Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Objective judgement of cheese varieties by multivariate analysis of HPLC profiles Smith, Anita Mohler 1987

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


831-UBC_1987_A6_7 S64.pdf [ 4.26MB ]
JSON: 831-1.0097082.json
JSON-LD: 831-1.0097082-ld.json
RDF/XML (Pretty): 831-1.0097082-rdf.xml
RDF/JSON: 831-1.0097082-rdf.json
Turtle: 831-1.0097082-turtle.txt
N-Triples: 831-1.0097082-rdf-ntriples.txt
Original Record: 831-1.0097082-source.json
Full Text

Full Text

OBJECTIVE JUDGEMENT OF CHEESE VARIETIES BY MULTIVARIATE ANALYSIS OF HPLC PROFILES By ANITA MOHLER SMITH B. Sc., Oregon State University, 1979  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DECREE OF MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES Department of Food Science  We accept this thesis as conforming to the required standard  THE UNIVERSITY OF BRITISH COLUMBIA June 1987 © A n i t a Mohler Smith, 1987  In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives.  It is understood that copying or  publication of this thesis for financial gain shall not be allowed without my written permission.  Department The University of British Columbia 1956 Main Mall Vancouver, Canada V6T 1Y3  ii  ABSTRACT  An objective analytical method was developed to characterize the taste profiles of five cheese varieties. Nonvolatile water extracts of Cheddar, Edam, Gouda, Swiss, and Parmesan cheeses were analyzed by high performance liquid chromatography (HPLC) with a reversed phase column. The HPLC operating conditions were determined with Mapping Super-Simplex followed by Centroid Mapping Optimization. A ternary gradient elution system was used with an Adsorbosphere C8 column to resolve a maximum number of components. The optimum solvent volume ratio was 96.8 : 1.2 : 2.0 for trifluoroacetic acid (0.1%), acetonitrile, and methanol, with a flow rate of 1.0 mL/min. Over 50.3 min this ratio was changed to 56.3 : 30.3 : 13.4. Multivariate statistical analyses including principal component and discriminant analyses were applied to 55 peak areas from 106 cheese chromatograms. Principal component analysis reduced the dimensionality of the "data from 55 to 17 principal components, which arecombinations of the original variables, with a 26% loss of explained sample variation. Discriminant analysis on data from a single HPLC column was able to correctly classify cheeses by variety at a greater than 90% success rate. This grouping rate dropped to 64% when data from all four HPLC columns was combined, implicating large between column variations. A semi-trained sensory panel correctly classified cheeses by variety at a 63% rate. This objective method provides a lasting fingerprint of cheese products.  iii TABLE O F  C O N T E N T S  ABSTRACT . :  ii  TABLE OF CONTENTS  iii  LIST OF TABLES  vi  LIST OF FIGURES  vii  LIST OF APPENDICES  ix  ACKNOWLEDGEMENT  x  I. INTRODUCTION  1  II. LITERATURE REVIEW  3  A. Development of Cheese Flavor  3  B. Cheddar Cheese Flavor  5  1. Amino Acids and Peptides  5  2. Free Fatty Acids  6  3. Carbonyl Compounds  7  4. Sulfur Compounds  7  (a) Hydrogen Sulfide  7  (b) Methanethiol  7  (c) Dimethyl Sulfide  8  C. Edam and Couda Cheese Flavor  8  1. Peptides and Amino Acids  8  2. Volatile Compounds  9  D. Swiss Cheese Flavor  10  1. Acids in Swiss Cheese  10  2. Miscellaneous Volatile Compounds  12  3. Peptides and Amino Acids  12  4. Sulfur Compounds  13  E. Parmesan Cheese Flavor  13  iv F. Reversed Phase HPLC  14  1. Theory  14  2. Applications of HPLC  15  3. Optimization of HPLC Conditions  16  (a) Columns  16  (b) Temperature  16  (c) Mobile Phase  17  G. Multivariate Analyses  18  1. Principal Component Analysis ;  20  2. Discriminant Analysis  20  3. Applications  21  111. MATERIALS AND METHODS  22  A. Cheese Samples  22  B. Sample Preparation  22  1. Water-Soluble Extract  22  2. pH  24  C. HPLC Analysis  24  1. HPLC Conditions  24  2. Internal Standard  25  D. Optimization  26  E. Sensory Evaluation  26  F. Statistical Analyses  27  1. HPLC Data  27  (a) Principal Component Analysis  28  (b) Discriminant Analysis  28  2. Sensory Data IV. RESULTS AND DISCUSSION  28 29  V  A. Sample Preparation  29  B. HPLC Conditions  30  C. Optimization  -31  D. HPLC Statistical Analyses  45  1. Principal Component Analysis  45  2. Discriminant Analysis  47  E. Sensory Analysis  55  F. Modified Extraction Procedure  65  V. CONCLUSIONS  67  VI. REFERENCES  69  VII. APPENDIX  77  vi LIST OF TABLES  Table 1. 2.  3.  4. 5.  6. 7. 8.  9.  10.  11.  Page Reproducibility of HPLC peak elutions with a Swiss cheese extract . .  32  Starting factor ranges for the optimization of HPLC operating parameters by Mapping Simplex optimization (MSO) and Centroid Mapping Optimization (CMO)  39  Eigenvalue, proportion of variance explained, and cumulative proportion of total variance in a principal component analysis using chromatographic data  46  Percent correct classification rate of cheese variety by discriminant and sensory analysis  49  Coefficients of discriminant functions from principal components for separation of 106 cheeses by variety  51  Summary table for peaks entered in stepwise discriminant analysis  53  Coefficients for discriminant functions from HPLC peaks  54  Eigenvalue, proportion of variance explained, and cumulative proportion of total variance in a principal component analysis using chromatographic data from column 1  56  Eigenvalue, proportion of variance explained, and cumulative proportion of total variance in a principal component analysis using chromatographic data from column 2  57  Eigenvalue, proportion of variance explained, and cumulative proportion of total variance in a principal component analysis using chromatographic data from column 3  58  Eigenvalue, proportion of variance explained, and cumulative porportion of total variance in a principal component analysis using chromatographic data from column 4  59  vii LIST OF FIGURES  Figure 1. 2. 3. 4. 5. 6. 7. 8.  9.  10.  11.  12. 13.  14.  Page Method to determine peak resolution , P=f/g, adapted from Morgan and Deming (1975)  19  Extraction procedure of water-soluble cheese components for HPLC analysis  23  Representative HPLC profile of water-soluble components of Cheddar cheese  33  Representative HPLC profile of water-soluble components of Edam cheese  34  Representative HPLC profile of water-soluble components of Gouda cheese  35  Representative HPLC profile of water-soluble components of Swiss cheese  36  Representative HPLC profile of water-soluble components of Parmesan cheese  37  Mapping responses of experiments to optimize peak resolution. Factor 1, initial methanol concentration  40  Mapping responses of experiments to optimize peak resolution. Factor 2, final methanol concentration  41  Mapping responses of experiments to optimize peak resolution. Factor 3, initial acetonitrile concentration  42  Mapping responses of experiments to optimize peak resolution. Factor 4, final acetonitrile concentration  43  Mapping responses of experiments to optimize peak resolution. Factor 5, time of HPLC run  44  Canonical plot of 106 cheese samples grouped by variety using proportional prior probabilities  50  Canonical plot of cheese samples from column one grouped by variety  60  viii 15. 16. 17. 18.  Canonical plot of cheese samples from column two grouped by variety  61  Canonical plot of cheese samples from column three grouped by variety  62  Canonical plot of cheese samples from column four grouped by variety  63  Modified extraction procedure of water soluble cheese components for HPLC analysis  66  ix  LIST OF APPENDICES  Appendix 1.  Eigenvectors for principal component analysis of chromatographic data  Page  77  X  ACKNOWLEDGEMENTS  I would like to express my appreciation to Dr. Shuryo Nakai for his support and encouragement throughout the course of my study. I would also like to thank the members of my graduate committee, Dr. W. Powrie, Dr. B. Skura, Dr. P. Townsley, and Dr. R. Peterson for their helpful comments. Thanks also to Sherman Yee who keeps the lab running. I am forever grateful to my husband, Nicholas, for his unending support, inspiration, and invaluable help with computers.  1  I.  INTRODUCTION  The chemical basis of taste and aroma in cheeses is not well understood.  Cheese  maturation is a complex process in which the curd is broken down by proteolysis, lipolysis, and other enzyme catalyzed reactions to yield a flavor and texture typical of the varieties.  Proteolysis products, including peptides and amino  acids, are thought to  involved with taste and aroma directly, or as precursors in subsequent reactions. releases free fatty acids which may be important  many be  Lipolysis  in subsequent flavor forming reactions  (Stadhouers and Veringa, 1973). Proteolysis is important to both textural and flavor changes in ripening cheeses. most hard cheeses proteolysis leads to a softer and less elastic body.  In  Microbial proteolysis is  influenced by the remaining coagulant in the cheese curd, in addition to starter proteinase activity.  Milk is low in free amino acids, but the starter culture's proteinases and peptidases  allow them to utilize protein bound amino acids for growth (Law, 1982).  Chymosin was  responsible for the degradation of 0i -] -casein and partial attack of /3-casein after one month s  aging in C o u d a cheese (Visser and de  Groot-Mostert,  1977).  Thus rennet  proteolysis  produces large and medium peptides which are in turn degraded to small peptides and amino acids by starter proteinases.  These- proteolysis products are involved with taste and aroma  directly, or as precursors in subsequent reactions. The importance of the nonvolatile water extractable fraction has been demonstrated in both Cheddar and Swiss cheeses.  M c G u g a n et al. (1979) and Aston and Creamer (1986)  showed that the water-soluble fraction was important to Cheddar flavor intensity. interpreted  as a direct  effect  of  demonstrated that the small peptide  proteolysis products.  Biede  and  Hammond  This was (1979b)  fraction was responsible for the sweet and the brothy-  nutty flavor in Swiss cheese. Traditionally trained graders and sensory panels have been used to assess the degree of flavor develoment in cheeses.  D u e to the time involved in training a panel, and the  2 subjective nature of the method, a more objective analysis is desired. Pham and Nakai (1984) and Amantea (1984) successfully used reverse phase high performance liquid chromatography (RP/HPLC) and multivariate analysis to classify Cheddar cheese samples by age. Water soluble cheese extracts containing proteins, peptides, amino acids, and salts were used in the analysis. The use of H P L C and gas liquid chromatography ( C C ) to measure taste and aroma c o m p o u n d s in foods has generated a large number of measurements, necessitating the use of pattern recognition techniques.  The objectives of multivariate analysis include data reduction,  grouping, determining relationships among variables, and using these relationships to predict the value of other variables. The goals of the present research were: 1.  T o find an objective method to characterize the taste profiles of Cheddar, Edam,  Gouda, Swiss, and Parmesan cheeses. 2.  T o use a computer-aided  optimization  procedure, to determine  the H P L C  operating parameters. 3.  T o apply multivariate analyses to the HPLC peaks to characterize and separate  cheese varieties. 4. T o compare the analytical results with sensory data.  3  II. LITERATURE REVIEW  A.  D E V E L O P M E N T O F CHEESE FLAVOR  Worldwide there may be up to 2000 varieties of cheeses but these can be related to 20 basic varieties (Kosikowski, 1985). content.  Cheeses can be grouped according to their moisture  Soft cheeses may have up to 80% moisture, semi-hard varieties have approximately  50% moisture, and hard cheeses contain 40% moisture (Law, 1981). The hard cheeses can be further divided  into 4 groups; those with a relatively simple microflora such as Cheddar; those  inoculated with mold spores and allowed to germinate such as Roquefort; the Swiss varieties that use thermophilic lactic acid bacteria as starters; and the Italian cheeses that use added lipases to develop the characteristic rancid flavor (Law, 1981). The varieties of Cheddar, Edam, G o u d a , Swiss, and Parmesan can thus be classified as hard natural cheeses. The development of cheese flavor is very complex involving a heterogeneous milk source,  a changing population  pathways.  of  microorganisms, and  interactions  between  metabolic  Many reviews on the formation of flavor and aroma compounds in cheeses have  been published but the exact mechanisms are not yet understood (Law, 1982; Adda et al., 1982; Law, 1981; Forss, 1979; and Mulder, 1952). The composition of a cheese has a great influence on its flavor.  Mulder (1952)  indicated that the flavor of Edam and G o u d a can be related back to its constituents, fat, protein, lactose, salt, and water. and contributes to texture. cheese.  Water acts as a diluting medium for all kinds of substances  Salt blends with other flavors and enhances the piquant flavor of  Lactose is transformed to lactic acid which gives cheese a refreshing flavor and  releases aldehydes, ketones, alcohols, and esters.  Breakdown products of lactic acid include  acetic, butyric, and propionic acids. All of the breakdown products contribute to the general cheese taste.  Proteins do not have much taste but do contribute to the texture of cheese.  4 Proteolytic enzymes both added and indigenous are key to the transformation of milk into a ripened cheese.  Reviews of the role of proteolysis in cheese maturation  numerous (Visser, 1981; Crappin et al., 1985; Rank et al., 1985). active in cheese are the bacterial  contaminants;  indigenous milk  the  milk-clotting  proteinases; the enzymes for  are  Four proteinase categories  endogenous proteinases  cheese making; and the  from  enzymes  produced by the starter cultures (Crappin et al., 1985). The  clotting  of  milk  by proteolytic  enzymes  in the  represents one of the oldest practices in f o o d technology. used in the manufacture  process of  making  cheese  Traditionally chymosin (rennin) is  of cheese, but due to shortages of the enzyme other sources  including proteinases from microorganisms and other mammals are being used. The enzymatic coagulation of milk consists of 3 stages. milk protein, k-casein, is attacked by proteinases, yielding 2  During the primary stage, the  peptides, the glycomacropeptide  which is hydrophilic, soluble and will be lost in the cheese whey, and moiety,  which is hydrophobic and remains on the micelle.  the para-k-casein  The progressive hydrolysis of k-  casein during the primary phase leads to the micelle surface charge changing from a net negative to a positive.  U p o n 80% destruction of the k-casein, the micelles aggregate.  aggregation is the secondary stage of the process.  This  The tertiary stage is less clearly defined,  but involves the expulsion of water from the cheese curd due to structural rearrangement and general proteolysis df the caseins in the curd ( Dalgleish, 1982). During cheesemaking, most of the milk clotting enzyme is removed in the  whey.  Matheson (1981) found that chymosin remaining in the curd did not decrease over a 3 month period in Cheddar cheese. chymosin activity. 1981)  Swiss cheese, due to its high heat treatment, contained no  Rennet is thought to cause only limited breakdown of the caseins (Visser,  resulting in larger  molecular weight  peptides.  Enzymes from  the  starter  bacteria  subsequently produce smaller peptides and free amino acids. Lipolysis due to starter bacteria plays a limited role in the development of flavor in hard cheeses.  Starter bacteria produce free fatty acids from m o n o - and diglycerides but not  5 from triglycerides. These m o n o - and diglycerides are most likely formed by the natural lipases of milk and the lipases of Gram-negative rods (Stadhouders and Veringa, 1973).  Thus more  fatty acids result from partly hydrolyzed milk fat than from fresh milk fat. The C o m p o n e n t Balance Theory as originally proposed by Mulder (1952) explains the flavor of cheese as a mixture of c o m p o u n d s balanced to yield a typical flavor.  W h e n any one  c o m p o n e n t is lacking, or in excess, the flavor equilibrium is upset and an atypical cheese results.  Mulder (1952) further stated that the flavor forming substances are so well balanced  that it is difficult to recognize them individually.  B. C H E D D A R FLAVOR  1. Amino acids and peptides Cheddar cheese flavor has been extensively studied for over 30 years yet the chemical basis of flavor is still mostly unknown.  Free amino  acids have been studied as flavor  contributing c o m p o u n d s . Aston et al. (1983) found the level of free amino acids, measured as phosphotungstic acid-soluble amino nitrogen, strongly correlated with the degree of flavor development in control and accelerated ripened Cheddar cheeses. acids increased throughout ripening in all the cheeses.  The levels of free amino  These free amino acids are most likely  flavor precursors rather than flavor c o m p o u n d s themselves. Proteolysis  products were  shown  to be directly  responsible for Cheddar  intensity by M c G u g a n et al. (1979) and Aston and Creamer (1986).  Mild and aged Cheddar  cheeses were separated by M c G u g a n et al. (1979) into water-soluble, fractions.  flavor  residue,  and fat  The nonvolatile water soluble fraction containing peptides, amino acids, and salts  contributed most to Cheddar flavor intensity.  The residue had n o cheese flavor and the  volatiles contributed to the overall Cheddar flavor but did not influence flavor  intensity.  Similar results were found by Aston and Creamer (1986) w h o further fractionated the watersoluble component.  The sub-fraction containing salt, free methionine, free leucine, and  6 several peptides produced the most Cheddar flavor.  A mixture  of amino acids, salt, and  calcium lactate lacked a full flavor, implicating peptides as a flavor contributor. The role of peptides in Cheddar flavor is similar to that of free amino acids. peptides may be  a source of bitterness in Cheddar cheese.  Yet  Czulak (1959) proposed that all  bitter peptides were produced by residual rennet in cheese. Rennet has a lower optimum p H than the starter cultures enzymes, thus rennet releases peptides from the casein including bitter peptides.  As the protein breakdown continues, the pH rises and bitter peptides are  gradually  broken  down  nonbitter  peptides  by the  starter  proteinases.  More  likely,  high molecular  weight  caused by the rennet hydrolysis of casein are subsequently hydrolyzed by  starter to bitter low molecular weight Thomas (1980) showed that  peptides (Lowrie  and Lawrence, 1972).  Mills and  cell-wall associated starter proteinases produce bitter peptides.  Intracellular peptidase acitvity may then reduce these bitter peptides to non-bitter  peptides  (Law et al., 1976). 2.  Free fatty acids Lipolysis was shown to be the major source of volatile fatty acids in Cheddar cheese.  Dulley and Grieve (1974) found that skim milk cheese produced a fatty acid level less than 20% of the control. All of the fatty acids except acetic increased at a much slower rate during aging in the skim milk cheeses versus the controls. The role of free fatty acids in Cheddar flavor is not well accepted.  Forss (1979)  claimed that butyric acid was selectively hydrolyzed and was a major contributor  to flavor with  caproic acid playing a minor role. relationship between  free  Law and Sharpe (1977) and Law et al.(1976) found no  fatty acid levels and flavor.  excessive levels of free fatty acids result in rancidity.  Yet  most researchers agree that  Levels 3 to 10 times higher than normal  resulted in rancid defects in Cheddar flavor (Law et al., 1976).  Thus the role of FFA in  Cheddar flavor seems to be as a contributor to the overall background flavor.  7 3. Carbonyl c o m p o u n d s Cheddar propionaldehyde,  cheese  contains  butyraldehyde,  many  carbonyl  acetone,  compounds  2-butanone,  including  2-pentanone,  nonanone, diacetyl, acetoin, and pyruvic acid (Aston and Dulley, 1982). carbonyl c o m p o u n d s in Cheddar flavor is unknown.  acetaldehyde,  2-heptanone,  2-  The role of these  Manning (1978) found ethanol and  butanone levels were independent of the age of the cheese while methanol, acetone, and 2pentanone levels increased with aging.  Day et al.(1960) found no direct correlation between  typical Cheddar aroma and a mixture of ten carbonyl compounds.  Thus volatile carbonyls  contribute to the total or overall Cheddar flavor. 4.  Sulfur c o m p o u n d s (a)  Hydrogen sulfide ( H 2 S )  The first volatile sulfur c o m p o u n d found to relate to Cheddar flavor was hydrogen sulfide. Originally increased up  Kristoffersen and Nelson (1955) found that -SH groups and free H 2 S  to 6 months.  The cheese with the highest flavor intensity score also contained  the highest level of free H S . 2  Later Kristoffersen and Gould (1960) found that H 2 S levels  fluctuated during a 12 month aging period with no consistent trend appearing.  Aston and  Douglas (1983) examined volatile sulfur products in control and accelerated ripened cheeses. The levels of H S increased up to 6 months of age, then decreased. 2  also found no correlation between H S 2  Manning et al. (1976)  and Cheddar flavor intensity, nor was the level of H S 2  responsible for the sulfide flavor defect. (b)  Methanethiol (CH^SH)  Methanethiol in Cheddar cheese is most likely produced from the decompostiion of Lmethionine which is released by starter enzymes (Law and Sharpe, 1977). The importance of methanethiol to Cheddar flavor is not clear.  Manning et al.(1976) found that methanethiol  correlated with flavor intensity but not flavor quality in cheeses aged up to 12 months. Aston and Douglas (1983) showed that methanethiol levels increased up to 6 months of ripening  8 then declined. Correlations with total flavor, mature flavor and estimated age were low.  Thus  methanethiol is not a g o o d indicator of flavor development, (c)  Dimethyl sulfide ( C H ) S 3  2  Dimethyl sulfide was found to be a part of Cheddar aroma by Patton et al. (1958). G o o d quality Cheddar aroma was thought to be directly related to the level of dimethyl sulfide.  Later Aston and Douglas (1983) showed that levels of dimethyl sulfide remained  constant during the aging period or decreased after 6 months of ripening.  Thus dimethyl  sulfide levels did not correlate with flavor development in Cheddar cheese.  C.  E D A M A N D C O U D A CHEESE FLAVOR  Edam and C o u d a , originating from the Netherlands, are quite similar in flavor.  Gouda  may be ripened to produce a range of flavor intensities from mild to aged, whereas Edam is usually ripened to the mild stage. basis) respectively, and is therefore 1977).  Edam contains less fat than C o u d a , 24 and 28.5% more firm (Campbell and Marshall, 1975  (wet  ; Kosikowski,  G o u d a and Edam flavors are far from understood and much research remains to be  done on characterizing the cheeses. 1.  Peptides and amino acids Protein breakdown  products including free  (1952) to contribute to cheese flavor. and 30% with a broth-like flavor.  amino acids were thought  by Mulder  Casein contains 20% amino acids with a sweet taste  Bitter tasting amino acids are also present in cheese.  Proteolysis products resulting from the action of residual rennet, the enzymes of starter bacteria, and native milk protease were characterized by cheese.  Rennet was found to produce most of the  corresponds to  high and low molecular weight  Visser (1977a,b) in G o u d a  soluble-N in C o u d a cheese which  peptides.  Very little amino-acid N  produced by rennet regardless of rennet levels (Visser, 1977b).  was  Cheeses made with only  rennet or natural milk proteases did not develop a cheese flavor, but rennet was able to  9 produce bitter cheeses (Visser, 1977a).  Starter bacteria and milk protease produced only small  amounts of soluble-N. Starter bacteria contributed significantly to the lower molecular weight (<1400) peptides and free amino acids whereas milk protease liberated these c o m p o u n d s in small amounts.  The nonbitter starter  bitter strains (Visser, 1977b).  strains produced higher levels of amino-acid N than the  In cheeses produced with starters, the nonbitter starters yielded  characteristic G o u d a flavor, whereas the flavor.  Nonbitter  (Visser, 1977a).  bitter starters  starters were able to breakdown  developed noticeabily less cheese  bitter peptides to nonbitter  products  Thus rennet influences the extent of soluble-N c o m p o u n d s and the starter  influences the production of amino-acid N. Amino acids themselves do not exhibit a strong smell or taste yet all hard cheeses contain the amino acids liberated from casein. Free amino acids were found to increase with ripening of Edam cheeses but not  linearly due to the decomposition of free amino acids.  Ali  and Mulder (1961) found that a mixture of amino acids representing those found in a ripened cheese, when added to a young neutral flavored cheese, did not yield a full flavored Edam. Isoleucine produced a bitter taste, alanine and proline brothy taste.  a sweet taste, and glutamic acid a  Thus amino acids are thought to contribute to the basic cheese taste (Ali and  Mulder, 1961). 2. Volatile c o m p o u n d s Fatty acids in G o u d a cheese enhance the typical flavor and give the cheese body (Mulder, 1952). Volatile fatty acids contribute to G o u d a flavor and aroma but the specifics are unknown (Badings and Neeter, 1980).  G o u d a cheeses aged 9 to 12 months were found by  Sloot  bis(methylthio)methane,  and  Harkes  (1975) to  alkylpyrazines in the aroma.  contain  anethole,  and  a series of  These volatile components were present in low amounts but  were thought to contribute to the overall G o u d a flavor.  10 D.  SWISS CHEESE FLAVOR  There are many types of Swiss cheese but the original was produced in the Emmental river valley in Switzerland (Kosikowski, 1977).  Swiss cheese is characterized by its eyes, or gas  holes, and a sweet nutty flavor. Reviews of Swiss cheese manufacture and flavor have been published by M o c q u o t (1979) and Langsrud and Reinbold (1973a,b,c,1974). cooked at a relatively high temperature  The curds are  (50C) thus thermophilic starters are used, as well as  propionibacteria. Mixed  cultures  are added  to  cheese  milk  in the  Streptococcus thermophilus and Lactobacillus helveticus or acid from lactose.  Propionibacteria contribute  manufacture  of  Swiss cheese.  Lactobacillus lactis produce lactic  to later cheesemaking stages (Law,  1981).  After pressing and brining, the cheese is stored for 7 to 14 days in the cold room transfered to a hot room (21-25C) to allow eye and flavor formation. stored at 2 to 5 C for further Kosikowski, 1977).  then  The cheese is then  curing for 2 to 9 months (Langsrud and Reinbold, 1973c;  Swiss cheese has a lower fat content. 30.5 versus 33% wet basis, and a  higher protein content, 26.1  versus 25.8%,  compared to Cheddar cheese (Campbell  and  Marshall, 1975). 1.  Acids in Swiss cheese Volatile fatty acids contribute most to Swiss cheese flavor but other fatty acids and  keto acids contribute to the background flavor. propionibacteria caused eye formation  Sherman (1921) was the first to show that  and the characteristic sweet flavor in Swiss cheese.  These bacteria metabolize lactose and lactate and yield propionic acid, acetic acid, and carbon dioxide. volatile  Cheeses with the typical sweet flavor were found to contain a higher content of acids than bland cheeses (Babel and Hammer,  1939).  Propionic and acetic acids  themselves are not sweet but calcium and sodium propionates added to cheese lacking flavor produced the sweet note.  11 G o o d quality Swiss flavor was found to contain large amounts of acetic and propionic acid but low levels of butyric acid (Langlois and Parmelee, 1963; Krett and Stine, 1951; Harper, 1959).  Flat flavored cheeses contained normal levels of acetic acid and little or no propionic  and butyric acids (Krett and Stine, 1951).  A burned flavor was apparent when butyric acid  levels were greater than propionic acid levels. Kurtz et al. (1959) disagreed with Babel and Hammer (1939) in the role of propionic acid in sweetness, by thinking not the sweet characteristic.  propionic acid contributed a "rich" flavor to Swiss cheeses but Other unidentified c o m p o u n d s were thought to be responsible  for the sweet and nutty flavors. Butyric, propionic, and higher fatty acids were found in Swiss cheese but levels were uncorrelated with the age of the cheeses (Hintz et al., 1956).  Cysteic acid, tauric, and valeric  acid were found for the first time in some Swiss cheeses.  Hintz et al. (1956) discovered that  a g o o d Swiss flavor required a minimum concentration  of 5.0 mg propionic acid per gm  cheese and 2.0 mg proline per gm cheese. Thus an interrelationship exists between  different  chemical classes of c o m p o u n d s , i.e. fatty acids and amino acids, similar results may exist in other cheese varieties (Harper,  1959).  Mitchell (1981) found that levels of acetic acid,  propionic acid and proline rose during a 4 month aging period.  Yet a mixture of these  compounds in levels typical of cheese, produced a Swiss-like flavor but was lacking some compounds of a high quality Swiss cheese. Paulsen et al. (1980) studied the role of starter bacteria in the production of free fatty acids.  Cheeses produced with _S_. thermophilus contained more free fatty acids of 6 to 10  chain lengths while cheeses made with fatty acids. lengths.  Shermanii contained more of the 12 and 14  carbon  Cheeses inoculated with _L helveticus released fewer fatty acids of all chain  Flavor in cheeses with a single inoculum were low as was the combination of  thermophilus and  shermanii.  Cheeses inoculated with  were cheesy in flavor but were not characteristic of Swiss.  thermophilus and _L helveticus  12 Keto acids were not related to the age of Swiss cheese but are present (Langsrud and Reinbold, 1973c).  Bassett and Harper (1958) found pyruvic and alpha-ketoglutaric acids to be  present in larger amounts while oxalsuccinic, oxalacetic and alpha-acetolactic were present in smaller amounts. 2.  Misc. volatile c o m p o u n d s Volatile c o m p o u n d s including alcohols, esters, lactones, and hydrocarbons have been  identified  in Swiss cheese (Langler et al., 1967; Langsrud and Reinbold, 1973c; Biede and  H a m m o n d , 1979a).  Alcohols present include ethanol and 1-propanol  levels that they most likely d o not contribute directly to Swiss flavor. esters with fatty acids (Langsrud and Reinbold, 1973c).  They may form flavorful  Aldehydes include acetaldehyde which  exceeds the flavor threshold of 0.4 mg/kg (Harvey, 1960) and thus flavor (Langsrud and Reinbold, 1973c).  but are in such low  may contribute to Swiss  Esters include methyl hexanoate and ethyl butanoate  in Swiss cheese and may contribute to the overall flavor.  Diacetyl plays a key role in the flavor  of cultured dairy products and is present in Swiss cheese.  Diacetyl was found to correlate  with sweetness in the water-soluble volatile fraction of Swiss cheese by Biede and Hammond (1979b).  Mitchell (1981) showed that diacetyl levels were highest in fresh curd and declined  after 3 weeks to a consistent level. 3.  Peptides and amino acids As in other cheese varieties, proteolysis plays a key role in the development of flavor  and texture in Swiss cheese.  Hintz et al. (1956) examined the levels of free amino acids in  Swiss cheeses aged from 1 to 11 months.  Levels of the free amino acids varied among the  cheeses but cysteic acid, threonine-serine, glutamic acid, and tyrosine-phenylalanine occured in all samples.  There was a trend of histidine to increase with age. Kosikowski and Dahlberg  (1954) also found that stronger flavored Swiss cheeses tended to have higher levels of free amino acids.  Hollywood and Doelle (1984) showed that nitrogen levels increased in a high  moisture Swiss cheese aged up to 63 days.  13 Biede and H a m m o n d (1979a,b) fractionated Swiss cheese into oil soluble, soluble  components.  The  nonvolatile  water-soluble  additionally contained burned, bitter, and nutty flavors.  fraction  was  the  and water-  sweetest  and  The sweet flavor was thought to be  due to the interaction of small peptides and amino acids with calcium and magnesium ions. The brothy-nutty flavor was also due to small peptides.  Medium peptides were resposible for  the burned and bitter flavor notes. 4.  Sulfur containing c o m p o u n d s Sulfur containing c o m p o u n d s may contribute to Swiss flavor (Langsrud and Reinbold,  1973c).  Dimethyl sulfide is present in concentrations of 0.056 to 0.183  1967) which is above flavor thresholds and thus adds to the taste.  ppm (Langler et al.,  Singh and Kristoffersen  (1971) found that Swiss curd slurries ripened at 30C produced a typical Swiss flavor at 4 to 5 days but beyond this a putrid unclean flavor was formed due to excess dimethyl sulfide.  E. PARMESAN FLAVOR  The two main types of Crana cheese, Parmigiano Reggiano and Grana Padano, were originally made in the Italian valley of Po (Reinbold, 1963).  The Grana class of cheeses are  c o o k e d , hard, and require long aging (Battistotti et al., 1984).  Grana refers to the granular  structure of the cheese (Kosikowski, 1977). The cheese is generally ripened for 2 years, or not less than  10 months in the United States, and contains not less than 32% fat (dry basis) and  not more than 32% moisture (Kosikowski, 1977). M u c h of the literature concerning Parmesan flavor is in Italian and the references in English are scarce.  As in the other cheese varieties, proteins in Parmesan cheese are broken  down releasing free amino acids.  Several amino acids including asparagine, proline, glutamic  acid, valine, methionine, isoleucine, tyrosine, phenylalanine, lysine, and histidine were found to increase with ripening in Grana Padano (Bianchi st al., 1974).  These amino acids were  found to be concentrated in granules and spots in the cheese. Tyrosine was present in high  14 levels in the granules followed by phenylalanine and glutamic acid.  The spots contained  mostly leucine and iso-leucine. Sulfur compounds  important in Cheddar flavor, including hydrogen sulfide and  methanethiol, were not found in significant amounts in Parmesan cheese (Manning and Moore, 1979).  Ethanol was the largest peak found by headspace  analysis using a gas  chromatograph.  F. REVERSED PHASE HIGH PERFORMANCE LIQUID CHROMATOGRAPHY  1. Theory Reverse phase high performance liquid chromatography (RP\HPLC) involves a nonpolar stationary phase with a polar mobile phase. The range of solvents and column types available allow a variety of compounds to be separated based on polarity.  Bonded  phase columns,  where the stationary phase is chemically attached to a support result in weak surface energies. The nonpolar group is often an octyl or an octadecyl group. Thus attractive forces between the stationary phase, solute molecules, and the mobile phase are weak and analysis time is short (Krstulovic and Brown, 1982). The exact mechanism of separation on RP columns is unknown, partly due to an incomplete understanding of the surface structure of the bonded phase (Krstulovic and Brown, 1982).  The solvophobic, or hydrophobic, mechanism  seems to be the main  mechanism of solute retention (Scobie and Brown, 1983; Hearn, 1983; Hancock and Sparrow, 1983; Krstulovic and Brown, 1982). The solvophobic model explains the interaction between solute molecules and the nonpolar packing as depending on weak dispersion forces. When a solute with nonpolar regions is placed in a polar mobile phase, the subsequent hydrophobic interactions force the binding of the solute to the stationary phase.  The driving force is not  the nonpolar attraction between the solute and stationary phase, but the repulsion between the solute and the mobile phase (Krstulovic and Brown, 1982). An organic solvent modifier, such as methanol or acetonitrile, will reduce the surface tension of the water molecules and  15 thus reduce the repulsive forces between the solvent and nonpolar region of the solute.  With  sufficient reduction in surface tension, the solute dissolves, or is eluted, in the mobile phase (Hancock and Sparrow, 1983).  The retention of a sample on a reverse phase column is thus  related to its surface polarity. Proteins were shown to adsorb on a RP column in a monolayer by Di Bussolo and Cant (1985).  Proteins of differing hydrophobicity eluted in order of hydrophobicity in an  organic-lean mobile phase.  W h e n 20% acetonitrile was added as an organic modifier to the  mobile phase, proteins competed for the hydrophobic sites on the stationary phase.  Proteins  were able to displace those injected earlier of a lesser hydrophobicity, but not those of a greater hydrophobicity. 2. Applications of HPLC Over the past decade H P L C has b e c o m e an invaluable tool to biochemical analysis due in part to its speed, high resolution, excellent recovery rate, and flexibility in expermental conditions (Hearn, 1983).  There are numerous articles using HPLC in the analysis of proteins  and peptides, and excellent review articles by Hearn (1983) and Hancock and Sparrow (1983). These authors found that generally a mobile phase of low pH produced better separations and peak shapes.  Ashoor and Knox (1985) applied water extracts of fish species to a RP column  and were able to distinguish between species by the retention times of protein peaks relative to bovine serum albumin. Whey proteins have been separated by a number of researchers using HPLC. phase columns, RP-8 and  C 6 , were used by Diosady et al. (1980) and by Pearce (1983)  respectively, while an HPLC-gel exclusion separate whey proteins.  Reverse  column was used by Bican and Blanc (1982) to  Olieman and van den Bedem (1983) applied an H P L C gel permeation  column to the detection of the glycomacropeptide in skim milk samples. In the past four years H P L C has been applied to the detection of cheese ripening products, including peptides and amino acids (Champion and Stanley, 1982; Pham and Nakai, 1984; Polo et al., 1985; Marsili, 1985; Kaminogawa et al., 1986; Aston and Creamer, 1986).  16 Champion and Stanley (1982) separated 71 peptides from Cheddar cheese with a C18 RP column, some of these compounds were found to be bitter. RP  C8  column  to  separate  chromatographic patterns.  Cheddar  cheeses  by  age  Pham and Nakai (1984) used a based  on  peak  The peaks were peptides and amino acids.  areas  of  the  Aston and Creamer  (1986) also studied the water soluble fraction of Cheddar cheeses with a RP-18 column. Protein breakdown products during the aging of C o u d a were followed by Kaminogawa et al. (1986). Mahon cheese.  Polo et al. (1985) examined the increase in free amino acids released during aging of cheese while Marsili (1985) followed  amino acid production in ripening Cheddar  Chang et al. (1985) applied HPLC analysis to determining levels of 3 biogenic amines  in Swiss and Cheddar cheeses. Thus HPLC analysis appears to be a key to the understanding of the complexity of cheese ripening. 3.  Optimization of HPLC conditions (a)  Columns  C o l u m n packings commonly used in RP/HPLC are C18, C 8 , C N , or phenyl (Hancock and Sparrow,  1983).  A C8 packing has lower surface loadings and hydrophobicity,  c o m p o u n d s are not retained compared  three types  of  as strongly as on a C18 column.  columns and found  a C8  Pham and Nakai (1984)  Adsorbosphere produced the  resolution and largest number of peaks with a cheese extract.  thus  best  Wilson et al. (1981) found  comparable peptide elution patterns on a LiChrosorb C8 and C18 column. (b)  Temperature  Temperature is less important in HPLC compared to gas chromatography due mostly to the low boiling points of  many of the HPLC solvents (Krstulovic and Brown, 1982).  phase composition may be altered rather than temperature compounds.  Mobile  to control the capacity factor of  Increased column temperatures will decrease the viscosity of solvents, reduce  the capacity factor and increase column efficiency (Hearn, 1983; Krstulovic and Brown, 1982). Higher temperatures  may also lead to solute stationary phase degradation (Hearn,  Wilson et al. (1981) found no change in elution  patterns of peptides over the  1983).  temperature  17 range of 25-55C.  Kolbe et al. (1985) also found no overall pattern change in the elution of 25  amino acids over the range of 25-40C, but higher temperatures produced shorter retention times compared to those at lower temperatures, (c)  Mobile phase  Snyder (1974,1978) classified numerous solvents into similar selectivity groups based on functionality.  Hydrogen bonding and dipole interactions determined whether the solvents  belonged to similar selectivity groups.  Eight groups are arranged in triangular fashion by the  degree of proton-donor, proton-acceptor, and dipole-dipole interactions they can undergo. By using solvents from the apices of the triangle, larger differences in chemical selectivity are possible (Snyder, 1978; Clajch et al., 1980). Before the mobile phase can be optimized, a method of determining the quality of a separation must be established.  The resolution of two or more peaks is influenced by the  capacity factor, k', the selectivity factor, CY, and the efficiency factor, N.  These three terms  relate to the resolution, R , of two peaks in the following manner (Snyder and Kirkland, 1974): g  R  s  = 1/4^-1) • (<fN) • (k/(1+k'))  selectivity factor  efficiency factor  capacity factor  The three terms may be varied independently in the optimization of peak separation.  The  capacity factor is the ratio of the retention volume of a c o m p o u n d to the elution volume of a nonretained c o m p o u n d . The selectivity, or separation factor, between two peaks is a ratio of capacity factors:  k  2  /k^ where k  2  is the value of the c o m p o u n d with the longer retention  time.  C o l u m n efficiency depends on the number of theoretical plates (Krstulovic and Brown,  1982)  which in turn depends on column length, particle size, and flow rate (Clajch et al.,  1980).  The k'value is easily altered by changing the mobile phase solvent strength, while O! is  changed by altering solvent composition (Clajch et al., 1980). One  of  the  more  simple chromatographic  response functions was  proposed by  Morgan and Deming (1975) as: CRF = 2ln(Pj) where Pj represents the peak separation of the  18 i*  n  pair of peaks. The separation between two peaks is P = f/g where f represents the length  of the depression below a straight line joining two consecutive peak maxima and g measures the height of the straight line above the baseline at the depression (Figure 1).  The published  list of criteria used to judge the separation of peaks is numerous; Debets (1985) in a review article listed ten different response functions.  Berridge (1982) weighted the function to have  maximum number of peaks as the most important  factor, while D'Agostino et al. (1985)  included analysis time in the function. Simplex optimization and its various modifications has been used to determine  the  best set of operating conditions for many chemical analyses (Routh et al., 1977; Amantea, 1984; Berridge, 1982; Morgan and Deming, 1974). simplex defined by n-l-1 1974).  Simplex optimization starts with an initial  vertices in the n-dimensional factor space (Morgan and Deming,  The idea is to move away from the worst response towards an optimum set of  experimental conditions (Nakai et al., 1984). Amantea (1984) applied Super Modified Simplex (Routh et al., 1977) with Mapping Super-Simplex (Nakai et al., 1984) to determine optimum operating parameters for H P L C analysis of cheese extracts.  G. Multivariate Analysis  Multivariate  statistical techniques are becoming invaluable to the f o o d researcher  dealing with subjective-objective properties of foods. This is evident in the increasing number of articles concerned with sensory and chemical data analyses (Stungis,1976; Martens and Russwurm,1983;  Bertucciolo et  al.,1986).  Multivariate  techniques reveal  the  underlying  structure and relationships of large tables of data, allowing an interpretation of the variables (Martens, 1983).  These methods are easily carried out with the use of computer packages  such as SAS (1985) and B M D P (1985).  Figure 1.  Method to determine peak resolution, P=f/g, adapted from and Deming (1975).  20 1.  Principal component analysis When  between  a large number of variables are measured, the relationships, or covariance,  variables  essential.  is soon  beyond  Principal component  comprehension, thus  analysis seeks to  reduce  a data the  reduction  number  technique  of variables while  maintaining a majority of the original information (Dillon and Goldstein, 1984). principal component extracted is a linear combination  is  The first  of the original variables and accounts  for as much of the total variance as possible. The second principal component (PC) is uncorrelated with the first  PC and accounts for as much of the  remaining variance  not  explained by the first PC. This may continue until there are as many PC's as original variables. Usually only a few principal components are necessary to describe the data or account for the sample variation  (Daultrey,  1976;  Dillon and  components to use for data interpretation,  Goldstein, 1984).  In  deciding  how  many  a c o m m o n rule of thumb is to retain principal  components with eigenvalues greater than one (Dillon and Goldstein, 1984; Daultrey, 1976). Thus a large number of correlated variables is transformed to a few uncorrelated principal components.  These components may disclose relationships that were unexpected.  Principal  component analysis may only be the first step in data analysis, as the PC's may be used in regression, cluster, and discriminant analyses (Johnson and Wichern, 1982). 2.  Discriminant analysis Discriminant  analysis considers a data set  consisting of  a categorical  dependent  variable and a set of independent variables. The categorical variable assigns each case to predefined groups.  The objective is to derive a function, or functions, that will discriminate  among the groups.  These functions can then be used to classify new cases into the groups  (Dillon and Goldstein, 1984).  The discriminant function minimizes the misclassification rate  with linear combinations of the  independent  variables.  The between-group  variance is  maximized, while the within-group variance is minimized (Dillon and Goldstein, 1984).  21 3. Applications Multivariate analysis has been applied to characterizing a variety of foods including olive oil (Forina et al.,1983), frozen peas (Martens,1986), and rancid butter (Woo and Lindsay, 1984). Aishirna (1983) used multiple regression, discriminant, principal component, and cluster analyses to determine the relationship between gas chromatography (GC) profiles and sensory data of soy sauce. Based on 39 peak areas, 8 brands of soy sauce were classified into superior, moderate,and low quality groups.  Aishirna (1985) later applied similar  techniques to the characterization of Worcestershire sauces. Multivariate analyses have been used by several groups of researchers to evaluate cheese ripening and flavor development (Pham and Nakai,1984; Rothe et al.,1982; Santa-Maria et al.,1986; Aishirna and Nakai, 1987). Rothe et al.(1982) characterized the flavor of Blue cheese with 15 sensory attributes and 58 instrumental values.  Based on correlations with  sensory data, the instrumental measures were reduced to 20. Discriminant functions based on analytical data were able to describe the intensity of eight sensory attributes, including rancid, fruity, and stinky. Pham and Nakai (1984) and Amantea (1984) applied principal component and discriminant analyses to Cheddar cheese HPLC profiles. The cheese extracts for HPLC analysis consisted of peptides and amino acids.  Amantea (1984) was able to reduce the  number of HPLC peaks, necessary to classify the cheeses by age, from 48 to 8. Aishirna and Nakai (1987) applied discriminant analysis of GC profiles to classify cheeses by variety. SantaMaria et al. (1986) also used proteolysis products including total nitrogen, tyrosine, and tryptophan to characterize Manchego cheese by age. Eighteen variables were used to classify 30 cheeses as fresh, medium, or aged. Bertuccioli et al. (1986) applied PCA and partial least squares (PLS) analysis to sensory and GC data collected on Provolone cheese. Seven peaks characterized the cheese ripening while PLS analysis indicated the correlation between total aroma and the peaks.  22  III. MATERIALS AND METHODS  A. CHEESE SAMPLES  Commercial brands of Cheddar, Couda, Edam, Swiss, and Parmesan cheeses were purchased from local supermarkets.  In addition, samples of Cheddar cheese were received  from Dairyland Foods (Burnaby, B. C.) and Canada Packers (Toronto, Ont).  Swiss and  Cheddar cheeses of varying ages were received from Kraft (Beaconfield, Que).  All cheeses  were stored at 5C.  B. SAMPLE PREPARATION  1.  Water-soluble extract Extraction of the water soluble fraction was a simplification of the method of  McCugan et al. (1979).  The initial separation was carried out by grating 4g of cheese and  centrifuging at 27,000 X g for 30 min at 25C (Figure 2). discarded.  The fat was pipetted off and  The residue was transfered to a stoppered test tube  methanol, 1.0mL methylene chloride, and 0.6mL water.  and  extracted with 1.0mL  The mixture was shaken well for 15  sec and centrifuged again at 27,000 X g for 30 min at 25C.  The methanol-water layer was  removed and volatiles were evaporated in a Silli Therm heating module (Pearce Chemical Company, Box 117, Rockford, IL) set at a temperature of 45C under a flow of nitrogen. After 20 min of evaporation, 1mL of water was added to the extract. continued for a total of 60 min.  The evaporation was  The sample was diluted to a final volume of 2.0 mL and  filtered through a 0.45^un Durapore HVLP Millipore filter (Bedford, MA). All cheese samples were extracted in duplicate, then combined into one final extract. until analyzed.  The samples were frozen  Frozen extracts were thawed at 5C and filtered through a 0.45yum Durapore  HVLP Millipore filter prior to injection into the HPLC.  4 g CRATED CHEESE CENTRIFUGE (27,000 X g, 26C, 30min)  RESIDUE  FAT  I  STOPPERED TUBE Add: 1mL MeOH ImL C H C L 0.6mL H 0 2  2  2  SHAKE (15 sec)  CENTRIFUGE (7,000 X g, 25C, 30min)  C H C I LAYER 2  2  1  M e O H / H 0 LAYER 2  CONCENTRATE (NITROGEN STREAM, 45C, 1hr) (ADD 1mL H 0 AFTER 20 MIN) 2  FILTER THROUGH 0.45;um PORE SIZE  HPLC ANALYSIS  Figure 2.  Extraction procedure of water-soluble cheese components for HPLC analysis.  24 2. pH The final extract was measured for pH.  An Accumet Model 420 digital pH meter  (Fisher Scientific Co., Ottawa, Ont.) was used throughout the study.  Approximately 1 mL of  the final cheese extract was used to determine pH.  C. HPLC ANALYSIS  1.  HPLC conditions Chromatography was performed on a Spectra-Physics  8700 liquid chromatograph  combined with a SP 8400 variable wavelength detector, and a SP 4100 computing integrator. (Spectra-Physics, Santa Clara, CA). The detector was run at a wavelength of 220 nm.  A  modification to the SP system was the use of a de-bubbler from Terochem Laboratories (P. O. Box 8188, Stn. F., Edmonton, Alta) placed between the ternary proportioning valve and the pump. A reverse phase column (250 X 4.5 mm I. D.) packed with Adsorbosphere C8 (5^um) purchased from Alltech ( Applied Science Labs, Deerfield, IL ) was used for chromatographic analysis.  The sample loop had a size of 50  JLKL.  A guard column was placed before the  Adsorbosphere C8 column. The guard column was packed with similar material as the main column.  A RP-8 Spheri-10 RP-CU guard cartridge (Brownlee Labs, Santa Clara, CA.) was  changed monthly during the analyses.  Four HPLC columns were required to complete this  study. The reproducibility of the HPLC analysis was examined with a Swiss cheese extract. The extract was injected into the column  once each day for 8 days.  The standard deviation  and coefficient of variation of the retention times of the major peaks were determined. A ternary gradient system was used to elute the water soluble compounds from column.  An optimum initial solvent volume  ratio of 96.8  trifluoroacetic acid (0.1%), acetonitrile, and methanol.  : 1.2  : 2.0 was used  the for  Over 50.3 min this ratio changed to  25 56.3 : 30.3 : 13.4. conditions. used.  A further 30 min was necessary to gradually return to the starting elution  Thus a single analysis took 80.3 min. A solvent flow rate of 1.0 mL per min was  All HPLC runs were carried out at ambient temperature. Methanol and acetonitrile were of HPLC grade (BDH Chemical C o . , Toronto, Ont).  Distilled water filtered through a Norganic cartridge (Millipore, Bedford, MA) was used to prepare the trifluoroacetic acid.  All solutions were degassed for 15 min prior to use.  During  gradient elution, the solutions were maintained in a degassed state with a slow stream of helium. A SP 4100 computing integrator, under BASIC control was used to calculate individual and total peak areas.  A chart speed of 0.5 cm/min was used with an attenuation of 16 to  record the chromatograms. 2.  Internal standard To  follow  the  deterioration  of  the  HPLC  column  and  day-to-day  variations, or  variations within a day's run an internal standard was mixed with the cheese extracts. compounds  were  examined  as possible standards,  including phthalic  acid,  Several  tryptophan,  tyrosine, benzoic acid, p-nitroaniline, p-dimethylaminobenzaldehyde, rutin, vanillin, tannic acid, ninhydrin, and salicylaldehyde. was  chosen  as  an  Based on the retention time, p-dimethylaminobenzaldehyde  internal  standard.  A  stock  solution  of  100  ppm  p-  dimethylaminobenzaldehyde in acetonitrile was prepared. The solution was filtered through a 0.45/A.m Durapore HVLP Millipore filter., The standard stock solution was stored in the dark at 10C. A cheese extract was mixed with 20 ppm of the standard prior to analysis. The  purity  of  the  p-dimethylaminobenzaldehyde  chromatography (TLC) and by H P L C .  was  checked  by  thin  layer  Two solvent systems were used to elute the c o m p o u n d  for TLC: 1.  95% benzene-5% methanol  2.  chloroform  26 The p-dimethylaminobenzaldehyde sample was prepared by dissolving  approximately 0.01 g  in a few drops of methylene chloride. Three concentrations of the sample were spotted on each of two TLC plates pre-coated with Silica Gel 60 (BDH Chemicals, Toronto, Ont.). The spots were allowed to dry and the plates were eluted for 3 hrs in their respective chambers. The plates were checked under uv illumination and stained in an iodine chamber for 15 min. The purity of the p-dimethylaminobenzaldehyde standard was examined secondly by the ternary gradient elution on the HPLC.  A sample of 10yuL of the stock solution in  acetonitrile, was injected into the HPLC.  D. OPTIMIZATION  METHODS  A blend of the new Mapping Super-Simplex optimization (new MSO) and  Kaneko,1985) with the Centroid Mapping Optimization method  1986) was used to determine the best conditions for HPLC resolution.  method (Nakai  (Aishima and Nakai, The idea is to move  away from the worst response towards an optimum set of conditions. Peak resolution was measured by the method of Morgan and Deming (1975) during the optimization procedure . Peak separation Pj of the i  pair of peaks in a chromatogram is  defined as: Pj = f/g where f represents the depth of the depression below a straight line joining two adjacent peak maxima, and g represents the height of the straight line above the baseline at the depression (Figure 1).  The sum of Pj is the resolution or response used in the optimization procedure.  E. SENSORY EVALUATION  All cheese samples were evaluated by a semi-trained taste panel. The members were initially screened for their ability to distinguish the four basic tastes, sweet, sour, salty, and bitter. The panel consisted of 7 university students, who were present during the time of the  27 study and who liked cheese. A minimum of 5 panelists rated the cheeses. Taste panels were held twice a week. Training consisted of familiarizing the panelists with the 5 varieties of cheeses being evaluated.  Discussion occured among the panelists as to intensity scores for the four  parameters being evaluated. Cheese samples were prepared in a standard manner. The outer portions of a block of cheese were removed and cubes of approximately 2 cm^ were cut. The cheeses were held at room temperature for 30 min prior to sensory evaluation. At each session, 4 or 6 cheese samples were evaluated under red lights to eliminate color differences.  Cheeses were identified as Cheddar, Edam, Couda, Swiss, or Parmesan and  rated on a scale from 1 to 5 for taste intensity, preference, and bitterness. ranged from a 1 for very low to a 5 for very strong.  Taste intensity  Preference ranged from a 1 for dislike  very much to a 5 for like very much. Bitterness ranged from a 1 for none present to a 5 for strongly bitter.  F. STATISTICAL ANALYSIS  1. HPLC data Each peak on the chromatogram, measured as area, was used for analysis.  The  internal standard peak, p-dimethylaminobenzaldehyde, was used to normalize the remaining peaks for each sample.  Multivariate analyses including, principal component and discriminant  analyses, were used to interpret the HPLC data. Statistical packages used were SAS programs of STEPDISC, DISCRIM, AND CANDISC (SAS institute Inc, Cary, NC). computer was used to perform the analyses.  An Amdahl 470 V/8  In multivariate analysis each peak from a  chromatogram was considered a variable for a given cheese sample . Thus HPLC profiles with p peaks for each of n cheese samples, can be thought of as p variables for n cases or observations.  28 (a)  Principal component analysis  The main goal of principal component analysis is to reduce the total sample data variance.  The variance of the original data matrix was explained by " p " components, but  often much of this variability can be explained by a smaller number " m " of the principal components.  The principal components consist of  variables and are uncorrelated.  linear combinations of the  original  Principal components with eigenvalues greater than one were  used in discriminant analysis. (b)  Discriminant analysis  Discriminant  analysis  deals  with  separating  distinct  subsequently allocating new observations to the defined groups. the  sets  of  observations  and  In addition, to representing  data with discriminant functions, discriminant analysis can be used to reduce the  p-  dimensional data space to 2 or 3 dimensions. The number of H P L C peaks " p " represent the dimensionality of the  data.  Plots of the  means of the  reduced linear combinations, or  canonical variables, more easily show the relationships between the groups.  Discriminant  analysis was performed with both the principal components and the original peak areas. 2.  Sensory data The frequency of correctly identifying cheeses by variety was determined  FREQ.  by SAS  The parameters of correct choice of variety, taste intensity, preference, and bitterness,  were evaluated by multivariate analysis of variance using SAS C L M (SAS, 1985).  29 IV. RESULTS AND DISCUSSION  A. SAMPLE PREPARATION  A total of 106 cheese samples were purchased from local supermarkets or received from three Canadian cheese manufacturers. The cheeses were from ten different countries with unknown histories for the most part.  The breakdown by cheese variety was:  40  Cheddars, 21 Edams, 28 Coudas, 13 Swiss, and 4 Parmesans. The Cheddars and a few Coudas were labeled by age, but the remaining cheeses were ripening.  not identified as to their length of  Fifteen of the Cheddars analyzed on the first column were from an accelerated  ripening study using enzymes and elevated temperatures. The extraction of the water soluble components was a modification of the long tedious procedure of McGugan et al. (1979).  Pham and Nakai (1984) and Amantea (1984)  modified the procedure for a smaller sample size.  McGugan et al. (1979) designed the  extraction steps to completly separate the fat from the non-fat fractions, thus there were many repetitions of the solvent extraction step. Methylene chloride, methanol and water were used to separate the volatile from the non-volatile fractions. The goal of the current study was to develop a quick extraction that could be used for quality control purposes by cheese manufacturers. Thus a rapid single step extraction was used (Figure 2). The water-soluble fraction indicates changes that occur due to proteolysis during cheese ripening (Rank et al.,1985; Noomen,1977; Kuchroo and Fox,1982a,b). Noomen (1977) developed a water-soluble extraction method adjusted to the conditions of the cheese. Kuchroo and Fox (1982 a,b) showed the water-soluble fraction to be a mixture of peptides produced by enzymatic action of the coagulant, starter, and other bacteria.  Large peptides  from casein were not present in the water extract as demonstrated by gel electrophoresis (Kuchroo and Fox, 1982a). Pham and Nakai (1984) determined that water soluble extracts of Cheddar cheese were protein breakdown products based on a positive Ninhydrin reaction.  30  The inclusion of methanol and water in the extraction procedure selects for more hydrophobic compounds versus water alone (Rank et al.,1985).  Harwarkar and Elliott (1971)  used a cholorform-methanol-water extraction of Cheddar cheese. Rank et al.(1985) found the only difference between a water extraction and the Harwarkar and Elliott method was a slight increase in the peptide patterns at an absorbance of 280nm.  Aston and Creamer (1986) also  noted minor differences in HPLC patterns between the McGugan et al. (1979) method and their water extraction. The Durapore filter used to remove particles larger than 0.45yum had no effect on the extracts.  This was demonstrated by analyzing filtered and unfiltered HPLC grade methanol.  There was no extra peak with the filtered methanol, indicating the purity of the filter paper. The  internal standard, p-dimethylaminobenzaldehyde, was  added to the  water-  methanol extract prior to HPLC analysis rather than during the extraction procedure. This was necessary as otherwise the standard was discarded with the methylene chloride layer.  B. HPLC CONDITIONS  An Adsorbosphere C8 reversed phase column was used to elute the non-volatile cheese components, since Pham and Nakai (1984) found this column produced the largest number of peaks and best resolution with Cheddar extracts. C8  Amantea (1984) also used this  column to elute Cheddar cheese compounds with a ternary gradient.  An RP-8 guard  cartridge was used between the injection port and the analytical column. The guard column prolongs the life of the C8 column by removing contaminants, highly retained solutes and particulate matter that may be in solvents (Anon, 1986). The guard cartridge was replaced at 2 to 4 week intervals. The reproducibility of the HPLC profiles was determined over a 15 day period with a Swiss cheese extract.  Nine major peaks were selected from the chromatographic profiles.  Means, standard deviations, and coefficients of variation of the nine peaks for 8 injections  31 areshown in Table 1.  The results indicate less variability between runs early and late in the  elution pattern, and a general increase in variability during the middle of the elution. Representative chromatograms for the five cheese varieties, Cheddar, Edam, G o u d a , Swiss, and Parmesan, are shown in Figures 3-7 . The internal standard peak in all cheeses was the  latest  eluted  large  peak.  The  purity  of  the  p-dimethylaminobenzaldehyde  was  demonstrated by the elution of one peak with HPLC analysis and one spot by T L C .  The  standard eluted late in the run and thus did not interfer with any cheese peaks of interest. A ternary gradient was used to elute a variety of components from the HPLC column. Methanol, acetonitrile, and water are each from different Snyder (1978) selectivity groups, and thus would allow different c o m p o u n d s , based on their ability to undergo hydrogen bonding and dipole interactions, to be separated. agent.  Trifluoroacetic acid (TFA)  acts as an ion-pairing  At a low p H of 2.0, the basic side chains of amino acids are fullly charged and  associate  with  Sparrow,1983).  the  oppositely  charged  TFA anions  (Acharya  et  al.,1983;  Hancock  Trifluoroacetic acid is one of the most commonly used anions and allows a  greater recovery of larger relatively nonpolar peptides (Hearn, 1983).  The presence of TFA  increased the retention times of histidine-containing tryptic peptides by making them hydrophobic (Acharya et al.,1983). (Hearn,1983).  and  more  This ion pairing effect resulted in improved peak shapes  Amantea (1984) used 0.1% TFA to improve the resolution of peptides and  amino acids in Cheddar cheese.  C.  OPTIMIZATION  Optimization has b e c o m e necessary in analytical procedures such as H P L C and G C where many parameters are important to the analysis and may interact.  The trial and error or  one-factor-at-a-time  may vary from  approaches  require  operator to another in the results. computer optimization is a must.  numerous  experiments  and  one  W h e n a number of factors are necessary to an analysis,  Advantages of computerized optimization include improved  Table 1. Reproducibility of HPLC peak elutions from a Swiss cheese extract, n = 8.  Retention Time (min)  Mean  Standard Deviation  Coefficient of Variation  3.82  0.072  1.9  4.80  0.084  1.8  8.10  0.850  10.5  8.79  0.772  8.8  11.03  0.936  8.5  17.74  0.937  5.3  27.74  1.300  4.7  32.56  0.928  2.8  36.89  1.401  3.8  33  E c  o CM CM <  JL 1 0  20  30  40  50  T I M E - M I N  Figure 3.  Representative HPLC profile of water-soluble components of Cheddar cheese.  34  1  °  20 T I  Figure 4.  30  M E -  4 0  50  M I N  Representative H P L C profile of water-soluble components of Edam cheese.  35  E c o CM CM <  10  20  30  40  50  TI ME- MIN  Figure 5.  Representative HPLC profile of water-soluble components of G o u d a cheese.  36  E c  o  CM CM <  1 0  20  30  40  50  T l M E - MIN  Figure 6.  Representative H P L C profile of water-soluble components of Swiss cheese.  20nm  37  CN  <  i  1 Vj J i -  1  •  10  »-  20 TI  Figure 7.  M E -  M  »  »  30  40  MIN  Representative HPLC profile of water-soluble components of Parmesan cheese.  1  50  38 research efficiency, while mapping the results graphically shows the response surface (Nakai et al.,1984) The selection of the ternary gradient elution conditions were determined by the new Mapping  Simplex  Optimizataion  (MSO)  of  Nakai  and  Kaneko (1985) and The  Mapping Optimization ( C M O ) procedures of Aishima and Nakai (1986). Centroid procedure is the possibility of  Centroid  A drawback to the  reaching a local optimum rather than the overall  optimum if the initial factor ranges are set too narrow.  Thus the new M S O procedure was  used to narrow the ranges of the factors. The initial factor ranges were selected based on the optimum solvent conditions of Amantea (1984), where the organic modifier varied from 6% to 37% (Table 2).  Ten experiments were carried out followed by graphing of the resolution  responses against the factor ranges. The trend towards the optimum allowed the selection of new narrower factor ranges that were used with C M O to rapidly find the optimum  HPLC  parameters. A total of 24 experiments, including the 10 from M S O , were necessary to reach the optimum HPLC separation.  Mappings of the evolution towards the optimum for the five  factors is shown in Figures 8-12.  C M O indicates which points are to be connected, showing  the trend towards the optimum.  The figures indicate experimental conditions corresponding  to vertex 20 as the optimum with a response of 21.825.  For factor 5, time of H P L C run,  vertex 20 corresponded to 67.4 min which was higher than most of the other vertices, with a range of 45 to 55 min (Figure 12).  The additional time of elution may have produced more  peaks than a shorter run, and thus showed a higher response. Vertex 24 showed the second highest response of 21.147, and corresponded to a more reasonable elution time of 50.3 min. Vertex 20 ranged from 2.0% to 26.2% acetonitrile while vertex 24 ranged from 1.2% to 30.3%; the methanol concentrations were similar. Thus vertex 20 had a shallower gradient than vertex 24.  Vertex 24 also had 5% more organic modifier by the end of the run and therefore would  be expected to elute more hydrophobic c o m p o u n d s than vertex 20.  39  Table 2.  Starting factor ranges for the optimization operating parameters by Mapping Simplex ( M S O ) and Centroid Mapping Optimization ( C M O ) .  FACTOR  F  2  :  3F  4  :  "5  :  of HPLC Optimization  STARTING RANGES MSO CMO  Initial methanol cone  0-10%  1-5%  Final methanol cone.  20-50%  25-35%  Initial acetonitrile cone.  0-10%  I- 3%  Final acetonitrile cone.  10-20%  II- 14%  Time of H P L C run  40-70 min  45-55min  40  Figure 8.  Mapping responses of experiments to optimize peak resolution. Factor 1, initial methanol concentration.  41  Figure 9.  Mapping responses of experiments to optimize peak resolution. Factor 2, final methanol concentration.  42  Figure 10.  Mapping responses of experiments to optimize peak resolution. Factor 3, initial acetonitrile concentration.  43  Figure 11.  Mapping responses of experiments to optimize peak resolution. Factor 4, final acetonitrile concentration.  44  n  40  i  i  1  1  1  45  50  55  60  65  FACTOR 5,  Figure 12.  ;  TIME  Mapping responses of experiments to optimize peak resolution. Factor 5, time of H P L C run.  45 Optimization produced a 55% improvement from the worst response to the best.  The  number of peaks separated increased 17%.  Vertex 24 was chosen as the optimum, which  corresponded to a solvent volume ratio of:  96.8 : 1.2 : 2.0 for TFA (0.1%), acetonitrile, and  methanol.  Over 50.3 min this  ratio was changed to 56.3 : 30.3 : 13.4.  An additional 30 min  was used to bring the solvent ratio back to the initial ratio, thus one cheese extract was analyzed in 80.3 min.  D.  HPLC STATISTICAL ANALYSES  Peak areas from each cheese profile were used for statistical analysis. Each profile was examined for the presence of 55 peaks. These 55 peaks were selected after examining relative retention times, in relation to the internal standard, of many varietal profiles. peaks were often absent in the chromatograms.  Several of the  To moderate or normalize large differences  in peak areas among the 55 peaks, each peak area was divided by the area of the internal standard peak.  The data was checked for multivariate normality and a l o g  e  transformation of  the peak areas improved the fit. 1.  Principal component analysis Principal component analysis (PCA) seeks to reduce the total sample data variance  (Johnson and Wichern, 1982). matrix,  The data matrix in this study can be considered a 106 by 55  corresponding to 55 variables, or HPLC peaks, for each of 106  Therefore the variance of this data matrix was explained by the 55 peaks.  cheese samples.  55 components corresponding to  Principal component analysis of the HPLC data resulted in 17 components with  eigenvalues greater than 1.0, which is a c o m m o n statistical cutoff point (Daultrey, 1976; Dillon and Goldstein, 1984).  The principal components consist of a few linear combinations of the  original variables and are uncorrelated. sample variation (Table 3).  The 17 components explained 74%  of the total  Thus the dimensionality of the data was reduced from 55 to 17  with a 26% loss of explained variation.  The proportion of variance explained by any one  principal c o m p o n e n t was not greater than 13% and most were only 2 to 4%.  This is fairly low  46 Table 3.  PRINCIPAL COMPONENT  Eigenvalue, proportion of variance proportion of total variance, in analysis using chromatographic data.  EIGENVALUE  PROP. VAR. EXPLAINED  explained, and a principal  cumulative component  CUMULATIVE VARIANCE  Prin 1  7.0521  0.1306  0.1306  Prin 2  5.5089  0.1020  0.2326  Prin 3  3.7983  0.0703  0.3029  Prin 4  2.8967  0.0536  0.3566  Prin 5  2.3333  0.0432  0.3998  Prin 6  2.1928  0.0406  0.4404  Prin 7  2.0181  0.0374  0.4778  Prin 8  1.7948  0.0332  0.5110  Prin 9  1.7381  0.0322  0.5432  Prin 10  1.5777  0.0292  0.5724  Prin 11  1.5161  0.0281  0.6005  Prin 12  1.4827  0.0274  0.6280  Prin 13  1.3879  0.0257  0.6536  Prin 14  1.2045  0.0223  0.6760  Prin 15  1.1714  0.0217  0.6977  Prin 16  1.1178  0.0207  0.7184  Prin 17  1.0920  0.0202  0.7386  47 and indicates why the individual peak loadings on the components were relatively uniform. The peaks are unknown and the components did not load heavily on any one or several peaks, thus the loadings are  shown in Appendix 1.  A second principal component analysis included the additional data of p H of extracts, HPLC column number, and country of origin of the cheese samples.  the  In this case 18  components had eigenvalues greater than 1.0 and these explained 75% of the total sample variation.  Since HPLC column number and country of origin were indicated as categorical  measurements, these two variables were eliminated from the analyses. The 55 peak areas with pH yielded again 18 eigenvalues that explained 75% of the variation.  The proportion of  variance explained by the eigenvalues and loadings on the principal components were similar to HPLC peak data alone and therefore are not shown. 2.  Discriminant analysis Discriminant  analysis  deals  with  separating  distinct  sets  subsequently allocating new observations to the defined groups. data can reduce the  dimensionality to  2 or  of  observations  and  Linear combinations of the  3 dimensions, that  more  easily show  the  relationships between the cheeses, this is shown in a plot of the canonical variables. Factors that  must be considered for an optimum classification are prior probabilities  of group membership and the cost of misclassifying a case. population accordingly.  than  another,  the  prior  probabilities  for  If one class contains a larger  classification should be  weighted  Similarly if certain misclassifications represent a more serious error than others,  the cost associated with these misclassifications should be considered (Johnson and Wichern, 1982). Prior probabilities for classification as any one  variety were set proportional to the  sample size for each variety. This was necessary since the class sizes ranged from 40 Cheddars to 4 Parmesans.  Misclassification costs were not considered, as one type of error was not  worse than another.  48 Discriminant analysis is flexible allowing the use of the original variables for calculating the functions, or the use of principal components after such an analysis. variables may be used in determining the discriminant function.  All of the predictor  If there are a large number of  variables, a stepwise selection method can reduce the number to a few significant variables necessary for classification.  Significance levels are used to add to or eliminate  predictor  variables from the discriminant function. The results from discriminant analysis can be visually represented by a canonical plot. This dimension reduction technique shows with 2 or 3 axes the groupings of the cases. Canonical variables are linear combinations of the original variables that have the  highest  multiple correlations with the groups, or best indicate differences between the groups (SAS, 1985). Canonical variables are uncorrelated with each other. Discriminant  analysis was  firstly  performed  on the  17  principal components  and  yielded a total percent correct classification rate of 64%, with a high of 85% for Cheddar and a low of 25% for Parmesan (Table 4). A plot of the first and second canonical variables (Figure 13) indicates the grouping of cheese samples by variety. which correlates with its 85% classification rate.  Cheddar forms a fairly distinct group  The remaining varieties of Edam, G o u d a ,  Swiss, and Parmesan, are mixed and do not form unique groups. Discriminant functions can be derived for each cheese variety and these can be used to classify unknown samples. The general form of the function for Cheddar is: Cheddar:  a-^prin 1) + a2(prin 2) + .... a - ^ p r i n  17)-constant  where prin 1....prin 17 are eigenvalues of the principal components for Cheddar, and a ^ . . . . a ^ are coefficients of the function.  The values for the coefficients for each cheese variety are  shown in Table 5. Thus the actual discriminant function for Cheddar cheese would Cheddar:  0.56(prin  1)+0.016(prin  + 0.41 (prin  6)-0.26(prin  2)-0.93(prin 7)-0.27(prin  3)-0.45(prin 8)-0.095(prin  be:  4)-0.22(prin  9)-0.24(prin10)  + 0.018(prin11)-0.30(prin12)-0.031(prin13)-0.19(prin14)-0.06(prin15) + 0.084(prin16) + 0.18(prin17)-1.55.  5)  49  Table  4.  Percent correct classification rate of cheese variety by discriminant and sensory analysis.  DESCRIPTION O F ANALYSIS  17 PC  CHEESE VARIETY* 3  CHEDDAR  EDAM  3  COUDA  SWISS  PARM  TOTAL  85  52  64  31  25  64  72  57  54  77  100  72  18 PC + 3var  90  57  75  50  50  73  18 PC + p H  80  57  71  67  25  71  8 peaks  82  48  57  42  50  63  Column 1  100  100  100  100  NA  100  Column 2  100  100  100  NA  NA  100  Column 3  100  100  89  88  NA  93  Column 4  100  60  100  NA  100  93  SENSORY  86  42  43  74  76  63  17 PC,eq  a. PC = principal components; equal = equal prior probability; 3 var = p H , country of origin, column number. b.  N. A., not analyzed on the column.  •  • o  o o  oo o oo © o ^ © o o o o co o o o  o  v  o  •  •o • o  u  • ^ °  vV  • v•  •  o% ao  •  6^7 a  v  v  V O CHEDDAR V EDAM •  GOUDA  •  SWISS  O PARMESAN _i  -  !  4  -  ,  3  -  !  2  -  1  ,  0  ,  ,  !  ,  1  2  3  4  1 ST CANONICAL VARIABLE Figure 13.  Canonical plot of 106 cheese samples grouped by variety using proportional prior probabilities.  51 Tabie  5.  Coefficients of discriminant functions from principal components for separation of 106 cheeses by variety.  VARIABLE  CHEDDAR  EDAM  GOUDA  SWISS  PARMESAN  Constant  -1.551  -2.453  -1.881  -2.960  -5.675  Prin 1  0.559  -0.296  -0.097  -0.704  -1.069  Prin 2  0.016  0.374  -0.116  -0.517  0.371  Prin 3  -0.933  0.656  0.876  -0.038  -0.123  Prin 4  -0.446  0.276  0.553  -0.165  -0.323  Prin 5  -0.222  0.412  0.219  -0.450  -0.013  Prin 6  0.408  -0.529  -0.427  0.415  0.340  Prin 7  -0.258  0.454  -0.090  0.323  -0.229  Prin 8  -0.268  0.211  0.201  0.144  -0.305  Prin 9  -0.095  -0.086  0.101  0.330  -0.377  Prin 10  -0.236  0.534  -0.185  0.424  -0.534  Prin 11  0.018  0.162  -0.181  -0.193  0.867  Prin 12  -0.303  0.549  0.105  -0.160  -0.073  Prin 13  -0.031  -0.021  0.068  -0.401  1.248  Prin 14  -0.186  0.243  0.059  0.022  0.108  Prin 15  0.061  0.362  -0.319  -0.024  -0.194  Prin 16  0.084  -0.002  -0.102  -0.204  0.550  Prin 17  0.182  0.510  -0.291  -0.485  -0.892  52 Similar functions can be obtained with the remaining varieties.  An unknown cheese sample  could be classified by variety by substituting values for the principal components in each equation for the 5 varieties.  The unknown sample would be allocated to the variety with the  highest discriminant function. If prior probabilities were set equally, an overall classification rate of 72% was obtained (Table  4).  This increase was due to a large improvement  Parmesan and Swiss cheeses, which were weighted analysis.  in the classification rates for  much higher in the equal  probability  Cheddar and G o u d a were grouped at lower rates, while Edam was the same.  Thus  sample size may alter the results significantly. Discriminant analysis of the 18 PC's from peak areas, p H , and the 2 categorical data produced an overall 73% correct classification rate. The 18 PC's resulting from peak areas and p H yielded a 71% correct discrimination rate (Table 4). Discriminant analysis may also be performed on the original H P L C peaks.  A stepwise  discriminant analysis was carried out with the 55 peaks where a significance level to enter and stay in the function was 0.05.  The analysis added nine peaks to the function and removed  one before stopping. A summary table showing the peak selection at each step can be seen in Table 6. The discriminant function for Cheddar cheese is: Cheddar:  19.75(X ) + 2 6 . 5 9 ( X ) - 0.37(X ) + 1 1 . 5 5 ( X ) + 6 9 . 8 7 ( X ) + 0.010(X ) 6  28  9  10  50  39  -1.38(X ) + 3 . 3 6 ( X ) - 10.35. 21  15  Similar functions can be derived from the data in Table 7 for the remaining 4 varieties. Thus to classify an unknown cheese sample, the peak areas would be substituted into the functions and the variety with the highest score would be the identity of the unknown sample.  The  overall classification rate using the eight peaks was 63% with a high of 82%. for Cheddar, and a low of 42% for Swiss (Table 4).  This rate is comparable to that using the  17 principal  components. Discriminant analyses thus far indicated a similar or slightly higher variety classification than that of the sensory panel.  Table 4  rate of cheese  shows an overall 63% correct  53 Table 6. Summary table for peaks entered in stepwise discriminant analysis.  STEP  VARIABLE ENTERED REMOVED  F STATISTIC  1  X6  6.986  2  X28  6.202  3  X9  4.978  4  X10  4.703  5  X50  3.711  6  X39  3.060  7  X17  2.697  8  X21  2.518  9  X15  3.186  10  X17  2.296  Table 7. Coefficients for discriminant functions from eight HPLC peaks.  VARIABLE  CHEDDAR  EDAM  COUDA  SWISS  PARMESAN  Constant  -10.349  - 7.762  - 8.533  - 6.425  -10.966  X6  19.753  11.209  15.545  10.869  17.873  X28  26.588  12.571  7.313  4.670  13.101  X9  -0.371  29.500  -0.105  9.842  7.089  X10  11.550  11.702  20.103  4.725  6.810  X50  69.870  42.805  76.940  27.781  21.612  X39  0.010  0.172  -0.701  5.686  3.377  X21  -1.378  -32.250  -5.132  -13.882  127.163  X15  3.364  13.610  4.054  6.558  -16.393  55 classification rate by the taste panel, with a high of 86% for Cheddar and a low of 42 and 43% for Edam and C o u d a respectively. A variable refered to as HPLC column number, mentioned previously, was in reference to the four HPLC columns that were used to analyze the 106 cheese extracts.  The columms  were all the same type and from the same manufacturer, but ordered at varying times of the year. Throughout the life-time of a column, the chromatographic pattern was compressed and the peak resolution decreased.  This was shown by the elution pattern of the test mixture  that came with each column, and the retention time of the internal standard peak. To  determine  if column-to-column differences  existed,  principal component  and  discriminant analyses of cheese variety were performed separately on samples from each HPLC column.  For each column, 14 or 15 principal components were able to explain from 88 to  99% of the total sample variation (Tables 8 to 11).  Each principal component explained a  larger amount of the sample variation than with the earlier combined data analysis.  The  component loadings were again relatively low for most of the HPLC peaks and are not shown. Discriminant analysis by column of the PC's yielded a total correct varietal classification of 93 to 100% (Table 4).  Columns one and two grouped cheeses by variety at a 100% rate.  This clear separation by variety can be seen in the canonical plots in Figures 14 to  17.  Parmesan was only analyzed on column 4 and Swiss on columns one and three.  E. SENSORY D A T A  Each cheese sample was tasted by 5 to 7 semi-trained judges. The panelists rated the cheeses on a scale from 1 to 5 for taste intensity, preference, and bitterness. identified the cheese by variety.  These results should not be generalized to the public at  large, as they reflect the tastes of this single panel. differing results, thus the data is not absolute. cheeses labeled as  The judges also  Another panel of judges may indicate  It was difficult lo "train" the panelists since  a specific variety varied tremendously in the type of taste.  Domestic  56 Table  8.  PRINCIPAL COMPONENT  Eigenvalue, proportion of variance proportion of total variance, in a using chromatographic data from column 1.  EIGENVALUE  PROP. VAR. EXPLAINED  explained, and cumulative principal component analysis  CUMULATIVE VARIANCE  Prin 1  9.1690  0.1698  0.1698  Prin 2  6.5973  0.1222  0.2920  Prin 3  5.7970  0.1074  0.3993  Prin 4  4.0034  0.0741  0.4734  Prin 5  3.6187  0.0670  0.5405  Prin 6  3.2443  0.0601  0.6006  Prin 7  2.5710  0.0476  0.6482  Prin 8  2.1203  0.0392  0.6874  Prin 9  1.9767  0.0366  0.7240  Prin 10  1.8894  0.0350  0.7590  Prin 11  1.8416  0.0341  0.7931  Prin 12  1.4855  0.0275  0.8206  Prin 13  1.2708  0.0235  0.8442  Prin 14  1.0894  0.0202  0.8643  Prin 15  1.0545  0.0195  0.8839  57  Table  9.  PRINCIPAL COMPONENT  Eigenvalue, proportion of variance proportion of total variance, in a using chromatographic data from column 2.  explained, and cumulative principal component analysis  PROP. VAR. EXPLAINED  CUMULATIVE  EIGENVALUE  Prin 1  7.4347  0.1377  0.1377  Prin 2  6.6257  0.1227  0.2604  Prin 3  6.2795  0.1163  0.3767  Prin 4  5.6778  0.1051  0.4818  Prin 5  4.9778  0.0922  0.5734  Prin 6  4.1152  0.0762  0.6502  Prin 7  3.5952  0.0666  0.7168  Prin 8  3.1557  0.0584  0.7752  Prin 9  2.6337  0.0488  0.8234  Prin 10  2.4810  0.0459  0.8699  Prin 11  1.9786  0.0366  0.9066  Prin 12  1.7696  0.0328  0.9393  Prin 13  1.3892  0.0257  0.9651  Prin 14  1.1513  0.0213  0.9864  VARIANCE  58 Table 10.  PRINCIPAL COMPONENT  Eigenvalue, proportion of variance proportion of total variance, in a using chromatographic data from column 3.  EIGENVALUE  PROP. VAR. EXPLAINED  explained, and cumulative principal component analysis  CUMULATIVE VARIANCE  Prin 1  9.2034  0.1704  .0.1704  Prin 2  8.0748  0.1495  0.3200  Prin 3  5.7410  0.1063  0.4263  Prin 4  4.4100  0.0817  0.5079  Prin 5  3.2848  0.0608  0.5688  Prin 6  2.6238  0.0459  0.6174  Prin 7  2.3639  0.0438  0.6611  Prin 8  2.1926  0.0406  0.7017  Prin 9  2.1783  0.0404  0.7421  Prin 10  1.9920  0.0369  0.7790  Prin 11  1.6539  0.0306  0.8096  Prin 12  1.6045  0.0297  0.8393  Prin 13  1.2973  0.0240  0.8633  Prin 14  1.1711  0.0217  0.8850  Prin 15  1.1253  0.0208  0.9059  59 Table 11.  PRINCIPAL COMPONENT  Eigenvalue, proportion of variance proportion of total variance, in a using chromatographic data from column 4.  EIGENVALUE  PROP. VAR. EXPLAINED  explained, and cumulative principal component analysis  CUMULATIVE VARIANCE  Prin 1  10.2611  0.1900  0.1900  Prin 2  7.0130  0.1299  0.3199  Prin 3  4.9030  0.0908  0.4107  Prin 4  4.5140  0.0836  0.4943  Prin 5  3.6005  0.0667  0.5610  Prin 6  3.1023  0.0574  0.6184  Prin 7  2.4305  0.0450  0.6634  Prin 8  2.2988  0.0425  0.7060  Prin 9  2.2297  0.0413  0.7473  Prin 10  1.6892  0.0313  0.7786  Prin 11  1.5680  0.0290  0.8076  Prin 12  1.4294  0.0265  0.8340  Prin 13  1.2770  0.0236  0.8577  Prin 14  1.2529  0.0232  0.8809  4  o  o °  o  8 °  o  93 o  o  z  o  •  • •  o o  -2H  -4H  -6  -4  Figure 14.  O CHEDDAR  vw  V EDAM  -2  0  2  •  GOUDA  •  SWISS  4  1 ST CANONICAL VARIABLE  Canonical plot of cheese samples from column one grouped by variety.  ©  6 •  4H  •  V V  CD <  V V  or  V V  _i <  o z o z: < o Q Z CN  -2H -4H  -6H  -8 -8  Figure 15.  —\—  -6  T  1  1  1  r  -2 0 2 4 1 ST CANONICAL VARIABLE  -4  8  Canonical plot of cheese samples from column two grouped by variety.  O  CHEDDAR  V  EDAM  •  GOUDA  3n  2-  VV  UJ _i CO  V  <  $  • • •  V  o-  •  •  •  _l <  • -1-  < CN  -2-  O CHEDDAR  -3-  V EDAM -4  -5 -4  -2  0  2  4  1 ST CANONICAL VARIABLE  Figure 16.  Canonical plot of cheese samples from column three grouped by variety.  •  GOUDA  •  SWISS  4-i  2H  o o  •  o o  • •  Z  v  Q  o  -2H O CHEDDAR -4  V EDAM •  o  -6-  1  -2  O PARMESAN 1  1  0  2  - i  6  1 ST CANONICAL VARIABLE  Figure 17.  GOUDA  Canonical plot of cheese samples from column four grouped by variety.  64 versus imported cheeses, of a given variety, account for some of the diversity. The panelists were familiar with Cheddar cheese taste but varied in their exposure to the varieties of Edam, G o u d a , Swiss, and Parmesan. cheeses by variety.  At times they expressed frustration in the identification of  W h e n this occured, samples of each cheese variety were made available  for varietal confirmation. The overall percent correct classification rate by sensory analysis was 63%. classification rate by variety was:  The  Cheddar 86%, Edam 42%, G o u d a 43%, Swiss 74% and  Parmesan 76% (Table 4). The four parameters measured by the taste panel were analyzed by a multivariate analysis of variance ( M A N O V A ) .  Univariate  test statistics showed a highly significant (<A  = 0.0001) difference between cheeses for taste intensity, preference, and correct choice of variety (1=yes, 2 = no) but means  no difference for bitterness.  Waller-Duncan k-ratio t-tests of the  indicated Parmesan taste intensities were higher than the remaining 4 varieties; Edam  had a lower taste intensity than the other varieties; and Cheddar, G o u d a , and Swiss cheeses were perceived at an equal level of taste intensity.  Means for preference scores showed that  Cheddar, Edam, and G o u d a were equally liked, while Swiss was preferred less, and Parmesan even less than Swiss.  Means for the correct choice of identification revealed that Edam and  G o u d a were identified less often than Cheddar, Swiss, or Parmesan. The main objective of the sensory results was to obtain a classification rate by cheese variety.  A s e c o n d goal was to use the taste intensity ratings to evaluate the age of the  cheeses, but since few of the cheeses themselves were marketed by degree of ripening, this became impossible. The parameters of preference and bitterness were recorded to see if they influenced the selection by variety,  ln general the stronger flavored cheeses, Swiss and  Parmesan, were preferred less than the milder varieties of Cheddar, Edam, and Gouda. follows the general trend in America where  This  Cheddar is the most popular cheese consumed.  Battistotti et al. (1980) indicated that 80% of the cheese made in America is the Cheddar type.  65 F. MODIFIED EXTRACTION PROCEDURE  The relatively short life span of the H P L C columns in this study is of concern. extraction  method  could  be  lengthened  to  produce  an  extract  with  less  The  interfering  compounds, or a correlation factor must be determined to correct for column-to-column variation.  An objective of this study was to develop a quick simple extraction procedure that  could be used for quality monitoring by cheese manufacturers.  Thus an extended extraction  procedure to extend the HPLC column life would be counter productive. Lipid material in the cheese extracts may be responsible for the rapid deterioration of the columns. A second solvent extraction step and the use of a C8 extraction tube was used to clean up the  extracts (Figure  18).  The evaporation of methanol with nitrogen was  eliminated in the modified procedure to save time, but a precipitate formed in the extracts. The extracts were refrigerated overnight and filtered through a 0.45 j u m filter to remove the precipitate prior to analysis. The decrease in peak resolution and shortening of the retention times still occured. These modifications were only examined with a medium Cheddar cheese. Rather than  lengthen  the  extraction  procedure, the  column variability  could be  monitored and corrected. A standard cheese extract run on each new HPLC column would indicate differences and allow subsequent modifications in the detection or interpretation of peaks.  The use of two internal standards such as alanine and methionine would also allow  better detection of the column's condition and of any changes in retention times.  66 4g CRATED CHEESE  CENTRIFUGE (27,000 X g, 25C, 30 min)  r  RESIDUE  FAT  I  STOPPERED TUBE 1mL MeOH 1mL C H C I 0.6mL H 0 2  2  2  2 TIMES  SHAKE 15 sec.  CENTRIFUGE (27,000 X g, 25C, 20min)  CH CI 2  2  RESIDUE  MeOH/H 0 2  1mL C H C I 2  CH CI 2  2  1  2  MeOH/H 0 2  C18 EXTRACTION TUBE REFRIGERATE OVERNIGHT FILTER THROUGH 0.45pm PORE SIZE HPLC ANALYSIS  Figure 18.  Modified extraction procedure of water-soluble cheese components for HPLC analysis.  67  V. CONCLUSIONS  Cheese ripening is a complex process in which both aroma and taste contribute to the overall flavor.  Volatiles, such as free fatty acids and carbonyl c o m p o u n d s , are important to  aroma, whereas protein breakdown products, such as peptides and amino acids, have been shown to be responsible for cheese taste.  These proteolysis products are found in the  nonvolatile water soluble fraction of cheeses. ln this study, water soluble extracts of over 100 cheeses representing five varieties were analyzed with a reversed phase high performance liquid chromatograph (RP/HPLC).  A  quick single step extraction that could be used by cheese manufacturers was developed. The nonvolatile fraction was collected in a water-methanol layer while the fat soluble components, were eliminated in a methylene chloride layer.  The simplicity of the extraction most likely  allowed some lipid material in the water-soluble fraction, which shortened the column life, as four H P L C columns were required to analyze the cheeses. The conditions used to elute the taste c o m p o u n d s were determined by a combination of Mapping Super-Simplex and Centroid Mapping Optimization .  Optimization resulted in a  ternary gradient elution, using trifluoroacetic acid (0.1%), methanol, and acetonitrile, which yielded a maximum number of resolved peaks consisting of a range of c o m p o u n d s of differing polarity. Multivariate  statistical  chromatographic profile  analyses  in order to  C o u d a , Swiss, and Parmesan.  were  performed  characterize the  on  55  cheese varieties  peaks  from  each  of Cheddar, Edam,  Principal component analysis was successful in reducing the  dimensionality of the data from 55 peaks to 17 principal components consisting of linear combinations of the peaks. interpretation  Since the identity  of the results is difficult.  of the  individual peaks are unknown, an  Discriminant analysis performed o n "both the  17  68 p r i n c i p a l components and the original peaks yielded a 64% correct classification rate by cheese variety.  These data were comparable to the 63% rate from a semi-trained taste panel.  C o l u m n variability was a problem that must be resolved.  Chromatographic data  analyzed by a single HPLC column was able to discriminate by cheese variety at a greater than 90% rate.  A standard cheese extract should be analyzed on each new column to check for  variation.  Cheeses used in this study were both Canadian and imported and thus represented  a range of flavors and histories within each variety. unlikely encounter the diversity found here.  A n individual cheese producer would  Therefore this method would be useful to  register a fingerprint of each cheese production lot.  O n c e a history of the H P L C profiles was  established by a cheese manufacturer, problems or variations from day-to-day would b e c o m e apparent.  69  REFERENCES  Acharya, A. S., Donato, A. D., Manjula, B. N., Fischetti, V. A., and Manning, J. M. Influence of trifluoroacetic acid on retention times of histidine-containing peptides in R/P HPLC. Int. J. Peptide Protein Res. 22: 78. Adda, J., C r i p o n , J. C , and Vassal, L. 1982. cheese. Food C h e m . 9: 115.  1983. tryptic  The chemistry of flavor and texture generation in  Aishima, T. 1983. Relationships between gas chromatographic profiles of soy sauce volatiles and organoleptic characteristics based on multivariate analysis. In "Instrumental Analysis of Foods," Vol. 1, p. 37. Academic Press, NY,NY. Aishima, T. 1985. Application of multivariate analysis to silica capillary C C profiles to differentiate the aroma characteristics of Worcestershire sauces. In "The Shelf Life of Foods and Beverages," p.755. Elsevier Sci. Pub., Amsterdam,Holl. Aishima, T., and Nakai, S. 1986. Centroid mapping optimization: A optimization for f o o d research and processing. J. F o o d Sci. 51: 1297.  new  efficient  Aishima, T., and Nakai, S. 1987. Pattern recognition of C C profiles for classification of cheese variety. J. Food Sci. In press. Ali, L. A. M., and Mulder, H. 1961. Taste and flavor-forming compounds in cheese. amino acid content of Edam cheese. Neth. Milk and Dairy J. 15: 377.  The  Amantea, C . F. 1984. Assessment of bacteriophage-insensitive culture bacteria for Cheddar cheese making and subsequent discriminant analysis for objective flavor evaluation. Ph. D. thesis, University of British Columbia, Vancouver, B. C . Anon.  1986. Extending the lifetime of liquid chromatography columns by using guard columns and scavenger columns (precolumns). Chromatography International 16: 16.  Ashoor, S. H., and Knox, M. J. 1985. Identification of fish species by high-performance liquid chromatography. J. Chromatogr. 324: 199. Aston, J. W., and Dulley, J. R. 1982.  Cheddar cheese flavor. Aust. J. Dairy Tech.  37: 59.  Aston, J. W., and Douglas, K. 1983. The production of volatile sulfur compounds in Cheddar cheese during accelerated ripening. Aust. J. Dairy Tech. 38: 66. Aston, J. W., and Creamer, L. K. 1986. Contribution of the components of the water-solublefraction to the flavor of Cheddar cheese. N. Z. J. Dairy Sci. Tech. In press. Aston, J. W., Durward, I. C , and Dulley, J. R. 1983. Cheddar cheese. Aust. J. Dairy Tech. 38: 55. Babel, F. ]., and Hammer, B. W. 4:81. Badings, H. T., and Neeter, R. Milk Dairy J. 34: 9.  1939.  1980.  Proteolysis and flavor development in  Flavor contributors of Swiss-type cheese.  F o o d Res.  Aroma c o m p o u n d s of milk and dairy products.  Neth.  70 Bassett, E. W . , and Harper, W . J. 1958. Isolation and identification of acidic and carbonyl c o m p o u n d s in different varieties of cheese. J. Dairy Sci. 91: 1206.  neutral  Battistotti, B., Bottazzi, V., Piccinardi, A., and Volpato, C . 1984. " C h e e s e : World of Cheese and Cheesemaking," Facts on File Inc., NY,NY.  to the  Berridge, J. C . 1982. Unattended optimization Chromatogr. 224: 1.  A Guide  using the modified simplex algorithm.  J.  Bertuccioli, M., M o n t e d o r o , C , and Clementi, S. 1986. Chemometric investigation on ripening and storage of Provolone cheese by instrumental and sensory evaluation of aroma. In "The Shelf Life of Foods and Beverages," Elsevier Sci. Pub., Amsterdam, Holl. Bianchi, A., Beretta, C , Caserio, G . , and Ciolitti, G .  1974.  Amino acid composition of  granules and spots in Grana Padano cheese. J. Dairy Sci. 57: 1504. Bican, P., and Blanc, B. 1982. Milk protein analysis-a high performance study. Milchwissenschaft. 37: 592. Biede, S. L., and H a m m m o n d , E. G. Dairy Sci. 62: 227. Biede, S. L., and H a m m o n n d , E. C . Dairy Sci. 62: 238. B M D P , 1985.  1979a.  1979b.  Swiss cheese flavor:  Swiss cheese flavor:  II.  I.  chromatography  Chemical analysis.  ).  Organoleptic analysis. J.  " B M D P Statistical Software Manual," University of California Press, Berkeley, C A .  Campbell, J. R., and Marshall, R. T. Hill Inc. NY,NY.  1975.  "The Science of Providing Milk for M a n , " McGraw-  Champion, H. M., and Stanley, D. W . 1982. HPLC separation Cheddar cheese. Can. Inst. F o o d Sci. Technol. J. 15: 283.  of bitter peptides  from  Chang, S. F., Ayres, J. W . , and Sandine, W . E. 1985. Analysis of cheese for histamine, tyramine, tryptamine, histidine, tyrosine, and tryptophane. J. Dairy Sci. 68: 2840. Czulak, J. 1959.  Bitter flavor in cheese. J. Dairy Technol. 14:177.  D'Agostino, G . , Castagnetta, L., Mitchell, F., and O'Hare, M. J. 1985. Computer-aided mobile phase optimization and chromatogram simulation in high-performance liquid chromatography. J. Chromatogr. 338: 1. Dalgleish, D. G . 1982. The enzymatic coagulation of milk. In "Developments Chemistry-1 Proteins," p. 157. Applied Sci. Pub., Essex, England.  in Dairy  Daultrey, S. 1976. Principal components analysis comcepts and techniques in geography. N o . 8. G e o Abstracts Ltd., Norwich, England.  modern  Day, E. A., Bassette, R., and Keeney, M. 1960. Identification of volatile carbonyl c o m p o o u n d s from Cheddar cheese. J. Dairy Sci. 43: 463. Debets, H. J. G . 1985.  Optimization methods for HPLC.  J. Liq. Chromatogr.  8: 2725.  71 Di Bussolo, J. M., and Cant, J. R. 1985. Visualization of protein retention and migration in reversed-phase liquid chromatography. J. Chromatogr. 327: 67. Dillon, W . R., and Goldstein, M.  1984.  "Multivariate Analysis Methods and Applications,"  John Wiley and Sons, NY,NY. Diosady, L. L , Bergen, I., and Harwalker, V. R. 1980. High chromatography of whey proteins. Milchwissenschaft. 35:671. Dulley, J. R., and Grieve, P. A. 1974. J. Dairy Tech. 29: 120.  performance  liquid  Volatile fatty acid production in Cheddar cheese. Aust.  Forina, M., Armanio, C , Lanteri, S., and Tiscornia, E. 1983. Classification of olive oils from their fatty acid composition. In " F o o d Research and Data Analysis," p. 189. Applied Sci. Pub., Essex, England. Forss, D. A. 1979. Mechanisms of formation of aroma c o m p o u n d s in milk and milk products. J. Dairy Res. 46: 691. Clajch, J. L , Kirkland, J. J., Squire, K. M., and Minor, J. M. 1980. Optimiztion of solvent strength and selectivity for R/P liquid chromatography using an interactive mixturedesign statistical technique. J. Chromatogr. 199:57. Grappin, R., Rank, T. C , and O l s o n , N. F. 1985. Primary proteolysis of cheese proteins during ripening. A review. J. Dairy Sci. 68: 541. Hancock, W. S., and Sparrow, J. T. 1983. The separation of proteins by R/P HPLC. In "HighPerformance Liquid Chromatography Advances and Perspectives," Vol 3. p.50. Academic Press, NY, NY. Harper, W . J. 1959.  Chemistry of cheese flavor.  J. Dairy Sci. 4 2 : 2 0 7 .  Harvey, R. J. 1960. Production of acetone and acetaldehyde by lactic streptococci. Res. 27: 41.  J. Dairy  Harwalkar, V. R., and Elliott, J. A. 1971. Isolation of bitter and astringent fractions from Cheddar cheese. J. Dairy Sci. 54: 8. Hearn, M . T. W . 1983. High-performance liquid chromatography of peptides. In " H i g h Performance Liquid Chromatography," Vol. 3. p 87. Academic Press, NY, NY. Hintz, P. C , Slatter, W . L., and Harper, W. J. 1956. A survey of various free amino and fatty acids in domestic Swiss cheese. J. Dairy Sci. 39: 235. Hollywood, N. W . , and Doelle, H. W . 1984. The effect of sampling position and age on levels of propionibacteria and flavor-related c o m p o u n d s in Swiss-type cheese. Aust. J. Dairy Technol. 39: 80. Johnson,  R. A., and Wichern,  D.  W.  1982.  Prentice-Hall Inc., Englewood Cliffs, NJ.  "Applied Multivariate Statistical Analysis,"  72 Karninogawa, S., Yan, T. R., Azuma, N., and Yamauchi, K. 1986. Identification of low molecular weight peptides in Couda-type cheese and evidence for the formation of these peptides from 23-N-terminal residues of Q! -| -casein by proteinases of Streptococcus cremoris H61. J Food Sci. 51: 1253. s  Kolbe, H. V. J., Lu, R. C , and Wohlrab, H. 1985. R-P HPLC separation and quanitiation of phenylthiohydantoin derivatives of 25 amino acids, including those of cysteic acid, 4hydroxyproline, methionine sulfone, s-carboxy methylcysteine, and s-methylcysteine. J. Chromatogr. 327: 1. Kosikowski, F. V. Ml.  1977.  "Cheese and Fermented Milk Foods," Edwards Bros. Inc., Ann Arbor,  Kosikowski, F. V.  1985.  Cheese. Scientific American. May: 88.  Kosikowski, F. V., and Dahlberg, A . C . 1954. A quantitative appraisal of the free amino acids in foreign type cheese. J. Dariy Sci. 37: 167. Krett, O . )., and Stine, ). B. Sci. 34: 476.  1951.  The role of the lower fatty acids in Swiss cheese.  J. Dairy  Kristoffersen, T., and Nelson, R. E. 1955. The relationship of serine deamination and hydrogen sulfide production by Lactobacillus casei to Cheddar cheese flavor. J. Dairy Sci. 38: 1319. Kristoffersen, T., and G o u l d , I. A. 1960. Cheddar cheese flavor II. Changes in flavor quality and ripening products of commercial Cheddar cheese during controlled curing. ]. Dairy Sci. 43: 1202. Krstulovic, A. M., and Brown, P. R. 1982. "Reversed-Phase High-Performance Liquid Chromatography Theory, Practice and Biomedical Applications," John Wiley and Sons, NY, NY. Kuchroo, C . N., and Fox, P. F. 1982a. Fractionation of the water-soluble-nitrogen Cheddar cheese: Chemical methods. Milchwissenschaft. 37:651. Kuchroo, C . N., and Fox, P. F. 1982b. Soluble nitrogen in Cheddar cheese: extraction procedures. Milchwissenschaft. 37: 331. Kurtz,  from  Comparison of  F. E., Hupfer, E. A., Corbin, R. E., Hargrove, R. E., and Walter, H. E. 1959. Interrelationships between p H , populations of Propionibacterium shermanii II, levels of free fatty acids, and the flavor ratings of Swiss cheese. J. Dairy Sci. 42: 1008.  Langler, J. E., Libbey, L. M., and Day. E. A. 1967. Identification and evaluation of selected c o m p o u n d s in Swiss cheese flavor. J. Agr. Food C h e m . 15: 386. Langlois, B. E., and Parmelee, C . E. 1963. Chemical changes in Swiss-type cheese during manufacture and curing. J. Dairy Sci. 46: 604. Langsrud, T., and Reinbold, C . W . 1973a. Flavor development and microbiology of Swiss cheese-A review I Milk quality and treatments. J. Milk Food Technol. 36: "487.  73 L a n g i i u c l , T., and Reinbold, G . W . 1973b. Flavor development and microbiology of Swiss cheese-A review II. Starters, manufacturing processes and procedures. J. Milk Food Technol. 36: 531. Langsrud, T., and Reinbold, G. W . 1973c. Flavor development and microbiology of Swiss cheese- A review III. Ripening and flavor production. J. Milk Food Technol. 36:593. Langsrud, T., and Reinbold, G . W . 1974. Flavor developments and microbiology of Swiss cheese-A review IV. Defects. J. Milk Food Technol. 37: 26. Law, B. A.  1981.  The formation of aroma and flavor compounds in fermented dairy products.  Dairy Sci. Abstracts. 43: 143. Law, B. A. 1982.  Flavor compounds in cheese.  Perfumer and Flavorist.  7: 9.  Law, B. A., and Sharpe, M. E. 1977. The influence of the microflora of Cheddar cheese on flavor development. Dairy Ind. Int'l. 4 2 : 1 0 . Law, B. A., Castanon, M . J., and Sharpe, M. E. 1976. The contribution of starter streptococci to flavor development in Cheddar cheese. J. Dairy Res. 43: 301. Lowrie, R. J., and Lawrence, R. C . 1972. N. Z. J. Dairy Sci. Technol. 7: 51. In Visser, F. M. W . 1977. Contribution of enzymes from rennet, starter bacteria anc milk to proteolysis and flavor development in G o u d a cheese. 2. Development of bitterness and cheese flavor. Neth. Milk Dairy J. 31: 188. Manning, D. J. 1978. Cheddar cheese flavor studies I. Production development of flavor during ripening. J. Dairy Res. 45: 479.  of volatiles  and  Manning, D. J., and M o o r e , C . 539.  J. Dairy Res.  46:  1979.  Headspace analysis of hard cheese.  Manning, D. J., Chapman, H. R., and Hosking, Z . D. 1976. The production of sulfur compounds in Cheddar cheese and their significance in flavor development. J. Dairy Res. 43: 313. Marsili, R. 1985. Monitoring chemical changes in Cheddar cheese during aging by high performance liquid chromatography and gas chromatography techniques. J. Dairy Sci. 68: 3155. Martens, H.  1983.  Understanding f o o d research data.  In " F o o d Research and Data Analysis,"  p. 5. Applied Sci. Pub., Essex, England. Martens, H., and Russwurm, H. Essex, England.  1983.  " F o o d Research and Data Analysis," Applied Sci. Pub.,  Martens, M . 1986. Sensory and chemical/physical quality criteria of frozen peas studied by multivariate data analysis. J. F o o d Sci. 51: 599. Matheson, A. R.  1981.  The immunochemical determination  of chymosin activity in cheese.  N. Z. J. Dairy Sci. Technol. 15: 33. M c G u g a n , W . A . , Emmons, D. B., and Larmond, E. 1979. fractions on intensity of Cheddar cheese flavor.  Influence of volatile and nonvolatile  J. Dairy Sci. 62: 398.  74 Mi!!:, O . E., and Thomas, T. D. 1980. Bitterness development in Cheddar cheese: the level of starter proteinase. N. Z. J. Dairy Sci Technol. 15: 131.  Effect of  Mitchell, C . E. 1981. The production of selected c o m p o u n d s in a Swiss-type cheese and their contribution to cheese flavor. Aust. J. Dairy Tehcnol. 36: 21. M o c q u o t , C . 1979. Reviews of the progress of dairy science: Res. 46: 133. Morgan, S. L , and Deming, S. N. Anal. C h e m . 46: 1170.  1974.  Morgan, S. L , and Deming, S. N. 1975. liquid chromatographic methods. Mulder, H.  1952.  J. Dairy  Simplex optimization of analytical chemical methods.  Optimization strategies of the development of gasJ. Chromatogr. 112: 267.  Taste and flavor forming substances in cheese. Neth. Milk Dairy J. 6:157.  Nakai, S., and Kaneko, T. Sci.  Swiss-type cheese.  1985.  Standardization of mapping simplex optimization.  J. Food  50: 845.  Nakai, S., Koide, K., and Eugster, K. 1984. A sew mapping super-simplex optimization for f o o d product and process development. ]. Food Sci. 49: 1143. N o o m e n , A . 1977. A rapid method for the estimation of the dissolved and the undissolved nitrogen c o m p o u n d s in cheese. Neth. Milk Dairy J. 31: 163. Olieman, C , and van den Bedem, ]. W. 1983. A sensitive H P L C method of detecting and estimating rennet whey total solids in skim milk powder. Neth. Milk Dairy J. 37: 27. Patton, S., W o n g , N. P., and Forss, D. A. cheese. ]. Dairy Sci. 41: 857.  1958.  Some volatile components of Cheddar  Paulsen, P. V., Kowalewska, J., H a m m o n d , E. C , and Clatz, B. A. 1980. Role of microflora in production of free fatty acids and flavor in Swiss cheese. J. Dairy Sci. 63: 912. Pearce, R. J. 1983. Analysis of whey proteins by high performance liquid chromatography. Aust. J. Dairy Technol. 38: 114. Pham, A - M , and Nakai, S. 1984. Application of stepwise discriminant analysis to H P L C profiles of water extract for judging ripening of Cheddar cheese. J. Dairy Sci. 67: 1390. Polo, C , Ramos, M . , and Sanchez; R. 1985. Free amino acids by high performance liquid chromatography and peptides by gel electrophoresis in Mahon cheese during ripening. F o o d C h e m . 16: 85. Rank, T. C , Crappin, R., and O l s o n , N. F. ripening: A review. Reinbold, C . W .  1963.  "Italian C h e e s e Varieties,"  Rothe, M., Engst, W . , and Erhardt, V. Die Nahrung.  1985.  Secondary proteolysis of cheese during  J. Dairy Sci. 68; 801.  26: 591.  1982.  Chas. Pfizer and C o . Inc., NY, Ny.  Studies on characterization of Blue cheese flavor.  75 Routh, M. W., Swartz, P. A., and Denton, M . B. simplex. Anal. C h e m . 49: 1422.  1977.  Performance of the super modified  Santa-Maria, C , Ramos, M., and O r d o n e z , J. A. 1986. Application of linear discriminant analysis to different proteolysis parameters for assessing the ripening of M a n c h e g o cheese. F o o d C h e m . 19:225. SAS Institute Inc.  1985.  "SAS User's Guide: Statistics," SAS Institute Inc., Cary N C .  Scoble, H. A., and Brown, P. R. 1983. Reversed-phase chromatography of nucleic acid fragments. In "High-Performance Liquid Chromatography Advances and Persepctives," Vol. 3, p. 1. Academic Press, NY, NY. Sherman, J. M. 1921. The cause of eyes and characteristic flavor in Emmental or Swiss cheese. J. Bacteriol. 6: 379. Singh, V. K., and Kristoffersen, T. Sci. 54: 349. Sloot, D., and Harkes, P. D. C h e m . 23: 56.  1971.  1975.  Accelerated ripening of Swiss cheese curd. J. Dairy  Volatile trace components in G o u d a cheese. J. Agr. F o o d  Snyder,  L. R. 1974. Classification of Chromatogr. 92: 223.  the  solvent  properties  of  c o m m o n liquids.  J.  Snyder,  L. R. 1978. Classification of Chromatogr. Sci. 16: 223.  the  solvent  properties  of  c o m m o n liquids.  J.  Snyder, L. R., and Kirkland, J. J. Wiley and Sons, NY, NY.  1974.  Stadhouders, J., and Veringa, H. A. Neth. Milk Dairy J. 27: 77.  "Introduction to M o d e r n Liquid Chromatography,"  1973.  Fat hydrolysis by lactic acid bacteria in cheese.  Stungis, C . E. 1976. Overview of applied multivariate analysis. A S T M STP 594. for Tesing and Materials. Philadelphia, PA.  American Soc.  Visser, F. M . W . 1977a. Contribution of enzymes from rennet, starter bacteria and milk to proteolysis and flavor development in G o u d a cheese. 2. Development of bitterness and cheese flavor. Neth. Milk Dairy J. 31: 188. Visser, F. M. W . 1977b. Contribution of enzymes from rennet, starter bacteria and milk to proteolysis and flavor development in G o u d a cheese. 3. Protein breakdown: analysis of the soluble nitrogen and amino acid nitrogen fractions. Neth. Milk Dairy J. 31: 210. Visser,  S. 1981. Proteolytic enzymes Netherlands Milk Dairy J. 35: 65.  and their  action  on  milk  proteins.  A  review.  Visser, F. M . W . , and de Croot-Mostert, A. E. A. 1977c. Contribution of enzymes from rennet, starter bacteria and milk to proteolysis and flavor development in G o u d a cheese. 4. Protein breakdown : a gel electrophoretical study. Neth. Milk Dairy J. 31: 247.  76 Wilson, K. J., Honegger, A., Stotzel, R. P., and Huges, C . ). 1981. The behavior of peptides on R/P supports during high-pressure liquid chromatography. Biochem. J. 199: 31. W o o , A. H., and Lindsay, R. C. 1984. Concentrations of major free fatty acids and flavor development in Italian cheese varieties. J. Dairy Sci. 67: 960.  Appendix 1.  E i g e n v e c t o r s f o r p r i n c i p a l component a n a l y s i s o f chromatographic d a t a  P R I N C I P A L COMPONENT A N A L Y S I S EIGENVECTORS  X3 X4  xs  X6 X7 XB X9 X10 XII X12 XI3 X14 X IS X16 X IT X18 X19 X20 X21 X22 X23 X24 X2S X26 X27 X28 X29 X30 X31 X32 X33 X34 X39 X3S X37 X3B X39 X40 X41 X42 X43 X44 X49 X46 X4T X48 X49  xso X51 XS2 X54 X9S  X56 XST  PRIN1  PRIN2  PR INS  PRIN4  0 041453  0.063780 .18O610 ,030137 .140509 ,102451 . 123993 .058342 . 158743 .082001 .102753 193441 .186919 232743 .207794 248937 . 253005 201028 214759 . 207656 .064517 .286212 , 197459 121 ISO . 149716 .086383 .017531 026361 026900 .015409 .184606 076538 .008831 059565  0.054334 -.113506  0.O575O5 006825 059492 025982 033275 043054 .049447 015729 .028946 .050359 .179538 ,047658 .081033 ,119097 ,119566 ,094240 .132863 .168726 .1B0899  177279 0 . 166300 0 . 077552 086986 226204 0 . 228795 0 . 144232 0 . 267655 0 . 201728 0 . 044342 0 . 07B254 069913 045551 081439 002394 110345 045685 048675 082493 076586 108772 015140 O6208I 052867 254304 254071 246543 183039 053342 055175 132118 203440 040991 O.035600 051879 192836 089337 076793 0 . 132469 060009 088349 0 . I99B60 089190 030983 0 . 097407 0 . 016224 o. 134086 059323 154598 106970 0 , 172577 0 . 199305  - .183057 - .151732  0. 1 7 3 6 9 8 0. 1 7 8 7 5 4  .016825 .078775 ,057280 015826 144363 .059996 0 .. 0 7 9 8 4 7 .054314 0 .. 1 5 7 7 5 B .039829 .171820 0 .,158092 .033001 0 .. 0 1 9 3 9 7 0. 048241 .005376 .108107 .056392 .13602S .010173 .082902 o. 0 3 9 3 6 7 0 . 147773 200932 0 209144 0 . 128319 16(366 .165457 .045224 0. .231603 .272859 .188221 o. .195744 0 . 226626 .206396 053437 .101460 0 .. 1 4 0 5 6 3 029172 051069 .193662 0 .184222 .167096 .112215 0 .207311 0 .227639 0 .087551 0 .171023 0 .022572 0 .153658 0 . 126133  0. 0.  0. 0. 0. 0.  0. 0 •0. 0. 0.  0. 0 0, 0. 0. 0. 0. 0. 0. 0 0. 0 0  0. 1 2 7 1 6 7 0. 133741  0. 0. 0. 0. 0 0 0 0. 0. 0. 0. 0. 0. 0. 0. 0 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.  0.  0. 0 0.  -  0.  0.  0.  0. 0. 0. 0. 0.  0. -, 0. 0. 0. 0. 0. 0.  0.  0. 1 15387 -.272346 0.002566 -.049784 -.045557  0.145221 0.169383 0.065576 0.205796  0.020432 -.005374 0.028614  0.128722 0.293709 0.249423  0.066471 0.24237S -.088399 -.233869 0.075218 -.113786 -.239986 O.025950 -.068188  0.13247S -.097106 -.069556 -.354889  0.121377  -.132763 -.031308 - . 124765 -.016233  0.024273 -.082670 -.112650 -.182957 0.049374 -.192529 0.015068 0.012328  0.177734 0.174731 0.058802 0.176186 0.050895 -.120691 0.021780 0.014394 0.065582  0.030951 0.045367  PR INS 0 . 159709 0 .018937 0 . 142937 - .0O9BB8 - .026635 - .029328 0 0O0391 .096194 - .177211 - .329861 0 .004429 - .138936 0 .081710 .027543 .040718 0 ., 0 2 0 7 4 3 0 . ,134718 ,005000 .000267 ,040362 .064792 0 . . 155791 0 ,. 1 5 3 8 0 3 .O705S9 .057036 .045229 .065741 0 . 148337 0 . 188236 ,061004 .052838 .002561 0 .277212 0 .136206 0 ,. 2 2 7 2 3 9 ,075278 ,107017 ,316206 ,265073 0 .121734 0 .234742 0 . 2 1 1998 0 .. 0 7 4 3 3 5 0 ,029544 .111134 .041970 0 .004164 0 .144399 0 .056579 0 .239243 0 .066621 . 133469  -  -  -  - .081032 - .208169  PRIN6  PRIN7  182430 055313 062544 134026 124829 341959 125385 275833 062053 071972 0 . 218920 0 . 092614 0 . 161979 0 . 047096 10097 1 0 . 093245 0 . 105552 059574 0 . 113963 210026 0 . 183168 0.040003 0 . 163730 -.023120 0 . 015496 -.031372 087885 -.090950 -.093079 0 . 179018 -.013551 089772 161463 142953 0.075839 0 . 019137 0 . 159205 0 . 042556 -.045937 O.0304O5 0 . 058424 0. 018333 198997 0.076702 102129 0 . 017661 0 . 117227 0 . 194612 0 . 094222 0 . 002178 0 . 289772 0 . 221639  0 .132599 009363 0 .. 2 8 4 6 0 0 .104 186 0 . 168121 .067880 0 ,. 2 7 6 7 4 2 .030618 .087745 .068545 .119512 ,112520 0 120509 0 . 054021 007503 041671 051357 ,139564 005783 231068 0 . 100575 0. 046993 0 ,102583 0.067524 0 069252 .056340 0 . 109189 .226437 .024222 0.001981 0 O0B50O 0 . .038721 .070922 0 . 144949 0. 021098 0 . 110595 0 . 230821 0 . 075041 0.025444 .398403 0 .054730 092481 0 . ,074021 0 . .191644 0 ., 2 5 9 0 8 9 0 . 317176 .009155 0 .. 0 2 0 8 5 3 .037991 0. 08S32O .132813 .011164  0 . 103067 061670 012806 077896 073949 .049980 0 . 141764 0 055449 .123406 0 . 0 7 1591 002153 0. 024917 0 . 154473 0 . 000544 O. 2 0 6 9 1 3 0 . 021458 0 . 127922 0 . 040117 0 . 009012 054682 007433 264668 167443 0 . 122782 0 . 038413 0 . 0O6978 091972 0 . 173028 0.035292 ,038181 . 185884 063119 .134908 125952 0 . 121586 206613 128758 0 . 11064 1 0 . 196309 241953 340896 0. 004406 121058 0. 005429 ,166535 097711 0. 089545 0 . 341029 0. 222933 0 271813 .005594 .1052 30  .133573 .138323  0.031940 .016210  0. 0. 0. 0.  0 .241496 0 .097195  -  -  _  -  PRIN8  0.  -  PRIN9 0 .022933 0 .086822 0 .034790 O .166322 - .025082 0 .187681 .132784 0 .023933 0 .192988 0 .032896 .089042 .105931 0 .021721 - .045271 0 . 151908 .052492 .156304 0 .203380 .067938 .036726 0 .147990 0.025486 0 .005051 .113344 .00OO58 .047614 .O12O10 - .214886 .084621 0 .101933 0 .096797 0 .065470 0 .225239 .211783 .133529 0 .102926 0 .171157 .023118 .300824 .267483 .235585 0.051467 0 .248779 0.062243 .134858 .231028 .047772 0 .26(000 0 .050611 0 .162496 .003198 .012978  -  -  .-  0. 1 3 S S B 4 .0330?6  PRIN10 0 .080488 .314731 0 .091747 .105987 .08179B 0 .034299 0 .119666 0 .026434 .003879 .065543 .161708 0 .120386 0 . 148827 0 .. 0 0 5 0 5 0 - .033924 0 .033716 043759 .091037 0 .075246 0 .121768 . 174126 0 .059236 0 .168723 o . 128662 - .O99609 0 .065204 - .045076 0 .114082 . 108429 .181656 0 .009289 .074933 0 .087259 202388 - ..0887 29 0 .078859 0 ,409991 0 .42824 1 0 .039812 0 .130878 0 .149879 .096701 0 . 1 13873 .O33210 .179910 .180477 .052556 0 .019848 0 .017636 0 . 1 146t9 0 .00O854 . 143557  -  0  -  -  -  _  .000762  - .111320  PRINCIPAL COMPONENT ANALYSIS EIGENVECTORS PRIN11 X3 X4 XS X6 X7 X8 X9 X10 XII xia X13 X14 X1B X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X2T X2S X2B X30 X31 X32 X33 X34 X3S X36 X3T X38 X39 X40 X41 X42 X43 X44 X4S X46 X47 X48 X49 X30 X31 XS2 X54 X55 XS6 XS7  074S68  0. 063692  105176 026077 034150 0. 034471 0. 099197 0. 0694 18 208970 298334 272120 219061 095469 0. 271084 0. 0. 110154 0. 1 15003 0. 353398 0. 027514 0. 170877 0. 358lOO -.045875 015350 0. 061204 0. 201563 0. 040277 0O3B87 0. -.062867 142273 091181 -.080594 024347 0. OO2072 0.064321 0. 078451 102476 0. 112867 039994 0. 088567 0. 036449 06 1599 071163 112110 0. 068138 058546 189180 0 .059424 .067393 154417 .182536 .077493 0 .101167 0 0 .129010 0. 247140 0. 285373  PRIN13  PRIN12 316533 081006 029014 0. 205148 0. 164 203 055682 0. 0. 288668 0. 104288 0. 019981 0. 016754 007177 0. 165017 163806 0. 1 12101 022222 048605 285732 089774 0. 041899 099747 0. 050749 0. 329182 0. 026506 0. 258683 0. 050336 OB 1087 -.097461 0.009924 169757 035796 063542 0. 0. 019969 0. 123624 ,103859 0. 126308 0..040050 .274 183 .186919 0..074079 0.067968 0 .032310 .219376 .077629 0 ,1S9860 .139883 .051148 .004BOO 0 .219761 .033768 0.051329 0.024395 .069365 0. 017288 0. 064941  -  -  0.0O8659 0.136673 - .349016 0.129105 0.070853 - .015620  0. 183029 .014151 0.159097 0.071149 .216266 .154421 - .11267B .262488 .102440 0.039455 0.129758 0.O01796 0.355627 .058580 .070102 0.012823 0.446023 0.021667 .053109 .055709 0.074189 .049429 .013383 .054270 0.185933 .050606 .068493 .115676 .145768 .056046 .147593  -  -  0.072713 0.165741  - .004650 .075264  0 0.089333 .002287 0.033419  - .059693  0 0.003921 0.040847 23 0.1607 0.140853 .129761 0.021744  - . 1 18669 - .099962  0.005715  _  PRIN14  PRIN13  PRIN16  PRIN17  PRIN18  .072506  0. 189066  0. 337488  0.314909 -.124602  0.023118 0. 297049 0. 177907 0. 337167 207803 0. 058082 0. 106718 164073 0. 017664 0. 020841 0. 070105 102160 044019 0. 190187 149823 0. 069436 1224B4 101698 0. -.034854 0. 106561 112109 005183 159097 127161 0. 110422 0. 144485 o. 233199 112287 0. 066720 12526B 104021 -.035480 211451 196617 105949 139638 0. 036651 O19990 031035 047576 0. 098495 0. 085257 208299 172236 0. 049351 0.  0.. 106733  .085602 0.029019 0.. 3S4208 0 .010538 0 088275 .005800 0..216132 ,087747 0.,137127 0..095950 0.086446 .121366 .103402 0.164323 .037921 .201853 .016897 090622 0..040697 -..061244 .128133 .198472 0..090342 0..126400 .044036 0.160391 0.368224 0.016213 0..086977 .043624 0.049021 0.074844 .137896 0.245498 0.063430 .112143 ,048311 .216560 0.032203 .243283 - .177139 - .057662 .164009 .070B97 0.010615 .063642 .068217 0.226410 .034599 .119985 .070236 0..172797  -  -  -  . 191628 ,095366 ,058246 0 .013841 , 134377 0., 1O07O6 0.146754 0..144B54 0..012869 ,185079 ,157470 0.076300 , 172723 0.,326511 , 121223' .040128 0 ,271987 .157821 .158285 0.031765 003342 0..150627 ,007083 .086053 0.210944 212906 0,,089266 0 .010566 0.006526 , 164144 0..011632 0.032938 .061584 13 0.,0896 0,,124889 .058842 0. .022590 .187051 0.,047453 0.169876 130059 - ..399681 .094476 .033474 0.033726 0.118899 .089292 0.015431 .052463 0.016735 0.109641 o. 120940 0..088148  -  -  -.054697 213093 0. 017583 0. 281630 0. 052039 120855 0. 036234 264013 166686 024593 0. 092041 031243 231786 0. 089127 068381 0. 011401  0.044222 0. 012492 120793 0. 058156 0. 23S065 0. 045328 022253 0. -.078107 008462 0. - .038008 0. 152078 132196 130893 0. 292455 0. 097242 0. 105 107 037508 271066 120O64 115577 001789 0. 192381 0. 059577 229525 0. 083261 131867 0. 210790 104328 106253 057632 0. 105030 0. 056283 093802 001079 0. 002654 039384 005468  0  -.  -.  0.071485 -.027007 0. 199355 -.O04911 0. 153567 -.015438 -.047206 0.199592 0.029524 -.026843 -.107927 0.292353 -.047447 0.064365 0.036090 -.318279 -.059635 0. 124249 0.052236 -.152580 0.O07770 -.180572 0.129581 0.089477 0.331568 •-.025834 -.267973 0.188297 0.002427 0.103330 0.031738 -.143584 -.025671 - .071056 -.138143 -.130642 -.034363 0.046824 0.148671 0.037504 0.097191 -.226613 -.078812 -.016063 0.010117 0.114013 0.103263 0.019444 0.040O18 -.127714 0.150935 0. 112318  0.059709 102035 0. 040412 0. 0352O6 173814 0. 300402 0. 012651 1524 13 055382  -. -.  PRIN19  PRIN20  -.234374  -.021472 -.140946 -.242596  0.203534 -.052234 -.347832 0.278778 -.038881 -.154036 -.006328 -.123474 0.131897 -.054283 -.201208 0.036420 0. 1319O0 -.1O6070 -.061269 0.136402 0.081037 -.080395 0.062429 0.046951 0.156558 -.011659 0.035116 -.058874  0.091699 -.OB7742 -.019693 0.335541 -.079409 -.060220 -.165143 0.022683 -.085623 0.005202 0.035524 0.095098 0.089682 -.043969 -.071067 0.172014 -.127777 0.062684 0.166113 -.377658 -.066633 -.093934  0.249397 0.293773 0.044636 0.170761 -.000794 -.095351 -.053256  0.054862 -.061319 0.145111 -.090537  0.053397 -.016623 -.108159 -.216941  0.133776 0.105333 - . 122930  0. 154928 0.086834 -.009088 0.120443 -.068657 -.019247 -.004362 0.141456 0.181639 -.191041 -.180509 0.O03I50 0.204599 0. 115326 -.084097 0.003131 O.090O63 -.053776  0.160683 -.015610 -.200778  0.177303 0.039345 -.019548 -.052262 0.036083 0.014121 0.020248 0.102136 -.084386 0.368466 -.172722 -.062583 0.07 1459 0.277273 -.130532 0.028636 0.127081 -.007767 0.018979  


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items