MONITORING FLAVOR QUALITY, COMPOSITION AND RIPENING CHANGES OF CHEDDAR CHEESE USING FOURIER-TRANSFORM INFRARED SPECTROSCOPY
DISSERTATION
Presented in Partial Fulfillment of the Requirements for Degree Doctor of Philosophy in the Graduate School of The Ohio State University
By
Anand S. Subramanian, M.S. *****
The Ohio State University 2009
Dissertation Committee: Approved by Dr. Luis E. Rodriguez-Saona, Advisor
Dr. W. James Harper ______Dr. V.M Balasubramaniam Advisor Dr. David B. Min Food Science and Nutrition Graduate Program Copyright by Anand Swaminathan Subramanian 2009 ABSTRACT
Cheese flavor develops during the ripening process, when complex changes take place, leading to the formation of flavor compounds, including amino acids and organic acids. However, cheese ripening is not completely understood, making it difficult to produce cheese of uniform flavor quality. Currently, cheese characteristics are determined using sensory panels and chromatography, which are expensive, time- consuming, and laborious. The objective was to develop a rapid and cost-effective technique based on Fourier-transform infrared spectroscopy (FTIR) for simultaneous analysis of cheese composition and flavor quality and identify some of the biochemical changes occurring during ripening.
Cheddar cheese samples were obtained from a cheese manufacturer, along with their composition, age and flavor quality data. The samples were powdered using liquid nitrogen and water-soluble compounds were extracted using water, chloroform and ethanol. The extracts were analyzed by reverse phase HPLC for 3 organic acids, GC-FID for 20 amino acids, and FTIR to collect the mid-infrared spectra (4000-700 cm-1). The collected spectra were correlated with flavor quality and composition data to develop classification models based on soft independent modeling of class analogy (SIMCA) and regression models based on partial least squares (PLS).
The infrared spectra of the samples were well defined, highly consistent within each sample and distinct from other samples. All the cheese samples formed well separated clusters based on their flavor quality (fermented, unclean, low flavor, sour, good, etc.) in SIMCA classification plot. The PLS models showed excellent fit with coefficient of correlation (r-value) >0.83 and could simultaneously determine the pH, ii moisture, fat, salt, age, 20 amino acids and 3 organic acids in less than 20 min per sample. The estimated standard errors of prediction were less than 5%. Absorptions from amino acids, organic acids and short-chain fatty acids (1800-900 cm-1) were found to be very important factors influencing the models. Lactic acid, glutamic acid, leucine, asparagine, phenylalanine and valine are some of the compounds that exhibited significant changes during ripening.
This method shows great promise as a rapid tool for simultaneous analysis of cheese characteristics, which can save time, labor and operational costs for the industry and cheese research.
iii Dedicated to My parents M. Subramanian and S. Savithri Thank you for your love, support and encouragement.
iv ACKNOWLEDGEMENTS
I thank my parents and my brother, whose support and encouragement were instrumental to all my achievements throughout my academic life.
My heartfelt thanks to Veena, whose patience, support and motivation during every step of my Ph.D. and help with my experiments and dissertation, greatly helped me achieve this and more.
I sincerely thank Dr. Luis Rodriguez-Saona for his intellectual support, enthusiasm, believing in my capabilities, eagerly extending help whenever needed and supporting me not only with my research and academic activities but also several extracurricular activities.
I wish to thank Dr. W. James Harper for his insightful suggestions, discussions and thoughts that helped in steering my research in the right direction. I am also grateful to Dr. David B. Min and Dr. V.M. Balasubramaniam for serving in my research committee and providing their valuable comments to improve my research.
Thanks to my colleagues Annegret Männig, Cecilia Shiroma-Kian, Nathan Baldauf, Francisco Parada-Rabell and Jian He for their help at various stages of my Ph.D. Special thanks to my lab mate Geng Chen for his help with my experiments.
I am thankful to my friends Raghupathy Ramaswamy and Mallikarjuna Rettiganti for their support and help with my academic and non-academic activities.
v This research was supported by grants from Ohio Agricultural Research and Development Center (OARDC), Midwest Advanced Food Manufacturing Alliance (MAFMA; USDA Cooperative State Research, special grant number 2006-34328-17149) and DairiConcepts LP (Springfield, MO). I particularly thank Kris Clements of DairiConcepts for his support and involvement in the project.
I wish to express my gratitude to the Department of Food Science and Technology for providing the opportunity to pursue my Ph.D., and financial assistance. I also thank the department staff for their help with activities related to my Ph.D. studies.
vi VITA
December 06, 1980 ...... Born – Chennai, India
1998 – 2002...... B.Tech. Food Process Engineering, Tamil Nadu Agricultural University.
2003 – 2005...... M.S. Agricultural and Biological Engineering, Pennsylvania State University.
2005 – Present...... Ph.D. Food Science and Nutrition, The Ohio State University.
PUBLICATIONS
Subramanian, A., Harper, J. W, ad Rodriguez-Saona, L. E. 2009. Rapid prediction of composition and flavor quality of Cheddar cheese using ATR-FTIR spectroscopy. J. Food Sci. 74(3): C292-C297.
Subramanian, A., Harper, J. W, and Rodriguez-Saona, L. E. 2009. Cheddar cheese classification based on flavor quality using a novel extraction method and Fourier transform infrared spectroscopy. J. Dairy Sci. 92:87-94.
Subramanian, A., and Rodriguez-Saona, L. E. 2008. Fourier Transform Infrared (FTIR) Spectroscopy. Pages 145-178 in Infrared Spectroscopy for Food Quality Analysis and Control. D. W. Sun, ed. Academic Press, New York, NY.
FIELDS OF STUDY
Major Field: Food Science and Nutrition
vii TABLE OF CONTENTS
ABSTRACT...... ii ACKNOWLEDGEMENTS...... v VITA...... vii LIST OF TABLES...... xi LIST OF FIGURES ...... xii
CHAPTER 1 ...... 1
LITERATURE REVIEW ...... 1 1.1 CHEDDAR CHEESE...... 1 1.1.1 Cheddar Cheese Manufacture...... 3 1.1.2 Cheddar Cheese Ripening...... 5 1.1.3 Biochemical Changes during Ripening ...... 6 1.1.3.1 Metabolism of Lactose, Lactate and Citrate ...... 6 1.1.3.2 Lipolysis and Catabolism of Free Fatty Acids...... 8 1.1.3.3 Proteolysis and Catabolism of Amino Acids...... 9 1.1.4 Factors Affecting Ripening and Flavor Development...... 13 1.1.5 Cheddar Cheese Flavor...... 15 1.1.5.1 Component Balance Theory ...... 15 1.1.5.2 Contribution of Volatile Phase...... 15 1.1.5.3 Contribution of Aqueous Phase ...... 16 1.1.6 Monitoring Cheese Ripening and Flavor...... 17 1.2 FOURIER-TRANSFORM INFRARED SPECTROSCOPY ...... 19 1.2.1 Brief History ...... 19 1.2.2 Fourier-Transform Mid-Infrared Spectroscopy ...... 20 1.2.3 Construction of FTIR Spectrometers...... 21 1.2.4 Collection of Spectra...... 23 1.2.5 Advantages and Limitations ...... 26 1.2.6 Chemometric Analyses of Spectroscopic Data...... 27 1.2.6.1 Principal Component Analysis ...... 27 1.2.6.2 Soft Independent Modeling of Class Analogy (SIMCA) ...... 28 1.2.6.3 Partial Least Squares Regression (PLSR)...... 29
viii 1.2.7 Cheese Analysis by FTIR ...... 30 1.2.8 Cheese Flavor Analysis by FTIR...... 31
CHAPTER 2 ...... 33
CHEDDAR CHEESE CLASSIFICATION BASED ON FLAVOR QUALITY USING A NOVEL EXTRACTION METHOD AND FOURIER TRANSFORM INFRARED SPECTROSCOPY ...... 33 2.1 ABSTRACT...... 33 2.2 INTRODUCTION ...... 34 2.3 MATERIALS AND METHODS...... 36 2.3.1 Cheddar Cheese Samples...... 36 2.3.2 Extraction of Flavor Compounds for FTIR Analysis ...... 37 2.3.3 FTIR Spectroscopy ...... 37 2.3.4 Multivariate Analysis...... 38 2.4 RESULTS AND DISCUSSION...... 38 2.4.1 FTIR Spectra of Cheddar Cheese ...... 38 2.4.2 FTIR Spectra of Water Soluble Fractions...... 39 2.4.3 Classification of Cheddar Cheese Based on Flavor...... 42 2.4.4 Identification of IR Bands Responsible for Classification ...... 47 2.5 CONCLUSION...... 50
CHAPTER 3 ...... 51
RAPID PREDICTION OF COMPOSITION AND FLAVOR QUALITY OF CHEDDAR CHEESE USING ATR-FTIR SPECTROSCOPY...... 51 3.1 ABSTRACT...... 51 3.2 INTRODUCTION ...... 52 3.3 MATERIALS AND METHODS...... 54 3.3.1 Cheddar Cheese Samples...... 54 3.3.2 Sample Preparation and FTIR Analysis...... 55 3.3.3 Multivariate Analyses ...... 55 3.4 RESULTS AND DISCUSSION...... 56 3.4.1 FTIR Spectra of Cheddar Cheese Extract...... 56 ix 3.4.2 PLSR Prediction Models...... 57 3.4.3 SIMCA Classification Based on Flavor Quality...... 61 3.4.4 Infrared Bands Responsible for Classification ...... 63 3.5 CONCLUSION...... 63
CHAPTER 4 ...... 65
RAPID MONITORING OF AMINO ACIDS, ORGANIC ACIDS AND BIOCHEMICAL CHANGES DURING CHEDDAR CHEESE RIPENING USING INFRARED SPECTROSCOPY ...... 65 4.1 ABSTRACT...... 65 4.2 INTRODUCTION ...... 66 4.3 MATERIALS AND METHODS...... 68 4.3.1 Cheddar Cheese Samples...... 68 4.3.2 FTIR Spectroscopy ...... 69 4.3.3 Gas Chromatography ...... 69 4.3.4 High Performance Liquid Chromatography ...... 70 4.3.5 Multivariate Analyses ...... 71 4.4 RESULTS AND DISCUSSION...... 71 4.4.1 Prediction of Amino Acids and Organic Acids ...... 73 4.4.1.1 Prediction of Amino Acids ...... 73 4.4.1.2 Prediction of Organic Acids...... 75 4.4.2 Biochemical Changes during Ripening ...... 80 4.4.2.1 Changes in Amino acids and Organic acid Concentrations...... 83 4.4.2.2 Spectral Changes during Ripening...... 86 4.5 CONCLUSION...... 94 LIST OF REFERENCES...... 95
x LIST OF TABLES
Table Page
1.1 Amino acids contributing specific tastes...... 17
2.1 Interclass distances of Cheddar cheese samples from Plant #1. Greater the distance between two samples the greater is the difference between them...... 45
2.2 Interclass distances of Cheddar cheese samples from Plant #2. Greater the distance between two samples the greater is the difference between them...... 48
3.1 Predicted and actual pH, fat, moisture, salt and flavor quality score for the 67-day old test samples...... 59
4.1 Retention times, correlation coefficient and residual standard deviation of amino acid standards determined by GC-FID...... 74
4.2 PLSR model parameters for determination of amino acids by GC-FID...... 78
4.3 Retention times, correlation coefficient and residual standard deviation of organic acid standards determined by HPLC...... 79
4.4 PLSR model parameters for determination of organic acids by HPLC...... 80
4.5 One-way analysis of variance (ANOVA) of amino acid and organic acid concentration at various stages of ripening...... 85
4.6 Interclass distances between classes in SIMCA classification of the cheese samples based on age. Greater the distance between two clusters the greater is the chemical difference between them...... 89
xi LIST OF FIGURES
Figure Page
1.1 Total cheese production in the United States by production year...... 2
1.2 Cheese production in the United States by type of cheese...... 2
1.3 General protocol for Cheddar cheese manufacture...... 4
1.4 General overview of the biochemical changes during cheese ripening...... 7
1.5 Pathways for metabolization of lactate during ripening...... 8
1.6 Pathways for lipolysis and catabolism of free fatty acids...... 10
1.7 Overview of proteolysis and amino acid catabolism during cheese ripening...... 12
1.8 Optical layout of a typical FTIR spectrometer...... 22
1.9 Typical interferogram of a modern FTIR spectrometer...... 24
1.10 An illustration of how a mid-infrared spectrum is obtained from the interferogram...... 25
2.1 Raw and derivatized spectra of Cheddar cheese and Cheddar cheese extract...... 40
2.2 Soft independent modeling of class anology (SIMCA) classification plot of Cheddar cheese from the two production plants...... 43
2.3 Soft independent modeling of class analogy (SIMCA) classification plot of Cheddar cheese from Plant #1...... 44
2.4 Soft independent modeling of class analogy (SIMCA) classification plot of Cheddar from Plant #2...... 46
2.5 Discriminating power plot for classification of Cheddar cheese samples...... 49
3.1 Typical raw and 2nd derivative FTIR spectra of Cheddar cheese extract...... 57
xii 3.2 Partial least squares regression (PLSR) models for prediction of pH, fat, salt, moisture, and flavor quality score of cheese samples...... 60
3.3 Soft independent modeling of class analogy (SIMCA) classification plot for discrimination of cheese samples based on flavor quality...... 62
3.4 Discriminating power plot for classification of Cheddar cheese samples...... 64
4.1 Overlaid spectra of Cheddar cheese sample at various stages of ripening...... 72
4.2 GC-FID chromatogram of a Cheddar cheese water soluble fraction showing the amino acid profile...... 76
4.3 PLSR model showing the correlation between alanine (nmols/g cheese) predicted by GC-FID and FTIR...... 77
4.4 HPLC chromatogram of a Cheddar cheese water soluble fraction showing the organic acids profile...... 79
4.5 PLSR model showing the correlation between lactic acid (μmols/g cheese) predicted by HPLC and FTIR...... 81
4.6 PLSR model showing the correlation of age of the cheese (days) to FTIR spectrum...... 82
4.7 Concentration of total free amino acids and organic acids at various stages of ripening...... 84
4.8 SIMCA classification plot showing the overall compositional difference between samples at various stages of ripening...... 88
4.9 Discriminating power plot for classification of Cheddar cheese samples based on age...... 90
4.10 SIMCA classification plot showing the overall compositional difference in a sample at various stages of ripening...... 91
4.11 Biochemical changes in Cheddar cheese during ripening...... 93
xiii CHAPTER 1
LITERATURE REVIEW
1.1 CHEDDAR CHEESE
The name “Cheddar” comes from the name of a village in Somerset, England and its method of cheese making. During cheese making the curd was piled into heaps to prevent the temperature from falling. The higher curd temperature thus achieved resulted in better acid production, lower pH and hence better quality (Lawrence et al., 2004). These piles came be to known as “Cheddars” and the cheese thus made was named Cheddar cheese. Although current methods of cheese manufacturing do not involve the traditional cheddaring step the quality and the name “Cheddar” have remained. Cheddar production has significantly increased since the time first Cheddar cheese factory was established in New York in 1861 (Lawrence et al., 2004). According to National Agricultural Statistics Service (NASS) of U.S. Department of Agriculture (USDA), cheese production has been increasing over the years (Figure 1.1). The total cheese production in 2007 was 9.7 billion lbs. Cheddar cheese is one of the most popular cheeses and in 2007 it accounted for 31.5% (3.1 billion lbs) of the total cheese produced in the U.S (Figure 1.2). This data clearly shows the status of Cheddar cheese in the U.S.
The USDA describes Cheddar cheese as, “cheese made by the cheddaring process or by another procedure which produces a finished cheese having the same physical and chemical properties as the cheese produced by the Cheddar process and is made from cow’s milk with or without the addition of coloring matter and with common salt, contains not more than 39% of moisture, and in the water-free substance, 1 10 9.7 9.53 9.5 ) s
b 9.15 l
n o
i 9 8.87 l l i b ( 8.55 8.6 n o i
t 8.5
c 8.26 8.26 u d o r
P 7.94
8 e s e e
h 7.49 C 7.5
7 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year
Figure 1.1: Total cheese production in the United States by production year (NASS, 2008).
Cheddar, 31.5%
Swiss, 3.2% Other Italian, 8.6%
Other Mozzarella, American, 34.1% 8.5%
All Other, 14.1% Figure 1.2: Cheese production in the United States by type of cheese (NASS, 2008).
2 contains not less than 50% of milk fat.” (USDA, 1956). If the ingredients used for making Cheddar cheese are not pasteurized the cheese must be ripened for at least 60 days at a temperature not less than 35°C (FDA 21 CFR 133.113, 2003).
1.1.1 Cheddar Cheese Manufacture
The manufacture of Cheddar cheese has seen several major developments and improvements over the century. One of the major changes has been in the availability and reliability of starter cultures. Apart from the starter culture, the process itself has undergone changes and improvements to support large scale production in automated production plants. Despite the improvements and developments the final quality of the Cheddar cheese depends on the ability of the process to precisely control moisture expulsion and development of acidity (Lawrence et al., 2004). Controlling Cheddar cheese production is especially difficult due to the typical long ripening times and relatively high residual lactose content, which can be conducive to the formation of off- flavors.
The manufacturing process is very similar for almost all varieties of cheese, and minor modifications are done at various steps to achieve desired characteristics. The processing steps, their control and effect on cheese quality have been explained in detail by several authors (Lawrence et al., 2004; Fox et al., 2000a). The general process for the production of Cheddar cheese is shown in Figure 1.3. Milk (pasteurized or unpasteurized) is mixed with starter culture, rennet and color in a cheese vat. Lactococcus lactis subsp. lactis, Lactococcus lactis subsp. cremoris, and Streptococcus salivaris subsp. thermophilus are some of the commonly used starter cultures for Cheddar. Starter cultures reduce the pH of the milk by producing lactic acid from lactose. Rennet, which is more active at lower pH, coagulates the protein (casein). Optionally, calcium chloride
(CaCl2) around 0.01% may be added since calcium plays a major role in coagulation of milk by rennet. After the coagulum is formed (around 40 min), it is cut and cooked at around 39°C until the final pH is between 6.1 and 6.4 (up to 1 hour).
3 Figure 1.3: General protocol for Cheddar cheese manufacture (National Dairy Council, 2009). 4 After draining the whey, the curd is matted and then milled (while stirring to avoid re-matting) to small regular cubes. In dry-salted cheeses like Cheddar, salt (1 to 3%) is sprinkled and mixed thoroughly. Salt improves flavor, controls fermentation, and reduces the moisture content due to osmosis. The cubes are then pressed into blocks, packaged and aged/ripened. Ripening period varies from 2 weeks in the case of mild Cheddar to 2 years or more in the case of sharp Cheddar (McSweeney, 2004a,b). Typical ripening temperatures for Cheddar range from 6 to 8°C (Fox and Cogan, 2004). The effects of the above steps and the chemical composition on cheese quality and texture have been reviewed in detail by Lawrence et al. (2004). Complex physical, chemical, and microbiological changes take place during the ripening process, leading to the development of characteristic texture and flavor of Cheddar cheese.
1.1.2 Cheddar Cheese Ripening
Cheese ripening is a complex process that involves several concurrent changes and interaction between different pathways. As mentioned earlier, the ripening period for rennet coagulated cheese may vary range from 2 weeks to 2 years. Considerable changes in texture and flavor occur during ripening as a result of several interlinked biochemical events. The complexity and the obscurity of the ripening process have attracted the attention of several researchers and numerous reviews have been published summarizing the current knowledge of cheese ripening (Collins et al., 2003a, 2004; Curtin and McSweeney, 2004; McSweeney, 2004a,b; McSweeney and Fox, 2004; Upadhyay et al., 2004; Marilley and Casey, 2004; Yvon and Rijnen, 2001; Fox et al., 2000b; Fox and Wallace, 1997; Fox, 1989). Some of the general changes taking place during cheese ripening are as follows: Death and lysis of starter cells Growth of adventitious microflora (Non-starter Lactic Acid Bacteria; NSLAB) Softening of texture due to hydrolysis of casein micelles by proteolysis Changes in pH Flavor formation through breakdown of proteins, fats and lactose
5 1.1.3 Biochemical Changes during Ripening
The biochemical changes in Cheddar cheese during ripening can generally be grouped in to two: 1) primary events consisting of metabolism of residual lactose, lactate and citrate, lipolysis, and proteolysis and 2) secondary events including metabolism of fatty acids and amino acids, leading to the development of volatile flavor (McSweeney, 2004a,b; Harper and Kristoffersen, 1956). A general overview of the biochemical pathways involved in the cheese ripening is shown in Figure 1.4. Elaborate discussions of all the biochemical changes during cheese ripening have been published in several books and review articles (Collins et al., 2003a, 2004; Curtin and McSweeney, 2004; McSweeney, 2004a,b; McSweeney and Fox, 2004; Upadhyay et al., 2004; Marilley and Casey, 2004; Yvon and Rijnen, 2001; Fox et al., 2000b; Fox and Wallace, 1997; Fox, 1989).
1.1.3.1 Metabolism of Lactose, Lactate and Citrate
During cheese production lactose is converted to lactic acid by starter lactic acid bacteria. Most of the remaining lactose is lost in whey during whey drainage. However, Cheddar cheese curd still has typically about 0.8 to 1% lactose (McSweeney and Fox, 2004; McSweeney, 2004a). From the quality standpoint, complete fermentation of lactose is essential to avoid growth of undesirable secondary organisms (Fox et al., 1990). Depending on the salt-in-moisture level and temperature the residual lactose is converted to lactate (McSweeney, 2004a). Any unfermented lactose is converted to D-lactate and L- lactate by NSLAB and racemization, respectively (Fox et al., 2000b; McSweeney, 2004a). In most cheese varieties the conversion of lactose to lactic acid mainly happens after the curds have been placed in moulds. Conversely, in Cheddar-type cheeses most of the acidification occurs prior to the salting step (Fox et al., 2000b). The extent and the rate of acidification have a significant influence on texture through demineralization of casein micelles, which in turn influences proteolysis and flavor formation (Creamer et al., 1985, 1988; Fox et al., 1990). Lactate can undergo reactions by several pathways as shown in Figure 1.5. It can be oxidized by lactic acid bacteria (LAB) in cheese to 6 ) McSweeney, 2004b Figure 1.4: General overview of the biochemical changes during cheese ripening ( Figure 1.4: General overview of the biochemical changes during
7 Figure 1.5: Pathways for metabolization of lactate during ripening (McSweeney and Fox, 2004).
acetate, ethanol, formic acid and carbon dioxide at rate dependent on oxygen availability (Fox et al., 2000b). Other pathways include conversion to propionate, acetate, water and carbon dioxide by Propioni bacterium spp., carbon dioxide and water by Penicillium spp. yeasts and butyric acid and hydrogen by Clostridium spp.
Citrate is also an important precursor for the flavor compounds in cheeses made using mesophilic starter cultures (McSweeney and Fox, 2004; McSweeney 2004a). The source of citrates in cheese is milk. Cheddar cheese curd typically contains about 0.2 to 0.5% citrate (McSweeney and Fox, 2004). NSLAB metabolize citrates to acetoin, acetate, butanediol, and diacetyl (Cogan and Hill, 1993; Palles et al., 1998).
1.1.3.2 Lipolysis and Catabolism of Free Fatty Acids
Lipolysis is considered to be an important biochemical event during cheese ripening as it results in the generation of several flavor compounds (Collins et al., 2003a, 2004; McSweeney and Sousa, 2000). Its effect on the flavor of blue cheese and hard 8 Italian cheese has been studied extensively and the mechanism of lipid breakdown and pathways of flavor production were reviewed in detail recently by Collins et al. (2004). Breakdown of fats in cheese is primarily by hydrolysis of triglycerides by the enzyme lipase. Oxidation is very limited due to low oxygen availability. Lipases in cheese may come from milk, rennet paste, starter cultures, secondary microorganisms, NSLAB, or exogenous lipase preparations (Collins et al., 2004). Milk is high in short-chain fatty acids, which have low flavor thresholds. Formation of these short-chain free fatty acids by lipolysis is a desirable character (McSweeney, 2004a). Free fatty acids also serve as precursors to the formation of several volatile flavor compounds (Collins et al., 2003a,b; McSweeney and Sousa, 2000). An overview of lipolysis and catabolism of fatty acids is shown in Figure 1.6. Contribution of lipolysis to flavor and quality of Cheddar is limited. Excessive lipolysis is in fact considered undesirable in Cheddar cheese (McSeeeney, 2004a).
Catabolism of free fatty acids, which is a secondary event in the ripening process, involves the breakdown of the fatty acids to form smaller compounds. Volatile flavor compounds such as lactones, thioesters, ethyl esters, alkanols, and hydroxyl fatty acids are formed as shown in Figure 1.6. Unlike mold ripened cheeses, these reactions do not form a significant part of Cheddar cheese flavor development.
1.1.3.3 Proteolysis and Catabolism of Amino Acids
Proteolysis is the most important of the three primary events during cheese ripening. Due to the complexity of proteolysis, catabolism of amino acids and their contribution to cheese flavor, this topic has been the focus of several studies. A comprehensive review of available literature and current knowledge on proteolysis and amino acid breakdown was published recently by Upadhyay et al. (2004). In addition several other researches on this topic are worth mentioning (Curtin and McSweeney, 2004; McSweeney and Sousa, 2000; Fox et al., 2000b; Yvon and Rijnen, 2001;
9 Figure 1.6: Pathways for lipolysis and catabolism of free fatty acids (Collins et al., 2004).
10 Fox, 1989; Aston et al., 1983). A summary of the proteolytic reactions and amino acid catabolism is shown in Figure 1.7.
Proteolysis contributes significantly to texture and flavor of cheese. It softens the texture by 1) hydrolyzing casein matrix, 2) decreasing water activity through changes in the water binding by the new carboxylic acid and amino groups formed from hydrolysis, and 3) indirectly by increasing the pH due to the formation of ammonia from amino acids during proteolysis. It also plays a significant role in the flavor development. Some of the short peptides produced during proteolysis are flavorful. Furthermore, amino acids released from proteolysis serve as substrates for formation several important volatile flavor compounds. It also enables release of sapid compounds from the cheese matrix. Proteinases and peptidases from coagulant, milk, starter LAB, NSLAB, secondary microorganisms, molds (in mold ripened cheeses) and Gram + bacteria (in smear type cheeses), are the proteolytic agents (Upadhyay et al., 2004). Residual rennet after cheese manufacture is the primary proteinase. Milk enzymes, primarily plasmin and to some extent cathepsins, also contribute to proteolysis. Enzymes from the microorganisms also influence the degree and the rate of proteolysis.
Initial proteolysis is due to the residual coagulant activity and plasmin. They hydrolyze casein to large and intermediate-sized peptides. The proteinases and peptidases metabolize the peptides to shorter peptides and amino acids. The pathways involved in the catabolism of amino acids to volatile flavor compounds are not fully unknown. A detailed summary of the various studies on the role of catabolism of amino acids in cheese flavor development was published by Curtin and McSweeney (2004). Two major pathways have been suggested: 1) aminotransferase or lyase activity and 2) deamination or decarboxylation. Aminotransferase activity results in the formation of α-ketoacids and glutamic acid. The α-ketoacids are further degraded to flavor compounds such as hydroxy acids, aldehydes, and carboxylic acids. α-Ketoacids from methionine, branched–chain amino acids (leuine, isoleucine, and valine) or aromatic amino acids (phenylalanine,
11 Figure 1.7: Overview of proteolysis and amino acid catabolism during cheese ripening (McSweeney, 2004b)
12 tyrosine and tryptophan) serve as precursors to volatile flavor compounds (Yvon and Rijnen, 2001). Volatile sulfur compounds are primarily formed from methionine. Methanethiol, which at low concentrations, contributes to the characteristic flavor of Cheddar cheese, is formed from the catabolism of methionine (Curtin and McSweeney, 2004; Weimer et al., 1999). Furthermore, bacterial lyases also metabolize methionine to α-ketobutyrate, methanethiol, and ammonia (Tanaka et al., 1985; Soda et al., 1983). On catabolism by aminotransferase aromatic amino acids yield volatile flavor compounds such as benzaldehyde, phenylacetate, phenylethanol, phenyllactate, etc. Deamination reactions also result in α-ketoacids and ammonia, which add to the flavor of cheese. However, decarboxylation of amino acids can cause flavor defects by producing biogenic amines.
1.1.4 Factors Affecting Ripening and Flavor Development
As evident from the above discussions the biochemical changes and the pathways of flavor development are still very obscure and difficult to comprehend. Several parameters have to be precisely controlled to produce a cheese of acceptable quality. Following is an outline of some of the important parameters that could influence the rate and extent of biochemical reactions during the ripening process, summarized from previous publications by Fox et al. (2000c), Lawrence et al. (2004), McSweeney (2004a), and Fox and Cogan (2004): 1. Milk a. Processing (heat treatment, bactofugation, microfiltration, standardization, etc.) b. pH and composition (fat, protein, enzymes, citrates, and lactose) c. Microbiological quality 2. Method of acidification a. Bacterial cultures i. Amount and activity (extent and rate of acid production) ii. Contribution of enzymes (plasmin and cathepsins) iii. Thermal stability of cultures (mesophilic/thermophilic) iv. Ripening properties 13 v. Rate of lysis vi. Resistance to bacteriophage attack vii. Secondary adjunct cultures (type, amount and activity) b. Acidulant or Acidogen i. Type (lactic acid, hydrochloric acid, glucano δ-lactone, etc.). ii. Amount 3. Coagulant a. Rennet i. Type (rennet paste or chymosin) ii. Amount and activity b. Acid/Heat (citric acid, lactic acid, glucano δ-lactone, etc.) 4. Optional ingredients (calcium chloride) 5. Processing a. Cutting (curd particle size influences syneresis) b. Cooking temperature and time (influences syneresis, enzyme activity, composition, and retention of coagulant) c. Extent of whey removal d. Agitation/Stirring (influences syneresis and curd composition) e. Amount of salt (more specifically salt-in-moisture ratio) f. Milling (size of the cubes) g. Cheddaring h. Pressing i. Final pH (influences texture and rate of whey removal) 6. Ripening a. Amount and activity of residual rennet b. Amount of residual enzymes from milk, starter cultures, and secondary organisms c. Activity of enzymes (proteinases, peptidases, lipases, etc.) d. Amount and activity of starter and secondary microorganisms e. Adventitious microflora (NSLAB) activity and amount f. Composition (lactose, fat, salt, citrates, and moisture) g. Conditions (temperature, humidity and time)
14 1.1.5 Cheddar Cheese Flavor
1.1.5.1 Component Balance Theory
Over 200 small peptides and amino acids and more than 200 volatile compounds constitute cheese flavor. Numerous reviews have been published in the last several decades on cheese flavor chemistry and the role of different flavor compounds in cheese flavor (Singh et al., 2003; Fox et al., 2000d). Characteristic flavor of each cheese type is due to different proportions of these compounds. One of the major hurdles in analyzing Cheddar cheese is the non-existence of a clear definition for an acceptable flavor (Fox et al., 2000d). This makes it difficult to chemically define its flavor. Early research in cheese flavor analysis concentrated on correlating Cheddar cheese flavor to a single compound or single class of compounds (Aston and Dulley, 1982). In the 1950s, Mulder (1952) and Kosikowski and Mocquot (1958) proposed the “component balance theory”, which attributed the flavor of cheese to a wide range of sapid and aromatic compounds. While this theory is still being accepted widely, many researchers have placed emphasis on monitoring ripening based on smaller groups of compounds. On general terms the type of cheese flavor compounds can be classified into two: 1) those that are soluble in aqueous phase and contribute to taste and 2) those that are volatile and contribute to the aroma (Fox et al., 2000d).
1.1.5.2 Contribution of Volatile Phase
Hundreds of aroma compounds have been isolated from Cheddar cheese over the years. Maarse and Vischer (1989) published a survey of literature on identification of aroma compounds in Cheese. This survey included a list of 213 volatile compounds identified in 40 studies on Cheddar cheese flavor. More recently, Singh et al. (2003) have listed several aroma compounds and discussed their role in cheese flavor. The general groups of aroma compounds include hydrocarbons, alcohols, aldehydes, ketones, acids, esters, lactones, amines, and sulfur compounds. Methanethiol is one of the very important aroma compounds, which is believed to be the major source of Cheddar cheese aroma (Urbach, 1997). Manning and Robinson (1973) reported that a headspace analysis of 15 volatile fractions emanating from Cheddar cheese mainly contained hydrogen sulfide, methanethiol and dimethyl sulfide. Esters contribute to fruity aroma and the volatile acids such as short-chain free fatty acids contribute to acidic aroma. The pathways for formation of several of these compounds are still unclear and characterization of volatile fraction and correlating it to cheese flavor is being pursued by several researchers.
1.1.5.3 Contribution of Aqueous Phase
The contribution of water-soluble fraction (WSF) to cheese flavor has been an active area of research and a number of authors have found a strong correlation between amino acid content and cheese flavor (Salles, et al., 1995; Aston and Creamer, 1986; Mulder, 1952;). McGugan et al. (1979) fractionated the volatile and non-volatile fractions of Cheddar cheese, assessed their contribution to overall flavor, and concluded that the non-volatile WSF contributed the most to the flavor intensity. The flavor of WSF was described by the taste panel as cheesey, brothy and slightly bitter. Similar observations were previously made by Harper (1949). The amino acids contributing to the base flavor of cheese included glutamic acids, leucine, isoleucine, valine, glycine, phenylalanine, cysteine, lysine, tyrosine and alanine. Organic acids such as succinic acid, acetic acid, propionic acid, lactic acid, pyruvic acid and α-ketoglutaric acid were also found to be associated with the taste of cheese (Dudley et al., 2005). Thus, non-volatile
WSF containing amino acids, organic acids, short-chain fatty acids (< C9:0), and peptides are the most important contributors to the flavor intensity of Cheddar cheese. Volatile aroma compounds add to and modify the base flavor.
Due to difficulties and lack of methods to isolate and examine smaller peptides (less than 5 amino acids), several of them may still have not been identified. But more than 200 peptides, the smallest with 5 to 6 amino acids, have been isolated from Cheddar cheese. They are believed to contribute to a brothy, bitter and savory flavor Cheddar cheese. Fox et al. (2000d) have categorized free amino acids according to their taste descriptor. γ-Glutamyl peptides have been identified to contribute to some of the sour,
16 umami, slightly sour, salty or metallic tastes. These peptides and amino acids have a low flavor threshold. Non-polar amino acids or amino acids with hydrophobic side chains such as isoleucine, lysine and tryrosine contribute to bitter flavor. A brief version of their report is presented in Table 1.1. Additionally, acetic acid, lactic acid, succinic acid, pyruvic acid and butyric acid contribute to acidity in addition to aroma.
Taste Amino Acids Bitter Histidine, Methionine, Valine, Arginine, Isoleucine, Phenylalanine, Tryptophan, Leucine and Tryosine Sweet Alanine, Glycine, and Serine Sweet and Bitter Lysine and Proline Sour Aspartic Acid, Glutamic Acid, and Aparagine Flat Glutamine Umami Glutamate Na and Aspartate Na
Table 1.1: Amino acids contributing to specific tastes.
1.1.6 Monitoring Cheese Ripening and Flavor
As new techniques and analytical methods emerge, new information is being uncovered about cheese ripening. Cheese ripening is understood only in general terms. It is still a very active area of research with many questions about flavor formation pathways and their interactions remaining unanswered. While studies on cheese ripening and flavor formation started more than a century ago, significant progress was made only after the development of chromatographic methods. Prior to chromatography, protein/peptide solubility in water, ethanol, buffers, etc. were used as indices to study proteolysis. These techniques have been reviewed in detail by Fox (1989) and McSweeney and Fox (1993, 1997). These tests mainly provided information about non- volatile products of protein breakdown, which contribute to the taste. The development of 17 techniques such as electrophoresis and reverse phase high performance liquid chromatography (HPLC) enabled analysis of large peptides and amino acids (Fox et al., 2000d). Free fatty acids are determined by titration, spectrophotometry, and adsorption chromatography. Organic acids such as lactic, acetic, pyruvic and succinic acids are more commonly determined using liquid chromatography and enzymatic reaction kits.
Since the 1950s chromatography has been the most used analytical tool for isolation of cheese components. Development of detectors like mass spectrometry (MS) simplified identification of compounds. Some recent manuscripts describe the instrumental and chemical methods available for monitoring cheese ripening (Le Quéré et al. 2004; Upadhyay et al., 2004; Marilley and Casey, 2004; Singh et al., 2003; Fox et al., 2000d,e; Schieberle 1995). Liquid or gas chromatography, mass spectrometry, and electrophoresis have been the most commonly applied techniques for analysis of lipolysis, proteolysis and lactose breakdown. Several publications are available on using these methods to monitor proteolysis (Izco et al., 2002; Otte et al., 1999; Ardo and Grippon, 1991; Singh et al., 1999; Butikofer and Ardo, 1999; Creamer, 1991) and lipolysis (Kilcawley et al., 2001; Innocente et al., 2000; Partidario, 1999; Partidario et al., 1998; Fernandez-Garcia et al., 1994).
More recently, instruments known as electronic noses, which try to imitate the functionality of the biological nose, have attracted some attention due to its speed of detection of volatile compounds. Although these techniques have made significant contributions to cheese science, they have several disadvantages:
1. Extensive use of solvents and gases that are expensive 2. Disposal problems involved in hazardous solvent wastes 3. High initial and operating costs 4. Requirement of specific accessories for different analytes 5. Requirement of extensive sample preparation to obtain pure and clean samples. 6. Labor intensive operation
18 These disadvantages prompt for constant evaluation of new and alternative techniques. Fourier-transform infrared (FTIR) spectroscopy has recently caught on as a valuable tool for analysis of food products including cheese. It overcomes several of the limitations of the conventional techniques and may be a good alternative quality control tool for cheese analysis.
1.2 FOURIER-TRANSFORM INFRARED SPECTROSCOPY
1.2.1 Brief History
Several publications and articles are available on the history of Fourier-transform infrared (FTIR) spectroscopy. Most recently, Ferraro published a very detailed report on the history of FTIR spectroscopy (Ferraro, 1999). Additionally, several books and publications present the history of FTIR spectroscopy in great detail (Christy et al., 2001; Johnston, 1991; Griffiths, 1986; Bell, 1972). All the publications agree that the root of IR spectroscopy is the discovery IR radiation by Sir William Herschel (1800). Herschel then developed prism based techniques for measuring IR spectra. The concept of FTIR spectroscopy is known for more than a century. It began with the invention of interferometer by Michelson in the 1880s (Michelson, 1891; Michelson, 1892). Despite many developments FTIR spectroscopy saw a decline in the decades prior to World War II, mainly due to lack of computer and instrumentation technology. Many other competing techniques, including dispersive techniques were being investigated and developed.
In the years following World War II, as fortuitous coincidence, several streams of advancements converged to promote FTIR spectroscopy. One of the major breakthroughs in the field of Fourier-transformation computation occurred in 1965 with the invention of fast-Fourier transformation (FFT) or Cooley-Tukey algorithm by James Cooley and John Tukey (Cooley and Tukey, 1965). Soon Forman (1966) explored its application to infrared spectroscopy. This algorithm significantly increased the resolution while reducing the analysis time. FTIR instrumentation also saw notable improvements in the
19 1960s with the introduction of helium-neon (He-Ne) lasers, better IR detectors, and analog-to-digital converters. The first commercial FTIR spectrometers were available in the late 1960s. Various instrumental and computational improvements, including digital signal processing, FTIR software, diagnostic features, etc., occurred in the following decades that improved the resolution, signal-to-noise ratio, sensitivity and speed of detection. The introduction of IR microscope in the 1980s made microsampling and analysis possible. Miniaturization and lower cost of production then extended its applications to industry and quality control labs. Today, due to rapid commercial development and extensive research, FTIR spectroscopy is considered as one of the most powerful techniques for chemical analysis. Because of its simplicity, sensitivity, versatility and speed of analysis its applications in biological analysis, including food, are growing at a rapid pace. The growing number of research papers on applications of FTIR spectroscopy in food is a testimony to this. Although applications in the field of food science were not widespread in the early years of FTIR spectroscopy, recent decades have seen numerous novel fundamental and applied researches in this area.
1.2.2 Fourier-Transform Mid-Infrared Spectroscopy
Fourier-transform mid-infrared (FT-MIR) spectroscopy monitors the fundamental vibrational and rotational stretching of molecules, which produces a chemical profile of the sample. The MIR (4000-400 cm-1) is a very robust and reproducible region of the electromagnetic spectra in which very small differences in composition of samples can be reliably measured (Subramanian and Rodriguez-Saona, 2008a). Molecules absorb MIR energy and exhibit stretching, bending, twisting, rocking and scissoring motions at one or more locations in the spectra depending on several factors including bond configuration, location, etc. It is rich in information that helps in analyzing the composition and determining the structure of chemical molecules. Several publications discuss the basics of FT-MIR in great detail and are suggested as additional references (Stuart, 2004; Chalmers and Griffiths, 2002; Mantsch and Chapman, 1996; Smith, 1996)
20 FT-MIR spectra reflect the total biochemical composition of the sample, with bands due to major cellular constituents such as water, lipids, polysaccharides, acids, etc. An example FT-MIR spectrum of a Cheddar cheese WSF extract can be found in Chapter 2 (Figure 2.1). The region from 4000 to 3100 cm-1 consists of absorbance from O-H and N-H stretching vibrations of hydroxyl groups and Amide A of proteins, respectively. Protein bands also appear in the regions 1700 – 1550 cm-1 (amide I and amide II) and -1 1310 – 1250 cm (amide III). The C-H stretching vibrations of –CH3 and >CH2 functional groups appear between 3100 and 2800 cm-1. The spectral range 1250 – 800 cm-1 consists of signals from phosphodiesters and carbohydrates. The region from 1200 to 600 cm-1 is called as the “fingerprint region” as it contains signals that are distinct between each sample and highly conserved within each sample. An exhaustive list of functional groups and their mid-infrared absorbances have been published by Coates (2000). Chemometric analysis is commonly required to draw meaningful inferences.
1.2.3 Construction of FTIR Spectrometers
An optical layout of an FTIR spectrometer is shown in Figure 1.8. The major components of an FTIR spectrometer are the IR source, beamsplitter, detector, several mirrors to direct light and a time-reference laser. The beam splitter, along with the stationary and moving mirrors is called an interferometer. It is the heart of most FTIR spectrometers. It was first constructed by Albert Abraham Michelson (Michelson, 1891), the first American to win the Nobel Prize. It is an optical device that enables controlled generation of interference patterns or interferograms. The construction and working of the interferometer can be found elsewhere (Smith, 1996; Johnston, 1991; Griffiths, 1986).
21 Figure 1.8: Optical layout of a typical FTIR spectrometer (Subramanian and Rodriguez- Saona, 2008a).
22 The source emits light in the IR region when electricity is passed through it. The light from the source passes through the aperture wheel and hits a mirror that directs the light onto the beam splitter. The beamsplitter, as the name suggests, serves to split the incident IR light into two. The mirrors are aligned so as to reflect the light waves in a direction that would allow recombination of the waves at the beam splitter. The movable mirror is capable of moving along the axis, away from and towards the beamsplitter. One half of the light passes through the beamsplitter and is reflected by stationary mirror back to the beam splitter. The other half of the light is reflected on to the moving mirror, which in turn reflects the light back to the beamsplitter. The two reflected beams from the mirrors recombine at the beamsplitter. The recombined light from the interferometer is then directed by mirrors into the sample compartment and is finally detected by the detector.
1.2.4 Collection of Spectra
The difference in distance travelled by the two light beams, created due to the movement of the mirror, is called as the optical path difference (OPD) or optical retardation. The recombined beam passes through the sample and is finally detected by the detector. Based on the OPD, the infrared light waves interfere constructively providing highest intensity or destructively providing lowest intensity or both constructively and destructively providing intermediate intensities. The plot of the intensity of light (in voltage) over the OPD is known as the interferogram. A typical interferogram obtained using an FTIR spectrometer is shown in Figure 1.9. An interferogram represents one forward motion of the mirror until the point of maximum intensity (centerburst) and backward motion to the initial position. The regions on either side of the centerburst are called wings of the interferogram, where both constructive and destructive interference take place at varying levels. Ideally the wings of the interferogram, which carry the signal from the sample, should be identical on either side of the centerburst. The interferogram is the raw output from a spectrometer and it has to
23 be mathematically transformed into a more meaningful and easily interpretable form, the spectrum.
Most FTIR instruments are single beam. The background and the spectrum are collected at different times. The spectrum of the sample obtained by Fourier-transforming the interferogram is called the single beam spectrum. This spectrum represents the signal from the sample as well as the instrument and environment. This single beam spectrum is ratioed against the background spectrum obtained without the sample, to obtain the actual spectrum of the sample. A schematic diagram of the sequence of steps involved in obtaining a spectrum of a sample is shown in Figure 1.10. Typically multiple scans are co-added together to reduce noise, which is random positive and negative signals (Smith, 1996).
Wing Wing t l o V
Centerburst
-0.012 -0.008 -0.004 0.000 0.004 0.008 0.012 0.016 Optical Retardation (cm) Figure 1.9: Typical interferogram of a modern FTIR spectrometer (Subramanian and Rodriguez-Saona, 2008a). The total intensity of the source is shown by the centerburst, which does not contain any signal from sample. The wings of the interferogram contain signal from the sample.
24 Without Sample With Sample
Interferogram
Fast Fourier Transform
Single Beam Spectrum
Ratioing Truncation (4000-700 cm-1)
Mid-infrared spectrum
Figure 1.10: An illustration of how a mid-infrared spectrum is obtained from the interferogram (Subramanian and Rodriguez-Saona, 2008a). 25 1.2.5 Advantages and Limitations
FTIR spectroscopy offer several advantages that have attracted several researchers to adopt this technique. A summary of advantages and disadvantages obtained from various sources are listed below (Stuart, 2004; Chalmers and Griffiths, 2002; Mantsch and Chapman, 1996; Smith, 1996; Johnston, 1991). Advantages include: 1. Simplicity, sensitivity, and speed of detection.
2. High-throughput (ability analyze several samples in a short time).
3. Possibility of non-destructive analysis depending on the application.
4. Requirement of relatively low sample volumes.
5. Less use of hazardous solvents that pose environmental and health hazards.
6. Relatively low operational cost.
Despite these advantages the application of FTIR has some shortcomings that may limit its application to certain types of analyses: 1. FTIR cannot detect atoms and monatomic ions, elements, and inert gases such helium and argon.
2. They cannot detect diatomic molecules such as N2 and O2. However, in certain cases this can be seen as an advantage as it eliminates the need for vacuum for analysis.
3. Biological samples including food are complex mixtures and hence their FTIR spectra are complicated with overlapping peaks and signal masking.
4. Most biological samples contain water, which has a strong absorption band that can mask certain important signals. Often times sample preparation procedures are required to reduce the effect of water.
5. Since most FTIRs are single beam instruments change in environments (carbon dioxide and water vapor) can occur during the experiment, causing uncertainties in the spectra.
26 1.2.6 Chemometric Analyses of Spectroscopic Data
Interpretation of spectroscopic data commonly requires the use of chemometrics to draw conclusions. In spectroscopy, chemometrics involves relating the spectrum to properties of the sample through the application of mathematical or statistical methods. Two general types of chemometric applications in spectroscopy are 1) multivariate classification and 2) multivariate calibration. The type of multivariate analysis is application- and data-dependent. Most commonly used classification methods are K- nearest neighbors, discriminant analysis, artificial neural networks, cluster analysis, and soft independent modeling of class analogy (Ballabio and Todeschini, 2008; Lavine, 2000). For calibration, classical least squares, inverse least squares, principal component analysis, principal component regression, partial least squares regression, and artificial neural networks are some of the options (Romía and Bernàrdez, 2008). In this research, soft independent modeling of class analogy (SIMCA) and partial least squares regression (PLSR) were used for analyses. These two methods, as they applies to spectroscopic data anlaysis are described below.
1.2.6.1 Principal Component Analysis
Principal component analysis (PCA) is a statistical procedure that extracts a small number of latent factors to explain the majority of variation in the data set. Basic information on PCA can be found in publications by Romía and Bernàrdez (2008), Jolliffe (2002) and Mark (2001). Essentially, it reduces the data set to a small number of latent factors, typically less than 5, called principal components (PCs). The PCs are orthogonal to each other and hence are uncorrelated. Therefore, the main advantages of PCA are it:
1. Reduces the size of the data set from numerous variables to a set of few PCs and removes useless information.
2. Eliminates correlation between the data by grouping original data into few orthogonal PCs.
27 The PCs are extracted in a hierarchical manner. This means that the first PC is determined in the direction that explains the maximum variance. The next PC is derived from the residuals in the data matrix after excluding the information included in PC1. The process continues until 100% of the variance in the data is accounted for. The number of PCs required for explain 100% of the variance is equal to the number of original variables. The data reduction is achieved by choosing the number of PCs that explain an acceptable level of variance in the data set.
1.2.6.2 Soft Independent Modeling of Class Analogy (SIMCA)
Soft independent modeling of class analogy or SIMCA was first introduced by Svante Wold (Wold, 1976). SIMCA is based on PCA and it allows for grouping multivariate data sets based on similarities. In simple terms, it involves performing a PCA on each class (or group) in the data (Ballabio and Todeschini, 2008; Lavine, 2000). Appropriate numbers of PCs are retained to account for most of the variation within each class. Each class is represented by a PC model. In order to determine the appropriate number of PCs retained in each class, a procedure known as cross-validation is normally performed. This involves leaving out a segment of the data during the each PCA. The left out data is predicted using the model and compared with the actual values. The PC model that yields the minimum prediction error for the omitted data is retained. The process is repeated for each class. Cross-validation helps finding out the appropriate number of PC for each class while ensuring high signal-to-noise ratio by excluding any PC that contains noise.
Unknown samples are predicted by comparing its residual variance with the average residual variance of the samples in a class. The comparison of average residual variation are made statistically using F-statistic. The class-membership of each sample is expressed probabilistically. Graphically, the samples are projected on to the PC space. Classes with close average residual variance will lie close to each others while those with a large difference in their average residual variance will lie far in the SIMCA
28 classification plot. SIMCA does not force assignment of a class to an unknown sample. If the sample’s residual variation does not fall into the confidence intervals of any class then no class is assigned to the unknown sample. SIMCA also provides additional features such as total modeling power and discriminating power that help improve analysis. The total modeling power shows the contribution of a wavenumber to explanation of variance by the PCs. Discriminating power shows how a particular wavenumber is helping the PCs to classify samples in the data set. Wavenumbers with low total modeling power and low discriminating power are eliminated from the analysis to reduce noise. SIMCA allows for working with smaller sample sizes (number of samples less than number of variables) without concerns of collinearity and classification by chance (Lavine et al., 1988).
1.2.6.3 Partial Least Squares Regression (PLSR)
Partial least squares regression or PLSR was developed by Herman Wold (1975) to analyze data involving collinearity problems. A detailed description of PLSR as it applies to analysis of spectroscopic data can be found in the book chapter by Romía and Bernàrdez (2008). There are several algorithms used to implement PLSR. Non-linear iterative partial least squares (NIPALS) is an algorithm that is very commonly implemented in PLSR (Romía and Bernàrdez, 2008). In PLSR, the y-variables (dependent variables) are related to x-variables (independent variables or infrared wavenumbers) through auxiliary variables called latent variables or PLSR factors. These factors are linear combinations of x-variables (wavenumbers) and are very similar in concept to the PCs computed by PCA. Another analysis type, known as principal component regression (PCR), exists that is based on PCA. The difference between PCR and PLSR is that PLSR attempts to include as much predictive information as possible in the first few factors. It uses the whole data matrix (y-variables vs. x-variables) during calibration and reduces the data appropriately to explain as much variance as possible in both y and x. In doing so, PLSR reduces the effect of irrelevant variations in the x- variables (noise vibrations in the spectrum). Like PCA, the data reduction is achieved by specifying appropriate number of factors that explain an acceptable level of variance
29 between y-variables (example: concentration) and x-variables (infrared wavenumbers). Cross-validation with leave-one-out approach can be applied to PLSR as well to improve model performance.
1.2.7 Cheese Analysis by FTIR
Cheese analysis using FTIR spectroscopy has been around for several years. Many of the early applications of infrared spectroscopy in food and dairy products analysis employed near-infrared (NIR) spectroscopy due to its simplicity. Even today several applications on cheese analysis are based on NIR spectroscopy. Some of the areas researched include determination of shelf-life of Crascenza cheese (Cattaneo et al., 2005), determination of fat, protein and total solids in cheese (Rodriguez-Otero et al, 1995), assessment of sensory properties (Sørensen and Jepsen, 1998), modeling of sensory and instrumental texture parameters in processed cheese (Blazquez, et al., 2006), prediction of free amino acids during cheese ripening (Skeie et al., 2006), prediction of maturity and sensory attributes of Cheddar cheese (Downey et al., 2005), and determination of moisture and fat content in processed cheese (Adams et al., 1999). Further information on NIR spectroscopy and its applications can be found elsewhere (Burns and Ciurczack, 2001; Osborne et al., 1993; Osborne and Fearn, 1986).
The capabilities of mid-infrared spectroscopy were recognized close to the 1980s. The ability of MIR to monitor fundamental vibrations of several functional groups provided a new tool for researchers to look at minor compounds in cheese. The advent of new sampling techniques like attenuated total reflectance (ATR), which allowed for analysis of solids and liquids while overcoming uncertainties due to sample thickness, simplified and diversified the applications of FTIR spectroscopy. Some of its early applications were focused on analysis of macromolecules in cheese such as fat, moisture and protein (McQueen et al., 1995; Chen et al., 1998; Chen and Irudayaraj, 1998). Shelf- life analysis (Cattaneo et al., 2005), analysis of cheese related microorganisms (Chen, 2008; Lefier et al., 2000), chemical parameters of cheese (Martín-del-Campo et al.,
30 2007), compositional analysis (Rodriguez-Saona et al., 2006), monitoring ripening of Swiss cheese (Martín-del-Campo et al., 2009) and determination of geographic origin (Karoui et al., 2004) were investigated more recently.
1.2.8 Cheese Flavor Analysis by FTIR
Almost all compositional studies on analysis of cheese by FTIR have been directed towards monitoring major food components such as protein and fat. Limited research has been conducted on the use of FTIR spectroscopy for analysis of minor components that contribute to the flavor. This is mainly because of difficulties in the sampling procedures and heterogeneous nature of cheese (McQueen et al., 1995). In order to fully tap the potential of FTIR spectroscopy it is essential to eliminate the effect of matrix as much as possible. Several extraction methods that extract essential flavor compounds from cheese while eliminating the effect of the cheese matrix have been discussed and researched before (Tieleman and Warthesen, 1991; McSweeney and Fox, 1993). However, none of them were applied to cheese analysis by FTIR spectroscopy until Koca et al. (2007) analyzed the WSF of Swiss cheese to determine short-chain fatty acids. A comparison of analysis of WSF with direct analysis of cheese clearly showed a significant increase in predictive capability of PLSR models. A correlation coefficient of >0.90 was reported for prediction of acetic, propionic and butyric acid contents and the FTIR spectra.
The objective of this research was to develop a simple sample preparation procedure that will provide water soluble extracts that have well defined and consistent spectra, which when analyzed through multivariate statistical classification and regression techniques could:
1. Rapidly classify Cheddar cheese based on flavor quality
2. Accurately predict composition and flavor compounds such as amino acids and organic acids.
3. Identify some of the flavor-related biochemical changes taking place during ripening. 31 In the early phases of this project a simple 10 min extraction procedure was developed to extract the essential flavor compounds from cheese with minimum interfering compounds (Subramanian and Rodriguez-Saona, 2008b). This procedure allowed for classification of Cheddar cheese based on flavor quality (Subramanian et al., 2009a). Additionally, several projects were generated based on this novel extraction method, including the study of accelerated ripening of Swiss cheese (Féliz-Pérez, 2007), monitoring the effect of adjunct cultures during Swiss cheese ripening (Chen et al., 2009), and correlation of Swiss cheese descriptive sensory analysis to FTIR spectra (Kocaoglu-Vurma et al., 2009; Eliardi et al., 2008). Subramanian et al. (2009b) further extended this technique to rapid analysis of Cheddar cheese composition and chemical properties. Furthermore, changes observed in the infrared spectra of Cheddar cheese were correlated to the biochemical events occurring during ripening. The chapters following this review detail the developments and findings of this research project, along with the applications on several aspects of cheese analysis that were explored.
32 CHAPTER 2
CHEDDAR CHEESE CLASSIFICATION BASED ON FLAVOR QUALITY USING A NOVEL EXTRACTION METHOD AND FOURIER TRANSFORM INFRARED SPECTROSCOPY
Anand Subramanian, W. James Harper, and Luis Rodriguez-Saona
2.1 ABSTRACT
Analysis of Cheddar cheese flavor using trained sensory and grading panels is expensive and time consuming. A rapid and simple solvent extraction procedure in combination with Fourier-transform infrared spectroscopy was developed for classifying Cheddar cheese based on flavor quality. Fifteen Cheddar cheese samples from two commercial production plants were ground into powders using liquid nitrogen. The water-soluble compounds from the cheese powder, without interfering compounds such as fat and protein, were extracted using water, chloroform and ethanol. Aliquots (10 µL) of the extract were placed on a zinc selenide crystal, vacuum dried and scanned in the mid-infrared region (4000 to 700 cm-1). The infrared spectra were analyzed by Soft Independent Modeling of Class Analogy (SIMCA) for pattern recognition. Sensory flavor quality of these cheeses was determined by trained quality assurance personnel in the production facilities. The SIMCA models provided 3D classification plots in which all the 15 cheese samples formed well separated clusters. The orientation of the clusters in 3D space correlated well with their cheese flavor characteristics (fermented, unclean, low flavor, sour, good Cheddar, etc.). The discrimination of the samples in the SIMCA plot was mainly due to organic acids, fatty acids and their esters, and amino acids (1450-1350 33 and 1200-990 cm-1), which are known to contribute significantly to cheese flavor. The total analysis time, including the sample preparation time, was less than 20 min per sample. This technique can be a rapid, inexpensive, and simple tool to the cheese industry for predicting the flavor quality of cheese.
2.2 INTRODUCTION
About 9.7 billion pounds of cheese was produced in the US in 2005. Around 31.5% of the total production was Cheddar cheese (National Agricultural Statistics Service, 2008). One of the most important parameters that decides the marketability of the cheese is its flavor. Undesirable flavors can reduce the acceptance and the price of the cheese significantly. The characteristic flavor, aroma, texture, and appearance of cheese develop during ripening. Cheese ripening or maturation is a slow process (2 weeks to 24 months) characterized by series of complex physical, chemical, and microbiological changes affecting the principal components of the cheese (Singh et al., 2003). The complex nature of the cheese ripening process makes it a challenge to produce cheese of uniform sensory properties, especially flavor.
The flavor profiles of cheese are complex and variety- or type-specific (Akalin et al., 2002) and are determined to a great extent by proteolysis, lipolysis and glycolysis during ripening (Bachmann et al., 1999; Singh et al., 2003). The effect of the manufacturing process, the composition of milk (such as protein and fat level), and the biochemical events that occur during ripening will influence the final quality of the cheese (Chen et al., 1998). The principle compounds that contribute to the flavor include organic acids, amino acids, sulfur compounds, lactones, methyl ketones, alcohols and phenolic substances (Seitz, 1990; Urbach, 1993). Currently, cheese flavor and grades are determined by trained panels. This approach is complex, expensive and time-consuming. Furthermore, it is practically difficult to taste and evaluate each batch of cheese. Hence, there is a need for rapid and reliable instrumental methods to determine the flavor quality
34 of cheese. A rapid method apart from saving time and money for the cheese industry will also help in ensuring better product quality.
The “component balance theory” put forth by Mulder (1952) and Kosikowski and Mocquot (1958), suggests that Cheddar cheese flavor is produced by a correct balance and concentration of a wide range of sapid and aromatic compounds. Hundreds of compounds have been identified in cheese flavor (McSweeney and Sousa, 2000). Hence, an instrumental method that is capable of simultaneously monitoring multiple compounds and their functional groups is necessary. Furthermore, cheese composition is very complex with many interfering components. Sample heterogeneity is also very common. Attempts have been directed at finding and evaluating microbiological and biochemical parameters by which cheese could be classified and whereby uniform cheese quality could be established (Adda et al., 1982; Farkye and Fox, 1990; Seitz, 1990; Singh et al., 2003). Methods such as high performance liquid chromatography (Akalin et al., 2002; Lues and Bekker, 2002), gas chromatography (Partidario et al., 1998; Thierry et al., 1999) and mass spectrometry (Alli et al., 1998) have been investigated for analysis of cheese components. These studies have made significant contributions to the understanding of the ripening process. However, they are complicated, time-consuming, require different conditions and accessories for analyzing different classes of compounds and have limited applications as routine quality control methods for overall cheese flavor quality. Due to the above reasons, no reliable instrumental methods exist for rapid analysis of flavor quality.
Fourier-transform infrared (FTIR) spectroscopy is a simple, rapid and reliable technique that has been widely researched and applied for analysis of food components. It is based on the principle that different chemical functional groups require different amounts of energy (different wavelengths) for excitation (Subramanian and Rodriguez- Saona, 2008a). FTIR spectroscopy (4000 to 700 cm-1) monitors the absorbance of infrared light by functional groups in the sample to provide a chemical fingerprint (spectrum) of the sample that shows the overall chemical composition of the sample. 35 FTIR spectroscopy combined with multivariate analysis has been suggested for rapid analysis of cheese by many researchers. Examples of its applications in cheese analysis include determination of fat, moisture and protein analysis (McQueen et al., 1995; Chen et al., 1998; Chen and Irudayaraj, 1998), shelf-life analysis of Crescenza cheese (Cattaneo et al., 2005), compositional analysis of Swiss cheese (Rodriguez-Saona et al., 2006) and determination of geographic origin of cheese (Karoui et al., 2004).
Almost all studies on analysis of cheese by FTIR have been directed towards monitoring major food components such as protein and fat. Limited research has been conducted on the use of FTIR spectroscopy for analysis of minor components that contribute to the flavor. This is mainly because of difficulties in the sampling procedures and heterogeneous nature of cheese (McQueen et al., 1995). Koca et al. (2007) reported a correlation coefficient of >0.90 between acetic, propionic and butyric acid contents and the FTIR spectra. Due to interferences from fat and protein, a complex water soluble extraction method was followed to prepare the samples for FTIR analysis. The objective of this research was to develop a simple sample preparation procedure that will provide water soluble extracts that have well defined and consistent spectra, which when analyzed through multivariate statistical classification techniques could rapidly classify Cheddar cheese based on flavor quality. To the best of our knowledge this is the first research on rapid flavor quality analysis of cheese using FTIR spectroscopy.
2.3 MATERIALS AND METHODS
2.3.1 Cheddar Cheese Samples
Fifteen different Cheddar cheese samples, ripened for 70 days, were obtained from two production plants of a commercial cheese manufacturer. Five cheese samples were obtained from a commercial cheese manufacturing facility (Plant #1) and 10 samples were obtained from another production plant (Plant #2). The sensory flavor quality of these cheeses was earlier analyzed by trained quality assurance personnel in the
36 production facilities and provided along with the cheese samples. Cheese samples were vacuum-packaged and stored at -40°C until analysis.
2.3.2 Extraction of Flavor Compounds for FTIR Analysis
A novel extraction procedure using water, chloroform, and ethanol was used to prepare the samples (Subramanian and Rodriguez-Saona, 2008b). About 20 g of frozen cheese sample was cut into small pieces. The pieces were mixed with about 20 mL of liquid nitrogen to freeze them rapidly. The hard frozen pieces were cryogenically ground in a blender and the powered cheese was then kept frozen (-40°C) until extraction. For extraction exactly 0.1 g of the powdered cheese was weighed and 0.5 mL of distilled water was added and mixed well. The mixture was sonicated using an ultrasonic dismembrator (Fisher Scientific, Pittsburgh, PA) for 10 sec to breakdown clumps and improve extraction of water soluble components from cheese powder. An equal amount (0.5 mL) of chloroform was added to the cheese powder-water mixture (to separate complex fat), mixed well, and centrifuged at 15,700 x g and 25°C for 3 min. Exactly 200 µL of the resulting supernatant was mixed with an equal amount of 100% ethanol (to precipitate complex proteins), and centrifuged at 15,700 x g and 25°C for 3 min. Supernatant (100 μL) was used for FTIR analysis.
2.3.3 FTIR Spectroscopy
FTIR spectroscopy of the cheese samples was carried out on a Varian 3100 FTIR spectrometer (Varian Inc., Palo Alto, CA) equipped with PermaGlow™ mid-IR source, extended range potassium bromate infrared beam splitter and deuterated triglycine sulfate detector. Aliquots (10 µL) of the extracts were placed on a 3-bounce MIRacle™ attenuated total reflectance (ATR) accessory with an zinc selenide crystal (Pike Technologies, Madison, WI) and vacuum dried to form a thin film. Infrared spectra were recorded between 4000 and 700 cm-1 at a resolution of 8 cm-1, using the data collection software (Varian Resolutions Pro v4.05, Varian Inc., Palo Alto, CA). In order to improve the signal to noise ratio 128 scans were averaged for each spectrum. For each cheese,
37 three samples were collected from different locations and powdered. Five independent extractions were made for each powder and 3 spectra were collected per extract. Hence, for the 5 samples from Plant #1, a total of 225 (5 cheeses x 3 powders per cheese x 5 extractions per powder x 3 spectra per extract) spectra were collected. For the 10 samples from Plant #2, a total of 450 spectra were collected.
2.3.4 Multivariate Analysis
Multivariate analyses of the data were carried out using a commercially available comprehensive chemometrics modeling software called Pirouette® (v3.11, Infometrix Inc., Woodville, WA). For analysis, spectra were imported into Pirouette®, mean- centered, transformed into their second derivative using a Savitzky-Golay polynomial filter (five-point window) and vector-length normalized. For classifying the cheese samples based on their sensory flavor quality, the spectra were analyzed by soft independent modeling of class analogy (SIMCA). SIMCA is a classification algorithm based on principal component analysis. Based on specific fingerprint spectral information from infrared light absorbance by functional groups in cheeses, multivariate analysis modeled the relationships between large numbers of dependant variables to classify different cheeses. The data were projected onto the first three principal component axes to visualize clustering of samples in 3D space based on their flavor quality (fermented, sour, good Cheddar, etc.). The spectral regions influencing the classification of the cheeses were determined from the measure of variable importance (discriminating power).
2.4 RESULTS AND DISCUSSION
2.4.1 FTIR Spectra of Cheddar Cheese
FTIR spectra of Cheddar cheese samples were collected in the mid-infrared region (4000 to 700 cm-1), using a 3-bounce ATR crystal. In a 3-bounce crystal, infrared light bounces on the sample 3 times, increasing the absorbance and hence the signal. For analysis the raw spectra were normalized, standardized and transformed to their second
38 derivatives to remove the baseline shifts, improve the peak resolution, and reduce the variability between replicates (Kansiz et al., 1999). A typical mid-infrared spectrum of Cheddar cheese and its second derivative are shown by lines A and B, respectively in Figure 2.1.
The spectra of Cheddar cheese were marked by absorbance from complex fat and protein in the regions 3,000 to 2,800 and 1,800 to 1,000 cm-1, respectively, which were very similar to observations reported by several other researchers (Koca et al., 2007; Rodriguez-Saona et al., 2006; Chen et al., 1998; Chen and Irudayaraj, 1998). Asymmetric and symmetric stretching vibrations of C-H groups in methylene groups of long-chain fatty acids were observed in the region 3000 to 2800 cm-1. Strong signal from the C=O groups of fatty acid esters were also present at 1740cm-1. Broad Amide I and Amide II bands of proteins peaked at around ~1640 and ~1540 cm-1, respectively. The spectra also included several other bands in the region 1800 to 900 cm-1, primarily due to C-H bending, C-O-H in-plane bending and C-O stretching vibrations of lipids, organic acids, amino acids, and carbohydrate derivatives, that play a significant role in cheese flavor. The region from 1800 to 900 cm-1 exhibited poor peak definition due to multiple absorptions and signal masking by major compounds. In order to analyze Cheddar cheese flavor, interferences and variability caused by macromolecules such protein and fat have to be minimized and peak definition has to be improved. This emphasized the need for an extraction method to improve spectral information.
2.4.2 FTIR Spectra of Water Soluble Fractions
Water soluble fractions of Cheddar cheese has been reported to contribute significantly more to Cheddar cheese flavor intensity than complex fat, fat soluble volatiles or insoluble fractions by several researchers (Engel et al., 2002; Engels et al., 1997; Aston and Creamer, 1986). Several different extraction procedures with various solvents were investigated in order to obtain consistent spectra that would enable classification of the Cheese based on flavor. Extraction using water, chloroform
39 Esters and Amide I and Amide aliphatic C=O and C-C II of Proteins chains of -C-H stretching stretching -C=O of acids fatty acids in fatty acids modes of and esters acids -O-H stretching in hydroxyl groups
D e c n a b r o s b A C B
A
4000 3600 3200 2800 2400 2000 1600 1200 800 Wavenumber (cm-1)
Figure 2.1: Raw (A and C) and derivatized (B and D) spectra of Cheddar cheese (––––) and Cheddar cheese extract (– – –). Certain important functional groups and their region of absorbance are highlighted. Spectral data were obtained in the mid-infrared region (4000 to 700 cm-1) at resolution of 8 cm-1 by co-adding 128 scans. Cheddar cheese was scanned obtained by pressing 0.5 g of cheese on a diamond ATR crystal. Extracts were scanned by drying 10 µL of the extract on a zinc selenide ATR crystal.
40 and ethanol was found to provide extracts of consistent composition that enabled effective discrimination of Cheddar cheese based on flavor. Drying of the extract on the crystal resulted in the formation of a uniform film of sample. The presence of ethanol in the extract helped in faster drying of the samples on the crystal. The drying time per sample was 3 min. This sample preparation method allowed for the collection of high- quality spectra with distinct spectral features that were very consistent within each sample.
Raw and derivatized FTIR spectra of the Cheddar cheese extract are shown in Figure 2.1 (Lines C and D). Complex fat and protein that play a limited role in the flavor of Cheddar cheese were removed by sequential extraction with chloroform and ethanol. Chloroform solubilized long-chain fatty acids, both of which were removed by centrifugation due to their immiscibility with water. This was confirmed by the signal reduction in the region 3,000 to 2,800 and 1740 cm-1 in the spectra of cheese extract (Figure 2.1). Ethanol with its high affinity to water removed water from the protein causing proteins to form complexes and precipitate on centrifugation. As evident from Figure 2.1, the absorption from Amide I and Amide II bands of proteins reduced significantly in the spectra of cheese extract. Additionally, ethanol helped in extracting compounds with slight non-polarity, including short and medium chain fatty acids, from the cheese. The resulting extract contained water soluble components from the cheese which included organic acids, short and medium chain fatty acids and their esters, alcohols, amino acids, and small peptides, all of which are known to contribute significantly to cheese flavor (Singh et al., 2003; Seitz, 1990; Urbach, 1993). This extraction procedure significantly reduced interfering compounds and improved spectral definition in the region 1800 to 900 cm-1 that contains signal from several flavor-related compounds.
41 2.4.3 Classification of Cheddar Cheese Based on Flavor
Direct examination of Cheddar cheese had low discrimination and inconsistent spectra presumably due to presence of interfering compounds and heterogeneity in cheese samples (data not shown). Visual comparison of the raw spectra of Cheddar cheese water soluble fractions showed numerous differences between cheeses in the spectral regions 1800 to 900 cm-1. Using the derivatized spectra, classification models based on SIMCA were developed to discriminate cheese from the two production plants. The SIMCA models were developed by computing a small number of orthogonal variables (the principal components or PCs) that explained as much of the variation as possible between the samples, while preserving the relevant information and eliminating random noise (Mark, 2001). SIMCA classification plot, a projection of the original data onto the first three principal components, allowed the visualization of well-separated clustering among the samples, whose orientation in 3D space correlated with their flavor quality. SIMCA also provides a 95% probability cloud, which means that there is 95% probability that the samples within that cloud belong to the same flavor category.
Many factors such as ingredients, milk composition, bacterial starter culture used for fermentation, etc. influence the final flavor of Cheddar cheese (Bachmann et al., 1999; Chen et al., 1998). Furthermore many plants develop their own strains of non- starter lactic acid bacteria with time, which significantly influence the final composition and flavor of the cheese. Our research showed marked differences in the chemical composition of samples from the two production plants (Figure 2.2). This apart from indicating the possibility of identifying the production plant based on the spectra also suggests that a separate flavor classification model is needed for each production plant. The discrimination of the 5 samples from Plant #1 is shown in Figure 2.3. All the five samples formed tight and distinct clusters. The distance between the clusters in a SIMCA plot is represented by the interclass distance (ICD). Greater the distance between two clusters the greater is the difference in composition of samples belonging to those clusters.
42 PC3 Plant #2 Plant #1
PC2 PC1
95% Probability Cloud
Figure 2.2: Soft independent modeling of class anology (SIMCA) classification plot of samples from the two production plants. The mid-infrared spectra were transformed into their 2nd derivative, mean-centered and normalized prior multivariate analysis. The samples clustered distinctly indicating marked difference in the chemical composition of samples from the two production plants. The 95% probability cloud indicates the probability of a sample belonging to the cluster in which it is located.
43 Plant #1 (n=225) 1
PC3 5 PC1
4 PC2
2 3
Figure 2.3: Soft independent modeling of class analogy (SIMCA) classification plot of samples from Plant #1. The samples were projected against the first three principal components (PC) that explained the largest amount of variance among the samples. All the five samples formed distinct clusters in 3D space based on their flavor quality descriptors (1 – Fermented, 2 – Unclean, 3 – Slight Sour, 4 – Good Cheddar, and 5 – Slight Burnt).
44 As a rule of thumb, a distance of over 3 indicates that the samples are well separated and hence different (Kyalheim and Karstand, 1992). Table 2.1 shows the interclass distance values of the 5 samples from Plant #1. The samples with fermented flavor and unclean flavor, which are considered undesirable, clustered far from other samples. The ICD of fermented sample and unclean sample from good Cheddar were 28.21 and 13.96 units, respectively. Samples with slight burn and slight sour notes, which were not major flavor defects, clustered relatively close to the good Cheddar, with ICDs of 6.14 and 3.70, respectively. Samples from Plant #2 also exhibited very good clustering based on their flavor quality (Figure 2.4). The ICDs for samples from Plant #2 are shown in Table 2.2. Samples with fermented flavor and low flavor clustered far from good Cheddar (ICD > 3.0). Their ICDs from good Cheddar were 5.08 and 8.48, respectively.
Samplesa 1 2 3 4 5 1 0.00 2 4.75 0.00 3 6.14 3.70 0.00 4 29.62 31.56 28.21 0.00 5 12.49 13.99 13.96 15.85 0.00
aSample 1 – Slight Burn, Sample 2 – Sour, Sample 3 – Good, Sample 4 – Fermented, and Sample 5 – Unclean.
Table 2.1: Interclass distances of Cheddar cheese samples from Plant #1. Greater the distance between two samples the greater is the difference between them.
45 Plant #2 (n=450) 13
8, 9, 10, and 11 PC1
12
6 and 14
PC2
7 15 PC3
Figure 2.4: Soft independent modeling of class analogy (SIMCA) classification plot of samples from Plant #2. The orientation of the clusters in the SIMCA plot correlated with the flavor of the cheese sample (Sample 6 and 14 – Good Cheddar, Sample 7 – Fermented, Sample 8, 9, 10, and 11 – Sour and Slight Acid, Sample 12 – Slight Low Flavor and Slight Sour, Sample 13 – Low Flavor, and Sample 15 – Sulfide.)
46 Samples with minor defects such as slight sour, slight acid, sulfide flavor and slight low flavor clustered close to the good Cheddar, with relatively low interclass distance. In flavor categories that contained more than one sample (Good Cheddar, and Sour and Slight Acid), samples within the flavor category clustered close to each other (ICD < 3.0) and away from other categories (ICD > 3.0). All the samples in the Sour and Slight Acid category (Samples 8, 9, 10, and 11) had ICD less than 3.0 among themselves and greater than 3.0 for samples outside the cluster (Table 2.2). Similarly, the good Cheddar samples (Samples 6 and 14) had ICD of less than 3.0 between themselves and ICD greater than 3.0 from other clusters. These data clearly indicate that the developed extraction method combined with the FTIR has the potential for rapid quality control of cheese flavor.
2.4.4 Identification of IR Bands Responsible for Classification
The spectral wavenumbers and the associated chemical functional groups that were responsible for the classification of the cheeses in SIMCA plot can be identified using the discriminating power plot. In the discriminating power plot each wavenumber in the spectral range is plotted against its importance in discriminating the samples that are in the model. The higher the value of discriminating power, the greater is the role of that wavenumber in classifying the samples. In other words, the greater the discriminating power the greater is the difference between the samples in compounds or functional groups associated with that wavenumber. The spectral regions and the associated functional groups/compounds responsible for the differentiation of the Cheddar cheese samples are highlighted in Figure 2.5. The spectral range 1800 to 900 cm-1 was found to be important in the analysis of cheese flavor by FTIR. This region contained several wavenumbers that played significant role in discriminating the samples (indicated by their high discriminating power values). The major bands responsible for the classification were 1411, 1377 and 1354 cm-1. These bands mainly include stretching vibrations of carboxyl groups, C-H bending vibrations of methyl groups and C-N stretching vibrations of amines (Coates, 2000; Guillén and Cabo, 1997).
47 15 0.00 14 0.00 7.82 13 0.00 8.87 12.07 12 0.00 5.61 6.29 6.49 11 3.25 7.94 0.00 5.17 8.74 10 3.62 0.00 1.80 6.35 8.73 10.29 9 3.98 0.00 2.23 2.79 6.06 8.58 10.37 8 3.58 8.89 0.00 2.66 1.96 1.39 5.62 8.71 7 4.63 7.49 0.00 6.02 7.70 7.06 7.18 9.79 12.56 6 2.99 5.51 5.58 5.42 3.97 5.39 8.48 0.00 5.08 3.81 a 9 6 7 8 Sample 6 and 14 – Good Sample 7 – Cheddar, Fermented, Sample 8, 9, 10, – Sour and Slight and 11 Acid, Sample 12 – Slight Low Flavor and 13 14 15 10 11 12 Slight Sour, Sample 13 – Low Flavor, and Sample 15 – Sulfide. a Samples Table 2.2: Interclass distances of Cheddar cheese samples from Plant #2. Greater the distance between two samples the greater Cheddar cheese samples from Plant #2. Greater the distance between Table 2.2: Interclass distances of is the difference between them.
48 450 ) s
t 400 1411 i n U
y 350 r a r t i 300 995 1377 b r
A 1007 (
r 250 e 1076 w 1354 o 200 1150 P
g 1435 n i
t 150 a n i 1516 1666
m 100
i 1543 r c s
i 50 D
0 903 1096 1288 1481 1674 Wavenumber (cm-1)
Figure 2.5: Discriminating power plot for classification of Cheddar cheese samples. The regions of the FTIR spectra that contribute to the discrimination of the cheese samples based on their flavor are highlighted. Higher discriminating power at a particular wavenumber is related to greater difference between the samples in the functional group associated with that wavenumber.
49 Koca et al. (2007) associated 1412 cm-1 with C-H symmetric bending vibrations of short and medium chain fatty acids. Some organic sulfates have also been reported to absorb in this region (Coates, 2000). Stretching vibrations of C-O groups appear in the region from 1200 to 1000 cm-1 (Rodriguez-Saona et al., 2006). These signals could be attributed to C-O containing compounds including alcohols, organic acids, fatty acids, lactones, keto acids, etc. Carboxyl groups of acids also absorb at around 1435 cm-1. Several other bands, 995 cm-1 associated with C-H in-plane bends of aromatic compounds, 1543 and 1516 cm-1 associated with aromatic nitrogen compounds, and 1666 cm-1 associated with Amide I bands of peptides also influenced the discrimination of Cheddar cheese based on flavor.
2.5 CONCLUSION
A rapid, simple, and reliable sample preparation method and FTIR technique was developed for analysis of Cheddar cheese flavor quality. The infrared spectra could be correlated to specific flavor notes such as fermented, sour, unclean, etc. and differentiated using multivariate classification models. The total analysis time, including sample preparation time was less than 20 min per sample. The results indicate that this technique be used for detection of flavor quality defects in Cheddar cheese. Furthermore, differences in the composition of cheese from different production plants can also be elucidated. The developed technique shows great promise as a rapid and simple tool for application to cheese analysis method and can save time and money for the cheese industry. It will enable better quality control and rapid monitoring of ripening process to achieve cheese of desired flavor quality. Additionally, this method will also initiate and accelerate further studies on rapid compositional analysis and cheese ripening process using FTIR spectroscopy.
50 CHAPTER 3
RAPID PREDICTION OF COMPOSITION AND FLAVOR QUALITY OF CHEDDAR CHEESE USING ATR-FTIR SPECTROSCOPY
Anand Subramanian, W. James Harper, and Luis Rodriguez-Saona
3.1 ABSTRACT
Multiple methods are required for analysis of cheese flavor quality and composition. Chromatography and sensory analyses are accurate but laborious, expensive and time consuming. A rapid and simple instrumental method based on Fourier-transform infrared (FTIR) spectroscopy was developed for simultaneous analysis of Cheddar cheese composition and flavor quality. Twelve different Cheddar cheese samples ripened for 67 days were obtained from a commercial cheese manufacturer along with their moisture, pH, salt, fat content and sensory flavor quality data. Water soluble components were extracted from the cheese, dried on zinc selenide FTIR crystal and scanned (4000-700 cm-1). Infrared spectra of the samples were correlated with their composition and flavor quality data to develop multivariate statistical regression and classification models. The models were validated using an independent set of ten 67-day old test samples. The infrared spectra of the samples were well defined, highly consistent within each sample and distinct from other samples. The regression models showed excellent fit (r- value>0.92) and could accurately determine moisture, pH, salt, and fat contents as well as the flavor quality rating in less than 20 min. Furthermore, cheeses could also be classified based on their flavor quality (slight acid, whey taint, good Cheddar, etc.). The discrimination of the samples was due to organic acids, amino acids and short-chain fatty 51 acids (1800 to 900 cm-1), which are known to contribute significantly to cheese flavor. The results show that this technique can be a rapid, inexpensive, and simple tool for predicting composition and flavor quality of cheese.
3.2 INTRODUCTION
Cheese is a major fermented dairy product with a per capita consumption of around 32 pounds per year. It supplies various nutrients such as proteins and vitamins and minerals such as calcium and phosphorus. With an annual production of 3.1 billion pounds, Cheddar cheese is one of the major types of cheeses in the US (National Agricultural Statistics Service, 2008). The composition and flavor of Cheddar cheese are primary determinants of its price and application. Analytical determination of pH, moisture, fat, salt and flavor serve to evaluate the quality of cheese.
Several factors, including raw materials, manufacturing process, non-starter lactic acid bacterial, cheese type, and the biochemical reactions such as proteolysis, lipolysis, etc. influence the composition and flavor of cheese (Singh et al., 2003; Akalin et al., 2002; Bachmann et al., 1999; Chen et al., 1998). Heterogeneity of cheese, complexity of the maturation process and interference from matrix compounds complicate the analysis. In addition, simultaneous measurement of flavor-related volatile compounds such as short-chain fatty acids and non-volatile compounds such as amino acids and organic acids is essential. The current methods for cheese analysis, including chromatography and sensory analysis are laborious, time-consuming, expensive and complete characterization may require multiple accessories and methods (Subramanian et al., 2009a). Also, chromatographic methods generally require separation and concentration that can be selective for certain classes of compounds. Hence, there is a need for rapid and reliable instrumental method for simultaneous determination of composition and flavor quality of cheese. A rapid method apart from saving time and money for the cheese industry will also help in ensuring better product quality.
52 Various classes of compounds, more importantly water soluble non-volatile compounds, and their ratios determine the flavor (McSweeney and Sousa, 2000; Aston and Creamer, 1986; Kosikowski and Mocquot, 1958; Mulder, 1952). Simultaneous monitoring of flavor compounds and the composition require methods that are capable of monitoring multiple functional groups. Fourier-transform infrared (FTIR) spectroscopy is a simple and rapid technique that monitors the molecular vibrations exhibited by various compounds under infrared light (Subramanian and Rodriguez-Saona, 2008a). FTIR provides an overall chemical profile of the sample that varies with the composition and the concentration of the compounds. Advances in FTIR instrumentation, development of sampling techniques like attenuated total reflectance (ATR) and chemometrics have made possible to extract information related to composition and conformation of food components from the infrared spectra. Several developments in the field of statistical multivariate analysis have simplified routine analysis of spectroscopic data. Classification tools such as soft independent modeling of class analogy (SIMCA), can be used for characterization of biochemical composition of sample as well as compounds formed during ripening of the cheese. The concentrations of the compounds can be estimated using factor analysis methods such as partial least squares regression (PLSR).
Application of FTIR spectroscopy and multivariate analysis to cheese analysis has been presented by several researchers. Initial publications were primarily based on near- infrared spectroscopy (10000-4000 cm-1) applied to determination of macromolecules such as fat, protein and total solids (Adams et al., 1999; McQueen et al., 1995; Rodriguez-Otero et al., 1995) and sensory properties (Sørensen and Jepsen, 1998). In the last decade, applications of infrared spectroscopy in cheese have increased and diversified. Some recent applications include determination of geographic origin of cheeses (Karoui et al., 2005; Pillonel et al., 2003), sensory and textural properties (Blazquez et al., 2006; Downey et al., 2005), free amino acids (Skeie et al., 2006), and shelf-life (Cattaneo et al., 2005). Mid-infrared spectroscopy has also gained significant attention due to its ability to monitor specific functional groups and hence smaller
53 molecules. It has enabled monitoring changes during ripening (Chen et al., 1998), cheese- related microorganisms (Lefier et al., 2000), and protein structure and interactions during ripening (Mazerolles et al., 2001). Publications by Koca et al. (2007) and Rodriguez- Saona et al. (2006) have further extended the applications of FT-MIR spectroscopy to analysis of minor compounds like short-chain free fatty acids and composition in Swiss cheese, respectively.
Flavor analysis using infrared spectroscopy is complicated due to difficulties in sampling procedures and interference from matrix compounds. Our previous research demonstrated a novel sample preparation procedure that enabled flavor quality analysis by FTIR spectroscopy (Subramanian and Rodriguez-Saona, 2008b; Subramanian et al., 2009a). Specific flavor notes could be correlated to the infrared spectra and cheese could be classified based on their flavor quality. In this paper we present the capability of a simple sample preparation method and FTIR spectroscopy to simultaneously determine pH, fat, moisture, salt, and flavor quality of Cheddar cheese. As far as can be determined from accessible literature this is the first research on rapid and simultaneous analysis of composition and flavor quality of cheese using FTIR spectroscopy.
3.3 MATERIALS AND METHODS
3.3.1 Cheddar Cheese Samples
Twelve different Cheddar cheese samples (4 different production days and 3 different vats per production day) ripened for a period of 67 days were provided by a commercial cheese manufacturer. Moisture, fat, salt content and pH of the samples were determined by the manufacturer using standard methods. The flavor quality of the samples were rated on a scale of 1 to 10 (10 being the highest flavor quality rating) by trained quality assurance personnel in the production facility and provided along with the samples. Upon reception the samples were stored at -40°C until analysis.
54 3.3.2 Sample Preparation and FTIR Analysis
Samples were prepared for FTIR analysis following the method described by Subramanian et al. (2009a). Powders of the cheese samples were prepared by cryogenically grinding (with liquid nitrogen) 20 g of the cheese. Exactly 0.1 g of the powder was mixed with 0.5 mL of water and sonicated (Fisher Scientific, Pittsburgh, PA) for 10 seconds. The water soluble fraction from the mixture was partitioned by adding 0.5 mL of chloroform and centrifuging (15, 700 g, 25°C, 3 min). To 200 µL of the supernatant, equal volume of ethanol was added and centrifuged (15, 700 g, 25°C, 3 min). The resulting supernatant was used for spectral measurements in a Varian 3100 FTIR spectrometer (Varian Inc., Palo Alto, CA). The spectrometer was equipped with a PermaGlow™ mid-IR source (4000-700 cm-1), potassium bromate beam splitter and a deuterated triglycine sulfate detector. An infrared-transparent 3-bounce zinc selenide attenuated total reflectance (ATR) accessory (MIRacle™, Pike Technologies, Madison, WI) was used as sample holder. Exactly 7.5 µL of the extract was vacuum-dried on the crystal to form a film and scanned in the spectrometer in the mid-infrared region. Each spectrum was recorded by co-adding 64 scans, which theoretically yields a high signal- to-noise ratio of 8:1. Five independent extractions were performed for each sample and 3 spectra were collected for each extract, yielding at least 15 spectra per sample and a total of 180 spectra in the model. For the validation set, 10 samples were evaluated with a total of 150 spectra.
3.3.3 Multivariate Analyses
The classification and regression analyses of the spectral data were performed using Pirouette® (version 3.11, Infometrix Inc., Woodville, WA). The spectra of the cheese samples were mean-centered, derivatized (Savitzky-Golay polynomial filter with a 5-point window) and normalized prior to multivariate analysis. The spectra were then matched with the pH, composition (moisture, salt, and fat) and flavor quality rating data to develop prediction models based on partial least squares regression (PLSR). A nonlinear iterative partial least-squares (NIPALS) algorithm was employed. 55 Classification model to differentiate cheese based on their flavor was developed using soft independent modeling of class analogy (SIMCA). The data were projected onto the first three principal component axes to visualize clustering of samples in 3D space based on their flavor note (whey taint, slight acid, good Cheddar, etc.). The spectral regions influencing the classification of the cheeses were determined from the measure of variable importance (discriminating power). The developed prediction models were validated with ten 67-days old independent test samples.
3.4 RESULTS AND DISCUSSION
Extraction with organic solvents enabled removal of compounds that interfered with the detection of essential compounds such organic acids and amino acids that contribute to the flavor (Subramanian et al., 2009a). Drying of the extract on the crystal resulted in the formation of a uniform film of sample. The drying time per sample was 3 min. FTIR spectra of the extracts were collected in the mid-infrared region (4000-700 cm-1), using a 3-bounce zinc selenide ATR crystal. In a 3-bounce ATR-FTIR crystal, infrared light bounces on the sample 3 times, increasing the absorbance and hence the signal. The sample preparation method allowed for the collection of high-quality spectra with distinct spectral features that were very consistent within each sample.
3.4.1 FTIR Spectra of Cheddar Cheese Extract
The FTIR spectra reflect the total chemical composition of the cheese extract, with absorbance bands due to acids, esters, alcohols and peptides. The band intensities vary with the overall concentration of the chemical functional groups in the sample. The raw spectra were transformed into their second derivatives to remove the baseline shifts, improve band resolution, and reduce noise and variability between replicates (Kansiz et al., 1999). A typical FTIR spectrum of Cheddar cheese extract and its second derivative are shown in Figure 3.1. The region from 4000 to 3100 cm-1 consists of absorbance from O-H and N-H stretching vibrations of hydroxyl groups and Amide A of polypeptides and
amino acids, respectively. The C-H stretching vibrations of –CH3 and >CH2 functional
56 groups of fatty acids appear between 3100 and 2800 cm-1. The spectral range 1800 to 900 cm-1 contains signal from polypeptides, amino acids, carbonyl groups of fatty acids, hydroxyl groups, carboxylic acid groups and fatty acid esters (typically short-chain). Visual comparison of the raw spectra showed numerous differences between cheeses, especially in the spectral region 1800-900 cm-1.
Esters and Amide I and Amide aliphatic II of Proteins chains of -C=O of acids fatty acids and esters -C-H stretching in fatty acids C=O and C-C -O-H stretching stretching e
c in hydroxyl groups modes of acids n a b r o s b A
4000 3600 3200 2800 2400 2000 1600 1200 800 Wavenumber (cm-1) Figure 3.1: Typical raw (- - - -) and 2nd derivative (——) FTIR spectra of Cheddar cheese extract. Exactly 7.5 µL of the extract was dried on zinc selenide crystal and scanned in the mid-infrared region (4000 to 700 cm-1). Important functional groups and their region of absorbance are highlighted.
3.4.2 PLSR Prediction Models
The transformed spectra were correlated with the composition (fat, salt, and moisture), pH and quality rating and analyzed by partial least squares regression (PLSR) with cross-validation (leave-one-out approach) to generate calibration models. PLSR is a bi-linear regression analysis which determines the analyte’s concentration (Y-variable) 57 by regressing a small number of orthogonal factors that are linear combinations of variables (X-variable; infrared wavenumbers). These orthogonal factors, also called latent variables, explain as much covariance as possible between X and Y (Bjorsvik and Martens, 1992). PLSR provides information-rich data set of reduced dimensionality, good reproducibility and lesser noise and has been very successful in developing calibration models for spectroscopic data (Martens and Martens, 2001). The PLSR calibration models for pH, fat, salt, moisture and flavor quality score are shown in Figure 3.2. All the five models exhibited excellent correlation with coefficient of correlation (r) values greater than 0.92. The pH of the cheese could be predicted with a standard error of cross-validation (SECV) value of just 0.01 (Figure 3.2A). The SECV is an estimate of the error expected when independent samples are predicted using the model. Similarly, the models for predicting fat content, salt, and moisture of the cheese samples exhibited very low SECV values of 0.21% (Figure 3.2B), 0.19% (Figure 3.2C), and 0.15% (Figure 3.2D), respectively. The flavor quality scores and the IR spectra also exhibited very high correlation (Figure 3.2E) with a coefficient of correlation of 0.92 and SECV of 0.39. This indicates that it is possible to predict flavor quality score within an error of just 0.39. Currently, determination of composition and pH require the use of multiple techniques and several organic chemicals. Furthermore, these methods are complicated and expensive. The results emphasize the capability of FTIR spectroscopy to rapidly and reliably predict cheese characteristics.
The PLSR prediction models described above were validated with an independent set of ten 67-days old test samples. The pH, fat, salt, moisture content and the flavor quality score were predicted using the developed models. The values predicted using the models and the actual values are presented in Table 3.1. The models showed excellent predictive capability. The average percentage deviation of predicted values from actual values were 0.23, 0.57, 0.47, 1.23, and 3.1 for pH, fat, moisture, salt and flavor quality score, respectively. These data clearly indicate the reliability of the models and the potential of this technique for simultaneous analysis of cheese characteristics.
58 6.00 6.00 7.00 8.00 7.00 9.00 8.00 8.00 6.00 5.00 Actual 6.02 6.17 6.93 7.49 7.36 8.81 7.65 7.75 6.20 4.88 Flavor Quality Score Flavor Quality Predicted 1.86 1.84 1.78 1.83 1.78 1.73 1.75 1.78 1.74 1.81 Actual Salt (%) 1.83 1.82 1.77 1.79 1.80 1.78 1.76 1.80 1.76 1.81 Predicted 35.80 36.20 35.50 35.60 35.50 35.60 35.40 35.80 35.60 36.50 Actual Moisture (%) Moisture 35.77 35.65 35.45 35.59 35.71 35.78 35.47 35.86 35.86 36.23 Predicted 34.80 34.40 35.40 35.10 35.50 35.20 35.20 35.00 35.40 34.60 Actual Fat (%) 34.79 35.03 35.28 35.20 35.19 35.07 35.07 35.06 35.17 34.88 Predicted 5.10 5.19 5.22 5.18 5.12 5.11 5.17 5.11 5.07 5.10 Actual pH 5.13 5.20 5.23 5.10 5.14 5.11 5.08 5.11 5.18 5.13 Predicted I J F E B D A C G H Test Sample Table 3.1: Predicted and actual pH, fat, moisture, salt and flavor quality score for the 67-day old test samples. fat, moisture, salt and flavor quality score for the 67-day old Table 3.1: Predicted and actual pH,
59 5.35 38.0
) 37.5
5.30 % (
t
n 37.0 e
5.25 t H n p
o
d 36.5 C
e t t 5.20 a c i
F 36.0
d d e
r 5.15 e t P
35.5 c i R I d
5.10 e
r 35.0
r-Value: 0.97 P r-Value: 0.96
R 5.05 SECV: 0.01 I 34.5 SECV: 0.21% Number of Factors: 13 Number of Factors: 13 5.00 34.0 5.00 5.05 5.10 5.15 5.20 5.25 5.30 34.0 34.5 35.0 35.5 36.0 36.5 37.0 37.5 38.0 A pH B Fat Content (%)
1.95 36.5 ) % ( )
t
% 1.90 n (
36.0 e t t n n e t 1.85 o n C
35.5 o e r C
u t 1.80 l t a s
i 35.0 S
o d
1.75 M
e t d c e i 34.5 t d
1.70 c i e
r r-Value: 0.96 r-Value: 0.95 d P e
SECV: 0.19% r 34.0 SECV: 0.15% R P
I 1.65
Number of Factors: 13 R Number of Factors: 13 I 1.60 33.5 1.60 1.65 1.70 1.75 1.80 1.85 1.90 1.95 34.0 34.5 35.0 35.5 36.0 C Salt Content (%) D Moisture Content (%)
11
10
e 9 r o c S 8 d e t c i
d 7 e r P
R 6 I r-Value: 0.92 5 SECV: 0.39 Number of Factors: 17 4 4 5 6 7 8 9 10 E Quality Score Figure 3.2: Partial least squares regression (PLSR) models for prediction of A) pH, B) fat, C) salt, D) moisture, and E) flavor quality score of cheese samples. The spectra of 67- days old cheese were transformed into their second-derivative, mean-centered and normalized prior to multivariate analyses.
60 3.4.3 SIMCA Classification Based on Flavor Quality
The spectra were also analyzed the soft independent modeling of class analogy (SIMCA) with the aim of classifying the cheese based on their flavor notes. SIMCA is a multivariate statistical technique based on principal component analysis (PCA) that reduces the dimensionality of multivariate data sets. In SIMCA, training sets are assigned to classes and a principle component model is generated for each class with distinct confidence regions within them (De Maesschalck et al., 1999). A scores plot is constructed by projecting the actual data on the first three principal components that explain the most variance in the training data set. The variance explained by the class model describes the signal and the residual variance describes the noise in the data set and is used to define probability boundaries around the class. Comparison of the average residual variance of samples in a class and the residual variance of an unknown sample can help in identification of the unknown sample (Lavine, 2000). The performance of this method depends not only on the difference between classes, but also strongly on the training set for each class (Candofi et al., 1999). The SIMCA classification plot for discrimination of cheese based on flavor is shown in Figure 3.3 (only 8 of the 12 samples are shown for ease of visualization). All the eight samples formed tight clusters and the location of the clusters in 3D space correlated well with their flavor quality. The good samples clustered together (cluster 5 and 6) and away from the samples with defects. Clusters with defects in flavor notes clustered away from the “good creamy Cheddar” but close to samples with similar flavor note.
The distance between the clusters in a SIMCA plot is represented by the interclass distance (ICD). It is a measure of the quality of the data. Greater the distance between two clusters the greater is the difference in composition of samples belonging to those clusters. As a rule of thumb, a distance of over 3 indicates that the samples are well separated (Kyalheim and Karstand, 1992). Samples of similar flavor quality had ICD of less than 3 within themselves and an ICD of greater than 3 when compared to samples
61 95% Probability Cloud
PC3 1 2 8 5 7 4 6
PC1 3 PC2
Figure 3.3: Soft independent modeling of class analogy (SIMCA) classification plot for discrimination of cheese samples based on flavor quality: 1 and 2 – both slight acid and good creamy, 3 and 4 – both slight whey taint and medium creamy, 5 and 6 – both good creamy Cheddar, and 7 and 8 – low flavor and good creamy. The samples were projected against the first three principal components (PC) that explained the largest amount of variance among the samples.
62 with a different flavor quality (data not shown). This data supports our previous publication on the possibility of classifying cheese samples based on flavor notes (Subramanian et al., 2009a).
3.4.4 Infrared Bands Responsible for Classification
The spectral wavenumbers and the associated functional groups that were responsible for the classification of the cheeses in SIMCA plot can be identified using the discriminating power plot. In the discriminating power plot each wavenumber in the spectral range is plotted against its importance in discriminating the samples that are in the model. The higher the value of discriminating power, the greater is the influence of that wavenumber in classifying the samples. The discriminating power also indicates the quality of the data and variables with low discriminating power are usually deleted because they contribute only to noise in the data set (Lavine, 2000). The spectral regions and the associated functional groups/compounds responsible for the differentiation of the Cheddar cheese samples are highlighted in Figure 3.4.
The spectral range 1800 – 900 cm-1 was found to be important in the analysis of cheese flavor by FTIR. This region consists of signals from C-O and C=O (~1175 cm-1), C-H bending (~1450 cm-1), esters (1750-1700 cm-1) and C-O stretching (~1240 and 1170 to 1115 cm-1) (Rodriguez-Saona et al., 2006). In the case of Cheddar cheese extract compounds containing these functional groups include the organic acids, alcohols, short- chain fatty acids and their esters, amino acids and small water soluble peptides, all of which are important for flavor.
3.5 CONCLUSION
A rapid, simple, and reliable FTIR technique was developed for simultaneous analysis of Cheddar cheese composition and flavor quality. The fat, salt, moisture, pH and flavor quality score of the cheese could be predicted in less than 20 min. Additionally, the infrared spectra correlated well with the flavor notes. This technique
63 3000 Amino acids and Aliphatic Chains of Fatty Acids
2500 ) U A ( 2000 r e w
o Organic Acids P (C-O and C-C Stretching
g 1500
n Vibrations i t
a Esters of Fatty n i m i 1000 r c s i D 500
0 850 1050 1250 1450 1650 1850 Wavenumbers (cm-1)
Figure 3.4: Discriminating power plot for classification of Cheddar cheese samples. The regions of the FTIR spectra that contributed to the discrimination of the cheese samples based on their flavor are highlighted. Higher the discriminating power at a particular wavenumber the greater is the difference between the samples in the chemical groups associated with that wavenumber.
could be a rapid quality control tool for the cheese industry to utilize in providing more consistency to cheese analysis and grading and improving quality. Further research and development of the technique could facilitate understanding, monitoring and controlling ripening process. Application of this technique for Swiss cheese analysis yielded similar results. Therefore, this technique has the potential of being applied to not only Cheddar cheese but also other types of cheese. The developed technique provides wide scope for further research and application and shows great promise as a rapid tool for cheese analysis.
64 CHAPTER 4
RAPID MONITORING OF AMINO ACIDS, ORGANIC ACIDS AND BIOCHEMICAL CHANGES DURING CHEDDAR CHEESE RIPENING USING INFRARED SPECTROSCOPY
Anand Subramanian, W. James Harper, and Luis Rodriguez-Saona
4.1 ABSTRACT
Amino acids and some organic acids are important water soluble compounds that play a vital role in cheese flavor. Their exact role in cheese flavor and the influence of their concentration are only vaguely known. The potential Fourier-transform (FTIR) spectroscopy as a rapid, simple and cost effective alternative to predict the concentration of amino acids and organic acids in Cheddar cheese and monitor biochemical changes during ripening was investigated. Twelve different Cheddar cheese samples were ripened for a period of 73 days and samples were drawn on days 7, 15, 32, 46 and 73. The water soluble fractions (WSF) of the cheese were analyzed by gas chromatography for 20 amino acids, liquid chromatography for 3 organic acids, and FTIR to collect the mid- infrared spectra (4000-700 cm-1). The collected spectra were correlated with age, and amino acid and organic acid contents and analyzed by soft independent modeling of class analogy (SIMCA) and partial least squares regression (PLSR) to develop prediction models. The changes in the FTIR spectra during ripening associated with biochemical changes in cheese during ripening were identified. SIMCA could classify the cheeses based on their age due to differences in the infrared absorption patterns in the region 1800-900 cm-1 associated with components such as amino acids, peptides, short-chain 65 fatty acids, and organic acids. The PLSR models showed excellent fit with coefficient of correlation (r-value) >0.83 and could simultaneously determine the levels of amino acids and organic acids in cheese in less than 20 min. The estimated standard errors for predicting amino acids and organic acids were less than 5% of the maximum observed concentration. Lactic acid, glutamic acid, leucine, asparagine, phenylalanine and valine are some of the compounds that exhibited significant changes during ripening. FTIR spectroscopy shows good prospect as a rapid tool for simultaneous monitoring of cheese constituents and monitoring ripening changes.
4.2 INTRODUCTION
Cheese ripening is a complex process which involves several concurrent changes and interaction between different pathways. The ripening period for rennet coagulated cheese may vary range from 2 weeks to 2 years. Considerable changes in texture and flavor occur during ripening as a result of several interlinked biochemical events. The complexity and the obscurity of the ripening process have attracted the attention of several researchers and numerous reviews have been published summarizing the current knowledge of cheese ripening (Collins et al., 2003a, 2004; Curtin and McSweeney, 2004; McSweeney, 2004a,b; McSweeney and Fox, 2004; Upadhyay et al., 2004; Marilley and Casey, 2004; Yvon and Rijnen, 2001; Fox et al., 2000b; Fox and Wallace, 1997; Fox, 1989). The biochemical changes in Cheddar cheese during ripening can generally be grouped in to two: 1) primary events consisting of metabolism of residual lactose, lactate and citrate, lipolysis, and proteolysis and 2) secondary events including metabolism of fatty acids and amino acids, leading to the development of volatile flavor (McSweeney, 2004a,b; Harper and Kristoffersen, 1956).
Proteolysis is the most important and complex of the three primary events during cheese ripening. This topic has been the focus of numerous studies and reviews (Upadhyay et al., 2004; Curtin and McSweeney, 2004; McSweeney and Sousa, 2000; Fox et al., 2000b; Yvon and Rignen, 2001; Fox, 1989; Aston et al., 1983). Apart from its
66 textural effects, proteolysis plays a significant role in flavor development. Some of the short peptides produced during proteolysis are flavorful. Furthermore, amino acids released from proteolysis serve as substrates for formation of several important volatile flavor compounds. It also enables release of sapid compounds from the cheese matrix. Another event that is important to cheese flavor is the lactose metabolism, leading to formation of flavor compounds including organic acids such as lactate and formate (McSweeney and Fox, 2004). Additionally, the extent and the rate of formation of lactate from lactose greatly influence the texture through demineralization of casein micelles, which in turn influence proteolysis and flavor formation (Creamer et al., 1985, 1988; Fox et al., 1990).
The contribution of water-soluble fraction (WSF), which include amino acids and organic acids, to cheese flavor has been an active are of research and a number of authors have found a strong correlation between amino acid content and cheese flavor (Salles, et al., 1995; Aston and Creamer, 1986; Mulder, 1952;). The non-volatile water soluble fraction contributed the most to the flavor intensity (McGugan et al., 1979; Harper 1949). The amino acids contributing to the base flavor the cheese included glutamic acids, leucine, isoleucine, valine, glycine, phenylalanine, cysteine, lysine, tyrosine and alanine. Organic acids such as succinic acid, acetic acid, propionic acid, lactic acid, pyruvic acid and α-ketoglutaric acid were also found to be associated with the taste of cheese (Dudley et al., 2005). Thus, the non-volatile WSF containing amino acids, organic acids, short-
chain fatty acids (< C9:0), and peptides is the most important contributors to the flavor intensity of Cheddar cheese. Volatile aroma compounds add to and modify the base flavor.
In the 1950s, Mulder (1952) and Kosikowski and Mocquot (1958) proposed the “component balance theory”, which attributed the flavor of cheese to a wide range of sapid and aromatic compounds. Characteristic flavor of each cheese type is due to different proportions of several flavor compounds. One of the major hurdles in analyzing Cheddar cheese is the non-existence of clear definition for an acceptable flavor (Fox et 67 al., 2000d). This makes it difficult to chemically define its flavor. Furthermore, investigating complex ripening changes requires techniques capable of monitoring multiple components simultaneously. Due to its ability to monitor multiple functional groups, FTIR has been investigated for cheese analysis by several authors previously (McQueen et al., 1995; Chen et al., 1998; Chen and Irudayaraj, 1998; Martín-del-Campo et al., 2007; Rodriguez-Saona et al., 2006). However, its ability to monitor flavor-related minor compounds was explored only recently. Koca et al. (2007) analyzed the WSF of Swiss cheese to determine short-chain fatty acids. A comparison of analysis of WSF with direct analysis of cheese clearly showed a significant increase in predictive capability of regression models. A correlation coefficient of >0.90 was reported for prediction of acetic, propionic and butyric acid contents and the FTIR spectra.
Our previous studies has shown that FTIR spectroscopy combined with an appropriate sample preparation procedure and multivariate analyses can discriminate Cheddar cheese based on flavor quality (Subramanian et al., 2009a) as well as predict composition (moisture, fat, and salt) and pH (Subramanian et al., 2009b). The overall goal of this research was to investigate the potential of FTIR spectroscopy as rapid and simple tool to monitor Cheddar cheese ripening and flavor using a water-soluble extract of the cheese. The specific objectives were: 1) to examine the possibility of simultaneous determination of flavor-related important water soluble compounds such as amino acids and organic acids, 2) correlate the level of amino acids and organic acids to the flavor quality of cheese and 3) identify and characterize the spectral changes associated with biochemical changes during Cheddar cheese ripening.
4.3 MATERIALS AND METHODS
4.3.1 Cheddar Cheese Samples Cheese samples for the study were obtained from a commercial Cheddar cheese manufacturer, along with the flavor quality descriptors. Cheese from 4 different production days and 3 different vats per production day (12 in total) were ripened for a period of 73 days. Samples were collected on days 7, 15, 32, 46 and 73. Three plugs were 68 collected from different locations for each sample. On reception, the samples were kept frozen at -40°C until analysis.
4.3.2 FTIR Spectroscopy
Samples for FTIR analysis were prepared as described by Subramanian and Rodriguez-Saona (2008b) and Subramanian et al. (2009a). About 20 g of cheese sample was ground into powder with liquid nitrogen and partitioned with water and chloroform by centrifugation (15, 700 g, 25°C, 3 min). The supernatant (aqueous phase) was mixed with equal volume of ethanol and centrifuged (15, 700 g, 25°C, 3 min). The supernatant, which contains essential non-volatile flavor compounds including free amino acids, short peptides, short-chain free fatty acids and amino acids, was used for FTIR analysis. Exactly 10 µL of the prepared extract was placed on the MIRacle™ attenuated total reflectance (ATR) accessory with 3-bounce zinc selenide crystal (Pike Technologies, Madison, WI) and vacuum dried to form a thin film. In a 3-bounce crystal infrared light is bounced on the sample three times, increasing the absorption, thereby increasing the signal intensity. Infrared spectra was recorded between 4000 and 700 cm-1 at a resolution if 8 cm-1 on a Varian 3100 FTIR spectrometer (Varian Inc., Palo Alto, CA) equipped with PermaGlow™ mid-IR source, extended range KBr beam splitter and deuterated triglycine sulfate detector. Interferograms of the samples recorded by the data collection software (Varian Resolutions Pro version 4.05, Varian Inc., Palo Alto, CA), were Fourier- transformed to obtain single beam spectrum and ratioed against the background to obtain the absorbance spectrum. A total of 128 scans were co-added per spectra to improve the signal-to-noise ratio. Five 5 extractions were performed for each sample and the spectrum for each extract was obtained by averaging 3 spectra. This yielded a total of at least 5 independent spectra per cheese sample.
4.3.3 Gas Chromatography
As mentioned earlier proteolysis and catabolism of amino acids play a vital role in flavor development of Cheddar cheese. The amino acids present in the Cheddar cheese
69 samples were identified and quantified by gas chromatography (Agilent 6890, Agilent Technologies, Santa Clara, CA). WSF from the samples were extracted by sonicating (Ultrasonic Dismembrator, Fisher Scientific, Pittsburgh, PA) 100 mg of the powdered cheese with 1 mL of distilled water for 10 sec. The mixture was centrifuged at 15, 700 g and 25°C for 3 min. The aqueous phase was derivatized and prepared for gas chromatography (GC) analysis using EZ:faast™ amino acid analysis kit (KG0-7165, Phenomenex Inc., Torrance, CA). Calibration curves were developed using free amino acid standard mixtures (AG0-7184, Phenomenex, Torrance CA), with free amino acid concentration ranging from 2 to 27 nmols/mL. Exactly 2 μL of the supernatant was injected into the GC in split mode (1:15 at 230°C, μL). Helium at 80°C was used as the carrier gas at a constant flow rate of 10 mL/min. The GC oven temperature was programmed to hold at 80°C for 2 min and then increase to 230°C at a rate of 4°C/min. Amino acids were separated using the ZB-AAA amino acid analysis GC column of dimensions 10 m x 0.25 mm, included with the EZ:faast™ kit (Phenomenex, Torrance, CA) and detected with a flame ionization detector (Agilent Technologies, Santa Clara, CA). Independent duplicates were analyzed and averaged for each sample.
4.3.4 High Performance Liquid Chromatography
Some of the organic acids in the cheese were detected using the WSF from the cheese. A slightly modified extraction method was followed to extract the WSF. Exactly 1.5 g of the powdered cheese was first sonicated (Ultrasonic Dismembrator, Fisher Scientific, Pittsburgh, PA) with 5 mL of chloroform for 10 sec. Next, 2 mL of distilled water was added and centrifuged at 3400 g and room temperature for 5 min. The supernatant was collected and run sequentially through a C18 solid phase extraction column (ODS-4, Whatman Inc., Sanford, ME) and a 0.2 μm nylon filter (Whatman Inc., Clifton, NJ). The filtrate was analyzed for organic acids by reverse phase high performance liquid chromatography (HPLC) analysis. Standard curves of lactic acid, formic acid and oxalic acid were developed using the concentration ranges 22.3 to 222.9 μmols/mL, 21.8 to 174.3 μmols/mL, and 0.1 to 11.1 μmols/mL, respectively. Reverse
70 phase HPLC analysis was performed using a HP1050 (Agilent Technologies, Santa Clara, CA) equipped with a Prevail™ Organic Acid Column (Alltech Associates Inc., Deerfield, IL) of dimensions 150 x 4.6 mm x 5 μm thickness. Phosphate buffer (25 mM
KH2PO4) at pH 2.5 (adjusted using 12 N HCl) was used as the mobile phase. Exactly 10 μL of the prepared sample was injected into the HPLC. The flow rate was 1.5 mL/min and the detection was done at 200 nm, using a UV detector. Independent duplicates were analyzed and averaged for each sample.
4.3.5 Multivariate Analyses
Multivariate analyses of the data were carried out using commercially available chemometric-modeling software called Pirouette® (v3.11, Infometrix Inc., Woodville, WA). For analysis, spectra were imported as GRAMS (.spc) files into Pirouette®, mean- centered, transformed into their second derivative using a Savitzky-Golay polynomial filter (five-point window) and normalized. Discrimination of the cheese samples was carried out using classification analysis based on soft independent modeling of class analogy (SIMCA). The IR spectra of cheese were correlated with their age and projected onto principal component axes to visualize clustering of the samples. Furthermore, the amino acids and organic acids identified using GC and HPLC, respectively, were correlated with the FTIR spectra to develop partial least squares regression (PLSR) prediction models. A non-linear iterative partial least squares algorithm was followed. Spectral changes were associated with some of the most significant flavor-related biochemical changes occurring during Cheddar cheese ripening.
4.4 RESULTS AND DISCUSSION
Water soluble fractions from Cheddar cheese samples were analyzed by FTIR spectroscopy, GC and HPLC for infrared absorbance, amino acid content and organic acid content, respectively. As reported previously (Subramanian and Rodriguez-Saona, 2008b; Subramanian et al., 2009a; Subramanian et al., 2009b), extraction with water, chloroform and ethanol significantly reduced the effect of cheese matrix on the infrared
71 spectrum and improved signal from important water soluble flavor compounds such as amino acids, peptides, short-chain fatty acids and organic acids. The extracts provided well-defined and high-quality spectra of cheese samples, showed significant and prominent changes during the ripening process. Overlaid FTIR spectra of WSF of a cheese sample on days 7, 15, 32, 46 and 73 are shown in Figure 4.1.
Amide I and Amide II of Proteins -C=O of acids -C-H stretching 0.7 and esters Esters and aliphatic in fatty acids chains of fatty acids 0.6 C=O and C-C -O-H stretching stretching e
c 0.5 in hydroxyl groups modes of acids n a
b 0.4 r o s
b 0.3 A 0.2
0.1
0.0
4000 3600 3200 2800 2400 2000 1600 1200 800 Wavenumber (cm-1) Figure 4.1: Overlaid spectra of Cheddar cheese sample at various stages of ripening. A gradual increase in the absorbance was observed in the region (1800-900 cm-1) from day- 7 (bottom most spectrum) to day-73 (top most spectrum).
The spectral range 1800 to 900 cm-1, containing signal from polypeptides, amino acids, carbonyl groups of fatty acids, hydroxyl groups, carboxylic acid groups and fatty acid esters (typically short-chain), has been reported to be important to analysis of cheese flavor (Subramanian et al., 2009a; Subramanian et al., 2009b). Other absorbances in the spectra include O-H and N-H stretching vibrations of hydroxyl groups and Amide A of polypeptides and amino acids in the region 3600-3100 cm-1, and C-H stretching vibrations of –CH3 and >CH2 functional groups of fatty acids appear between 3100 and 2800 cm-1. Transformation of spectra to second derivative prior to multivariate analysis improved band resolution, reduced noise, variability between replicates and effects of 72 baseline shifts (Kansiz et al., 1999). The sample preparation procedures were slightly different for GC and HPLC analysis. The methods were modified to better suit the analytical conditions used for chromatography.
4.4.1 Prediction of Amino Acids and Organic Acids
The amino acids (20) and organic acids (3) present in the WSF of cheese samples were analyzed by GC and HPLC, respectively. The concentrations thus determined were correlated with the FTIR spectrum and analyzed by partial least squares regression (PLSR) analysis to develop prediction models. PLSR was developed by Herman Wold (1975) to analyze data involving collinearity problems. A detailed description of PLSR as it applies to analysis of spectroscopic data can be found in the book chapter by Romía and Bernàrdez (2008). In PLSR, the y-variables (dependent variables: amino acid/organic acid concentrations) are related to x-variables (independent variables or infrared wavenumbers) through auxiliary variables called latent variables or PLSR factors. These factors are linear combinations of x-variables (wavenumbers). PLSR attempts to include as much predictive information as possible in the first few factors. It uses the whole data matrix (y-variables vs. x-variables) during calibration and reduces the data appropriate to explain as much variance as possible in both y and x. In doing so, PLSR reduces the effect of irrelevant variations in the x-variables (noise vibrations in the spectrum). Cross- validation with leave-one-out approach was followed to improve model performance. Outliers were eliminated based on outlier diagnostics provided by PLSR.
4.4.1.1 Prediction of Amino Acids
Standard curves were developed using standard mixtures containing 21 amino acids. The sample preparation for GC-FID analysis of both standards and cheese WSF involved a derivatization step to impart volatility to amino acids. Excellent correlations with low residual standard deviations (RSD) were observed between the peak area and amino acid concentration. The retention times, correlation coefficient (r-value) and RSD values observed are summarized in Table 4.1. The r-values for all the amino acids were
73 Retention Time Correlation Residual Standard Amino Acids (min) Coefficienta Deviation Alanine 1.929 0.99 0.05 Glycine 2.035 0.99 0.04 Valine 2.226 1.00 0.04 Leucine 2.423 1.00 0.04 Isoleucine 2.480 1.00 0.04 Threonine 2.680 0.99 0.05 Serine 2.722 0.98 0.04 Proline 2.797 0.99 0.07 Asparagine 2.877 0.99 0.05 Aspartic Acid 3.373 0.99 0.06 Methionine 3.414 1.00 0.04 Hydroxyproline 3.542 0.99 0.05 Glutamic Acid 3.698 0.99 0.04 Phenylalanine 3.747 1.00 0.06 Glutamine 4.302 1.00 0.02 Arginine 4.647 1.00 0.05 Lysine 4.886 1.00 0.05 Histidine 5.067 1.00 0.05 Tyrosine 5.325 1.00 0.06 Tryptophan 5.622 1.00 0.07 Cystineb 6.247 1.00 0.05
aRounded to nearest 2 digits (value of 1.00 implies an r-value ≥ 0.995). bNot detected in cheese WSF.
Table 4.1: Retention times, correlation coefficient and residual standard deviation of amino acid standards determined by GC-FID.
74 ≥0.99 and RSDs were ≤0.07. The total run time per sample was about 6.5 min. The GC- FID chromatograms of water soluble extracts from cheese had distinct peaks with good resolution. A typical chromatogram is shown in Figure 4.2. The amino acid concentrations in cheese thus determined were matched with the corresponding FTIR spectrum of the sample and analyzed by PLSR to develop prediction models.
A PLSR model showing the correlation between GC-FID determination and FTIR prediction of alanine concentration in cheese WSF is shown in Figure 4.3. It is evident from the high coefficient of correlation (r-value) of 0.99 that the model linearly-fits the actual data well. The standard error of cross-validation (SECV), which is an estimate of the error expected when independent samples are predicted using the model, and the number of PLSR factors were 12.70 nmols/g cheese and 15, respectively. The r-values and SECV obtained for prediction of other amino acids in the cheese WSF, along with concentration range, number of PLSR factors used by the models and the total variance explained by the model are listed in Table 4.2. The r-values were ≥0.88 and the SECV were ≤96 nmols/g cheese, clearly indicating the potential of this technique to accurately determine amino acid concentrations in cheese. Furthermore, it is worth mentioning that the errors in PLSR models reported above are not entirely from FTIR spectroscopy alone and include errors from chromatographic analysis as well. The factors for all the models explained >99% of the variance.
4.4.1.2 Prediction of Organic Acids
Three organic acids, oxalic, formic and lactic acid, were identified in cheese WSF based on the retention times. Standard curves were developed for the three acids, correlating the peak area with concentration. Table 4.3 summarizes the retention times, r- values, and RSDs. All the r-values were ~1.00 (>0.995) and the RSDs were ≤0.03. The level of oxalic, formic and lactic acid in cheese WSF was determined based on the HPLC peak area and the developed standard curves. An example HPLC chromatogram of cheese WSF is shown in Figure 4.5. The total run time per sample was about 20 min.
75 6.46.4 6.16.1 5.85.8 2121 5.45.4 1313 – – Hydroxyproline Hydroxyproline (3.542) (3.542) 1414 – – Glutamic Glutamic Acid Acid (3.698)(3.698) 1515 – – Phenylalanine Phenylalanine (3.747) (3.747) 1616 – – Glutamine Glutamine(4.302)(4.302) 1717 – – Arginine Arginine (4.647) (4.647) 1818 – – Lysine Lysine (4.886)(4.886) 1919 – – Histidine Histidine (5.067) (5.067) 2020 – – Tyrosine Tyrosine(5.325)(5.325) 2121 – – Tryptophan Tryptophan (5.622) (5.622) 2020 5.15.1 1919 1818 4.84.8 1717 4.44.4 1616 4.14.1 66 – – Isoleucine Isoleucine (2.480) (2.480) 77 – – Threonine Threonine (2.680) (2.680) 88 – – Serine Serine (2.722) (2.722) 99 – – Proline Proline (2.797) (2.797) 1010 – – Asparagine Asparagine (2.877) (2.877) 1111 – – Aspartic Aspartic Acid Acid(3.373)(3.373) 1212 – – Methionine Methionine (3.414) (3.414) TimTim e e(m(m in) in) 3.83.8 1414 1515 1313 1212 3.43.4 1111 3.13.1 1010 99 2.82.8 88 11 –– Alanine Alanine (1.929) (1.929) 22 –– Glycine Glycine (2.035) (2.035) 33 –– Valine Valine (2.226) (2.226) 44 –– Norvaline Norvaline (2.333) (2.333) 55 –– Leucine Leucine (2.423) (2.423) 77 66 55 2.42.4 44 33 2.12.1 22 11 1.81.8
00
Figure 4.2: GC-FID chromatogram of a Cheddar cheese extract showing the amino acid profile. The amino acids and their of a Cheddar cheese extract showing the amino acid profile. The Figure 4.2: GC-FID chromatogram 6.5 min. Norvaline was used as the internal standard (IS). retention times are identified. The total run time was approximately 1010 2020 3030 4040 5050 6060 7070 8080
A A p p
76 400
350 ) e s e e
h 300 c
g / s l 250 o m n ( 200 R I T F
y 150 b
e n i
n 100 a l A 50
0 0 50 100 150 200 250 300 350 400 Alanine by GC-FID (nmols/g cheese)
Figure 4.3: PLSR model showing the correlation between alanine (nmols/g cheese) predicted by GC-FID and FTIR. The model shows high degree of linear correlation (r- Value = 0.99) and low standard error of cross-validation (SECV; 12.70 nmols/g cheese). A total of 15 PLSR factors were used to explain the variance.
77 Total SECV Range (nmols/g Variance Amino Acid r-Value (nmols/g Factors cheese)a Explained cheese) (%) Alanine 43.44 – 362.96 0.99 12.70 15 99.34 Glycine UD – 302.98 0.98 13.93 19 99.68 Valine 50.48 – 819.75 0.99 28.68 19 99.67 Leucine 115.34 – 1838.75 0.99 53.39 19 99.67 Isoleucine UD – 108.02 0.96 8.21 19 99.67 Threonine UD – 230.09 0.96 16.41 15 99.34 Serine UD – 295.47 0.95 23.07 16 99.43 Proline 73.02 – 283.78 0.88 19.05 19 99.67 Asparagine UD – 1270.74 0.98 68.35 19 99.67 Aspartic Acid 43.80 – 236.16 0.94 14.03 19 99.67 Methionine UD – 201.38 0.99 8.90 19 99.68 Hydroxyproline UD – 108.51 0.94 10.48 17 99.56 Glutamic Acid 192.75 – 2175.80 0.98 95.76 19 99.68 Phenylalanine 73.35 – 955.13 0.99 39.97 18 99.63 Glutamine UD – 613.27 0.97 27.07 19 99.68 Arginine 55.34 – 672.00 0.98 25.28 19 99.67 Lysine 85.97 – 672.40 0.97 25.09 19 99.67 Histidine UD – 133.99 0.83 13.83 18 99.59 Tyrosine 28.52 – 338.21 0.97 11.60 19 99.67 Tryptophan UD – 57.30 0.86 4.68 19 99.68
aUD – Undetectable.
Table 4.2: PLSR model parameters for determination of amino acids by GC-FID.
78 Retention Time Correlation Residual Standard Organic Acids (min) Coefficienta Deviation Oxalic 1.154 1.00 0.01 Formic 1.961 1.00 0.03 Lactic 2.791 1.00 0.01
aRounded to nearest 2 digits (value of 1.00 implies an r-value ≥ 0.995).
Table 4.3: Retention times, correlation coefficient and residual standard deviation of organic acid standards determined by HPLC.
400 1 1 – Oxalic Acid (1.154) 350 2 – Formic Acid (1.961) 300 3 – Lactic Acid (2.791) 250 U
A 200
m 2 3 150
100
50
0 0.0 2.7 5.3 8.0 10.7 13.3 16.0 18.7 Time (min)
Figure 4.4: HPLC chromatogram of a Cheddar cheese extract showing the organic acids profile. Oxalic, formic and lactic acid and their retention times are identified. The total run time was approximately 20 min.
The concentration of oxalic, formic and lactic acids in each sample, determined from the above standard curves, were assigned as the dependent variable to the corresponding FTIR spectra (with wavenumbers as independent variables) of the sample analyzed by PLSR analysis. The model parameters such as the concentration range, r- 79 value, SECV, number of factors and the variance explained are listed in Table 4.4 and the PLSR model for prediction of lactic acid is shown in Figure 4.5. The r-values were ≥0.95 and the SECVs were ≤12.80 μmols/g cheese. A total of 19, 18, and 15 PLSR factors were required to explain >99% of the model variance for prediction oxalic, formic and lactic acid, respectively.
Total Range SECV Variance Organic Acid (μmols/g r-Value (μmols/g Factors Explained cheese) cheese) (%) Oxalic 3.64 – 11.69 0.98 0.35 19 99.63 Formic 46.80 – 184.10 0.95 9.90 18 99.62 Lactic 120.60 – 331.93 0.96 12.80 15 99.35
Table 4.4: PLSR model parameters for determination of organic acids by HPLC.
4.4.2 Biochemical Changes during Ripening
Several biochemical changes occur during the ripening process, affecting the flavor, composition and texture of cheese. The length of ripening period (age) significantly influences the flavor of cheese. The possibility of rapidly predicting the endpoint of cheese ripening to achieve acceptable level of flavor development is a prospect that holds many advantages for the cheese-maker. To investigate the possibility of predicting the age of the cheese based on the composition of WSF, the FTIR spectra of cheese WSF were correlated to the age of the samples and analyzed by PLSR. It is evident from Figure 4.6 that a high degree of correlation (r-value > 0.99) exists between the age of the cheese and FTIR spectrum of WSF from cheese. The age of a cheese could be predicted within an estimated error of ± 1.42 days.
80 350 ) e s e
e 300 h c
g / s l o
m 250 m (
R I T F
y 200 b
d i c A
c i
t 150 c a L
100 100 150 200 250 300 350 Lactic Acid by HPLC (mmols/g cheese)
Figure 4.5: PLSR model showing the correlation between lactic acid (μmols/g cheese) predicted by HPLC and FTIR. The model shows high degree of linear correlation (r- Value = 0.96) and low standard error of cross-validation (SECV; 12.80 μmols/g cheese). A total of 15 PLSR factors were used to explain the variance.
81 90
80
) 70 s y a
D 60 (
e g A
50 d e t c
i 40 d e r P 30 R I T F 20
10
0 0 10 20 30 40 50 60 70 80 Actual Age (Days)
Figure 4.6: PLSR model showing the correlation of age of the cheese (days) to FTIR spectrum. The model shows high degree of linear correlation (r-Value > 0.99) and low standard error of cross-validation (SECV; 1.42 days). A total of 18 PLSR factors were used to explain the variance.
82 4.4.2.1 Changes in Amino acids and Organic acid Concentrations
Amino acid concentrations undergo significant changes during the ripening process due to formation of new amino acids from proteolysis and breakdown to amino acids to form volatile flavor compounds. Similarly organic acids change in concentrations due to several simultaneous and sequential reactions. A box plot of total amino acids (Figure 4.7A) and organic acids (Figure 4.7B) clearly shows an overall increase in the concentration during ripening. In general, most significant increase in free amino acid and organic acid concentration was observed between days 15 to 32 of ripening. The concentration of 20 amino acids determined using GC-FID and 3 organic acids determined by HPLC were analyzed using analysis of variance (ANOVA) to statistically compare the concentrations at various stages of ripening and identify compounds that showed significant changes during the ripening process. Comparisons were made by 1- way ANOVA analysis with Tukey’s pairwise comparison at a family error rate of 5%. For each compound, the mean concentration from all 12 samples, were compared on the 5 sampling days (7, 15, 32, 46 and 73). The individual confidence intervals while comparing the concentration of each compound on five sampling days was 99.35%. The results of such comparison for 20 amino acids and 3 organic acids are tabulated in Table 4.5. All amino acids and organic acids, except alanine and tryptophan, showed no significant change in concentration from day 7 to 15 and from day 32 to 46. However, the concentrations increased significantly from day 15 to 32 and from day 46 to 73. Alanine showed significant change in concentration from day 7 to 15 while tryptophan showed a significant increase in concentration only after 46 days of ripening.
Amino acids and organic acids that are present in significantly higher levels than others at the end of ripening were determined by performing a 1-way ANOVA of their concentrations in 73-days old cheese samples. The normality of the data, a prerequisite for comparison of means by ANOVA, was verified using normal probability plots.
83 600
500 ) g / s l o 400 m n ( s d i
c 300 A o n i
m 200 A l a t o T 100
0 7 15 32 46 73 A Age
600 )
g 500 / s l o m μ ( 400 s d i c A c i
n 300 a g r O l a t
o 200 T
100 7 15 32 46 73 B Age
Figure 4.7: Concentration of total free amino acids (A) and organic acids (B) at various stages of ripening. The total free amino acids included 20 amino acids that were determined in cheese. The total organic acids included oxalic, formic and lactic acids.
84 Change in Concentration during Ripeningb Compound P Valuea 7 to 15 15 to 32 32 to 46 46 to 73 Alanine ~0.000 S S NS S Glycine ~0.000 NS S NS S Valine ~0.000 NS S NS S Leucine ~0.000 NS S NS S Isoleucine ~0.000 NS S NS S Threonine ~0.000 NS S NS S Serine ~0.000 NS S NS S Proline ~0.000 NS S NS S Asparagine ~0.000 NS S NS S Aspartic Acid ~0.000 NS S NS S Methionine ~0.000 NS S NS S Hydroxyproline ~0.000 NS S NS S Glutamic Acid ~0.000 NS S NS S Phenylalanine ~0.000 NS S NS S Glutamine ~0.000 NS S NS S Arginine ~0.000 NS S NS S Lysine ~0.000 NS S NS S Histidine ~0.000 NS S NS S Tyrosine ~0.000 NS S NS S Tryptophan ~0.000 NS NS NS S Oxalic ~0.000 NS S NS S Formic ~0.000 NS S NS S Lactic ~0.000 NS S NS S
aA significant P value implies that the mean concentration of each compound on the 5 sampling days were not the same. bSignificant change (S) and no significant change (NS) were evaluated using Tukey's pairwise comparison with 95% simultaneous confidence intervals.
Table 4.5: One-way analysis of variance (ANOVA) of amino acid and organic acid concentration at various stages of ripening.
85 At a level of significance of 0.05 the mean concentrations of at least some amino acids and organic acids were found to be significantly different (P = ~0.000). Some of the amino acids that were present in significantly high concentrations were glutamic acid (65.85 nmols/g cheese), leucine (65.20 nmols/g cheese), asparagine (42.88 nmols/g cheese), phenylalanine (35.84 nmols/g cheese), valine (29.71 nmols/g cheese), arginine (20.84 nmols/g cheese), lysine (19.43 nmols/g cheese), glutamine (14.68 nmols/g cheese), alanine (14.49 nmols/g cheese) and glycine (10.18 nmols/g cheese). Glutamic acid is an important intermediate product of proteolysis and amino acid catabolism and plays a significant role in the development of cheese flavor. Its concentration was found to be the highest in all cheese samples and it showed the most significant increase in concentration in all the 12 samples during the ripening process. In the case of organic acids, lactic acid concentration was highest at 262.30 μmols/g cheese, followed by formic acid (137.74 μmols/g cheese) and oxalic acid (9.33 μmols/g cheese).
4.4.2.2 Spectral Changes during Ripening
The 12 samples as a group were significantly different in composition at various stages of ripening. The samples were grouped based on their age and analyzed by SIMCA, a classification method. SIMCA, which was introduced by Svante Wold (Wold, 1976), allows for grouping multivariate data sets based on similarities, in this case based on age. It is based on principal component analysis (PCA), which is a statistical procedure that extracts small number of latent factors to explain the major variation in the data set (Romía and Bernàrdez, 2008; Jolliffe, 2002; Mark, 2001). Essentially, it reduces the data set to a small number of latent factors, typically less than 5, called principal components (PCs). The PCs are orthogonal to each other and hence are uncorrelated. PCA reduces the size of the data set, removes useless information and eliminates correlation between the data. In simple terms, it involves performing a PCA on each class (age – 7, 15, 32, 46, and 73 days) in the data (Ballabio and Todeschini, 2008; Lavine, 2000). Appropriate numbers of PCs are retained to account for most of the variation within each class. Each class is represented by a PC model. Classes with close average 86 residual variance will lie close to each others while those with a large difference in their average residual variance will lie far in the SIMCA classification plot. SIMCA allows for working with smaller sample sizes (number of samples less than number of variables) without concerns of collinearity and classification by chance (Lavine et al., 1988). Figure 4.8 shows the SIMCA classification of the cheese samples based on their age. Evidently, the samples were different in their chemical composition at different stages of ripening. Samples of the same age formed fairly distinct clusters from samples of other ages, implying similarities in the composition within an age group and dissimilarities from other age groups.
The distance between the classes in a SIMCA plot is represented by the interclass distance (ICD). Greater the distance between two clusters the greater is the difference in composition of samples belonging to those clusters. As a rule of thumb, a distance of over 3 indicates that the samples are well separated and hence different (Kyalheim and Karstand, 1992). The ICDs shown in Table 4.6 clearly indicate that the average residual variance between each age cluster was significantly different (> 3.0) from the others to be considered chemically distinct. Among the four ripening periods (7-15, 15-32, 32-46, and 46-73), the period from 15 to 32 exhibited the greatest change in composition in all the 12 samples. This is derived from the facts that clusters 15 and 32 were very visually separate in Figure 4.8 and the ICD between cluster 15 and cluster 32 (9.4) was greater an the ICDs between 7 and 15 (4.2), 32 and 46 (4.0), and 46 and 73 (6.5). Similar observations were reported for rheological properties of Cheddar cheese by Baron (1949).
87 PC2
95% Confidence Cloud
PC3 PC1
73 46 32 15 7
Figure 4.8: SIMCA classification plot showing the overall compositional difference between samples at various stages of ripening. The FTIR spectra of 12 cheese samples as a group were compositionally different on days 7, 15, 32, 46 and 73 of ripening. The 95% confidence cloud represents the probability of a sample within the cloud actually belonging to the class covered by the cloud.
88 Ripening Day 7 15 32 46 73 7 0.0 15 4.2a 0.0 32 14.9 9.4b 0.0 46 18.8 12.7 4.0c 0.0 73 23.8 17.7 9.0 6.5d 0.0
a, b, c, and dStage 1 (day 7 to 15), 2 (day 15 to 32), 3 (day 32 to 46), and 4 (day 46 to 73) of ripening.
Table 4.6: Interclass distances between classes in SIMCA classification of the cheese samples based on age. Greater the distance between two clusters the greater is the chemical difference between them.
SIMCA also provides additional features known as discriminating power that helps improve analysis. The discriminating power shows how a particular wavenumber is helping the PCs to classify samples in the data set. Wavenumbers with high discriminating powers are important to the model while those with low total low discriminating power are eliminated from the analysis to reduce noise. The greater the discriminating power at a particular wavenumber, the greater is the difference in composition between the classes in functional groups associated with that wavenumber. The discriminating power plot for discrimination of the cheese samples based on their age is shown in Figure 4.9. The major discrimination was due to the wavenumber ranges 1500-1300 and 1750-1650 cm-1. The bands between 1450 and 1410 cm-1 contain absorbances from acidic amino acids such as glutamic acid and the aliphatic chains of fatty acids. This finding, combined with the previous observation that glutamic acid concentrations were the highest all samples, further emphasizes the importance of glutamic acid in Cheddar cheese ripening and flavor. The extent of proteolysis and fatty acid breakdown were also significant to the discrimination of cheese based on age. The samples also exhibited variations in the wavenumber range 1750-1650 cm-1, consisting of absorbance from esters of fatty acids. This can be attributed to differences in the level and rate of lipolysis at different stages of ripening.
89 4000 Amino acids and Aliphatic Chains of Fatty Acids
3500 Amide bonds 3000 of Peptides r e w
o 2500 P
Organic Acids
g Esters of n (C-O and C-C i t 2000 a Stretching Vibrations Fatty Acids n i m i
r 1500 c s i D 1000
500
0 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 Wavenumber (cm-1)
Figure 4.9: Discriminating power plot for classification of Cheddar cheese samples based on age. The regions of the FTIR spectra that contributed to the discrimination are highlighted. Higher discriminating power at a particular wavenumber is related to greater difference between the samples in the chemical groups associated with that wavenumber.
90 When the 12 cheese samples were compared separately on 5 sampling days, the samples exhibited significant change in composition during ripening. For example, a SIMCA classification plot of a sample on days 7, 15, 32, 46 and 73 of ripening is shown in Figure 4.10. All the five samples formed distinct clusters that were well separated in the principal component space. Similar, trends were observed in the remaining 11 cheese samples. The discrimination was due to absorbances from amino acids, peptides, short- chain fatty acids and organic acids in the region 1800 to 900 cm-1.
PC2
PC3
Day 30 Day 7
Day 73 PC1
Day 15
Day 45
Figure 4.10: SIMCA classification plot showing the overall compositional difference in a sample at various stages of ripening. The same sample at different stages of ripening had significant differences in its composition, leading to formation of distinct and separate clusters in the SIMCA plot.
91 A more detailed look at the changes occurring during ripening can be obtained by examining the changes occurring at every stage of ripening. This involves integrating results from SIMCA analysis of 4 stages of ripening: 1) 7 to 15, 2) 15 to 32, 3) 32 to 46 and 4) 46 to 73. Figure 4.11 shows the overlaid discriminating power plots for SIMCA analyses of a sample that attained a final flavor quality of “Good Creamy Cheddar”. Essentially, this plot highlights the compositional differences between day 7 and 15 (stage 1), 15 and 32 (stage 2), 32 and 46 (stage 3), and 46 and 73 (stage 4). The infrared bands that changed, along with the extent of change on a relative scale are shown in the figure. Between day 7 and 15, minor changes occurred in fatty acid and amino acid composition, which could potentially represent initial stages of fat and protein breakdown. The period from day 15 to 32 (stage 2) exhibited the greatest amount of changes, possibly due to heightened protein and fat breakdown (1650 – 1300 cm-1 and 1710 cm-1). The discriminating power showing the spectral changes during the second stage (represented on left axis), was at least 3 times the discriminating power observed during other ripening stages. This further confirms the fact that most significant changes in texture and composition occur close to 30 days of ripening. The third stage (day 32 to 46) shows the appearance of breakdown products, especially organic acids (1200-900 cm-1). The final stage (day 46 to 73) of Cheddar cheese ripening showed relatively less intense but numerous changes in the regions that correspond to amino acids, short-chain fatty acids and organic acids, which could signify formation of flavor related minor compounds. Thus, the described FTIR technique could provide valuable information on flavor related biochemical changes during ripening. Such information could be of help in understanding flavor formation during cheese ripening as well as monitoring and controlling cheese ripening process to achieve desired flavor quality.
92 70000 20000
18000 60000 ) s 2
16000 t 3 i Day 46 to73 n o t U
5
50000 y r 1 14000
a y r t a i D b
12000 r –
A ( r 40000
e r e w o Day 32 to 46 10000 w o P
P g 30000 n g i t 8000 n i a t n a i n i m
i 6000 m r
20000 i c r s c i s i D 4000 D 10000 Day 15 to 32 2000
Day 7 to 15 0 0 903 1096 1288 1481 1674 Wavenumber (cm-1)
Figure 4.11: Biochemical changes in Cheddar cheese during ripening. The spectral changes during the four ripening stages are highlighted: stage 1, 3, and 4 on the right axis and stage 2 on the left axis. The higher the discriminating power at a particular wavenumber the greater is the extent of the change in functional groups associated with that wavenumber.
93 4.5 CONCLUSION
Infrared technology has much to offer to quality control and analysis of cheese. This research cleared showcased the potential of FTIR spectroscopic analysis of water soluble fractions from cheese to simultaneously monitor age, amino acids and organic acids in cheese in less than 20 min per sample. FTIR spectroscopy also provided information on biochemical changes during Cheddar cheese ripening, that were correlated to chromatographic data. The ability FTIR to monitor multiple components simultaneously offers certain unique advantages over conventional methods, which can beneficial to rapid flavor quality analysis by spectroscopy. With its potential for prediction of cheese flavor quality and composition being previously established, good prospects exists for this technique as a rapid and simple quality control as well as research tool. Furthermore, the advent of hyphenated technologies such as GC-FTIR and HPLC-FTIR has opened new avenues for research on cheese ripening and flavor development.
94 LIST OF REFERENCES
Adams, M. J, Latham, K., Barnett, N. W., and Poynton, A. J. 1999. Calibration models for determining moisture and fat content of processed cheese using near-infrared spectrometry. J. Sci. Food Agric. 79:1232-1236.
Adda, J., Gripon, J. C., and Vassal, L. 1982. The chemistry of flavor and texture generation in cheese. Food Chem. 9:115-129.
Akalin, A. S., Gonc, S., and Akba, Y. 2002. Variation in organic acids content during ripening of pickled white cheese. J. Dairy Sci. 85:1670-1676.
Alli, I., Okoniewska, M., Gibbs, B. F., and Konishi, Y. 1998. Identification of peptides in Cheddar cheese by electrospray ionization mass spectrometry. Int. Dairy J. 8:643- 649.
Ardo, Y., and Gripon, J. 1991. Chromatographic methods used to measure proteolysis in cheese. Int. Dairy Fed. Bulletin. 261:29-34.
Aston, J. W. and Dulley, J. R. 1982. Cheddar cheese flavor. The Aust. J. Dairy Tech. 37:59-64.
Aston, J. W., Durwardi, I. G., and Dulley, J. R. 1983. Proteolysis and flavour development in Cheddar cheese. Aust. J. Dairy. Tech. 38:55-59.
Aston, J. and Creamer, L. K. 1986. Contribution of the components of water-soluble fraction to the flavor Cheddar cheese. New Zealand Dairy Sci. Technol. 21:229- 248.
95 Bachmann, H. P., Butikofer, U., and Meyer, J. 1999. Prediction of flavor and texture development in Swiss-type cheeses. Lebensm.-Wiss. U.-Technol. 32:284-289. Ballabio, D., and Todeschini, R. 2008. Multivariate classification for qualitative analysis. Pages 83-104 in Infrared Spectroscopy for Food Quality Analysis and Control. D. W. Sun, ed. Academic Press, New York, NY.
Baron, M. 1949. Further studies of rheological properties of cheese during manufacture and ripening. Dairy Ind. 14:146-151.
Bell, R. J. 1972 Introductory Fourier transform spectroscopy. Academic Press Inc., New York, NY. p 382.
Bjorsvik, H. R., and Martens, H. 1992. Data analysis: calibration of NIR instruments by PLS regression. Pages 150-180 in Handbook of Near-Infrared Analysis. D. A., Burns, ed. Taylor and Francis, New York, NY.
Blazquez, C., Downey, G., O’Callaghan, D., Howard, V., Delahunty, C., Sheehan, E., Everard, C., and O’Donnell, C. P. 2006. Modeling of sensory and instrumental texture parameters in processed cheese by near infrared reflectance spectroscopy. J. Dairy Res. 73:58-69.
Burns, D. A., and Ciurczak, E. W. 2001. Hand book of near-infrared analysis. Taylor & Francis, New York, NY. p 814.
Butikofer, U., and Ardo, Y. 1999. Quantitative determination of free amino acids in cheese. Int. Dairy. Fed. Bull. 337.24-32.
Candofi, A., De Maesschalck, R., Massart, D. L., Hailey, P. A., and Harrington, A. C. E. 1999. Identification of pharmaceutical excipients using NIR spectroscopy and SIMCA. J. Pharma. Biomed. Anal. 19:923-35.
Cattaneo, T. M. P., Giardina, C., Sinelli, N., Riva, M., and Giangiacomo, R. 2005. Application of FT-NIR and FTIR spectroscopy to study the shelf-life of Crescenza cheese. Int. Dairy J. 15:693-700.
Chalmers, J. M., and Griffiths, P. R. 2002. Handbook of vibrational spectroscopy. Wiley & Sons, New York, NY. p 4000.
96 Chen, G. 2008. Monitoring the effect of adjunct cultures during Swiss cheese ripening by infrared microspectroscopy. The Ohio State University. MS Thesis. p 69.
Chen, G., Kocaoglu-Vurma, N. A., Harper, W. J., and Rodriguez-Saona, L. E. 2009. Application of infrared microspectroscopy and multivariate analysis for monitoring the effect of adjunct cultures during Swiss cheese ripening. J. Dairy Sci. Forthcoming.
Chen, M., and Irudayaraj, J. 1998. Sampling technique for cheese analysis by FTIR spectroscopy. J. Food Sci. 63:96-99.
Chen, M., Irudayaraj, J., and McMahon, D. J. 1998. Examination of full fat and reduced fat Cheddar cheese during ripening by Fourier transform infrared spectroscopy. J. Dairy Sci. 81:2791-2797.
Coates, J. 2000. Interpretation of Infrared Spectra, a Practical Approach. In Encyclopedia of Analytical Chemistry, Meyers R R ed. pp. 10815-10837. John Wiley and Sons Ltd., Chichester, UK.
Cogan, T. M., and Hill, C. 1993. Cheese starter cultures. Pages 193-256 in Cheese: Chemistry, Physics and Microbiology, Second Edition – Volume 1: General Aspects. P. F. Fox, ed. Chapman & Hall, New York, NY.
Collins, Y. F., McSweeney, P. L. H., and Wilkinson, M. G. 2003a. Lipolysis and free fatty acid catabolism in cheese: a review of current knowledge. Int. Dairy. J. 13:841-866.
Collins, Y. F., McSweeney, P. L. H., and Wilkinson, M. G. 2003b. Evidence for a relationship between autolysis of starter bacteria and lipolysis in Cheddar cheese. J. Dairy Res. 70:105-113.
Collins, Y. F., McSweeney, P. L. H., and Wilkinson, M. G. 2004. Lipolysis and catabolism of fatty acids in cheese. Pages 373-389 in Cheese: Chemistry, Physics and Microbiology, Third Edition – Volume 2: Major Cheese Groups. P. F. Fox, ed. Chapman & Hall, New York, NY.
Cooley, J. W., and Tukey, J. W. 1965. An algorithm for the machine calculation of complex Fourier series. Math. Comp. 19:297-301.
97 Creamer, L. K., Lawrence, R. C., and Gilles, J. 1985. Effect of acidification of cheese milk on the resultant Cheddar cheese. NZ J. Dairy Sci. Technol. 20:185-203.
Creamer, L. K., Gilles, J., and Lawrence, R. C. 1988. Effect of pH on the texture of Cheddar and Colby cheese. NZ J. Dairy Sci. Technol. 23:23-35.
Creamer, L. K. 1991. Electrophoresis of cheese. Int. Dairy Fed. Bull. 261.14-28.
Christy, A. A., Ozaki, Y., and Gregoriou, V. G. 2001. Modern Fourier transform infrared spectroscopy. Elsevier, New York, NY. p 356.
Curtin, Á. C., and McSweeney, P. L. H. 2004. Catabolism of amino acids in cheese during ripening. Page 435-454 in Cheese: Chemistry, Physics and Microbiology, Third Edition – Volume 2: Major Cheese Groups. P. F. Fox, ed. Chapman & Hall, New York, NY.
De Maesschalck, R., Candofi, A., Masart, D. L., and Heuerding, S. 1999. Decision criteria for soft independent modeling of class analogy applied to near infrared data. Chemometr. Intell. Lab. Syst. 47:65-77.
Downey, G., Sheehan, E., Delahunty, C., O’Callaghan, D., Guinee, T., and Howard, V. 2005. Prediction of maturity and sensory attributes of Cheddar cheese using near- infrared spectroscopy. Int. Dairy. J. 15:701-9.
Dudley, E., and Steele, J. E. 2005. Succinate production and citrate metabolism by Cheddar cheese non-starter lactic acid bacteria. J. Appl. Microbiol. 98.4-23.
Eliardi, A. 2008. Rapid profiling of Swiss cheese by mid-infrared/attenuated total reflectance spectroscopy. The Ohio State University. MS Thesis.
Engel, E., Lombardot, J. B., Garem, A., Leconte, N., Septier, C., Le Quere, J. L., and Salles, C. 2002. Fractionation of the water-soluble extract of cheese made from goat’s milk by filtration methods: behavior of fat and volatile compounds. Int. Dairy J. 12:609-619.
Engels, W. J. M., Dekker, R., de Jong, C., Neeter, R., and Visser, S. 1997. A comparative study of volatiles compounds in the water-soluble fraction of various types of ripened cheese. Int. Dairy J. 7:255-263.
98 Farkye, N. Y., and Fox, P. F. 1990. Objective indices of cheese ripening. Trends Food Sci. Technol. 1:37-40.
FDA 21 CFR 133.113, 2003. Cheeses and related cheese products. Available at http://www.foodsafety.gov/~lrd/fcf133.html, Accessed on April 19, 2009.
Féliz-Pérez, D. J. 2007. Accelerated ripening by enzyme modified application in Swiss cheese. The Ohio State University. Thesis. p 134.
Ferdandez-Garcia, E., Lopez-Fandino, R., Alonsa, L., and Ramos, M. 1994. The use of lipolytic and proteolytic enzymes in the manufacture of Manchego type cheese. J. Dairy Sc. 77:2139-2149.
Ferraro, J. R. 1999. History of Fourier transform-infrared spectroscopy. Spectroscopy. 14:28-40.
Forman, M. L. 1966. Fast Fourier-transform technique and its application to Fourier spectroscopy. J. Opt. Soc. Am. 56:978-983.
Fox, P. F. 1989. Proteolysis during cheese manufacture and ripening. J. Dairy Sc. 72:1379-1400.
Fox, P. F., Lucey, J. A., and Cogan, T. M. 1990. Glycolysis and related reactions during cheese manufacture and ripening. CRC Crit. Rev. Food Sci. Nutr. 29:237-253.
Fox, P. F., and Wallace, J. M. 1997. Formation of flavour compounds in cheese. Adv. Appl. Microbiol. 45:17-85.
Fox, P. F., Guinee, T. P., Cogan, T. M., and McSweeney, P. L. H. 2000a. Overview of cheese manufacture. Pages 10-17 in Fundamentals of Cheese Science. Aspen Publication, Gaithersburg, MD.
Fox, P. F., Guinee, T. P., Cogan, T. M., and McSweeney, P. L. H. 2000b. Biochemistry of cheese ripening. Pages 236-278 in Fundamentals of Cheese Science. Aspen Publication, Gaithersburg, MD.
Fox, P. F., Guinee, T. P., Cogan, T. M., and McSweeney, P. L. H. 2000c. Factors that affect cheese quality. Pages 341-347 in Fundamentals of Cheese Science. Aspen Publication, Gaithersburg, MD. 99 Fox, P. F., Guinee, T. P., Cogan, T. M., and McSweeney, P. L. H. 2000d. Cheese flavor. Pages 282-303 in Fundamentals of Cheese Science. Aspen Publication, Gaithersburg, MD.
Fox, P. F., Guinee, T. P., Cogan, T. M., and McSweeney, P. L. H. 2000e. Analytical methods for cheese. Pages 523-543 in Fundamentals of Cheese Science. Aspen Publication, Gaithersburg, MD.
Fox, P. F., and Cogan, T. M. 2004. Factors that affect the quality of cheese. Pages 583- 608 in Cheese: Chemistry, Physics and Microbiology, Third Edition – Volume 2: Major Cheese Groups. P. F. Fox, ed. Chapman & Hall, New York, NY.
Griffiths, P. R., and de Haseth, J. A. 1986. Fourier transform infrared spectrometry. John Wiley & Sons, New York, NY. p 656.
Guillén, M. D., and Cabo, N. 1997. Infrared spectroscopy in the study of edible oils and fats. J Sci. Food Agric. 75:1-11.
Harper, W. J. 1949. Role of amino acids in the flavor of Cheddar Cheese. Ph.D Dissertation, University of Wisconsin, Madison, WI. p125.
Harper, W. J. and Kristoffersen, T. 1956. Biochemical aspects of cheese ripening. J. Dairy Sci. 39:1773-1775.
Herschel, W. 1800. Observations tending to investigate the nature of the Sun, in order to find the causes or symptoms of its variable emission of light and heat; with remarks on the use that may possibly be drawn from solar observations. Philoso. Trans. R. Soc. Lon. 1:49-52.
Innocente, N., Moret, S., Corradini, C., and Conte, L. S. 2000. A rapid method for the quantitative determination of short-chain free volatile fatty acids from cheese. J. Agric. Food Chem. 48:3321-3323.
Izco, J., Tormo, M., and Jiménez-Flores, R. 2002. Rapid simultaneous determination of organic acids, free amino acids, and lactose in cheese by capillary electrophoresis. J. Dairy Sci. 85:2122-2129.
Johnston, S. F. 1991. Fourier transform infrared: A constantly evolving technology. Ellis Horwood, New York, NY. p 340.
100 Jolliffe, I. T. 2002. Principal Component analysis. Springer, New York, NY. p 487.
Kansiz, M., Heraud, P., Wood, B., Burden, F., Beardall, J., and McNaughton, D. 1999. Fourier transform infrared microspectroscopy and chemometrics as a tool for the discrimination of cyanobacterial strains. Phytochemistry. 52:407-417.
Karoui, R., Dufour, E., Pillonel, L., Picque, D., Cattenoz, T., and Bosset, J. O. 2004. Determining the geographic origin of Emmental cheeses produced during winter and summer using a technique based on concatenation of MIR and fluorescence spectroscopic data. Eur. Food Res. Technol. 219:184-189.
Karoui, R., Dufour, E., Pillonel, L., Schaller, E., Picque, D., Cattenoz, T., and Bosset, J. O. 2005. The potential of combined infrared and fluorescence spectroscopies as a method of determination of geographic origin of Emmental cheeses. Int. Dairy. J. 15:287-98.
Kilcawley, K., Wilkinson, M., and Fox, P. F. 2001. A survey of lipolytic and glycolytic end-products in commercial Cheddar enzyme-modified cheese. J. Dairy Sci. 84:66-73.
Koca, N., Rodriguez-Saona, L. E., Harper, W. J., and Alvarez, V. B. 2007. Application of Fourier transform infrared (FTIR) spectroscopy for monitoring short-chain free fatty acids in Swiss cheese. J. Dairy Sci. 90:3596-603.
Kocaoglu-Vurma, N. A., Eliardi, A., Drake, M. A., Rodriguez-Saona, L. E., and Harper, W. J. 2009. Rapid profiling of Swiss cheese by attenuated total reflectance (ATR) infrared spectroscopy and descriptive sensory analysis. J. Food Sci. Forthcoming.
Kosikowski, F. V., and Mocquot, G. 1958. Advances in cheese technology. FAO Agric. Stud. Nr. 38, Food and Agriculture Organization (FAO). Rome, Italy.
Kvalheim, O. M., and Karstand, T. V. 1992. SIMCA – classification by means of disjoint cross validated principal components models. Pages 209-248 in Multivariate pattern recognition in chemometrics: illustrated by case studies. R. G. Brereton, ed. Elsevier, Amsterdam, Netherlands.
Lawrence, R. C., Gilles, J., Creamer, L. K., Crow, V. L., Heap, H. A., Honoré, C. G., Johnston, K. A., and Samal, P. K. 2004. Cheddar cheese and related dry-salted cheese varieties. Pages 71-102 in Cheese: Chemistry, Physics and Microbiology,
101 Third Edition – Volume 2: Major Cheese Groups. P. F. Fox, ed. Chapman & Hall, New York, NY.
Lavine, B. K., Jurs, P. C., and Henry, D. R. 1988. Chance classifications by nonparametric linear discriminant functions. J. Chemom. 2:1-10.
Lavine, B. K. 2000. Clustering and classification of analytical data. Pages 9689-9710 in Encyclopedia of Analytical chemistry: instrumentation and applications. R. A. Meyers, ed. John Wiley and Sons Ltd., New York, NY.
Le Quéré, J. L. 2004. Cheese flavour: instrumental techniques. Pages 489-510 in Cheese: Chemistry, Physics and Microbiology, Third Edition – Volume 2: Major Cheese Groups. P. F. Fox, ed. Chapman & Hall, New York, NY.
Lefier, D., Lamprell, H., and Mazerolles, G. 2000. Evolution of Lactococcus strains during ripening in Brie cheese using Fourier transform infrared spectroscopy. Lait. 80:247-54.
Lues, J. F. R., and Bekker, A. C. M. 2002. Mathematical expression for organic acids in early ripening of a Cheddar cheese. J. Food Compos. Anal. 15:11-17.
Maarse, H., and Visscher, C. A. 1989. Volatile compounds in food – Sixth Edition. TNO Division of Nutrition and Food Research, Zeist, The Netherlands.
Manning, D. J., and Robinson, H. M. 1973. The analysis of volatile substances associated with Cheddar cheese aroma. J. Dairy Res. 40:63-75.
Mantsch, H. H., and Chapman, D. 1996. Infrared spectroscopy of biomolecules. Wiley- Liss. Inc. New York, NY. p 359.
Marilley, L., and Casey, M. G. 2004. Flavours of cheese products: metabolic pathways, analytical tools and identification of producing strains. Int. J. Food Microbiol. 90:139-159.
Mark, H. 2001. Multilinear regression and principal component analysis. Pages 129-184 in Handbook of Near-Infrared Analysis. D. A. Burns, and E. W. Ciurczak. ed. Taylor and Francis, New York, NY.
102 Martens, H., and Martens, M. 2001. Multivariate analysis of quality: an introduction. John Wiley and Sons Ltd., New York, NY. p445.
Martín-del-Campo, S. T., Picque, D., Cosío-Ramírez, R., and Corrieu, G. 2007. Evaluation of chemical parameters in soft mold-ripened cheese during ripening by mid-infrared spectroscopy. J. Dairy Sc. 90:3018-3027.
Martín Del Campo, S. T., Bonnaire, N., Picque, D., and Corrieu, G. 2009. Initial studies into the characterization of ripening stages of Emmental cheeses by mid-infrared spectroscopy. Dairy Sci. Technol. 90:155-167.
Mazerrolles, G., Devaux, M. F., Duboz, G., Duployer, M. H., Riou, N. M., and Dufour, E. 2001. Infrared and fluorescence spectroscopy for monitoring protein structure and interaction changes during cheese ripening. Lait. 81:509-27.
McGugan, W. A., Emmons, D. B., and Armond, E. L. 1979. Influence of volatile and nonvolatile fraction on intensity of Cheddar cheese flavor. J. Dairy Sc. 62:398- 403.
McQueen, D. H., Wilson, R., Kinnunen, A., and Jensen, E. P. 1995. Comparison of two infrared spectroscopic methods for cheese analysis. Talanta. 42:2007-2015.
McSweeney, P. L. H. and Fox, P. F. 1993. Cheese: Methods of chemical analysis. Pages 341-388 in Cheese: Chemistry, Physics and Microbiology, Second Edition – Volume 1: General Aspects. P. F. Fox, ed. Chapman & Hall, New York, NY.
McSweeney, P. L. H., and Fox P. F. 1997. Chemical methods for characterization of proteolysis in cheese during ripening. Lait. 77:41-76.
McSweeney, P. L. H., and Sousa, M. J. 2000. Biochemical pathways for the production of flavour compounds in cheeses during ripening: A review. Lait. 80:293-324.
McSweeney, P. L. H. 2004a. Biochemistry of cheese ripening. Int. J. Dairy Technol. 57:127-144.
McSweeney, P. L. H. 2004b. Biochemistry of cheese ripening: introduction and overview. Pages 347-360 in Cheese: Chemistry, Physics and Microbiology, Third Edition – Volume 2: Major Cheese Groups. P. F. Fox, ed. Chapman & Hall, New York, NY.
103 McSweeney, P. L. H., and Fox, P. F. 2004. Metabolism of residual lactose and of lactate and citrate. Pages 361-371 in Cheese: Chemistry, Physics and Microbiology, Third Edition – Volume 2: Major Cheese Groups. P. F. Fox, ed. Chapman & Hall, New York, NY.
Michelson, A. A. 1891. On the application of interference-methods to spectroscopic measurements, -I. Philoso. Mag. 31:338-346.
Michelson, A. A. 1892. On the application of interference-methods to spectroscopic measurements, -II. Philoso. Mag. 34:280-299.
Mulder, H. 1952. Taste and flavor-forming substances in cheese. Neth. Milk Dairy J. 6:157-67.
National Agricultural Statistics Service. 2008. United States Department of Agriculture, Available at http://www.nass.usda.gov/Charts_and_Maps/Dairy_Products/index.asp, Accessed Mar. 31, 2009.
National Dairy Council. 2009. Steps in Cheddar cheese manufacture. Available at http://www.nationaldairycouncil.org/NR/rdonlyres/10F65F52-5156-4A17-8CE0- BDA4ABB72722/0/FIGURE03.pdf, Accessed on April 19, 2009.
Osborne, B. G., and Fearn, T. 1986. Near infrared spectroscopy in food analysis. Longman Scientific & Technical, New York, NY. p 200.
Osborne, G. B., Fearn, T., and Hindle, P. H. 1993. Practical NIR spectroscopy with applications in food and beverage analysis. Longman Scientific & Technical, New York, NY. p 227.
Otte, J., Ardo, Y., Weimer, B., and Sorensen, J. 1999. Capillary electrophoresis used to measure proteolysis in cheese. Int. Dairy Fed. Bull. 337:10-16.
Palles, T., Beresford, T., Condon, S., and Cogan, T. M. 1988. Citrate metabolism in Lactobacillus casei and Lactobacillus plantarum. J. Appl. Microbiol. 85:147-154.
Partidario, A. M., Barbosa, M., and Boas, L. V. 1998. Free fatty acids, triglycerides and volatile compounds in Serra da Estrela cheese-changes throughout ripening. Int. Dairy J. 8:873-881.
104 Partidario, A. 1999. Comparison of free fatty acids, volatile compounds and sensory characteristics of Serra de Estrela cheese at different ripening stages. Milchwissenschaft. 54:381-385.
Pillonel, L., Luginbühl, W., Picque, D., Schaller, E., Tabacchi, R., and Bosset, J. O. 2003. Analytical methods for the determination of the geographic origin of Emmental cheese: mid- and near-infrared spectroscopy. Eur. Food. Res. Technol. 216:174-8.
Rodriguez-Otero, J. L., Hermida, M., and Cepeda, A. 1995. Determination of fat, protein, and total solids in cheese by near-infrared reflectance spectroscopy. J. AOAC Int. 78:802-6.
Rodriguez-Saona, L. E., Koca, N., Harper, W. J., and Alvarez, V. B. 2006. Rapid determination of Swiss cheese composition by Fourier transform infrared/attenuated total reflectance spectroscopy. J. Dairy Sci. 89:1407-1412.
Romía, M. B., and Bernàrdez, M. A. 2008. Multivariate calibration for quantitative analysis. Pages 51-82 in Infrared Spectroscopy for Food Quality Analysis and Control. D. W. Sun, ed. Academic Press, New York, NY.
Salles, C., Septier, C., Roudeot-Algaron, F., Guillot, A., and Etievant, P. X. 1995. Sensory and chemical analysis of fraction obtained by gel permeation of water soluble Compté cheese extracts. J. Agric. Food Chem. 43:1659-1668.
Schieberle, P. 1995. New developments in methods for analysis of volatile flavor compounds and their precursors. Pages 403-432 in Characterization food: emerging methods. Elsevier Science, Amsterdam.
Seitz, E. W. 1990. Microbial- and enzyme-induced flavor in dairy foods. J. Dairy Sci. 73:3663-3691.
Singh, T., Gripon, J., and Fox. P. F. 1999. Chromatographic analysis and identification of peptides in cheese. Int. Dairy Fed. Bull. 337:17-22.
Singh, T. K., Drake, M. A., and K. R. Cadwallader. 2003. Flavor of Cheddar cheese: A chemical and sensory perspective. Comp. Rev. Food Sci. Food Saf. 2:166-189.
105 Skeie, S., Feten, G., Almøy, T., Østlie, H., and Isaksson, T. 2006. The use of near infrared spectroscopy to predict selected free amino acids during cheese ripening. Int. Dairy. J. 16:236-42.
Smith, B. C. 1996. Fundamentals of Fourier transform infrared spectroscopy. CRC Press, New York, NY. p 202.
Soda, K., Tanaka, H., and Esaki, N. 1983. Multifunctional biocatalysis: methionine γ- lyase. Trends Biochem. Sci. 8:214-217.
Sørensen, L., and Jepsen, R. 1998. Assessment of sensory properties of cheese by near- infrared spectroscopy. Int. Dairy J. 8:863-71.
Stuart, B. 2004. Infrared spectroscopy: fundamentals and applications. Wiley, New York, NY. p 242.
Subramanian, A., and L.E. Rodriguez-Saona, L.E. 2008a. Fourier transform infrared (FTIR) spectroscopy. Pages 145-178 in Infrared Spectroscopy for Food Quality Analysis and Control. D. W. Sun, ed. Academic Press, New York, NY.
Subramanian, A., and Rodriguez-Saona, L. E, inventors. 2008b. Rapid Extraction Method for Analysis of Cheese Flavor Using Infrared Spectroscopy. The Ohio State University, assignee. Provisional Appln. Ser. No. 61/059,890.
Subramanian, A., Harper, W. J., and Rodriguez-Saona, L. E. 2009a. Cheddar cheese classification based on flavor quality using a novel extraction method and Fourier transform infrared spectroscopy. J. Dairy Sci. 92:87-94.
Subramanian, A., Harper, J. W, and Rodriguez-Saona, L. E. 2009b. Rapid prediction of composition and flavor quality of Cheddar cheese using ATR-FTIR spectroscopy. J. Food Sci. 74(3): C292-C297.
Tanaka, H., Esaki, N., and Soda, K. 1985. A versatile bacterial enzyme; L-methionine γ- lyase. Enzyme Microbiol. Technol. 7:530-537.
Thierry, A., Maillard, M. B., and Le Quéré, J. L. 1999. Dynamic headspace analysis of Emmental aqueous phase as a method to quantify changes in volatile flavor compounds during ripening. Int. Dairy. J. 9:453-463.
106 Tieleman, A. E., and Warthesen, J. J. 1991. Comparison of three extraction procedures to characterize Cheddar cheese proteolysis. J. Dairy Sci. 74:3686-3694.
United States Department of Agriculture (USDA). 1956. United States Standards for Grades of Cheddar Cheese. Agricultural Marketing Services (AMS), U.S. Department of Agriculture (USDA). Available at http://www.ams.usda.gov/AMSv1.0/getfile?dDocName=STELDEV3004465. Accessed on April 15, 2009.
Upadhyay, V. K., McSweeney, P. L. H., Magoul, A. A. A., and Fox, P. F. 2004. Proteolysis in cheese during ripening. Pages 391-433 in Cheese: Chemistry, Physics and Microbiology, Third Edition – Volume 2: Major Cheese Groups. P. F. Fox, ed. Chapman & Hall, New York, NY.
Urbach, G. 1993. Relations between cheese flavor and chemical composition. Int. Dairy J. 3:389-422.
Urbach, G. 1997. The flavour of milk and dairy products. II. Cheese: contribution of volatile compounds. Int. J. Dairy Technol. 50:79-89.
Weimer, B., Seefeldt, K., and Diaz, B. 1999. Sulfer metabolism in bacteria associated with cheese. Antonie van Leeuwenhoek. 76:247-261.
Wold, H. 1975. Soft modeling by latent variables: The nonlinear iterative partial least squares approach. Pages 520-540 in Perspectives in Probability and Statistics. J. Gani, ed. Academic Press, London, UK.
Wold, S. 1976. Pattern recognition by means of disjoint principal components models. Pat. Recog. 8:127-139.
Yvon, M., and Rijnen, L. 2001. Cheese flavor formation by amino acid catabolism. Int. Dairy J. 11:185-201.
107