups present in a substance because different functional groups absorb different frequencies of radiation (Guillen & Cabo, 1997). A portion of the infrared region between 650-1500 cm-1 is known as the fingerprint region and is unique to each molecule, allowing for identification of similar substances (Günzler & Gremlich, 2002).

FT-IR spectroscopy offers several advantages to dispersive instruments such as improvement in the signal-to-noise ratio, higher resolution, nondestructive to the sample and rapid scanning. The use of an ATR accessory can overcome the common problems with traditional infrared spectroscopy including sample preparation and spectral reproducibility (PerkinElmer, 2005).

1.1.2 Near-Infrared Spectroscopy

Near-infrared (NIR) spectroscopy has recently become an increasingly important technique in the food science industry due to its abilities to quantify a sample in a non destructive way (McClure, 2007). NIR spectroscopy is a fast and accurate technique that measures chemical bonds based on overtones and combination bands of specific functional groups. NIR bands are 10-100 times less intense than the corresponding mid- infrared fundamental bands (Rodriguez-Saona, Fry, McLaughlin, & Calvey, 2001). In the

NIR region, overtones and combination bands of fundamentals are observed between

4,000 to 14,000 cm-1. It can be deduced that bands having a higher intensity in the NIR region are primarily from functional groups that contain a hydrogen atom attached to a carbon, nitrogen or oxygen which are common groups in major food components

(Wheling, 1998). Substances that contain proteins contain many N-H groups, fats and oils contain mostly C-H groups, while carbohydrates contain O-H groups. By making measurements at wavelengths known to absorb energy for these groups, it is possible to determine the amount of protein, lipids and other components present in the sample (Li-

Chan, Griffiths, & Chalmers, 2010b).

NIR spectroscopy is less suitable for detailed qualitative analysis than mid-infrared spectroscopy, which shows all fundamentals, overtones and combinations of low-energy vibrations (Günzler, & Gremlich, 2002). NIR spectra contain a series of successive overlapping bands which are difficult to assign to specific functional groups, therefore

NIR provides little interpretable structural information (Li-Chan, Griffiths, & Chalmers,

2010b).

A major advantage of NIR spectroscopy is its ability to measure the composition of solid food products directly by the use of diffuse reflectance techniques (Wehling, 1998).

When radiation strikes a material to be measured, part of the radiation is reflected from the sample surface and reflectance gives little useful information about the sample.

Another portion of the radiation will penetrate through the surface of the sample and be reflected off several sample particles before it exits. This internal reflection is referred to as diffuse reflectance and each time the radiation interacts with the sample it can absorb a

8 portion of the radiation. The amount of energy absorbed at specific wavelengths contains information about the chemical composition (Wehling, 1998).

The most common radiation source in a NIR instrument is a tungsten-halogen lamp with a quartz envelope. This lamp emits radiation in both the visible and NIR spectral regions.

Silicon detectors are used in the 700-1100 nm range and lead sulfite detectors are used in the 1100-2500 nm range. NIR instruments can use monochromators, or more recently,

Fourier transform technology and interferometers. NIR instruments use filters to direct individual wavelengths of radiation onto the sample (Wehling, 1998). Four requirements are necessary for the successful application of the NIR spectrometer. Accurate and reproducible reference analysis, excellent spectral quality, comprehensive sample assembly and excellent sample presentation to the instrument are necessary to ensure efficient calibration of the instrument. The more accurate and precise the reference data, the more precisely the spectral data will represent the sample and the more effective NIR analysis will be (Li-Chan, Griffiths, & Chalmers, 2010b).

NIR spectroscopy is most widely used in the grain, cereal, and oil processing industries

(Wehling, 1998). Unlike mid-infrared analysis, NIR diffuse reflectance is able to measure powdered sample with minimal sample preparation (Günzler & Gremlich, 2002). This technique has been successfully used to measure moisture, protein and fat in red meats, poultry and fish (Wehling, 1998). NIR spectroscopy is also used for quantitative analysis in a number of dairy products including fat and protein in cheese (Rodriguez-Otero,

Hermida, & Cepeda, 1995), as well as lactose, protein and moisture in milk and whey powders (Baer, Frank, & Loewenstein, 1983). NIR spectroscopy has made it possible to

9 analyze materials, such as wheat, that could not be analyzed before due to lack of convenience and the high cost of previously used methods (Williams, 2010). NIR techniques have also been successful in measuring total sugars and soluble solids in fruits and vegetables (Polshin, Lammertyn, & Nicolaї, 2010). Although NIR reflectance analysis was initially used for agricultural and food applications, in combination with chemometrics it is now applicable for many other areas. (Günzler & Gremlich, 2002).

More emerging uses of NIR spectroscopy has been in the use of pharmaceuticals

(MacDonald & Prebble, 1993) and chemical contaminants such as melamine (Mauer,

Chernyshova, Hiatt, Deering, & Davis, 2009) and acrylamide (Segtnan, Kita, Mielnik,

Jorgensen, & Knutsen, 2006).

1.1.3 Handheld Infrared Spectrometer

A more recent contribution in the world of spectroscopy is the development of portable handheld infrared systems. These systems operate using the same components and principals as a benchtop FT-IR spectrometer but offers unique properties. The Thermo

Scientific TruDefender FT is a handheld FT-IR system that is designed to analyze and identify unknown chemical substances directly in the field (Thermo Fisher Scientific, Inc.

2011). These systems were originally developed by the U.S. government for identification of illegal drugs, weapons materials and hazardous substances. (Rein, 2008).

These systems are lightweight yet rugged and allow for rapid identification of unknown chemicals and precursors (Figure 1.3) (Thermo Fisher Scientific, Inc. 2011). Systems

10 can either use external reflectance probe for hard, reflective surfaces or a diamond internal reflectance probe for soft, nonreflective samples (Rein, 2008).

Figure 1.3. TruDefender FT handheld portable infrared spectrometer (Thermo Fisher Scientific, Inc. 2011).

There have been several recent publications that employ the use of the portable FT-IR spectrometers. It has been used for high quality in-field and on-site analysis of quality parameters in plums (Perez-Marin, Paz, Guerrero, Garrido-Varo, & Sanchez, 2010). The application of the handheld for quantization of trans fat in edible oils (Birkel &

Rodriguez-Saona, 2011) has been successfully demonstrated as well as monitoring oil oxidative stability (Allendorf, Subramanian, & Rodriguez-Saona, 2011).

1.2 Risk of Chemical Hazards

Acrylamide is a compound that is mainly formed from the Maillard reaction between asparagine and carbohydrates. This neurotoxic compound is classified as a potential human and animal carcinogen and genotoxicant (Roach, Andrzejewski, Gay, Nortrup, &

Musser, 2003). In 2002, Swedish scientists surprised the world by finding significant 11 quantities of acrylamide in carbohydrate rich, heat-treated foods which generated an international health alarm (FDA, 2007). Acrylamide is an industrial chemical used in a variety of products from water purification to packaging. Even though acrylamide is used as an industrial chemical, it forms in food as a result of the heat-induced reaction between asparagines and reducing sugars (FDA, 2007). N-glycosylasparagine is a direct precursor of acrylamide formation. The open chain form of this precursor forming oxazolidin-5-one is the key step that allows decarboxylation of asparagines followed by the formation of acrylamide (Roach, Andrzejewski, Gay, Nortrup, & Musser, 2003). Potato crisps are in the group of food products with the highest levels of acrylamide due to the naturally high level of acrylamide precursors in potatoes (Segtnan, Kita, Mielnik, Jorgensen, &

Knutsen, 2006). Acrylamide is also found in french fries, cereal foods such as cookies, crackers, and toasted breads, as well as coffee (FDA, 2007). Up to 40% of all foods contain acrylamide based on the risk assessment completed by Peterson, Kleinow, Kraska and Lech (1985). Sorgel et al. (2002) found that 10-50% of dietary acrylamide in pregnant women is transferred though the blood to the fetus and breast milk was found to contain up to 18.8μg/L of acrylamide. The FDA estimates that the average U.S. consumer’s intake of acrylamide is 0.4μg/kg body weight. Current analytical methods involving gas chromatography (GC), high performance liquid chromatography (HPLC) and mass spectrometry (MS) are used for the monitoring of acrylamide content (FDA

2007). These common practices are expensive and time consuming (Pedreschi, Segtnan,

& Knutsen, 2010).

Shortly after the discovery of acrylamide in foods, the U.S. Food and Drug

Administration (FDA, 2007) discovered that furan, another cooking-related chemical, occurred in a larger amounts and in a larger variety of foods than had previously been reported. Furan is also an industrial chemical that is used in the production of other chemicals, for example, lacquers and agricultural products (Maga, 1979). In 2004, the

FDA reported finding furan in foods, particularly ones exposed to processing techniques such as retorting in cans and jars. The formation of furan is not well understood but several mechanisms are possible, such as the oxidation of polyunsaturated fatty acids or the breakdown of amino acids in presence or absence of reducing sugars (Perez Locas &

Yaylayan, 2004). The FDA reported finding levels between less than 2 to over 170 ppb and they estimate that the average intake for U.S. consumers was 0.2μg/kg. The

International Agency for Research on Cancer lists furan as a possible carcinogen to humans and is on the Department of Health and Human Services Report as a carcinogen based on animal testing (FDA, 2007).

The UN Food and Agricultural Organization (FAO) defines pesticides as a substance or mixture intended to prevent, destroy, repel or mitigate any pest, including insects, rodents and weeds (Armenta, Quintas, Garrigues, & de la Guardia, 2005). Under the provisions of the Food Quality and Safety Act, the Environmental Protection Agency (EPA) is responsible for regulating the pesticides that are used by growers to protect crops and for setting limits on the amount of pesticides that may remain in or on foods marketed in the

USA. They also limit the levels of pesticides residues left on foods which are called tolerances (Jelinek, 1981). The Food and Drug Administration (FDA) enforces tolerances

13 established for foods other than meat, poultry and some egg products, in which the US

Department of Agriculture (USDA) enforces these tolerances. This to ensure that the nation’s food supply is maintained safely.

Over $39.4 billion was spent on pesticide expenditures in the world in 2007. Herbicides account for the largest portion of total expenditures followed by insecticides, fungicides and other pesticides, respectively (Environmental Protection Agency, 2011). Chemical pesticides have been a major advantage to developing countries economic growth. But there is growing concern that the adverse health effects are being overlooked because there is a high dependency on their use (Ecobichon, 2001). Pesticide residues can be left on fresh produce (among other products) and there has been an increasing concern on the safety of foods with pesticides residues. Pesticides are designed to be toxic to living things, so naturally they come with risks. Pesticides are toxic to humans and animals and can cause serious health implications. Gas chromatography (GC) or high-performance liquid chromatography’s (HPLC) are the common quality control methods for measuring pesticide residues (Armenta, Quintas, Garrigues, & de la Guardia, 2005). Unfortunately these methods require several hours therefore there is an ongoing interest in the development of fast, nondestructive and simple analytical method for pesticide analysis to reply to the consumers concern and the environmental concerns (Hiroaki, Tsunetomo,

& Eiji, 1998).

Melamine (2,4,6-triamino-1,3,5-triazine) is a nitrogen rich compound that is commonly used to make melamine resin (a synthetic heat-tolerant polymer), plastics, fertilizer and other products and is not approved as an ingredient in food (Lin, He, Awika, Yang,

Ledoux, Li, et al., 2008). Due to its low cost and high nitrogen content (66%) it is added as a food adulterant in order to increase the apparent protein content (Wei, Lam, Cheng,

Lu, Ho, & Li, 2010). Protein content is estimated by determining the nitrogen content and adding melamine can increase the nitrogen content of the food without the test being able to detect the source of the nitrogen is coming from (Yan, Zhou, Zhu, & Chen, 2009). In the spring of 2007, melamine contaminated food killed or sickened thousands of cats and dogs in the United States. It was reported that the pet food was contaminated with melamine or melamine-related compounds (cyanuric acid, ammeline and ammelide). It was discovered that some wheat gluten and other protein ingredients from China were contaminated. It was later discovered that melamine was found in animal feeds raised for human consumption. In the fall of 2008, infant formulas in China were also illegally adulterated with melamine and it was discovered that over 50,000 children were affected and serious health complication arose (Jackson, 2009).

With all of the melamine adulterated incidents relating to human food and animal feeds, it is safe to say that there is considerable consumer anxiety and desire to control the presence of these compounds. Currently, HPLC coupled with mass spectroscopy (HPLC-

MS) is the principal analytical technique used by the FDA for the detection and quantification of melamine in foods. Even though the limit of detection for this method is as low as 10 ppb, this method is considerably time consuming and labor intensive (Lin,

He, Awika, Yang, Ledoux, Li, et al., 2008). Therefore, is an urgent need to establish a simple and effective method for the analysis of melamine (Yan, Zhou, Zhu, & Chen,

2009). A technology that provides rapid and accurate analysis would be highly beneficial for difficult testing areas such as fields and rural areas.

Mycotoxins are toxic secondary metabolites of fungi that can have serious health implications. When mycotoxins are present in foods in high levels, these fungal metabolites can have toxic effects that range from liver or kidney deterioration to liver cancer or even birth defects. Several mycotoxins, such as aflatoxins, have been classified by the International Agency for Research in Cancer as possible and probable human carcinogens (Kuiper-Goodman, 2004). Despite efforts to control fungal contamination, toxigenic fungi are ubiquitous in nature and occur regularly in the worldwide food supply chain (Murphy, Hendrick, Landgren, & Bryant, 2006). Thousands of mycotoxins exist but only a few present significant food safety challenges. At the farm level, mold growth can result in reduced crop yields and even livestock productivity can decrease from death or illness due to the animals consumption of contaminated feed. Currently in food manufacturing, detection is complicated due to limitations in analytical methodology.

HPLC and enzyme-linked immunosorbent assay (ELISA) are the common screening tools used for detection of aflatoxins in corn samples (Kuiper-Goodman, 2004). These methods however are inconvenient and impractical for routine production use. (Gordon,

Wheeler, Schudy, Wicklow, & Greene, 1998).

1.3 Vibrational Spectroscopy for Detection of Chemical Contaminants

Even though major advances have been made in the last century in ways for assessing and managing chemical hazards in foods, chemical contaminants in foods continue to be an area of concern. Contaminants can be added by mistake or intentionally, where some are added as economic adulterants or even maliciously. Identification of chemical hazards is necessary especially with their long-term carcinogenic potential. Although the emphasis in the past has been on microbial food safety issues, there has been increased attention on chemical food safety. (Jackson, 2009).

Foods are commonly adulterated with materials that are of a lower quality to reduce the cost for manufactures. Some of these adulterants have potentially harmful components.

Trans fat and acrylamide are byproducts of processing the product and melamine is an example of a chemical contaminant that was intentionally added. Improvements in analytical chemistry have made it possible to detect food adulterants that would have been previously impossible to detect (Jackson, 2009).

Vibrational spectroscopy provides useful, well-established analytical techniques for quantitative determination of major and minor components. Regan, Meaney, Vos,

MacCraith, and Walsh (1996) demonstrated measuring the pesticide levels in water using

ATR-FT-IR spectroscopy. They were able to detect chlorinated pesticides in the low ppm range by measuring the absorption bands specific to alachlor and atrazine. The specific absorption bands of alachlor at 1104 cm-1 and atrazine at 1577 cm-1, enabled simultaneous determination of these chlorinated pesticides in water (Regan, Meaney,

Vos, MacCraith, & Walsh, 1996). Hiroaki, Tsunetomo and Eiji (1998) have also proposed a diagnostic method for pesticide residues in agricultural products. It uses the

ATR of infrared light on the sample surface to measure multiple pesticide residues in minutes (Hiroaki, Toyonori, & Eiji, 2002). They also discuss another non destructive method; diffuse reflectance infrared spectroscopy, which is more applicable to uneven samples such as fruits and grains. From the spectra and the concentration of the pesticide residues, the calibration models are used to estimate the residual concentration of the pesticides by partial least squares regression of the spectra. In the calibration model, the cross-validation of the models is carried out by using the leave-one-out method.

Near- and mid-infrared spectroscopy combined with chemometrics allows for rapid detection and quantification of melamine in infant formulas. A study by Mauer,

Chernyshova, Hiatt, Deering and Davis (2009) compares the mid-IR spectra of unadulterated infant formula powder and melamine as shown in Figure 1.4. The mid-IR spectrum of melamine shows that the peaks at 3500-3000 cm-1 and 1700-1300 cm-1 are attributed to the stretching and bending vibrations of the amino groups present.

Figure 1.4. FTIR spectra of melamine and infant formula collected using ATR method (Mauer, Chernyshova, Hiatt, Deering, & Davis, 2009)

PLS models were developed from these spectra using multicative scatter correction and first-derivative preprocessing steps. These steps help correct the spectra for spectral noise and background effects due to variations in the particulate size or shape. The PLSR model using the FTIR-ATR spectral information combined with multivariate analysis validated that a linear relationship exists between the actual content of melamine and the predicted content of melamine in the infant formula powder with an R2 of 0.9931. An

SECV value of less than 1 and less than 10 factors were used which meets the criteria for a valid PLS model (Mauer, Chernyshova, Hiatt, Deering, & Davis, 2009). It was slightly more difficult to assign spectral peaks to specific functional groups in the near-IR spectra than in the mid-IR spectra. The PLSR validation models using the near-IR spectral information had a higher R2 value and a lower standard error of cross validation (SECV) value than the mid-IR. Both of these methods were rapid and accurate. A limit of 1 ppm 19 for melamine was set by the FDA for infant formula. The near- and mid-IR spectrometers were used to collect spectra of unadulterated infant formula and infant formula containing

1 ppm melamine. Both models were able to distinguish between the adulterated (1 ppm) and unadulterated infant formulas, with no misclassifications and in a short period of time

(Mauer, Chernyshova, Hiatt, Deering, & Davis, 2009). This study shows that near- and mid-IR method have shorter preparation times and shorter run times making vibrations spectroscopy a useful method for measuing melamine in infant formulas.

Due to its speed and ability to form contact-free measurements without destroying the samples, NIR spectroscopy is a great technique for the analysis of the proximate composition of complex foods (Segtnan, Kita, Mielnik, Jorgensen, & Knutsen, 2006).

NIR has the capability of detecting and quantifying many different compounds present in foods due to differences in the vibrational and rotational energies of bonds within the food matrix (Pedreschi, Segtnan, & Knutsen, 2010). The most pronounced differences that can be related to acrylamide levels in foods can be found in the 1900-2200 nm range.

The peak at 1934 nm was found to be the most significant. Using PLSR, predictive models could be generated using the spectral information from the ground chips in the spectral region from 400-2498 nm. The models had an R2 of approximately 0.95 and a mean prediction error of 256.6μg/kg. A large part of the visible region (400-600 nm) is also significantly related to the acrylamide content. Also both fat regions (1700-1800 and

2270-2370 nm) are also positively correlated with the acrylamide content. Changing processing parameters like frying and drying times will affect the NIR spectra so data should be collected for all production processes used. Overall, NIR spectral models

20 appear to give a good indication of the amount of acrylamide in potato crisps (Segtnan,

Kita, Mielnik, Jorgensen, & Knutsen, 2006).

FT-IR has the capability of expressing a specific “fingerprint” which makes it a great tool for identification of unknown microbial strains. This technique has great sensitivity, microbial sample preparation is simple, no reagents are needed and the data acquisition is faster than other physical-chemical techniques (Garip, Gozen, & Severcan, 2009). FT-IR has proved to be a powerful tool for differentiating fungi and its use for identifying specific mycotoxins is profound. Zotti et al. (2008) used FT-IR spectroscopy equipped with an ATR as a tool for characterizing and verifying the presence of fungal species on paper (Zotti, Ferroni, & Calvini, 2008). FT-IR when combined with microscopy appears to be a precise tool for localization and identification of fungal materials on wood. FT-IR microscopy improves chemical data quality and has the potential for monitoring chemical changes (Naumann, Navarro-Gonzalez, Peddireddi, Kues, & Polle, 2005).

Berardo et al. (2005) have used NIR spectroscopy as a practical spectroscopic procedure for the rapid identification and detection of micotoxigenic fungi and other toxic metabolites in naturally and artificially contaminated products. The results obtained indicated that NIR could accurately predict the incidence of maize kernels infected by fungi (Fusarium verticilloides in particular 59% infection) as well as the quantity of the mycotixin fumonisin B1 in the meal. Total kernel infection ranged from 28 to 100%, with a mean of 86% in naturally contaminated samples. Calibration models were obtained utilizing maize kernels and maize meals (R2=0.80). The bands at 1430, 1470, 1820, 2140 and 2180 nm are related to the total fungal infection could be assigned to the first

21 overtone of the OH stretching modes of glucose, NH in amino acids, and CH combination bands in cis unsaturation. The major absorption peaks for F. verticilliodes infection were observed at 1954 and 2378 nm, corresponding to the second overtone of the CH stretching mode of carbonyl compounds (Berardo, Pisacane, Battilani,

Scandolara, Pietri, & Marocco, 2005).

Grains are constantly under surveillance for mycotoxin contamination because of the risks to both humans and animals (Gordon, Wheeler, Schudy, Wicklow, & Greene,

1998). Gordon, Wheeler, Schudy, Wicklow, & Greene, (1998) have been able to detect fungal infections in corn by the use of FT-IR photacoustic spectroscopy. Because of its inherently surface-profiling signal, it is a sensitive detector of fungal infections on the surface of corn kernels. A. flavus displays typical infrared spectra which significantly differs from the spectra of corn which was also distinct from infected corn. There were several differences and peak shifts that made the spectra of infected corn unique. The increased protein content from fungal growth on infected corn causes the peak composed

-1 of overlapping OH and NH2 bands from 3400 to 3360 cm to shift to lower wavenumbers. The ratio of two methylene bands at 2853 and 2923 cm-1 appeared to be a measure of the level of infection in most samples (Gordon, Wheeler, Schudy, Wicklow,

& Greene, 1998). Other changes in the spectra due to fungal infections of the corn can be found in Gordon, Wheeler, Schudy, Wicklow, & Greene, 1998.

1.4 References

Armenta, S., Quintas, G., Garrigues, S., & de la Guardia, M. (2005). Mid-infrared and Raman spectrometry for quality control of pesticide formulations. Trac-Trends in Analytical Chemistry, 24(8), 772-781.

Allendorf, M., Subramanian, A., & Rodriguez-Saona, L. (2011). Application of a Handheld Portable Mid-Infrared Sensor for Monioring Oil Oxidative Stability. Journal of the American Oil Chemists Society., DOI 10.1007/s11746-011-1894-9.

Baer, R. J., Frank, J. F., & Loewenstein, M. (1983). Compositional analysis of whey powders using near infrared diffuse reflectance spectroscopy. Journal of Food Science, 66, 858-863.

Berardo, N., Pisacane, V., Battilani, P., Scandolara, A., Pietri, A., & Marocco, A. (2005). Rapid detection of kernel rots and mycotoxins in maize by near-infrared reflectance spectroscopy. Journal of Agriculture and Food Chemistry, 53(21), 8128-8134.

Birkel, E., & Rodriguez-Saona, L. (2011). Application of a Portable Handheld Infrared Spectrometer for Quantitation of trans Fat in Edible Oils. Journal of the American Oil Chemists Society, 88(10), 1477-1483.

Ecobichon, D. J. (2001). Pesticide use in developing countries. Toxicology, 160(1-3), 27- 33.

Environmental Protection Agency. (2011). Pesticides Industry Sales and Usage. [WWW page]. URL http://www.epa.gov/opp00001/pestsales/07pestsales/market_estimates2007.pdf.

Food and Drug Administration. (2007). Acrylamide, Furan, and the FDA. [WWW page]. URL http://www.fda.gov/Food/FoodSafety/FoodContaminantsAdulteration/ ChemicalContaminants/Acrylamide/ucm194482.htm#Authors.

Garip, S., Gozen, A. C., & Severcan, F. (2009). Use of Fourier transform infrared spectroscopy for rapid comparative analysis of Bacillus and Micrococcus isolates. Food Chemistry, 113(4), 1301-1307.

Gordon, S. H., Wheeler, B. C., Schudy, R. B., Wicklow, D. T., & Greene, R. V. (1998). Neural network pattern recognition of photoacoustic FTIR spectra and knowledge-based techniques for detection of mycotoxigenic fungi in food grains. Journal of Food Protection, 61(2), 221-230.

Griffiths, P. R., & de Haseth, J. A. (2007). Fourier Transform Infrared Spectrometry (2nd Ed.) (pp 3-6, 75-88, 321-3329). Hoboken, NJ: John Wiley & Sons, Inc., Publication.

Guillen, M. D., & Cabo, N. (1997). Infrared spectroscopy in the study of edible oils and fats. Journal of the Science of Food and Agriculture, 75(1), 1-11.

Günzler, H., & Gremlich, H. (2002). IR Spectroscopy: An Introduction (pp 52-62). Germany: Wiley-VCH.

Hiroaki, I., Toyonori, N., & Eiji, T. (2002). Measurement of pesticide residues in food based on diffuse reflectance IR spectroscopy. Ieee Transactions on Instrumentation and Measurement, 51(5), 886-890.

Hiroaki, I., Matsuzawa, T., & Eiji, T. (1998). Semiquantative at-line measurement system for multiresidues of pesticide in food based on the ATR spectroscopy. IEE Intstrumentation and Measurement, 675-678.

Ismail, A. A., van de Voort, F. R., & Sedman, J. (1997). Fourier Transform Infrared Spectroscopy: Principles and Applications. In J. R. J. Paré & J. M. R. Bélanger (Eds.), Instrumental Methods in Food Analysis (pp.97-117). Amsterdam: Elsevier Science.

Jackson, L. S. (2009). Chemical food safety issues in the United States: past, present, and future. Journal of Agriculture and Food Chemistry, 57(18), 8161-8170.

Jelinek, C. (1981). Occurrence and Methods of Control of Chemical Contaminants in Foods. Environmental Health Perspectives, 39(Jun), 143-151.

Juanéda, P., Ledoux, M., & Sébédio, J. (2007). Analytical methods for determination of trans fatty acid content in food. European Journal of Lipid Science & Technology, 109, 901-917.

Kuiper-Goodman, T. (2004). Risk assessment and risk management of mycotoxins in food. In N. Magnan, & M. Olsen, Mycotoxins in food: Detection and control (pp. 3-26). St. Paul, MN: Woodhead Publishing Ltd and CRC Press LLC.

Li-Chan, E. C., Griffiths, P. R., & Chalmers, J. M. (2010a). Applications of Vibrational Spectroscopy in Food Science; Volume 1: Instrumentation and Fundamental Applications (pp.4-80). Chichester, UK: John Wiley & Sons, Ltd.

Li-Chan, E. C., Griffiths, P. R., & Chalmers, J. M. (2010b). Applications of Vibrational Spectroscopy in Food Science; Volume 2: Analysis of Food, Drink and Related Materials (pp.349-425). Chichester, UK: John Wiley & Sons, Ltd.

Lin, M., He, L., Awika, J., Yang, L., Ledoux, D. R., Li, H., & Mustapha, A. (2008). Detection of melamine in gluten, chicken feed, and processed foods using surface enhanced Raman spectroscopy and HPLC. Journal of Food Science, 73(8), T129- 134.

MacDonald, B. F., & Prebble, K. A. (1993). Some applications of near-infrared reflectance analysis in the pharmaceutical industry. Journal of Pharmaceutical and Biomed Analysis, 11(11-12), 1077-1085.

Maga, J. A. (1979). Furans in foods. CRC Critical Reviews in Food Science & Nutrition, 11(4), 355-400.

Mauer, L. J., Chernyshova, A. A., Hiatt, A., Deering, A., & Davis, R. (2009). Melamine detection in infant formula powder using near- and mid-infrared spectroscopy. Journal of Agriculture and Food Chemistry, 57(10), 3974-3980.

McClure, W. F. (2007). Introduction. In Y. Ozaki, W. F. McClure, & A. A. Christy (Eds.), Near-Infrared Spectroscopy in Food Science and Technology (pp.1-4). Hoboken, New Jersey: John Wiley & Sons, Inc.

Murphy, P. A., Hendrick, S., Landgren, C., & Bryant, C.M. (2006). Food Mycotoxins: An Update. Journal of Food Science, 71(5), R51-R65.

Naumann, A., Navarro-Gonzalez, M., Peddireddi, S., Kues, U., & Polle, A. (2005). Fourier transform infrared microscopy and imaging: Detection of fungi in wood. Fungal Genetics and Biology, 42(10), 829-835.

Pedreschi, F., Segtnan, V. H., & Knutsen, S. H. (2010). On-line monitoring of fat, dry matter and acrylamide contents in potato chips using near infrared interactance and visual reflectance imaging. Food Chemistry, 121(2), 616-620.

Penner, M. H. (1998). Basic Principals of Spectroscopy. In S. S. Nielson (Ed.), Food Analysis (2nd Ed.) (pp. 387-398). Gaithersberg, Maryland: Aspen Publishers, Inc.

Perez Locas, C., & Yaylayan, V. A. (2004). Origin and mechanistic pathways of formation of the parent furan--a food toxicant. Journal of Agriculture and Food Chemistry, 52(22), 6830-6836.

Perez-Marin, D., Paz, P., Guerrero, J. E., Garrido-Varo, A., & Sanchez, M. T. (2010). Miniature handheld NIR sensor for the on-site non-destructive assessment of post- harvest quality and refrigerated storage behavior in plums. Journal of Food Engineering, 99(3), 294-302.

PerkinElmer Technical Note. (2005). FT-IR Spectroscopy: Attenuated Total Reflectance (ATR). [WWW page]. URL http://shop.perkinelmer.com/content/technicalinfo/tch_ftiratr.pdf.

Petersen, D. W., Kleinow, K. M., Kraska, R. C., & Lech, J. J. (1985). Uptake, disposition, and elimination of acrylamide in rainbow trout. Toxicology and Applied Pharmacology, 80(1), 58-65.

Polshin, E., Lammertyn, J., & Nicolaї, B. M. (2010). In E. C. Li-Chan P. R. Griffiths, & J. M. Chalmers (Eds.), Applications of Vibrational Spectroscopy in Food Science; Volume 2: Analysis of Food, Drink and Related Materials (pp.349-352). Chichester, UK: John Wiley & Sons, Ltd.

Regan, F., Meaney, M., Vos, J. G., MacCraith, B. D., & Walsh, J. E. (1996). Determination of pesticides in water using ATR-FTIR spectroscopy on PVC/chloroparaffin coatings. Analytica Chimica Acta, 334(1-2), 85-92.

Rein, A. (2008). Handheld FT-IR spectrometers: Bringing the spectrometer to the sample. Supplement: FT-IR technology for today’s spectroscopists. Spectrocopy, 44-50.

Roach, J. A., Andrzejewski, D., Gay, M. L., Nortrup, D., & Musser, S. M. (2003). Rugged LC-MS/MS survey analysis for acrylamide in foods. Journal of Agriculture and Food Chemistry, 51(26), 7547-7554.

Rodriguez-Otero, J. L., Hermida, M., & Cepeda, A. (1995). Determination of fat, protein, and total solids in cheese by near-infrared reflectance spectroscopy. Journal of the American Oil Chemists Society International, 78(3), 802-806.

Rodriguez-Saona, L. E., Fry, F. S., McLaughlin, M. A., & Calvey, E. M. (2001). Rapid Analysis Of sugars in fruit juice by FT-NIR Spectroscopy. Carbohydrate Research, 336, 63-74.

Segtnan, V. H., Kita, A., Mielnik, M., Jorgensen, K., & Knutsen, S. H. (2006). Screening of acrylamide contents in potato crisps using process variable settings and near- infrared spectroscopy. Molecular Nutrition & Food Research, 50(9), 811-817.

Sorgel, F., Weissenbacher, R., Kinzig-Schippers, M., Hofmann, A., Illauer, M., Skott, A., & Landersdorfer, C. (2002). Acrylamide: increased concentrations in homemade food and first evidence of its variable absorption from food, variable metabolism and placental and breast milk transfer in humans. Chemotherapy, 48(6), 267-274.

Thermo Fisher Scientific, Inc. (2011). Thermo Scienctific TruDefender FT: Handheld FTIR for Unknown Chemical and Explosives Identification. [WWW page]. URL http://www.ahurascientific.com/download/pdf/TruDefender_FTi_SpecSheet_TM O_5_18_11.pdf van de Voort, F.R., Sedman, J., & Russin, T. (2001). Lipid Analysis by vibrational spectroscopy. European Journal of Lipid Science & Technology, 103(12), 815- 26.

Wehling, R. L. (1998). Infrared Spectroscopy. In S. S. Nielson (Ed.), Food Analysis (2nd Ed.) (pp. 413-424). Gaithersberg, Maryland: Aspen Publishers, Inc.

Wei, F., Lam, R., Cheng, S., Lu, S., Ho, D., & Li, N. (2010). Rapid detection of melamine in whole milk mediated by unmodified gold nanoparticles. Applied Physics Letters, 96(13), 133702.

Williams, P. (2010). The Analysis of Wheat by Near-Infrared Spectroscopy. In E. C. Li- Chan P. R. Griffiths, & J. M. Chalmers (Eds.), Applications of Vibrational Spectroscopy in Food Science; Volume 2: Analysis of Food, Drink and Related Materials (pp.349-352). Chichester, UK: John Wiley & Sons, Ltd.

Yan, N., Zhou, L., Zhu, Z., & Chen, X. (2009). Determination of melamine in dairy products, fish feed, and fish by capillary zone electrophoresis with diode array detection. Journal of Agriculture and Food Chemistry, 57(3), 807-811.

Zotti, M., Ferroni, A., & Calvini, P. (2008). Microfungal biodeterioration of historic paper: Preliminary FTIR and microbiological analyses. International Biodeterioration & Biodegradation, 62(2), 186-194.

Chapter 2: Characterization and Authentication of a Novel Vegetable Source of Omega-3

Fatty Acids, Sacha Inchi (Plukenetia volubilis L.) Oil

Natalie E. Maurera, Beatriz Hatta-Sakodab, Gloria Pascual-Chagmanb and

Luis E. Rodriguez-Saonaa

aDepartment of Food Science and Technology, The Ohio State University

2015 Fyffe Court, Columbus, OH 43210, USA. bFacultad de Industrias Alimentarias, Universidad Nacional Agraria La Molina, Av. La

Molina s/n. La Molina. Lima - Perú.

2.1 Abstract

Consumption of omega-3 fatty acids (ω-3’s), whether from fish oils, flax or supplements, can protect against cardiovascular disease. Finding plant-based sources of the essential ω-

3’s could provide a sustainable, renewable and inexpensive source of ω-3’s, compared to fish oils. Our objective was to develop a rapid test to characterize and detect adulteration in sacha inchi oils, a Peruvian seed containing higher levels of ω-3’s in comparison to other oleaginous seeds. A temperature-controlled ZnSe ATR mid-infrared benchtop and diamond ATR mid-infrared portable handheld spectrometers were used to characterize sacha inchi oil and evaluate its oxidative stability compared to commercial oils. Soft independent model of class analogy (SIMCA) and partial least squares regression (PLSR) analyzed the spectral data. Fatty acid profiles showed that sacha inchi oil (44% linolenic acid) had similar levels of PUFA as flax oils. PLSR showed good correlation coefficients

(R2>0.9) between reference tests and spectra from infrared devices, allowing for rapid determination of fatty acid composition and predicting oxidative stability. Oils formed distinct clusters allowing the evaluation of commercial sacha inchi oils from Peruvian markets and showed some prevalence of adulteration. Determining oil adulteration and quality parameters using the ATR-MIR portable handheld spectrometer allowed for portability and ease-of-use, making it a great alternative to traditional testing methods.

Keyword: FT-IR spectroscopy; ω-3; sacha inchi oil; pattern recognition analysis; adulteration

2.2 Introduction

Native to the Peruvian jungles, sacha inchi (Plukenetia volubilis L.), also known as the

“Inca peanut” or “wild peanut”, grows at altitudes between 200 and 1500 m (Follegatti-

Romero, Piantino, Grimaldi, & Cabral, 2009) and is a plant of the Euphorbiaceous family

(Hamaker et al., 1992). The flour and oil from the seeds are commonly used among the

Peruvian natives. The seeds contain approximately, on average, 48% oil and 27% proteins that are rich in cysteine, tyrosine, threonine and tryptophan (Guillén, Ruiz, Cabo,

Chirinos, & Pascual, 2003). Sacha inchi was highest in oil content between soybean and cottonseed, but comparable to sunflower and peanut (Hamaker et al., 1992). Sacha inchi seeds have a unique fatty acid composition containing a large amount of unsaturated fatty acids (about 85% polyunsaturation), comprising of approximately 34% linoleic acid (ω-

6) and 51% linolenic acid (ω-3) (Guillén, Ruiz, Cabo, Chirinos, & Pascual, 2003). These essential ω-6 and ω-3 fatty acids offer important health and nutritional benefits such as providing protection against cardiovascular disease (Guillén, Ruiz, Cabo, Chirinos, &

Pascual, 2003), which is the number one killer disease in the United States (CDC, 2010).

They also protect against rheumatoid arthritis, cancer and possibly the severity of viral infections (Fernandes & Venkatraman, 1993).

The primary source of the essential polyunsaturated fatty acids (PUFA’s) comes from fish oils which are rich in these beneficial fatty acids (Venegas-Caleron, Sayanova &

Napier, 2010). Fish oil contains ω-6 (0.9-12 g/100g oil) and ω-3 (11.9-35.3 g/100g oil)

(Rubio-Rodriguez et al., 2010) fatty acids. Production and processing of fish oils is a

30 costly manufacturing process aimed at removing color pigments, contaminants (dioxins, furans and or polyaromatic hydrocarbons) and volatile components responsible for the oil’s odor and flavor (Bimbo, 2011). There is concern regarding the sustainability of fish due to its decreasing population from decades of over-fishing (Pauly, Watson & Adler,

2005). Environmental pollution has resulted in the accumulation of dioxins and heavy metals in fish and must be tested to rule out dangerous levels of pollutants (Yokoo et al.,

2003). Due to potential contamination, the benefits of obtaining PUFA’s from fish are being questioned (Yokoo et al., 2003). It is cheaper and easier to remove oil from flax seed and is a renewable material whereas fish is a diminishing source. Nevertheless, α- linolenic acid (ALA) needs be converted to produce docosahexaenoic acid (DHA) and eicosapentaenoic acid (EPA) from flaxseed oil, which is an inefficient process (Huang,

Pereira, & Leonard, 2004).

Due to the nutritional benefits of oils containing high levels of ω-3’s, the potential for adulteration by unscrupulous manufacturers for economic gain (by using low cost ingredients for the replacement, partially or totally, of high-cost ingredients) is a concern

(Tay, Singh, Krishnan & Gore, 2002). Adulteration could threat the health of consumers so detecting adulteration is a priority. Adulteration of oils is becoming more common due to its economic profits, therefore, a more sophisticated and rapid method of detecting adulteration is needed compared to the traditional chromatographic methods (Gurdeniz &

Ozen 2009). The development of a robust and reliable system combined with multivariate analysis to monitor chemical processes occurring during lipid oxidation, detect

31 adulteration and determine fatty acid composition is of great importance, both in terms of public health and consumer protection.

ATR-MIR benchtop and portable handheld spectrometers are relatively recent applications of spectroscopy when combined with chemometric analysis for determining authenticity of oils. Fourier-transform infrared (FT-IR) spectroscopy equipped with a temperature controlled ZnSe-ATR (attenuated total reflectance) accessory is a great alternative for detecting adulteration compared to traditional methodologies including chromatographic methods (Gurdeniz & Ozen, 2009).

The objective of this study was to develop a rapid test combining ATR-MIR spectroscopy with chemometrics to characterize and detect adulteration in sacha inchi oils. In addition the oxidative stability performance of sacha inchi oils was compared with flax, high oleic sunflower and corn oils.

2.3 Materials and Methods

Vegetable oils (3 of each; corn, canola, flax, cottonseed, sunflower, high oleic sunflower and olive) were purchased from a local grocery stores in Columbus, OH. Nine sacha inchi samples (numbers 1-9) were supplied from UNALM, Peru.

2.3.1 Reference Methods

Peroxide value (PV) was determined using the AOCS official method Cd 8-53 (Firestone,

1998).

Free fatty acid (FFA) value was determined using the AOCS official method Ca 5a-40

(AOCS, 1993).

Methyl ester fatty acid analysis was performed to determine the fatty acid composition of the oils using a fatty acid methyl ester (FAME) procedure (Firestone, 1995; Rodriguez-

Saona, Barrett, Selivonchick, 1995). Esterification was achieved by adding 10mL of 4% methanolic-sulfuric acid to 0.5g oil samples plus 1 mL benzene in a glass tube with a teflon screw top cap. Methyl esters were extracted using a partition of 2 mL hexane and 1 mL distilled water. 1 mL of the hexane portion was collected in a 1.5 mL vial and evaporated under nitrogen. The samples were re-diluted using 0.5 mL iso-octane. Methyl esters were analyzed on a Agilent 6890 Series (Santa Clara, CA) GC equipped with a flame ionization detector (FID). An HP G1513A autosampler and tray were used to automate the injections. Separation of the components was done using an HP-FFAP 25 m x 0.32 mm x 0.5μm column using helium as the gas carrier. The injection volume was 1

μL with a split ratio of 20:1. The oven conditions were 110 oC (1 min), to 220 oC (5 oC/min) hold for 15 min. The injector temperature was 220 oC and the detector temperature was 250 oC. Fatty acids were identified by comparing the retention times against reference standards (NuChek Prep GC standard 15A, Nu-Chek Prep, Inc.,

Elysian, MN).

Oil samples were analyzed individually in duplicate and statistical significance of the differences between mean values was assessed by one-way ANOVA with Minitab® 14.0 statistical software (Minitab Inc., State College PA). The Tukey multiple comparison

33 procedure was used to compare the means for each reference test. A probability to p≤

0.05 was used to establish the statistical significance.

2.3.2 Monitoring Oxidative Stability

To prepare the accelerated oxidative stability conditions, 50 mL of each oil (corn, high oleic sunflower, flax and sacha inchi) were measured into ten glass jars with lids (total of

40 jars, 2 for each oil each day). Uncovered jars were heated to 65 ± 1 oC in an incubator to accelerate oxidative conditions. The jars were left uncovered to expose the oils to the circulating air and to facilitate oxidation. Every 5 days for 20 days, a jar of each oil was removed from the incubator and stored at -40 oC until analysis. PV (AOCS official method Cd 8-53), FFA (AOCS official method Ca 5a-40) and methyl ester fatty acid analysis (Firestone, 1995; Rodriguez-Saona, Barrett, Selivonchick, 1995) reference methods were used as well as ATR-MIR benchtop and portable handheld spectra collection with multivariate analysis to analyze the data.

2.3.3 FT-IR Spectroscopy

Infrared spectral data was collected on an ATR-MIR benchtop and portable handheld spectrometers. Oil samples were heated in an oven to 65 oC prior to data collection for each spectrometer and measured in duplicate. The TruDefenderTM FT infrared portable handheld spectrometer (Thermo Scientific), which was originally designed for field identification of unknown materials, is equipped with a diamond ATR crystal. The handheld unit had a spectral range of 4,000-650 cm-1 and spectral data was collected by

34 co-adding 4 scans at a resolution of 4 cm-1. Infrared spectra were also collected on a benchtop Varian Excalibur 3100 spectrometer (Varian, Palo Alto, CA, USA) equipped with a KBr beam splitter and deuterated triglycine sulfate (DTGS) detector. A single- bounce FatIRTM temperature controlled ATR ZnSe crystal (Harrick Scientific,

Pleasantville, NY, USA) was set to 65 oC. Spectra were collected over a range of 4,000-

700 cm-1 at 4 cm-1 resolution and an interferogram of 64 scans were co-added. Spectral data was displayed in terms of absorbance and viewed using Resolutions Pro Software.

The instrument was purged continuously with CO2 from a CO2RP140 dryer (Dominic

Hunter, Charlotte, NC, USA) to prevent interference in the spectra.

2.3.4 Data Analysis

The spectra were analyzed using multivariate statistical analysis software (Pirouette® version 4.0, Infometrix Inc., Woodville, WA, USA). The spectra were imported into the software from the instruments as GRAMS (.spc) files and analyzed by normalizing, mean-centering and taking the second derivative for each spectra. Soft independent modeling of class analogy algorithm (SIMCA) was used to classify the oil samples based on their type. SIMCA’s discriminating power plot was used to identify which infrared bands were responsible for classification. Second derivative partial least squares regression (PLSR) models with cross-validation were used to study the correlation between the infrared spectrum of oil and PV, FFA values and fatty acid composition.

Models were cross-validated using a leave-one-out approach.

In SIMCA, the principal components contain information about influential chemical components that define the classes. By determining the F-statistic, an upper limit for residual variance (noise) can be calculated for all samples belonging to each class, resulting in a set of probabilities of class-membership for each sample (boundary cloud).

Thus, an unknown sample can only be assigned to the class for which it has a high probability. If the residual variance of a sample exceeds the upper limit for the modeled classes in the data set, it is not assigned to any of the classes; either it is an outlier, or it comes from a class not represented in the data set (Lavine, 2000). PLSR is a bi-linear regression based on the extraction of latent variables (Bjorsvik & Martens, 1992). These orthogonal factors (latent variables) explain most of the covariance of the X (spectra) and

Y variables. PLSR reduces the dimensions contained in thousands of IR predictors into a few factors to explain variations in both the dependent variables and the spectral domain.

The ideal end-result is a linear model able to predict a desired characteristic based on a selected set of predictors (IR spectra). PLSR has been particularly successful in developing multivariate calibration models for the spectroscopy field because it reduces the impact of irrelevant X-variations (noise) in the calibration model (Bjorsvik &

Martens, 1992; Martens & Naes, 1989). This capability provides a more information-rich data set of reduced dimensionality and eliminates data noise that results in more accurate and reproducible calibration models. The quality of the final model was evaluated based on the number of latent variables, loading vectors, standard error of cross validation

(SECV), coefficient of determination (R-value), and outlier diagnostics. Sample residual and Mahalanobis distance were used to determine outliers.

2.4 Results and Discussion

2.4.1 Characterization and Authentication of Sacha Inchi Oils

The ATR-MIR spectrum of several commercial vegetable oils and sacha inchi

(Plukenetia volubilis L.) (Figure 2.1) oils provided unique information about different triglyceride substitution patterns due to degree and form of unsaturation of the acyl groups and their fatty acid chain length (Guillén & Cabo, 1997).

Figure 2.1. FT-IR mid-infrared spectra of high oleic sunflower, canola, corn, sunflower, cottonseed, olive, sacha inchi and flax oil using a temperature-controlled single-bounce ZnSe ATR mid-IR. 37

Major spectral differences were evident in the 3050-2800 cm-1 region associated with the stretching vibration of the cis olefinic =C-H double bonds (3010 cm-1) and the methylene asymmetrical and symmetrical stretching vibrations (2950 and 2845 cm-1). In addition, flax and sacha inchi oils showed unique spectral bands in the 1120-1000 cm-1 range and

725 cm-1 which resulted from the stretching vibrations of the –C-O groups of esters derived from primary and secondary alcohols and the rocking vibrations of methylene group overlapping with bending vibration on out-of-plane mode of alkenes with cis- disubstituted olefinic groups, respectively (Guillén and Cabo, 1997). Flax and sacha inchi oils showed a more intense peak at 3010 cm-1 due to their higher degree of polyunsaturated fatty acids. Flax and sacha inchi oils contained approximately 68% (53% linolenic acid (ω-3) and 15.5% of linoleic acid (ω-6)) and 78% (44% linolenic acid and

33.5% of linoleic acid) polyunsaturated fatty acids (Table 2.1), respectively, which is in good agreement with levels reported by Guillén, Ruiz, Cabo, Chirinos, & Pascual,

(2003). On the contrary, olive and high oleic sunflower oils showed the least intense band at 3010 cm-1 due to their predominance of oleic acid (~76%).

Table 2.1. Fatty acid composition for commercial vegetable oils (olive, canola, cottonseed, corn, high oleic sunflower, sunflower, flax and sacha inchi) and pure authentic sacha inchi oils using fatty acid methyl ester (FAME) procedure Fatty Acid (%)a Oil Sample Palmitic Stearic Oleic Linoleic Linolenic Olive 11.8 (+/-0.7) 2.8 (+/-0.06) 74.3 (+/-2.1) 8.4 (+/-1.0) 0.6 (+/-0.02) Canola 5.2 (+/-0.09) 2.3 (+/-0.03) 64.1 (+/-0.3) 20.2 (+/-0.3) 6.6 (+/-0.2) Cottonseed 22.4 (+/-0.9) 2.8 (+/-0.05) 18.4 (+/-0.5) 52.8 (+/-0.7) 0.5 (+/-0.03) Corn 11.2 (+/-0.5) 2 (+/-0.09) 28.5 (+/-0.5) 56 (+/-0.2) 0.9 (+/-0.1) High Oleic Sunflower 5.1 (+/-0.05) 2.7 (+/-0.05) 78.6 (+/-0.03) 11.6 (+/-0.06) 0.1 (+/-0.001)

Sunflower 4.7 (+/-0.05) 3.8 (+/-0.05) 60.2 (+/-0.03) 29 (+/-0.05) 0.37 (+/-0.0007) Flax 5.6 (+/- 0.2) 4.4 (+/-0.05) 20 (+/-0.8) 15.5 (+/-0.5) 53.4 (+/-0.007) Authentic Sacha Inchi 4.67 (+/- 0.3) 3.5 (+/- 0.1) 10.7 (+/- 0.6) 33.5 (+/- 1.0) 44 (+/- 1.3) Commercial Sacha Inchi Oilsb 4.8 (+/-0.3) 3.4 (+/-0.1) 10.1 (+/-0.5) 37.7 (+/-1.6) 42.4 (+/-1.3) Suspicious Sacha Inchi samples based on information from the infrared systemsc 2 12.6 3.8 18.8 53.5 7.1 4 4.5 3.2 11.0 37.2 42.3 9 9.7 4.1 19.6 49.6 14.5 1

a Relative % of fatty acid based on GC chromatographic peak area b Average of sacha inchi samples (1, 3, & 4-8) that were verified as authentic by the temperature controlled ZnSe ATR-IR and portable handheld diamond ATR-IR. c Sacha inchi samples that were not clusterd as authentic by either the temperature controlled ZnSe ATR-IR or portable handheld diamond ATR-IR.

FT-IR ATR combined with a multivariate analysis has been successfully implemented as

a reliable, quick and simple technique for analyzing food products, edible oils and lipids

(Birkel & Rodriguez-Saona 2011). Soft independent modeling of class analogy (SIMCA)

is a classification procedure based on principal component analysis (PCA) that groups

samples into their classes based on differences in their composition (infrared bands)

(Allendorf, Subramanian, & Rodriguez-Saona, 2011). A class space is built whose

boundaries discriminate between the samples fitting the class model and the samples that

cannot be considered as belonging to the studied class.

Pattern recognition analysis by the SIMCA method using the spectral data collected from a mid-infrared single-bounce ZnSe ATR-MIR benchtop (Figure 2.2A) and a diamond

ATR-MIR portable handheld system (Figure 2.2C) showed well-separated clustering for the different oils.

SIMCA reduces the dimensionality in multivariate data sets so that systematic variation can be separated for each class and the residual variance (noise) is used to define the class boundaries (Vogt & Knutsen, 1985) around each class model. Thus, objects with residual standard deviation above the critical value (95% confidence interval) are considered as outliers or identified as samples not belonging to a group (Vogt & Knutsen, 1985).

Models generated with the ATR-MIR benchtop and ATR-MIR portable handheld spectrometers had interclass distances (ICD) ranging from 8.4 to 141 and 1.66 to 23.6, respectively, with flax and sacha inchi oils having the lowest ICD in both models. An

ICD greater than 3 is used to determine if classes are significantly different among each other and the greater this ICD the greater the difference in their chemical composition

(Allendorf, Subramanian, & Rodriguez-Saona, 2011). The SIMCA discriminating power plots explained the differences in functional groups responsible for the separation of classes. The higher intensity of the bands, the greater its influence in discriminating the oil samples. Figure 2.2B shows that most of the variance in the ATR-IR benchtop model was explained with bands at 1114 cm-1 corresponding to asymmetric –C-O ester stretching vibrations. The ATR-MIR portable handheld (Figure 2.2D) had a peak in the same region (1104 cm-1) also corresponding to differences in esters derived from secondary alcohols present in the triglyceride molecule. In addition, both SIMCA models

-1 showed the influence of bands between 2922-3016 cm corresponding to –CH2 asymmetric stretching from differences in the fatty acid chain length between the oils and the cis- olefinic group =CH- assignable to the degree of polyunsaturated fatty acids, respectively.

The health and nutritional importance of the ω-3 polyunsaturated acyl groups provide protection against cardiovascular disease, rheumatoid arthritis and cancer (Fernandes &

Venkatraman, 1993), creating consumer demands for foods supplemented with ω-3 acyl groups. Sales of omega-3 products are expected to reach $8.2 billion by 2012 (Packaged

Facts, 2009) and 63% of consumers are trying to add omega-3 fatty acids into their diets.

Consequently, adulteration of high-cost ingredients (ω-3 polyunsaturated rich oils) with lower grade, cheaper substitutes could potentially lead to economic fraud and serious health implications to consumers (Lai, Kemsley, & Wilson, 1994). Our SIMCA classification using a set of pure sacha inchi oils, provided by the Universidad Nacional

Agraria (Lima, Peru), was used to authenticate commercial sacha inchi oils purchased from local markets in Lima (Peru). The SIMCA plots generated using the spectra collected from the ATR-MIR benchtop (Figure 2.2A) and portable handheld (Figure

2.2C) spectrometers were used to predict authenticity of 9 commercial sacha inchi oil samples. Predictions made from the ATR-MIR benchtop (Figure 2.3A) and portable handheld (Figure 2.3B) SIMCA models showed that samples #2 and #9 were clustered away from the pure sacha inchi oil class. Data from the GC methyl ester fatty acid composition reference method showed that sample #2 and #9 contained higher linoleic

(50 - 54%) and lower linolenic (7 - 14.5%) acid as compared to pure sacha inchi samples

(Table 2.1). Therefore, this data revealed that samples #2 and #9 are not authentic and likely have been adulterated with another vegetable oil. Interestingly, the same manufacturer made these two oils. The ATR-MIR portable handheld SIMCA prediction plot also showed that sample #4 lied outside the typical sacha inchi class boundary. The

42 fatty acid composition reference method showed that this sample had similar levels of polyunsaturated fatty acids to the pure sacha inchi samples. This false positive prediction

(non-authentic) by the ATR-MIR portable handheld spectrometer could be attributed to measurement uncertainty due to environmental or sampling factors, leading to increased noise and therefore, the false positive.

SI # 9 A SI # 9 B Corn Corn SI # 2 SI # 2 PC2 PC2

Cottonseed Sunflower SI # 1, 3, 5-8 Sunflower Cottonseed High Oleic PC1 PC1 Sunflower SI # Canola High Oleic Sunflower Canola SI # 4 1, 3-8 PC3 PC3 Olive Olive

Flax Flax

Figure 2.3. SIMCA prediction plots for authentication of sacha inchi oils using the mid- IR ATR benchtop (A) and mid-IR ATR portable handheld (B) spectrometers.

Partial least squares regression models were created by combining the infrared spectral data with the quantitative measurements obtained from the GC methyl esters fatty acid composition, peroxide value (PV) and free fatty acid (FFA) reference methods. The power of the PLSR method is based on its ability to mathematically correlate spectral changes to changes in the concentration of a component of interest (van de Voort, Ismail,

Sedman, Dubois, & Nicodemo, 1994). PLSR plots generated for linolenic (ω-3)

(Figure 2.4A & 2.4B), linoleic (ω-6) (Figure 2.4C & 2.4D) and oleic (ω-9) (Figure

2.4E & 2.4F) acids showed good correlation between the fatty acid levels and the estimated concentrations using the ATR-MIR systems. Table 2.2 shows that the regressions developed from using the spectral data obtained from the ATR-MIR benchtop had a lower SECV (0.9-2.0 %) in predicting %fatty acid composition than the ATR-MIR portable handheld spectrometer (SECV 1.7-2.7). Leave-one-out cross validation method was used for selection of the optimal number of factors in the PLSR models, ranging from 5-9, depending on the infrared system used to collect the spectrum. Overall, the benchtop system used more latent variables (factors) due to its higher signal-to-noise ratio, extracting more statistically significant variation contributions associated with combinations of different chemical components.

Table 2.2. PLSR model statistical analysis for determining fatty acid composition in oils (corn, high oleic sunflower, flax and sacha inchi) using a benchtop and hendheld infrered system.

MIR Concentration range (%)a Factors SECVc r Vald Technique ( IR regions (cm-1) )b

9.6 - 76.6 Temperature-Controlled Oleic (940-1860; 2770-3100) 9 2.0 1.00 Single-bounce ZnSe 7.4-56.0 Linoleic (920-1860; 2780-3090) 8 1.07 1.00 ATR- IR 6.5 - 56.5 Linolenic (950-1820; 2780-3100) 9 0.89 1.00 10.6-78.6 Portable Handheld Unit Oleic (940-1855; 2775-3080) 7 2.7 0.99 Single-bounce diamond 8.4-56 Linoleic (920-1870; 2770-3090) 7 2.65 0.99 ATR-IR 0.1-53.4 Linolenic (950-1830; 2760-3100) 5 1.71 1.00

a Relative % of fatty acid based on GC chromatographic peak area b Selected frequencies (cm-1) regions used to build the PLSR model c Standard Error of Cross Validation d Correlation of Cross Validation Model

Figure 2.4. Partial least squares regression (PLSR) of free fatty acids using the mid-IR ATR benchtop (A- linolenic acid, C- linoleic acid, E- oleic acid) and the mid-IR ATR portable handheld spectrometers (B- linolenic acid, D- linoleic acid, F- oleic acid).

2.4.2 Monitoring Oxidative Stability

The high content of unsaturated fatty acids makes the sacha inchi oils very susceptible to peroxidation under mild environmental conditions (Follegatti-Romero, Piantino,

Grimaldo, & Cabral, 2009). Exposure of oils (corn, high oleic sunflower, flax and sacha inchi) to 65 oC for 20 days resulted in a marked increased in peroxide value (Table 2.3).

Composition differences in the oils played a significant (p-value < 0.05) role in oxidation rates. Sunflower oil showed the highest oxidative stability while corn had high levels of peroxide formation at the end of 20 days (Table 2.3). Interestingly, corn oil showed the least oxidation in the first five days compared to the other oils, as shown in Figure 2.5.

This may be attributed to the presence of natural antioxidants, α-tocopherols, in corn oil that have been reported to inhibit the formation of hydroperoxides, but once the tocopherols were depleted the oxidation process in corn oil accelerated rapidly (Huang,

Frankel, & German, 1994). The rate of peroxide formation in high oleic sunflower and corn (after an induction period) oils increased linearly over the storage time as the rate of peroxide formation equals its decomposition during propagation reactions in the thermal lipid oxidation process. Sacha inchi and flax oils showed exponential trends in their peroxide value reaching a plateau after 15 days (Figure 2.5) as peroxide decomposition exceeds the rate of formation, resulting in decreasing peroxide values (Orlein, Risbo,

Rantanen, & Skibsted, 2006). Overall, sacha inchi and flax oils were the most oxidatively unstable due to their large amounts of polyunsaturated fatty acids making them more susceptible to oxidation. Of the four oils compared in this study, high oleic sunflower oil had the largest amount of saturated and monounsaturated fatty acids making it the most

47 stable. The level of FFA found in the oils and rates of formation also varied according to the type of oil. In commercial practices, the oil is usually discarded after the FFA level reaches 1-1.5%, as there will be sufficient breakdown material present to rapidly catalyze further oxidation (Tseng, Moreira, & Sun, 1996). FFA values did not significantly change after the first 5 days, thus, using FFA values alone for monitoring oxidation was not sufficient. Overall, there were not significant (p-value > 0.05) changes in saturated, monosaturated and polyunsaturated fatty acids during the accelerated oxidative study, as measured by the fatty acid methyl ester analysis, except for polyunsaturated levels in flax oil (p-value 0.015).

120.0

100.0

80.0 Corn 60.0 High Oleic Sunflower 40.0 Flax 20.0 Sacha Inchi

0.0 Peroxide value (meq/Kg oil) value(meq/Kg Peroxide 0 10 20 30 Storage time at 65 C (Days)

Figure 2.5. Peroxide value (PV) of corn, high oleic sunflower, flax and sacha inchi oils versus the storage time (in days) in a 65 oC oven

Table 2.3. Peroxide value (PV), free fatty acid (FFA), and fatty acid composition for the different oil samples studied during a 20 day oxidative stability test. Storage (Days) Oil Reference test 0 5 10 15 20 PV 1 4.5 42.3 82.6 111.4 Corn (meq/kg oil) (+/-0.001) (+/-0.4) (+/-2.8) (+/-1.4) (+/-0.1) 0.3 0.5 0.65 0.67 0.7 FFA (%) (+/-0.001) (+/-0.01) (+/-0.02) (+/-0.04) (+/-0.01) Saturated fat 13.7 13.8 14.2 14.3 14.3 (g/100g oil)a (+/-0.4) (+/-0.1) (+/-0.04) (+/-0.2) (+/-0.6) Monounsaturated fat 30.5 30.4 30.8 31.2 31.2 (g/100g oil)b (+/-0.5) (+/-0.3) (+/-0.2) (+/-0.09) (+/-0.5) Polyunsaturated fat 54.7 55.4 54.8 54.4 54.5 (g/100g oil)c (+/-0.1) (+/-0.3) (+/-0.1) (+/-0.2) (+/-0.7)

PV 4 21.2 41.5 54.9 71.7 High (meq/kg oil) (+/- 0.1) (+/-1.1) (+/-1.0) (+/-1.4) (+/-2.1) Oleic 0.38 0.45 0.5 0.54 0.52 Sunflower FFA (%) (+/-0.001) (+/-0.01) (+/-0.01) (+/-0.001) (+/-0.001) Saturated fat 6.5 6.5 6.6 6.5 6.4 (g/100g oil)a (+/-0.001) (+/-0.07) (+/-0.01) (+/-0.05) (+/-0.04) Monounsaturated fat 77.9 78.2 78.3 78.3 78.1 (g/100g oil)b (+/-0.02) (+/-0.2) (+/-1.5) (+/-0.5) (+/-0.2) Polyunsaturated fat 15.4 15.1 13.6 14.7 14.8 (g/100g oil)c (+/-0.02) (+/-0.07) (+/-1.4) (+/-0.1) (+/-0.1)

PV 9.06 42.5 69.7 78.6 81.6 Flax (meq/kg oil) (+/-0.01) (+/-2.1) (+/-2.1) (+/-3.5) (+/-0.001) 0.65 0.85 0.94 0.95 ( 0.95 FFA (%) (+/-0.01) (+/-0.01) (+/-0.03) +/-0.01) (+/-0.001) Saturated fat 9.4 9.9 10.1 10.2 6 (g/100g oil)a (+/-0.3) (+/-0.1) (+/-0.05) (+/-0.001) (+/-0.001) Monounsaturated fat 22.1 24.4 25.1 26.2 23.9 (g/100g oil)b (+/-0.8) (+/-0.6) (+/-0.7) (+/-0.1) (+/-0.2) Polyunsaturated fat 64.8 64.1 63.4 63.3 62.9 (g/100g oil)c (+/-0.4) (+/-0.2) (+/-0.1) (+/-0.2) (+/-0.3)

Sacha PV 3.4 64.3 89.7 101.4 104.6 Inchi (meq/kg oil) (+/-0.01) (+/-1.5) (+/-2.2) (+/-2.3) (+/-1.4) 0.36 0.58 0.7 0.7 0.75 FFA (%) (+/-0.02) (+/-0.03) (+/-0.01) (+/-0.03) (+/-0.001) Saturated fat 7.4 7.9 8.2 8.1 8.3 (g/100g oil)a (+/-0.2) (+/-0.02) (+/-0.08) (+/0.001) (+/-0.05) Monounsaturated fat 10 10.1 10.6 11 12.5 (g/100g oil)b (+/-0.4) (+/-1.3) (+/-0.09) (+/-0.9) (+/-0.1) Polyunsaturated fat 80.9 80.2 78.9 80 78.3 (g/100g oil)c (+/-0.4) (+/-0.09) (+/-0.09) (+/-1.6) (+/-0.8)

(Continued)

Table 2.3: Continued a Saturated fat refers to relative percent GC chromotographic peak area for palmitic plus stearic fatty acids b Monounsaturated refers to relative percent GC chromotographic peak area for oleic acid c Polyunsaturated refers to the relative percent GC chromotographic peak area for linoleic plus linolenic fatty acids All concentrations were determined using the GC-FAME method as reported in Table 1

After 20 days of stressing the oils at 65 oC, SIMCA projection plots showed spectral differences within oil samples clustering according to storage day (0-20) (Figure 2.6).

The discriminating power plots showed changes at bands at 1705-1766 cm-1 (depending on the oil type) due to their thermal oxidation, which correlated to the carbonyl-stretching band from the acyl-linked hydrocarbon chains (Borchman & Sinha, 2002). All of the oils followed similar patterns in the SIMCA projections plots for spectra collected using the

ATR-MIR benchtop (Figure 2.6A) or portable handheld spectrometer (Figure 2.6B), with the former technique yielding greater interclass distances.

PLSR was used to correlate oxidative reference tests with spectral data from the ATR-

MIR benchtop and portable handheld spectrometers. Table 2.4 gives the PLSR model statistical analysis for determining PV and FFA values for each of the four oils used in the oxidative stability test (corn, high oleic sunflower, flax and sacha inchi). The regressions developed from the spectral data obtained from the ATR-MIR benchtop had a lower SECV in predicting both the PV and FFA value than the ATR-MIR portable handheld spectrometer. Rancidity becomes noticeable in oils with peroxide values above

20meq/kg as noted in Gulla, Waghray & Reddy, (2010), therefore the PV is reported in

two concentration ranges, providing lower SECV values using the lower range. In this

study, most oil were greater than 20meq/kg by day 5, with the exception of corn oil.

Therefore our study supports combining PV and FFA for monitoring oxidative stability.

Table 2.4. Statistical analyses for the PLSR models developed to determine the PV and FFA of oils (corn, high oleic sunflower, flax and sacha inchi) during their oxidation process using a mid-IR ATR benchtop and mid-IR ATR portable handheld spectrometers. MIR Concentration range (%)a Factors SECVc r Vald Technique ( IR regions (cm-1) )b

Temperature-Controlled PV 1-66 8 2.06 1.0 (820-1850; 2745-3080) Single-bounce ZnSe (meq/kg oil) 1-112 9 4.13 0.99 (825-1855; 2740-3075) ATR- IR FFA (%) 0.028-0.96 8 0.06 0.94

(965-1800; 2805-3070)

Portable Handheld Unit PV 1-66 6 4.97 0.97 (850-1880; 2730-3100) Single-bounce diamond (meq/kg oil) 1-112 6 10.20 0.96 (860-1885; 2730-3100) ATR-IR FFA (%) 0.028-0.96 7 0.13 0.94

(960-1800; 2790-3100)

1 a Relative % of fatty acid based on GC chromatographic peak area b Selected wavelength (λ) regions used to build the PLSR model c Standard Error of Cross Validation dCorrelation of Cross Validation Model

Flax A B PC2 PC2

Sacha Inchi Day 20 Flax Day 15 PC1 PC3 PC1 Day 10 PC3 High Oleic High Oleic Day 5 Sunflower Sunflower Day 0 Corn Corn

Sacha Inchi

2.5 Conclusion

A temperature controlled ZnSe ATR-MIR benchtop and diamond ATR-MIR portable handheld spectrometers have proven to be fast, reliable and rapid techniques to characterize and authenticate sacha inchi oil. The ATR-MIR portable handheld gave comparable spectral information as the ATR-MIR benchtop but there was a slight increase in noise. Sacha inchi oil (44%) has a similar ω-3 composition to flax (53%), but it had approximately twice as much ω-6 fatty acids. Combining FT-IR spectroscopy with chemometrics has proven to be a great alternative to traditional testing methods. SIMCA projection plots formed distinct clusters for each oil class making prediction about oil fatty acid composition easy. PLSR showed good correlation coefficients (R2>0.9) between reference tests and spectra from infrared devices, allowing for rapid determination of fatty acid composition and prediction of oxidative stability.

2.6 Acknowledgement

The authors would like to acknowledge the Ohio Agricultural Research and Development

Center (OARDC) for their financial support of this research. We would also like to thank

Thermo Scientific (formerly Ahura Scientific) for providing the portable handheld spectrometer and technical support.

2.7 References

AOCS (American Oil Chemists’ Society). (1993). In Official Methods and Recommended Practices by the American Oil Chemists’ Society. Method Ca5a- 40, Free Fatty Acid. Champaign, IL.

Bimbo, A.P. (2011). The Production and Processing of Marine Oils: The American Oil Chemists’ Society Lipid Library. [WWW page]. URL http://lipidlibrary.aocs.org/processing/marine/index.htm

Birkel, E., Rodriguez-Saona, L. (2011). Application of a Portable Handheld Infrared Spectrometer for Quantitation of trans Fat in Edible Oils. Journal of the American Oil Chemists Society, 88:1477-1483.

Bjorsvik, H.R., & Martens, H. (1992). Data analysis: calibration of NIR instruments by PLS regression. In: D. Burns, & E. Ciurczak (Eds.), Handbook of Near-Infrared Analysis (pp. 15-180). New York: Dekker.

Borchman, D., & Sinha, S. (2002). Determination of Products of Lipid Oxidation by Infrared Spectroscopy. In D. Armstrong, Oxidative Stress and Antioxidant Protocols, (Vol 186, pp. 25). Totowa, New Jersey: Humana Press Inc.

CDC (Center for Disease Control and Prevention). (2010). Center for Disease Control and Prevention: America’s Heart Disease Burden. [WWW page]. URL http://www.cdc.gov/heartdisease/facts.htm

Fernandes, G., & Venkatraman, J.T. (1993). Role of Omega-3 Fatty Acids in Health and Disease. Nutrition Research, 13, S19-S45.

Firestone, D. (1995). In Official Methods of Analysis of AOAC International, 16th Ed. Method 969.33, Oils and fats. Association of Official Analytical Chemists, Arlington, VA.

Firestone, D. (1998). In Official Methods and Recommended Practices of the American Oil Chemists’ Society, 5th Edition. Method Cd 8-53, Peroxide value. American Oil Chemists Society, Champaign, IL.

Follegatti-Romero, L.A., Piantino, C.R., Grimaldi, R., & Cabral, F.A. (2009). Supercritical CO2 extraction of omega-3 rich oil from Sacha Inchi (Plukenetia volubilis L.) seeds. Journal of Supercritical Fluids, 49, 323-329.

Guillén, M.D., & Cabo, N. (1997). Infrared Spectroscopy in the Study of Edible Oils and Fats. Journal of Science of Food and Agriculture, 75, 1-11.

Guillén, M.D., Ruiz, A., Cabo, N., Chirinos, R., & Pascual, G. (2003). Characterization of Sacha Inchi (Plukenetia volubilis L.) Oil by FTIR Spectroscopy and 1H NMR. Comparison with Linseed Oil. Journal of the American Oil Chemists Society, 80, 755-762.

Gulla, S., Waghray, K., & Reddy, U. (2010). Blending of Oils-does it Improve the Quality and Storage Stability, an Experimental Approach on Soybean and Palmolein Based Blends. American Journal of Food Technology, 5, 182-194.

Gurdeniz, G., & Ozen, B. (2009). Detection of adulteration of extra-virgin olive oil by chemometric analysis of mid-infrared spectral data. Food Chemistry, 116, 519- 525.

Hamaker, B.R., Valles, C., Gilman, R., Hardmeier, R.M., Clark, D., Garcia, H.H., Gonzales, A.E., Kohlstad, I., Castro, M., Valdivia, R., Rodriguez, T., & Lescano, M. (1992). Amino Acid and Fatty Acid Profiles of the Inca Peanut (Plukenetia volubilis). Cereal Chemistry, 69, 461-436.

Huang, S., Frankel, E.N., & German, J.B. (1994). Antioxidant Activity of α- and γ- Tocohperols in Bulk Oils and in Oil-in-Water Emulsions. Journal of Agricultural and Food Chemistry, 42, 2108-2114.

Huang, Y.S., Pereira, S.L., & Leonard, A.E. (2004). Enzymes for transgenic biosynthesis of long-chain polyunsaturated fatty acids. Biochimie, 86, 793-798.

Lai, Y.W., Kemsley, E.K., & Wilson, R.H. (1994). Potential of Fourier Transform Infrared Spectroscopy for the Authentication of Vegetable Oils. Journal of Agricultural and Food Chemistry, 42, 1154-1159.

Lavine, B.K. (2000). Clustering and Classification of Analytical Data. In: R.A. Meyers, (Eds.), Encyclopedia of Analytical Chemistry (pp. 9689-9710). New York: John Wiley & Sons Ltd.

Martens, H., & Naes, T. (1989). Models for Calibration. In: H. Martens, & T. Naes, (Eds.), Multivariate Calibration (pp. 73-112). Great Britain: John Wiley & Sons Ltd.

Orlein, V., Risbo, J., Rantanen, H., & Skibsted, L.H. (2006). Temperature-dependence of rate of oxidation of rapeseed oil encapsulated on a glassy food matrix. Food Chemistry, 94, 37-46.

Packaged Facts. (2009). Omega-3,-6, and -9 Fatty Acids: Trends in the Worldwide Food and Beverage Markets, 2nd Edition. [WWW page]. URL http://www.packagedfacts.com/sitemap/product.asp?productid=1903782

Pauly, D., Watson, R., & Adler, J. (2005). Global trends in world fisheries: impacts on marine ecosystems and food security. Philosophical Transactions of the Royal Society London B: Biological Sciences, 360, 5-12.

Rubio-Rodriguez, N., Beltran, S., Jaime, I., M. de Diego, S., Sanz, M.T., & Carallido, J.R. (2010). Production of omega-3 polyunsaturated fatty acid concentrates: A review. Innovative Food Science and Emerging Technologies, 11, 1-12.

Rodriguez-Saona, L.E., Barrett, D.M., & Selivonchick, D.P. (1995). Peroxidase and lipoxygenase influence on stability of polyunsaturated fatty acids in sweet corn (Zea mays L.) during frozen storage. Journal of Food Science, 60, 1041-44.

Tay, A., Singh, R.K., Krishnan, S.S., & Gore, J.P. (2002). Authentication of olive oil adulteration with vegetable oils using Fourier transform infrared spectroscopy. LWT- Food Science and Technology, 35, 99-103.

Tseng, Y., Moreira, R., & Sun, X. (1996). Total frying-use time effects on soybean-oil deterioration and on tortilla chip quality. International Journal of Food Science and Technology, 31, 287-294. van de Voort, F.R., Ismail, A.A., Sedman, J., Dubois, J., & Nicodemo, T. (1994). The determination of peroxide value by Fourier transform infrared spectroscopy. Journal of the American Oil Chemists Society, 71, 921-26.

Venegas-Caleron, M., Sayanova, O., & Napier J.A. (2010). An alternative to fish oils: Metabolic engineering of oil-seed crops to produce omega-3 long chain polyunsaturated fatty acids. Progress in Lipid Research, 49, 108-119.

Vogt, N.B., & Knutsen, H. (1985). SIMCA pattern recognition classification of five infauna taxonomic groups using non-polar compounds analyzed by high resolution gas chromatography. Marine Ecology Progress Series, 26, 145-156.

Yokoo, E.M., Valente, J.G., Grattan, L., Schmidy, S.L., Platt, I., & Silbergeld, E.K. (2003). Low Level Methylmercury exposure affects neuropsychological function in adults. Environmental Health: A Global Access Science Source, 2, 8.

Chapter 3: Rapid Assessment of Quality Parameters in Cocoa Butter using ATR-IR

Spectroscopy and Multivariate Analysis

Natalie E. Maurer1 and Luis Rodriguez-Saona1

1Department of Food Science and Technology, The Ohio State University

2015 Fyffe Court, Columbus, OH 43210, USA.

3.1 Abstract

The development of sensitive and robust screening tool(s) for assuring the quality of incoming raw materials would supplement the assurances provided by food manufacturer vendor auditing programs. Our aim was to evaluate the ability of attenuated total reflectance mid-infrared (ATR-IR) spectroscopy in combination with multivariate analysis as a screening tool for the diverse cocoa butter supply. Forty different cocoa butter samples encompassing an acceptable range of compositional variability for the chocolate industry were included. Cocoa butters were characterized for their melt characteristics (DSC Hardness), triacylglycerol content and fatty acid composition (GC-

FAME). Soft independent modeling of class analogy (SIMCA) and partial least squares regression (PLSR) were used for classification and quantification analysis. SIMCA classified all cocoa butters in distinct clusters in a 3-dimensional space but no sample clustering patterns were associated with melt characteristics. Spectral differences responsible for the separation of classes were attributed to stretching vibrations of the ester (–C=O) linkage (1660-1720 cm-1). PLSR models showed correlation coefficients >

0.93 and prediction errors (SECV) of 1.5 units for melt characteristics, 0.2-0.3% and 0.4-

Key Words: Cocoa butter, FT-IR spectroscopy, attenuated total reflectance, authenticity, multivariate analysis

3.2 Introduction

Cocoa butter is derived from the seeds of the fruit from the Theobroma cacao tree, which grows in three main regions of the world including West Africa, Latin America and

South East Asia, with Côte d’Ivoire, West Africa being the world’s largest producer

(Crews, 2002). Cocoa butter is considered the most important by-product of the cocoa bean due to its unique physical and chemical characteristics. These characteristics produce specific functional properties that are very useful in the food industry (Liendo,

Padilla, & Quintana, 1997). Cocoa butter is primarily used in the confectionary industry but also has a place in cosmetics and pharmaceutical creams (Crews, 2002). Cocoa butter is of special interest in the food industry because it is the continuous phase of chocolate and is responsible for the transport and suspension of all other chocolate constituents. It is responsible for soft texture, plasticity, easy diffusion of taste and flavor and unusual melting properties (Liendo, Padilla, & Quintana, 1997). Cocoa butter is brittle at room temperature and quickly melts at body temperature, (Lipp & Anklam, 1998a), which makes it very desirable to most consumers.

The unique melting behavior of cocoa butter can be attributed to its triacylglycerol

(TAG) composition. Triacylglycerol molecules make up the majority of cocoa butters composition (Lipp & Anklam, 1998a). Symmetrical TAGs make up approximately 90% of the TAG species in cocoa butter consisting primarily of 2-oleyl glycerides of palimitic,

59 stearic and oleic acid (POP, POS, SOS) (Lipp & Anklam, 1998a, Lipp et al., 2001).

These TAGs are also responsible for providing the unique polymorphic crystal structure of cocoa butter (Lipp & Anklam, 1998a). The behavior of chocolate during the manufacturing process and the texture of finished chocolate are both attributed to the physical properties of the cocoa butter. The natural variation of these physical properties is derived from differences in fatty acid and subsequent triacylglycerol composition resulting from differences in growing regions and/or conditions (Liendo, Padilla, &

Quintana, 1997). For example the cocoa butters from Africa (Ghana, Ivory Coast) contain a significantly lower amount of oleic acid than the cocoa butters from South America

(Ecuador, Brazil) (Lipp & Anklam, 1998a).

Both fatty acid composition and triglyceride structure affect the melting profile of cocoa butter. Hardness is an empirical assessment directly derived from the melting profile of cocoa butter and is a way to evaluate the solid state characteristics of cocoa butter. One of the most common ways to assess hardness is through the use of differential scanning calorimetry (DSC) (Chaiser & Dimick, 1989).

The determination of ingredient authenticity is an important issue facing the global food industry. Recent melamine adulteration of wheat flour and diary ingredients has brought much attention to the risks associated with global ingredient supply chains (Jackson,

2009). Chocolate, an incredibly popular snack worldwide, requires ingredients sourced from many different places around the world. Ingredient authenticity is an important concern for chocolate manufacturers just as it is for many other food manufacturers. The potential financial rewards for substitution of high valued raw materials with cheaper

60 ingredients can cause potential risks for food manufactures and thus consumers (Che

Man, Syahariza, Mirghani, Jinap, & Bakar, 2005). Chromatographic methods are the traditional methods of choice for identification and determination of alternative lipid materials in cocoa butter. Traditional chromatographic methods have included both high performance liquid chromatography (HPLC) and gas chromatography (GC) (Lipp &

Anklam, 1998b). These methods are time consuming, require the use of hazardous solvents and would not be amenable to screening the volumes of ingredients used by a global food manufacturer. Additionally, these methods don’t allow for reliable, rapid quantification of components of cocoa butter (Lipp & Anklam, 1998b). In order to more effectively deter fraudulent practices within global ingredient supply chain for cocoa butter, a quick and reliable analytical technique for the authentication of cocoa butter would be of great value. Cocoa butter has been also been examined as a case study for the application of this technical approach to other food ingredients. The development of a robust and reliable system combined with multivariate analysis in order to authenticate as well as quantify specific components of cocoa butter is important for the protection of food manufactures and consumers.

Fourier transform infrared (FT-IR) spectroscopy is a robust and rapid analytical method that measures the vibrations of bonds within functional groups and therefore, can give quantitative information about the total bio-chemical composition of a food sample without destroying it (Goodacre & Anklam 2001). FT-IR spectroscopy equipped with a temperature controlled ZnSe-ATR (attenuated total reflectance) accessory combined with multivariate data analysis has been successfully implemented as a reliable, rapid and

61 simple technique for analyzing food products, edible oils and lipids (Birkel & Rodriguez-

Saona, 2011). Soft independent modeling of class analogy (SIMCA) is a classification procedure based on principal component analysis (PCA) that groups samples into classes based on differences in their composition (infrared bands) (Allendorf, Subramanian, &

Rodriguez-Saona 2011). A class space is built whose boundaries discriminate between the samples fitting the class model and the samples that cannot be considered as belonging to the studied class. Partial least squares regression (PLSR) is used to find the fundamental relationship between two factors and thus allows for predictions to be made on cocoa butter composition based upon specific infrared spectral characteristics.

The objective of this study was to use ATR-IR spectroscopy in combination with multivariate analysis to classify and characterize a wide variety of commercial cocoa butters applicable to chocolate manufacturing.

3.3 Materials and Methods

Cocoa butter samples (40 in total) used for this study were obtained from a leading chocolate manufacturer. The cocoa butters were initially characterized for macro- composition and key physical properties by primary analytical methods.

All reagents were obtained from Fisher Scientific (Pittsburgh, PA, USA).

3.3.1 Hardness

Hardness values were determined using differential scanning calorimetry (DSC).

Commercial cocoa butter samples of approximately 30g were completely melted at 60°C.

Molten samples were mixed well and 2-3 mg of cocoa butters were transferred to aluminum DSC pans using a micropipette. The samples were tempered on the DSC-4 head using a programmed temper cycle. A typical cycle consisted of sample melting at

60°C for 2 min, slow cooling to 0°C at 2°C/minute, and subsequently holding at this temperature for 2 minutes. Analysis was then conducted by heating the sample from 0 °C to 50 °C at the rate of 5 °C /min. The tempering step allowed cocoa butter to crystallize in a stable polymorph . The area under the melting endotherm was measured using a partial area program.

3.3.2 Free Fatty Acid Composition

The method was adapted from Ciucanu & Kerek (1984), and Christie (1982). Fatty acid analysis was performed to determine the fatty acid composition of the cocoa butters using a fatty acid methyl ester (FAME) procedure.

Diethyl ether (4mL) was added to 2-3 drops of each cocoa butter. After thoroughly mixing, 0.25mL of methanolic tetramethylammonium hydroxide (TMAH) solution was added to the vial and shaken for one minute. After four minutes sitting, 5mL of DI water was added very slowly. After the layers had separated, the top ether layer was transferred to another vial containing 1 gram of anhydrous sodium sulfate. After mixing, 2-3mL of the clear ether layer was used for the GC analysis. Methyl esters were analyzed on an

Agilent 6850 GC equipped with a flame ionization detector (FID). Separation of the components was done using a HP INNOWax Polyethylene Glycol 30m x 0.25mm x 0.25 micron column using helium as the gas carrier. The injection volume was 1μL with a split ratio of 100:1. The initial column temperature was set at 175 °C and then programmed at

5 °C min−1 to 250 °C, which was held for 20 minutes. Quantitative determination of the fatty acids was accomplished by using an external standard calibration employing Nu-

Chek GLC Reference Standard 68 oC.

3.3.3 Triacylglycerol (TAG) Composition

The methodology described was used to determine triacylglyceride composition of the cocoa butters. 8 +/- 1 mg of each cocoa butter was weighed and diluted to 10 mL with hexane. An Agilent 6890 Series (Santa Clara, CA) GC was used for analysis with an

Agilent DB-17HT (50%-phenyl)-methylpolysiloxane, 30m x 0.32mm x 0.15μm GC column using helium as the gas carrier. The injection volume was 2 μL with a flow rate of 2.0 mL-min.. The column temperature was maintained at 80°C for 2 min after injection then programmed at 50 °C/min to 300 °C, and again at 30 oC/min to 350 oC and held for 30 minutes, with a total run time of 40.07 minutes. The inlet temperature was set to the oven track and the detector temperature was set to 360 oC.

3.3.4 ATR-IR Spectroscopy

Each cocoa butter sample was heated in an oven to 65 oC to ensure a homogenously melted sample prior to data collection. A benchtop Varian Excalibur 3100 spectrometer

(Varian, Palo Alto, CA, USA) equipped with a KBr beam splitter and deuterated triglycine sulfate (DTGS) detector was used. A single-bounce FatIRTM temperature controlled ATR ZnSe crystal (Harrick Scientific, Pleasantville, NY, USA) was set to 65 oC and spectra were collected over a range of 4,000-700 cm-1 at 4 cm-1 resolution and an interferogram of 64 scans were co-added. Spectral data was displayed in terms of absorbance and viewed using Resolutions Pro Software (Varian, Palo Alto, CA, USA).

The instrument was purged continuously using a CO2RP140 dryer (Dominic Hunter,

Charlotte, NC, USA) to prevent moisture and CO2 interference in the spectra. All samples were measured in duplicate.

3.3.5 Multivariate Analysis

The spectra were analyzed using multivariate statistical analysis software (Pirouette® version 4.0, Infometrix Inc., Woodville, WA, USA). The spectra were imported into the software from the instruments as GRAMS (.spc) files and analyzed by normalizing, mean-centering and taking the second derivative for each spectra. SIMCA and PLSR are pattern recognition data-analysis methods that use the variance-covariance matrix to reduce the dimensionality of multivariate data sets by determining the principal components that best explain the systematic variation (De Maesschalck, Candofi, Masart,

& Heurding 1999). In SIMCA, the principal components contain information about influential chemical and/or biological systems defining the classes (Sjastram, Wold,

Lindbergh, Persson, & Martens 1983). By determining the F-statistic, an upper limit for the residual variance (noise) can be calculated for all samples belonging to each class,

65 resulting in a set of probabilities of class-membership for each sample. Thus, an unknown sample can only be assigned to the class for which it has a high probability. If the residual variance of a sample exceeds the upper limit for the modeled classes in the data set, it is not assigned to any of the classes; either it is an outlier, or it comes from a class not represented in the data set (Lavine, 2000). SIMCA was used to classify the cocoa butter samples based on their chemical characteristics. SIMCA’s discriminating power plot was used to identify which infrared bands were responsible for classification.

PLSR is a bi-linear regression based on the extraction of “latent variables (LVs)”. These orthogonal factors (latent variables) explain most of the covariance of the X (spectra) and

Y variables. It reduces the dimensions contained in thousands of IR predictors into a few factors to explain variations in both the dependent variables and the spectral domain. The ideal end-result is a linear model able to predict a desired characteristic (i.e. hardness) based on a selected set of predictors (IR spectra). PLSR has been particularly successful in developing multivariate calibration models for the spectroscopy field because it reduces the impact of irrelevant X-variations (noise) in the calibration model (Sjastram,

Wold, Lindbergh, Persson, & Martens 1983). This capability provides a more information-rich data set of reduced dimensionality and eliminates data noise that results in more accurate and reproducible calibration models. The quality of the final model will be evaluated based on the number of latent variables, loading vectors, standard error of cross validation (SECV), coefficient of determination (r-value), and outlier diagnostics.

Cross-validated (leave-one-out approach involving systematically removing one calibration sample at a time and employing the remaining samples for model building)

PLSR models were used to study the correlation between the infrared spectrum of cocoa butter samples and hardness, TAG and fatty acid profiles. The cross-validation algorithm is used to determine the number of latent variables (LV) that yields the minimum prediction error. Using the loadings plot for the statistically significant LV allowed identification of key infrared bands.

3.4 Results and Discussion

The ATR-IR spectrum of cocoa butter samples provided unique information about different TAG substitution patterns. The cocoa butter samples showed similar spectral characteristic absorption bands (Figure 3.1) as those of common vegetable oils (Safar,

Bertrand, Robert, Devaux, & Genot, 1994). Some of the most prominent absorption regions were in the 3050-2800 cm-1 region associated with the stretching vibration of the cis olefinic =C-H double bonds and the methylene asymmetrical and symmetrical stretching vibrations (-C=CH cis, CH2, and CH3). The degree of unsaturation of fats can be determined from the peak height ratio of the oleofinic (3030 cm-1) and aliphatic (2857 cm-1) C-H stretching vibration bands from ATR-IR spectra (Guillen & Cabo, 1997). At

~1746 cm-1, the C=O group of triglycerides showed a stretching vibration band. The fat content is usually estimated from this strong carbonyl absorption the carboxylic group of esters (Safar, Bertrand, Robert, Devaux, & Genot, 1994). Bending and rocking vibrations

-1 of CH, CH2 and CH3 groups were observed between 1465 and 1370 cm while bands between 1250 to 720 cm-1 region resulted from the overlapping of the methylene rocking

67 vibrations and the out-of-plane bending vibration of cis-disubstituted olefins (Che Man,

Syahariza, Mirghani, Jinap, & Bakar, 2005).

Figure 3.1. FT-IR mid-infrared spectra and band assignments of cocoa butter using a temperature-controlled single-bounce ZnSe ATR mid-IR.

Pattern recognition analysis by SIMCA using the spectral data collected from a mid- infrared single-bounce ZnSe ATR-MIR benchtop (Figure 3.2) showed well-defined clustering for the 40 different cocoa butters. The SIMCA classification plot allowed visualization of sample clustering by projecting the objects onto the first 3 principal components and the model was constructed using a probability threshold of 0.95. Overall, no evident pattern was found between hardness and the classification of cocoa butters in the SIMCA projection plot, except that the softest cocoa butter sample (DSC 58) was grouped farthest (interclass distance (ICD)=860) from the rest of the samples (Figure 68

3.2A). All of the interclass distances among cocoa butter samples were above 3 showing significant chemical differences among each other. The greater the ICD, the greater the difference among the cocoa butters composition (Allendorf, Subramanian, & Rodriguez-

Saona 2011). Figure 3.2B shows the clustering of samples excluding the soft cocoa butter (DSC 58). Among the remaining 39 samples, we included 2 authentic cocoa butters that were sourced directly from West Africa and Indonesia to obtain insight on the compositional differences (pointed out by ICD) of pure samples due to geographical origins.

A B PC PC 2 PC 1 PC a PC 2 1 3 PC

DSC 58

a Group of the 24 cocoa butters within an interclass distance (ICD) of less than 20 to either of the two authentic cocoa butter samples

Figure 3.2. Soft independent modeling of class analogy (SIMCA) 3D projection plots of second derivative transformed spectral data collected by ATR-MIR spectrometer for all 40 cocoa butter samples (A) and the cocoa butter samples disregarding the softest cocoa butter (DSC 58) (B).

The ICD among these authentic samples was 14.9 and there were 22 additional samples within an interclass distance of 20 (or less) from one of the two (or both) authentic samples (Figure 3.2B) with hardness values ranging from 81 to 94 DSC units. There were 11 cocoa butters that differed significantly from this main cluster with ICD 36-98 from the authentic samples and hardness ranging from 72-92 DSC units. No other clustering patterns were found between the SIMCA classification of cocoa butters and chemical composition (FAME and TAG) among cocoa butter samples.

The cocoa butters evaluated in this study contained on average 17.4% POP, 35.3% POS and 20.2% SOS, making up to 73% of the TAG composition (Table 3.1). In addition, the cocoa butters contained an average of 27.2% palmitic acid, 35.5% stearic acid, 31.9% oleic acid and 2.7% linoleic acid, (Table 3.1) (Lipp et al. 2001). Our results were in agreement with Lipp et al. (2001), reporting that ~ 85% of the TAG species in cocoa butter consist primarily of 2-oleyl glycerides of palmitic, stearic and oleic acid (POP,

POS, SOS), providing its unique melting properties (Lipp & Anklam, 1998a).

Table 3.1. Statistical data for the hardness, triacylglycerol (TAG) and fatty acid composition levels using the 40 commercial cocoa butter samples. Palmitic Stearic Oleic Linoleic POP POS SOS Property DSC % % % % % % % na 40 38 40 39 40 37 37 36

mean 85.0 27.2 35.5 31.9 2.7 17.4 35.3 20.2

Minimumb 72.0 24.8 34.3 30.7 2.4 14.8 32.3 17.9

Maximumc 94.0 28.7 36.6 33.8 2.9 21.6 40.2 22.9

Standard 4.4 1.1 0.6 0.8 0.1 2.0 2.6 1.4 deviation a Number of cocoa butter samples used in data analysis b Minimum level of the trait found in the cocoa butter samples c Maximum level of the trait found in the cocoa butter samples

The SIMCA discriminating power plots provided information regarding the main functional groups responsible for the separation of classes. The higher the intensity of the bands, the greater its influence in discriminating the cocoa butter samples. Figure 3.3A shows the effect of the softest cocoa butter (DSC 58) on the discriminating power, revealing that most of the variance in the model was explained with the band at 1710 cm-1 corresponding to asymmetric -C=O ester stretching vibrations associated with free fatty acids in samples (Guillen & Cabo 1997). Figure 3.3B also shows that this band (1710 cm-1) was a critical source of discrimination for the additional 39 cocoa butter samples, suggesting that the extent of hydrolytic rancidity (free fatty acids) contributes to the

SIMCA classification. In addition, absorption bands at 1150 cm-1 and in the region of

2800 to 3050 cm-1 associated with the O-C=O ester bond and the –C-H (and cis C=H) methylene group stretching vibrations of triacylglycerols (Goodacre & Anklam 2001),

71 respectively, were important for explaining the clustering patterns. Both SIMCA discriminating power plots (Figure 3.3A&B) showed these three main bands as a source of discrimination between the cocoa butter samples but the different intensities indicate differences in prevalence of these bands in the pattern recognition models.

A 1710

2924 40

1150

20 Discriminating Power +03)(E Discriminating Power +03)(E 0 1467 1911 2605 3010 Wavenumber (cm-1)

B 1710 2925 120 1150

Discriminating Power Discriminating Discriminating +02) +02) (E (E Power Power 0 1467 1911 2605 3010 Wavenumber (cm-1)

Figure 3.3. SIMCA discrimination plots for the ATR-MIR spectrometer for all 40 cocoa butter samples (A) and the cocoa butter samples disregarding the softest cocoa butter (DSC 58) (B).

PLSR analysis was used to developed calibration models by correlating the infrared spectral data with the measurements obtained from the reference tests. PLSR plots

(Figure 3.4) generated for hardness, FFA and TAG all showed good correlations (r >

0.93) between the reference test values and the estimated values using the ATR-IR system revealing that the regression models were accurate in predicting hardness, fatty acid and TAG levels. The number of factors in the PLSR algorithm, using a leave-one- out cross-validation method, were 8 for DSC and 12 for fat composition (Table 3.2).

PLSR selects the optimum number of factors that produces the lowest prediction residual error sum of squares based on the F-statistic test. Performance of models for hardness showed SECV, an estimate of the expected error when using the models for unknown samples, of 1.6 units (DSC) and explained 98.6% of the variance. SECV for FFA and

TAG composition ranged from 0.04-0.34 and 0.44-0.77, respectively, and explained >

98.3% of the variability depending on the test analyte. The models showed a strong relationship between the reference tests and infrared predicted values, indicating that it is possible to predict the melting index of cocoa butter based on the detection of specific functional groups.

Table 3.2. Summary of statistical multivariate analysis for determining hardness, triacylglycerol (TAG) and fatty acid composition levels Factors SECVa r Valb DSC 8 1.55 0.94 Palmitic Acid 12 0.34 0.95 Stearic Acid 12 0.2 0.94 Oleic Acid 12 0.27 0.95 Linoleic Acid 12 0.04 0.93 POP 12 0.61 0.95 POS 12 0.77 0.96 SOS 12 0.44 0.95 a Standard Error of Cross Validation b Correlation of Cross Validation Model

95 A

80 Predicted DSC Hardness DSC Predicted 75

75 80 85 90 95 Measured DSC Hardness

B 36.5

36.0 Acid

35.5 Stearic

35.0 Predicted 34.5

34.5 35.0 35.5 36.0 36.5 Measured Stearic Acid

20 Predicted SOS Predicted

18 19 20 21 22 23 Measured SOS

3.5 Conclusion

The capability of ATR-IR to fingerprint a variety of commercial cocoa butter samples allowed for discrimination of the samples using unique infrared spectral data. Clustering of the cocoa butters was not associated to any of the compositional traits that were evaluated, including hardness, fatty acid and triglyceride composition. The band at 1710 cm-1 was important in discriminating the samples and was related to free fatty acids which could indicate sample abuse or degradation. The spectral information was correlated to the reference tests values and regression models were able to predict the concentration levels for specific fatty acids, triglycerides and the hardness values. Based on the performance of the models for the cocoa butters we were able to obtain prediction errors for DSC (1.55), palmitic acid (0.34), stearic acid (0.2) and oleic acid (0.27) as well as TAG (POP (0.61), POS (0.77) and SOS (0.44)) which are major quality indicators for the cocoa butter samples. The findings of this study clearly demonstrate the ability of

ATR-IR spectroscopy combined with multivariate analysis to classify and predict melting characteristics of cocoa butter samples.

3.6 References

Allendorf, M., Subramanian, A., & Rodriguez-Saona, L. (2011). Application of a Handheld Portable Mid-Infrared Sensor for Monitoring Oil Oxidative Stability. Journal of the American Oil Chemists Society, DOI 10.1007/s11746-011-1894-9. . Birkel, E., & Rodriguez-Saona, L. (2011). Application of a Portable Handheld Infrared Spectrometer for Quantitation of trans Fat in Edible Oils. Journal of the American Oil Chemists Society, 88(10), 1477-1483.

Chaiser, D., Dimick, P. S. (1989). Lipid and Hardness Characteristics of Cocoa Butters from Different Geographical Regions. Journal of the American Oil Chemists Society, 66, 1771-1775.

Che Man, Y. B., Syahariza, Z. A. , Mirghani, M. S. , Jinap. S., Bakar, J. (2005). Analysis of potential lard adulteration in chocolate and chocolate products using Fourier transform infrared spectroscopy. Food Chemistry, 90, 815-819.

Christie, W. W. (1982). A simple procedure for rapid transmethylation of glycerolipids and cholesteryl esters. Journal of Lipid Research, 23(7), 1072-1075.

Ciucanu, I., & Kerek, F. (1984). Rapid and Simultaneous Methylation of Fatty and Hydroxy Fatty-Acids for Gas-Liquid-Chromatographic Analysis. Journal of Chromatography, 284(1), 179-185.

Crews, C. (2002). Authentication of cocoa butter. In: Jee M (ed) Oils and Fats Authentication (pp 66-97). Boca Raton, FL: CRC Press LLC.

De Maesschalck, R., Candolfi, A., Massart, D. L., & Heuerding, S. (1999). Decision criteria for soft independent modelling of class analogy applied to near infrared data. Chemometrics and Intelligent Laboratory Systems, 47(1), 65-77.

Goodacre, R., & Anklam, E. (2001). Fourier transform infrared spectroscopy and chemometrics as a tool for the rapid detection of other vegetable fats mixed in cocoa butter. Journal of the American Oil Chemists Society, 78(10), 993-1000.

Guillen, M. D., & Cabo, N. (1997). Infrared spectroscopy in the study of edible oils and fats. Journal of the Science of Food and Agriculture, 75(1), 1-11.

Jackson, L. S. (2009). Chemical food safety issues in the United States: past, present, and future. Journal of Agriculture and Food Chemistry, 57(18), 8161-8170.

Lavine, B. K. (2000). Clustering and Classification of Analytical Data. In: Meyers RA (ed) Encyclopedia of Analytical Chemistry, Chichester, UK: John Wiley & Sons Ltd.

Liendo, R., Padilla, F. C., & Quintana, A. (1997). Characterization of cocoa butter extracted from Criollo cultivars of Theobroma cacao L. Food Research International, 30(9), 727-731.

Lipp, M., & Anklam, E. (1998a). Review of cocoa butter and alternative fats for use in chocolate - Part A. Compositional data. Food Chemistry, 62(1), 73-97.

Lipp, M., & Anklam, E. (1998b). Review of cocoa butter and alternative fats for use in chocolate - Part B. Analytical approaches for identification and determination. Food Chemistry, 62(1), 99-108.

Lipp, M., Simoneau, C., Ulberth, F., Anklam, E., Crews, C., Brereton, P., de Greyt, W., Schwack, W., & Wiedmaier, C. (2001). Composition of genuine cocoa butter and cocoa butter equivalents. Journal of Food Composition and Analysis, 14(4), 399- 408.

Safar, M., Bertrand, D., Robert, P., Devaux, M. F., & Genot, C. (1994). Characterization of Edible Oils, Butters and Margarines by Fourier-Transform Infrared- Spectroscopy with Attenuated Total Reflectance. Journal of the American Oil Chemists Society, 71(4), 371-377.

Sjostrom, M., Wold, S., Lindberg, W., Persson, J. A., & Martens, H. (1983). A Multivariate Calibration-Problem in Analytical-Chemistry Solved by Partial Least-Squares Models in Latent-Variables. Analytica Chimica Acta, 150(1), 61- 70.

List of References

AOCS (American Oil Chemists’ Society). (1993). In Official Methods and Recommended Practices by the American Oil Chemists’ Society. Method Ca5a- 40, Free Fatty Acid. Champaign, IL.

Baer, R.J., Frank, J.F., & Loewenstein, M. (1983). Compositional analysis of whey powders using near infrared diffuse reflectance spectroscopy. Journal of Food Science, 66, 858-863.

Bimbo, A.P. (2011). The Production and Processing of Marine Oils: The American Oil Chemists’ Society Lipid Library. [WWW page]. URL http://lipidlibrary.aocs.org/processing/marine/index.htm

CDC (Center for Disease Control and Prevention). (2010). Center for Disease Control and Prevention: America’s Heart Disease Burden. [WWW page]. URL http://www.cdc.gov/heartdisease/facts.htm

Chaiser, D., Dimick, P. S. (1989). Lipid and Hardness Characteristics of Cocoa Butters from Different Geographical Regions. Journal of the American Oil Chemists Society, 66, 1771-1775.

Christie, W. W. (1982). A simple procedure for rapid transmethylation of glycerolipids and cholesteryl esters. Journal of Lipid Research, 23(7), 1072-1075.

Ciucanu, I., & Kerek, F. (1984). Rapid and Simultaneous Methylation of Fatty and Hydroxy Fatty-Acids for Gas-Liquid-Chromatographic Analysis. Journal of Chromatography, 284(1), 179-185.

Crews, C. (2002). Authentication of cocoa butter. In: Jee M (ed) Oils and Fats Authentication (pp 66-97). Boca Raton, FL: CRC Press LLC.

Ecobichon, D.J. (2001). Pesticide use in developing countries. Toxicology, 160(1-3), 27- 33.

Environmental Protection Agency. (2011). Pesticides Industry Sales and Usage. [WWW page]. URL http://www.epa.gov/opp00001/pestsales/07pestsales/market_estimates2007.pdf.

Fernandes, G., & Venkatraman, J.T. (1993). Role of Omega-3 Fatty Acids in Health and Disease. Nutrition Research, 13, S19-S45.

Firestone, D. (1995). In Official Methods of Analysis of AOAC International, 16th Ed. Method 969.33, Oils and fats. Association of Official Analytical Chemists, Arlington, VA.

Firestone, D. (1998). In Official Methods and Recommended Practices of the American Oil Chemists’ Society, 5th Edition. Method Cd 8-53, Peroxide value. American Oil Chemists Society, Champaign, IL.

Garip, S., Gozen, A.C., & Severcan, F. (2009). Use of Fourier transform infrared spectroscopy for rapid comparative analysis of Bacillus and Micrococcus isolates. Food Chemistry, 113(4), 1301-1307.

Gordon, S.H., Wheeler, B C., Schudy, R.B., Wicklow, D.T., & Greene, R.V. (1998). Neural network pattern recognition of photoacoustic FTIR spectra and knowledge-based techniques for detection of mycotoxigenic fungi in food grains. Journal of Food Protocol, 61(2), 221-230.

Griffiths, P.R., & de Haseth, J.A. (2007). Fourier Transform Infrared Spectrometry (2nd Ed.) (pp 3-6, 75-88, 321-3329). Hoboken, NJ: John Wiley & Sons, Inc., Publication.

Guillén, M.D., & Cabo, N. (1997). Infrared Spectroscopy in the Study of Edible Oils and Fats. Journal of Science of Food and Agriculture, 75(1), 1-11.

Günzler, H., & Gremlich, H. (2002). IR Spectroscopy: An Introduction (pp 52-62). Germany: Wiley-VCH.

Gurdeniz, G., & Ozen, B. (2009). Detection of adulteration of extra-virgin olive oil by chemometric analysis of mid-infrared spectral data. Food Chemistry, 116, 519- 525.

Huang, S., Frankel, E.N., & German, J.B. (1994). Antioxidant Activity of α- and γ- Tocohperols in Bulk Oils and in Oil-in-Water Emulsions. Journal of Agricultural and Food Chemistry, 42, 2108-2114.

Huang, Y.S., Pereira, S.L., & Leonard, A.E. (2004). Enzymes for transgenic biosynthesis of long-chain polyunsaturated fatty acids. Biochimie, 86, 793-8.

Ismail, A.A., van de Voort, F.R., & Sedman, J. (1997). Fourier Transform Infrared Spectroscopy: Principles and Applications. In J. R. J. Paré & J. M. R. Bélanger (Eds.), Instrumental Methods in Food Analysis (pp.97-117). Amsterdam: Elsevier Science.

Jackson, L.S. (2009). Chemical food safety issues in the United States: past, present, and future. Journal of Agriculture and Food Chemistry, 57(18), 8161-8170.

Jelinek, C. (1981). Occurrence and Methods of Control of Chemical Contaminants in Foods. Environmental Health Perspectives, 39(Jun), 143-151.

Juanéda, P., Ledoux, M., & Sébédio, J. (2007). Analytical methods for determination of trans fatty acid content in food. European Journal of Lipid Science & Technology, 109, 901-917.

Kuiper-Goodman, T. (2004). Risk assessment and risk management of mycotoxins in food. InN. Magnan, & M. Olsen, Mycotoxins in food: Detection and control (pp. 3-26). St. Paul, MN: Woodhead Publishing Ltd and CRC Press LLC.

Lavine, B.K. (2000). Clustering and Classification of Analytical Data. In: R.A. Meyers, (Eds.), Encyclopedia of Analytical Chemistry (pp. 9689-9710). New York: John Wiley & Sons Ltd

Li-Chan, E.C., Griffiths, P.R., & Chalmers, J.M. (2010a). Applications of Vibrational Spectroscopy in Food Science; Volume 1: Instrumentation and Fundamental Applications (pp.4-80). Chichester, UK: John Wiley & Sons, Ltd.

Li-Chan, E.C., Griffiths, P.R., & Chalmers, J.M. (2010b). Applications of Vibrational Spectroscopy in Food Science; Volume 2: Analysis of Food, Drink and Related Materials (pp.349-425). Chichester, UK: John Wiley & Sons, Ltd.

Liendo, R., Padilla, F. C., & Quintana, A. (1997). Characterization of cocoa butter extracted from Criollo cultivars of Theobroma cacao L. Food Research International, 30(9), 727-731.

Lin, M., He, L., Awika, J., Yang, L., Ledoux, D.R., Li, H., & Mustapha, A. (2008). Detection of melamine in gluten, chicken feed, and processed foods using surface enhanced Raman spectroscopy and HPLC. Journal of Food Science, 73(8), T129- 134.

Lipp, M., & Anklam, E. (1998a). Review of cocoa butter and alternative fats for use in chocolate - Part A. Compositional data. Food Chemistry, 62(1), 73-97.

Lipp, M., & Anklam, E. (1998b). Review of cocoa butter and alternative fats for use in chocolate - Part B. Analytical approaches for identification and determination. Food Chemistry, 62(1), 99-108.

MacDonald, B.F., & Prebble, K.A. (1993). Some applications of near-infrared reflectance analysis in the pharmaceutical industry. Journal of Pharmaceutical and Biomed Analysis, 11(11-12), 1077-1085.

Maga, J.A. (1979). Furans in foods. CRC Critical Reviews in Food Science & Nutrition, 11(4), 355-400.

Martens, H., & Naes, T. (1989). Models for Calibration. In: H. Martens, & T. Naes, (Eds.), Multivariate Calibration (pp. 73-112). Great Britain: John Wiley & Sons Ltd.

Mauer, L.J., Chernyshova, A.A., Hiatt, A., Deering, A., & Davis, R. (2009). Melamine detection in infant formula powder using near- and mid-infrared spectroscopy. Journal of Agriculture and Food Chemistry, 57(10), 3974-3980.

McClure, W.F. (2007). Introduction. In Y. Ozaki, W. F. McClure, & A.A. Christy (Eds.), Near-Infrared Spectroscopy in Food Science and Technology (pp.1-4). Hoboken, New Jersey: John Wiley & Sons, Inc.

Murphy, P. A., Hendrick, S., Landgren, C., & Bryant, C.M. (2006). Food Mycotoxins: An Update. Journal of Food Science, 71(5), R51-R65.

Orlein, V., Risbo, J., Rantanen, H., & Skibsted, L.H. (2006). Temperature-dependence of rate of oxidation of rapeseed oil encapsulated on a glassy food matrix. Food Chemistry, 94, 37-46.

Pedreschi, F., Segtnan, V.H., & Knutsen, S.H. (2010). On-line monitoring of fat, dry matter and acrylamide contents in potato chips using near infrared interactance and visual reflectance imaging. Food Chemistry, 121(2), 616-620.

Penner, M. H. (1998). Basic Principals of Spectroscopy. In S.S. Nielson (Ed.), Food Analysis (2nd Ed.) (pp. 387-398). Gaithersberg, Maryland: Aspen Publishers, Inc.

Perez Locas, C., & Yaylayan, V.A. (2004). Origin and mechanistic pathways of formation of the parent furan--a food toxicant. Journal of Agriculture and Food Chemistry, 52(22), 6830-6836.

Perez-Marin, D., Paz, P., Guerrero, J.E., Garrido-Varo, A., & Sanchez, M.T. (2010). Miniature handheld NIR sensor for the on-site non-destructive assessment of post- 85

harvest quality and refrigerated storage behavior in plums. Journal of Food Engineering, 99(3), 294-302.

PerkinElmer Technical Note. (2005). FT-IR Spectroscopy: Attenuated Total Reflectance (ATR). [WWW page]. URL http://shop.perkinelmer.com/content/technicalinfo/tch_ftiratr.pdf.

Petersen, D.W., Kleinow, K.M., Kraska, R.C., & Lech, J.J. (1985). Uptake, disposition, and elimination of acrylamide in rainbow trout. Toxicology and Applied Pharmacology, 80(1), 58-65.

Polshin, E., Lammertyn, J., & Nicolaї, B.M. (2010). In E.C. Li-Chan P.R. Griffiths, & J.M. Chalmers (Eds.), Applications of Vibrational Spectroscopy in Food Science; Volume 2: Analysis of Food, Drink and Related Materials (pp.349-352). Chichester, UK: John Wiley & Sons, Ltd.

Regan, F., Meaney, M., Vos, J.G., MacCraith, B.D., & Walsh, J.E. (1996). Determination of pesticides in water using ATR-FTIR spectroscopy on PVC/chloroparaffin coatings. Analytica Chimica Acta, 334(1-2), 85-92.

Rein, A. (2008). Handheld FT-IR spectrometers: Bringing the spectrometer to the sample. Supplement: FT-IR technology for today’s spectroscopists. Spectrocopy, 44-50.

Roach, J.A., Andrzejewski, D., Gay, M.L., Nortrup, D., & Musser, S.M. (2003). Rugged LC-MS/MS survey analysis for acrylamide in foods. Journal of Agriculture and Food Chemistry, 51(26), 7547-7554.

Rodriguez-Saona, L.E., Fry, F.S., McLaughlin, M.A., & Calvey, E.M. (2001). Rapid Analysis Of sugars in fruit juice by FT-NIR Spectroscopy. Carbohydrate Research, 336, 63-74.

Tseng, Y., Moreira, R., & Sun, X. (1996). Total frying-use time effects on soybean-oil deterioration and on tortilla chip quality. International Journal of Food Science and Technology, 31, 287-294. van de Voort, F.R., Ismail, A.A., Sedman, J., Dubois, J., & Nicodemo, T. (1994). The determination of peroxide value by fourier transform infrared spectroscopy. Journal of the American Oil Chemists Society, 71, 921-26. van de Voort, F.R., Sedman, J., & Russin, T. (2001). Lipid Analysis by vibrational spectroscopy. European Journal of Lipid Science & Technology, 103(12), 815- 26.

Wehling, R. L. (1998). Infrared Spectroscopy. In S. S. Nielson (Ed.), Food Analysis (2nd Ed.) (pp. 413-424). Gaithersberg, Maryland: Aspen Publishers, Inc.

Wei, F., Lam, R., Cheng, S., Lu, S., Ho, D., & Li, N. (2010). Rapid detection of melamine in whole milk mediated by unmodified gold nanoparticles. Applied Physics Letters, 96(13), 133702.