UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Headspace gas chromatography for quality assessment of canned Pacific salmon Girard, Benoit 1991

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

Item Metadata

Download

Media
831-UBC_1991_A1 G57.pdf [ 9.98MB ]
Metadata
JSON: 831-1.0100443.json
JSON-LD: 831-1.0100443-ld.json
RDF/XML (Pretty): 831-1.0100443-rdf.xml
RDF/JSON: 831-1.0100443-rdf.json
Turtle: 831-1.0100443-turtle.txt
N-Triples: 831-1.0100443-rdf-ntriples.txt
Original Record: 831-1.0100443-source.json
Full Text
831-1.0100443-fulltext.txt
Citation
831-1.0100443.ris

Full Text

HEADSPACE GAS CHROMATOGRAPHY FOR QUALITY ASSESSMENT OF CANNED PACIFIC SALMON by BENOIT GIRARD B.Sc, L'Universite Laval, 1984 M.Sc, The University of B r i t i s h Columbia, .1987 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY i n THE FACULTY OF GRADUATE STUDIES Department of Food Science We accept t h i s t hesis as conforming j to the required standard THE UNIVERSITY OF BRITISH COLUMBIA September 1991 © Benoit Girard, 1991 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 of The University of British Columbia Vancouver, Canada DE-6 (2/88) ABSTRACT The method currently established to assess the q u a l i t y of canned P a c i f i c salmon r e l i e s on sensory evaluation. Among the sensory a t t r i b u t e s of importance, odour plays a determining r o l e i n grade assignment. It would therefore be useful to obtain information about the v o l a t i l e components which can be i n d i c a t i v e of various q u a l i t y c r i t e r i a . This study was p r i m a r i l y undertaken: (1) to analyze the headspace v o l a t i l e s of canned salmon with a rapid method, (2) to apply multivariate s t a t i s t i c s on the headspace v o l a t i l e data for c l a s s i f y i n g canned salmon i n terms of species, sexual maturity, and degree of decomposition, and (3) to investigate the perceived odour of canned salmon v o l a t i l e s separated by dynamic headspace gas chromatographic methods. Sample weight, incubation temperature and time were studied to develop a s t a t i c headspace sampling method for v o l a t i l e s j i n canned salmon. A random-centroid optimization program (RCO) simultaneously searched for the optimal l e v e l s of other factors, namely, i n i t i a l oven temperature, column headpressure, and t o t a l flowrate. RCO was found to be an e f f e c t i v e optimization program which allowed the performance of several treatment runs at a time. Optimal conditions of operation permitted the detection of 80 v o l a t i l e compounds, 34 of which were i d e n t i f i e d including aldehydes, alkanes, aromatic compounds, sulfur-containing compounds, alkenes, ketones, several other compounds plus an alcohol and an ac i d . Forty-four selected headspace v o l a t i l e s from cans of 4 species of P a c i f i c salmon (chum, coho, pink, sockeye) , chum salmon at 3 stages of sexual maturity, and pink salmon of 3 q u a l i t y grades were q u a n t i t a t i v e l y determined using the s t a t i c headspace gas chromatographic (SHGC) method, and analyzed by multivariate s t a t i s t i c a l methods. P r i n c i p a l component analysis (PCA) and common factor analysis (CFA) f a c i l i t a t e d the int e r p r e t a t i o n and further handling of the c o l l e c t e d gas chromatographic data by transforming them into ten or fewer important dimensional f a c t o r s . Discriminant analyses (DA) were applied to the PCA scores for group c l a s s i f i c a t i o n . In l i g h t of non-compliance of s t a t i s t i c a l assumptions by the newly generated variables, e r r o r rates of l i n e a r , quadratic, and non-parametric functions computed by the res u b s t i t u t i o n and cro s s - v a l i d a t i o n methods were compared. The non-parametric (NPAR) Epanechnikov kernel method maintained a 90% rate or higher of effectiveness at segregating canned salmon of d i f f e r e n t species, stages of sexual maturity, and q u a l i t y l e v e l s . NPAR-DA also provided a high degree of discrimination at the i i beginning of refrigerated decomposition where the detection of spoilage by sensory grading i s uncertain. Ethanol and 3-methyl-l-butanol contributed s i g n i f i c a n t l y to c l a s s i f i c a t i o n of q u a l i t y grade of canned pink salmon. Dynamic headspace concentration by Tenax trap sampling/gas chromatography/mass spectrometry (TTS/GC/MS) and cryofocussing concentration sampling/gas chromatography/odour evaluation (CCS/GC/OE) were other means used tfo analyze v o l a t i l e components of canned pink salmon, grade A and reject, and canned late-run chum salmon. A t o t a l of 130 compounds were i d e n t i f i e d ; hydrocarbons and ketones were found i n large numbers followed by s u l f u r - c o n t a i n i n g compounds, nitrogen-containing compounds, alcohols, aldehydes, and acids. The headspace p r o f i l e of a l l analyzed samples possessed several odour a t t r i b u t e s which were associated with the chemical structures i d e n t i f i e d . No single compound was responsible for the c h a r a c t e r i s t i c aromas of canned pink salmon, grade A or reject. 2-Methyl-butanal and two lower b o i l i n g point unknowns had a hay or straw-like, cooked-malt odour t y p i c a l of canned chum salmon of spawning maturity. i v TABLE OF CONTENTS ABSTRACT i i TABLE OF CONTENTS v LIST OF TABLES v i i LIST OF FIGURES x ACKNOWLEDGEMENTS xv I. GENERAL INTRODUCTION 1 A. Quality of seafood products 5 1. Species 5 2. Diets 6 3. Spawning 7 B. Postmortem d e t e r i o r a t i v e factors 9 1. Enzymatic degradation 9 2. Microbial spoilage 10 3. Chemical spoilage 13 C. Formation of flavour v o l a t i l e s from muscle food 15 II. STATIC HEADSPACE GAS CHROMATOGRAPHIC METHOD TO ANALYZE VOLATILES IN CANNED SALMON 17 A. Introduction 17 B. Materials and methods 19 1. C o l l e c t i o n and preparation of samples 19 2. Sta t i c headspace gas chromatography (SHGC) 20 3. Examination of the GC factors 20 4. Gas chromatography-mass spectrometry (GC-MS) . . . . 21 C. Results and discussion 24 III. CANNED SALMON QUALITY CLASSIFICATION BY MULTIVARIATE ANALYSIS OF GAS CHROMATOGRAPHIC DATA. 36 A. Introduction 36 B. Materials and methods 42 1. C o l l e c t i o n and canning of salmon 42 2. Investigated treatments 42 3. Sensory assessment of raw and canned salmon 44 4. Static headspace gas chromatography (SHGC) 48 5. Static headspace gas chromatography-mass spectrometry (SHGC-MS) 49 6. Preparation of standard solutions 49 7. Data handling and s t a t i s t i c a l a n alysis 51 C. Results and discussion 53 1. MVA of v o l a t i l e s from canned salmon of d i f f e r e n t species 53 2. MVA of v o l a t i l e s from canned chum salmon at three stages of sexual maturity 80 v 3. MVA of v o l a t i l e s from canned pink salmon during r e f r i g e r a t e d decomposition 97 4. Performance of sensory evaluation and MVA 117 5. MVA of v o l a t i l e s from raw pink salmon during r e f r i g e r a t e d decomposition 127 IV. DYNAMIC HEADSPACE ANALYSIS OF VOLATILE FLAVOUR COMPONENTS IN CANNED SALMON 138 A. Introduction 138 B. Materials and methods 141 1. C o l l e c t i o n and canning of salmon 141 2. Tenax trap sampling/gas chromatography/mass spectrometry (TTS/GC/MS) 141 3. Cryofocussing concentration sampling/gas chromatography/odour evaluation (CCS/GC/OE) 142 C. Results and discussion 146 V. CONCLUSIONS 176 REFERENCES 179 i | v i LIST OF TABLES Table 1. I d e n t i f i c a t i o n of v o l a t i l e compounds detected in canned pink salmon 35 Table 2. Number of P a c i f i c salmon processed with respect to the species, stages of sexual maturity, and r e f r i g e r a t e d decomposition studies for two sampling years 43 Table 3. Grading guide for whole raw P a c i f i c salmon 46 Table 4. Grading guide for canned P a c i f i c salmon 47 Table 5. Operating conditions for the s t a t i c headspace gas chromatographic method used to analyze v o l a t i l e s i n canned P a c i f i c salmon . 50 Table 6. I d e n t i f i c a t i o n of v o l a t i l e compounds used in multivariate analysis of canned P a c i f i c salmon 55 Table 7. Correlation (loadings) of gas chromatographic peak variables from canned P a c i f i c salmon (chum, coho, pink, sockeye) with the f i r s t ten p r i n c i p a l components 57 Table 8. B a r t l e t t ' s tests f o r homogeneity of within group variance-covariance between canned P a c i f i c salmon (chum, coho, pink, sockeye) for the f i r s t ten p r i n c i p a l components 61 Table 9. Kolmogorov-Smirnov normality t e s t of canned P a c i f i c salmon (chum, coho, pink, sockeye) within p r i n c i p a l components. . 62 Table 10. Univariate and multivariate t e s t s t a t i s t i c s of discriminant analysis on the f i r s t ten p r i n c i p a l components from the v o l a t i l e s of canned P a c i f i c salmon (chum, coho, pink, sockeye) 64 Table 11. Standardized canonical variate c o e f f i c i e n t s for species discrimination of canned P a c i f i c salmon (chum, coho, pink, sockeye) 68 Table 12. C l a s s i f i c a t i o n matrix for actual and predicted group membership of canned P a c i f i c salmon (chum, coho, pink, sockeye) by l i n e a r discriminant analysis using the resubstitution method 71 Table 13. Comparison of error count estimation methods for d i f f e r e n t discriminant analysis (DA) applied to the f i r s t ten p r i n c i p a l components from canned P a c i f i c salmon (chum, coho, pink, sockeye) v o l a t i l e s 79 v i i Table 14. Correlation (loadings) of gas chromatographic peak variables from canned chum salmon tested at three sexual maturity stages with the f i r s t eight p r i n c i p a l components 82 Table 15. Univariate and m u l t i v a r i a t e test s t a t i s t i c s of discriminant analysis on the f i r s t eight p r i n c i p a l components from v o l a t i l e s of canned chum salmon tested at three sexual maturity stages 85 Table 16. Standardized canonical variate c o e f f i c i e n t s for discrimination of canned chum salmon tested at th'ree sexual maturity stages 88 Table 17. Bartlett's tests f o r homogeneity of within group variance-covariance between canned chum salmon tested at three sexual maturity stages f o r the f i r s t eight p r i n c i p a l components 91 Table 18. Kolmogorov-Smirnov normality test of canned chum salmon tested at three sexual maturity stages within the f i r s t eight p r i n c i p a l components 93 Table 19. Comparison of err o r count estimation methods for d i f f e r e n t discriminant analysis (DA) functions applied to the f i r s t eight p r i n c i p a l components from canned chum salmon tested at three sexual maturity stages . . . . 95 Table 20. Concentration ranges of three v o l a t i l e compounds from canned pink salmon of d i f f e r e n t q u a l i t y grades 100 Table 21. Loadings of the f i r s t nine varimax rotated factors from factor analysis of canned pink salmon v o l a t i l e s of the refrigerated storage study 101 Table 22. Bartlett's tests f o r homogeneity of within group variance-covariance between q u a l i t y grades of canned pink salmon for selected gas chromatographic peaks variables and the f i r s t ten p r i n c i p a l components 110 Table 23. Kolmogorov-Smirnov normality test of quality grades of canned pink salmon f or selected gas chromatographic peaks and the f i r s t ten p r i n c i p a l components I l l Table 24. Univariate and mul t i v a r i a t e test s t a t i s t i c s of l i n e a r discriminant analysis on the f i r s t ten varimax rotated p r i n c i p a l components from canned pink salmon during the refrigerated storage study 113 v i i i Table 25. Cross-validated e r r o r count estimates of l i n e a r discriminant functions c a r r i e d out on selected gas chromatographic peaks from canned pink salmon of the refrigerated storage study 115 Table 26. Comparison of cross-validated error count estimates for different discriminant analysis functions (DA) of selected gas chromatographic peaks and the f i r s t ten p r i n c i p a l components from canned pink salmon of the r e f r i g e r a t e d storage study 116 Table 27. Comparison of erro r rates for non-parametric discriminant functions (NPAR-DA) and sensory grading of canned pink salmon of the r e f r i g e r a t e d storage study 120 Table 28. Loadings of the f i r s t four varimax rotated p r i n c i p a l components of v o l a t i l e compounds from raw pink salmon of the r e f r i g e r a t e d storage study (year 2) 130 Table 29. C l a s s i f i c a t i o n by non-parametric discriminant analysis functions (NPAR-DA) of selected peaks (3 and 18) from the gas chromatographic analysis of raw pink salmon of the refrigerated storage study 135 Table 30. V o l a t i l e compounds t e n t a t i v e l y i d e n t i f i e d i n canned pink salmon of grades A and reject, and canned l a t e run chum, salmon by Tenax trap sampling/gas chromatography/mass spectrometry (TTS/GC/MS) 150 Table 31. Cryofocussing concentration sampling/gas chromatography/ odour evaluation (CCS/GC/OE) of v o l a t i l e components from canned pink salmon of good q u a l i t y (grade A) 161 Table 32. Cryofocussing concentration sampling/gas chromatography/ odour evaluation (CCS/GC/OE) of v o l a t i l e components from canned pink salmon of advanced decomposition (grade reject) 162 Table 33. Cryofocussing concentration sampling/gas chromatography/ odour evaluation (CCS/GC/OE) of v o l a t i l e components from canned chum salmon of advanced sexual maturity (spawning dark) 163 ix LIST OF FIGURES Figure 1. Flowchart of the random centroid optimization program . . 22 Figure 2. Effect of f i s h weight on-total chromatographic area (temperature of v i a l incubation, 75°C; time of v i a l incubation, 1 h; column headpressure, 60 kPa; s p l i t ratio, 50:1) 25 Figure 3. Effect of incubation temperature on t o t a l chromatographic area (time of v i a l incubation, 1 h; f i s h weight, lOg; column head-pressure, 60 kPa; s p l i t r a t i o , 50:1) 26 Figure 4. Effect of incubation time on t o t a l chromatographic area (temperature of v i a l incubation, 75°C; f i s h weight, lOg; column headpressure, 60 kPa; s p l i t r a t i o , 50:1) 27 Figure 5. Optimization mapping results of peak separation as a function of i n i t i a l oven temperature (temperature of v i a l incubation, 105°C; time of v i a l incubation, 1 h; f i s h weight, lOg) 29 Figure 6. Optimization mapping results of peak separation as a function of column headpressure (temperature of v i a l incubation, 105°C; time of v i a l incubation, 1 h; f i s h weight, lOg) 30 Figure 7. Optimization mapping results of peak separation as a function of t o t a l flowrate (tempearture of v i a l incubation, 105°C; time of v i a l incubation, 1 h; f i s h weight, lOg) 31 Figure 8. Chromatogram of v o l a t i l e s from canned pink salmon (temperature of v i a l incubation, 105°C; time of v i a l incubation, 1 h; f i s h weight, lOg; i n i t i a l oven temperature, 35°C; column headpressure, 95 kPa; t o t a l flowrate, 44 mL/min) 33 Figure 9. Chromatogram of v o l a t i l e s from canned pink salmon selected to carry out multivariate s t a t i s t i c a l analyses 54 Figure 10. Plot of the f i r s t two p r i n c i p a l component scores for the salmon species (A, pink-year 1; B, coho-year 1; C, chum-year 1; D, sockeye-year 1; E, pink-year 2; F, sockeye-year 2; G, coho-year 2; H, chum-year 2) 59 Figure 11. Estimated density d i s t r i b u t i o n s of canned P a c i f i c salmon (chum, coho, pink, sockeye) for p r i n c i p a l component 3 based on l i n e a r discriminant analysis 66 x Figure 12. Canonical representation of centroid means and dispersions for canned P a c i f i c salmon (C, coho; K, chum; P, pink; S, sockeye) 69 Figure 13. Estimated density d i s t r i b u t i o n s of canned P a c i f i c salmon (chum, coho, pink, sockeye) f o r p r i n c i p a l component 3 based on quadratic discriminant analysis 72 Figure 14. Comparison of two error count estimation methods against the smoothing parameter of the Epanechnikov kernel c l a s s i f i e r on p r i n c i p a l component scores of canned P a c i f i c salmon (chum, coho, pink, sockeye) v o l a t i l e s . . 75 Figure 15. Estimated density d i s t r i b u t i o n s of canned P a c i f i c salmon (chum, coho, pink, sockeye) f o r p r i n c i p a l component 3 based on non-parametric discriminant analysis 77 Figure 16. Plot of the f i r s t two p r i n c i p a l component scores f o r the canned chum salmon tested at three sexual maturity stages (A, s i l v e r - b r i g h t / y e a r 1; B, semi-bright/year 1; C, dark/year 1; D, s i l v e r - b r i g h t / y e a r 2; E, semi-bright/year 2; F, dark/year 2) 84 Figure 17. Plot of the two canonical v a r i a t e scores for canned chum salmon tested at three sexual maturity stages (S, s i l v e r bright; B, semi-bright; D, commercial dark) 8 9 Figure 18. Comparison of error count estimate method against the smoothing parameter of the Epanechnikov kernel c l a s s i f i e r on p r i n c i p a l component scores of canned chum salmon tested at three sexual maturity stages 94 Figure 19. Chromatogram of v o l a t i l e s from canned pink salmon during the r e f r i g e r a t e d storage of year 1. (A, day 0/grade A; B, day 8/grade B; C, day 13/grade reject) 98 Figure 20. Plots of v o l a t i l e s from canned pink salmon of the refrigerated storage study with high loadings f o r factor 3 102 Figure 21. Plots of v o l a t i l e s from canned pink salmon of the refrigerated storage study with high loadings f o r factor 4 104 Figure 22. Projection of gas chromatographic peak variable loadings on p r i n c i p a l components 4 and 5 for canned pink salmon of the re f r i g e r a t e d storage study 105 Figure 23. Plot of the scores of p r i n c i p a l component 4 over refrigerated storage time f o r canned pink salmon . . . . 106 x i Figure 24. Plot of the scores of p r i n c i p a l component 5 over refrigerated storage time for canned pink salmon . . . . 107 Figure 25. Plot of the scores of p r i n c i p a l component 5 against 4 for canned pink salmon of the r e f r i g e r a t e d storage study 109 Figure 26. Sensory c l a s s i f i c a t i o n before and a f t e r canning pink salmon of the r e f r i g e r a t e d storage study 118 Figure 27. Logarithmic r e l a t i o n s h i p s of sensory r a t i n g of canned salmon with r e f r i g e r a t e d storage time. D i g i t s represent the number of times a r a t i n g was encountered f o r each day of r e f r i g e r a t e d storage 123 Figure 28. Linearized r e l a t i o n s h i p of the polynomial q u a l i t y function using p r i n c i p a l components 4 (PC4) and 5 (PC5) over r e f r i g e r a t e d storage time 125 Figure 29. Linearized r e l a t i o n s h i p of the polynomial q u a l i t y function using peaks 3, 7, and 18 over r e f r i g e r a t e d storage time 12 6 Figure 30. Chromatograms of v o l a t i l e s from raw pink salmon during the r e f r i g e r a t e d storage of year 2. (A, day 0/grade A, day 10/grade B, day 21/grade reject) 128 Figure 31. Plot of the scores of p r i n c i p a l component 1 against 2 for raw pink salmon of the r e f r i g e r a t e d storage study of year 2 (A, grade A; B, grade B; R, grade r e j e c t ) . . . 132 Figure 32. Plots of v o l a t i l e s from raw pink salmon of the refrigerated storage study of year 2 with high loadings for p r i n c i p a l component 2 134 Figure 33. Can piercer f i x t u r e , valve, and s t a i n l e s s s t e e l tubing assembled to concentrate canned salmon v o l a t i l e s by cryofocussing (CCS) 144 Figure 34. Schematic representation of the v o l a t i l e desorption steps and the oven temperature program for cryofocussing concentration sampling/gas chromatography/odour evaluation (CCS/GC/OE) 145 Figure 35. Total ion chromatogram obtained by gas chromatography/ mass spectrometry (GC/MS) of headspace v o l a t i l e components from canned pink salmon of good q u a l i t y (grade A) concentrated using Tenax trap sampling (TTS). Compounds are i d e n t i f i e d by peak-numbers shown i n Table 30 147 x i i Figure 36. Total ion chromatogram obtained by gas chromatography/ mass spectrometry (GC-MS) of headspace v o l a t i l e components from canned pink salmon of advanced decomposition (grade reject) concentrated using Tenax trap sampling (TTS). Compounds are i d e n t i f i e d by peak numbers shown i n Table 30 148 Figure 37. Total ion chromatogram obtained by gas chromatography/ mass spectrometry (GC-MS) of headspace v o l a t i l e components from canned chum salmon of advanced sexual maturity (spawning dark) concentrated using Tenax trap sampling (TTS). Compounds are i d e n t i f i e d by peak numbers shown i n Table 30 149 Figure 38. Chromatogram obtained by gas chromatography/flame io n i z a t i o n detection (GC/FID) of headspace v o l a t i l e components from canned pink salmon of good q u a l i t y (grade A) concentrated using cryofocussing. Compounds are i d e n t i f i e d by peak numbers shown i n Table 30 and l e t t e r s refer to Table 31 158 Figure 39. Chromatogram obtained by gas chromatography/flame io n i z a t i o n detection (GC/FID) of headspace v o l a t i l e components from canned pink salmon of advanced decomposition (grade reject) concentrated using cryofocussing. Compounds are i d e n t i f i e d by peak numbers shown in Table 30 and l e t t e r s r e f e r to Table 32 159 Figure 40. Chromatogram obtained by gas chromatography/flame io n i z a t i o n detection (GC-FID) of headspace v o l a t i l e components from canned chum salmon of advanced sexual maturity (spawning dark) concentrated using cryofocussing. Compounds are i d e n t i f i e d by peak numbers shown i n Table 30 and l e t t e r s r e f e r to Table 33 160 x i i i ACKNOWLEDGEMENTS I wish to express my appreciation and gratitude to Dr. Shuryo Nakai for his support, council, and invaluable assistance throughout the course of t h i s research project. I would also l i k e to acknowledge and thank the other members of my supervisory committee, Dr. Brent J. Skura, Dr. Timothy D. Durance, Dr. William D. Powrie, and Dr. George K. Iwama for t h e i r time, e f f o r t s , and c r i t i c a l c o n t r i b u t i o n . The f i n a n c i a l assistance to t h i s project by the Science Council of B r i t i s h Columbia through the Graduate Research Engineering and Technology (GREAT) Award i s g r a t e f u l l y acknowledged. Furthermore, my appreciation i s extended to the F i s h e r i e s Council of B r i t i s h Columbia f o r t h e i r cooperation and collaboration, and B.C. Packers Ltd. who allowed me to use t h e i r f a c i l i t i e s . F i n a l l y , I would l i k e to thank my families, G i r a r d and Weintraub, for t h e i r continued support and understanding during the past several years. I want to acknowledge the fellow students and s t a f f i n the Department of Food Science whose friendship and enthusiasm provided rewarding and enjoyable times. Most of a l l , a s p e c i a l thought i s d i r e c t e d to Sara Weintraub for her encouragement, i n s p i r a t i o n , motivation, patience, and love. x i v I. GENERAL INTRODUCTION 1 Canada's f i s h i n g grounds y i e l d around 70 marketable species of f i s h and seafood. In economics, f i s h are second only to grain, the country's most valuable food export. In 1987, the Canadian commercial f i s h i n g industry harvested 1.5 m i l l i o n tonnes with an estimated value of $1.64 b i l l i o n . This landed value was transformed into 839,800 tonnes of seafood products, worth $3.3 b i l l i o n s (Department of Fi s h e r i e s and Oceans, 1988). Approximately 80% of the value of the Canadian f i s h e r y production i s currently exported. Salmon and herring are the main export products from the P a c i f i c coast. Commercial f i s h i n g constitutes a major factor i n the economies of the coastal provinces and northern communities, although i t contributes only a small portion of our Gross National Product (0.5%). In B r i t i s h Columbia, commercial f i s h i n g ranks fourth among the primary industries, and f i s h processing accounts f o r over 25% of a l l food manufacturing a c t i v i t i e s . More than h a l f of the $1 b i l l i o n wholesale value of BC f i s h production came from ocean-caught salmon i n 1988; 40% of which was sold i n the form of canned salmon. Pink salmon accounted for 62%, chum 18%, sockeye 17%, and coho 2% of the number of 48-lb cases of cans (Aquaculture and Commercial Fi s h e r i e s Branch, 1989) . Along with i t s economical s i g n i f i c a n c e , f i s h occupies an important place as a part of the human d i e t . In order to maintain a high l e v e l of quality, i t i s e s s e n t i a l to ensure that f i s h products going to market are 2 wholesome, meet c e r t a i n commercial requirements, and are a e s t h e t i c a l l y acceptable to the consumer. Good manufacturing practices, q u a l i t y assurance and inspection are parts of the strategy. The method c u r r e n t l y established to evaluate the q u a l i t y of fresh and canned salmon i s l a r g e l y based on sensory evaluation. Among the sensory attributes of texture, colour, taste, and odour, the l a t t e r two play important roles i n the quality assurance of f i s h products. Rejections of canned salmon, that may be classed as poor q u a l i t y , are mostly due to the presence of off-odours i n d i c a t i v e of spoilage. Regulatory agencies presently r e l y on t r a i n e d assessors to ensure a minimum l e v e l of q u a l i t y for canned salmon. The procedure i s considered quick; one e s s e n t i a l c r i t e r i o n of a method needed to judge q u a l i t y of numerous lots Ojf canned products during busy f i s h i n g seasons. Although a very useful t o o l , sensory flavour tests possess some l i m i t a t i o n s . Sensory tasters are used as measuring instruments, somewhat variable over time and among themselves, and prone to bias due to p h y s i o l o g i c a l and psychological f a c t o r s . Extensive t r a i n i n g aimed at removing possible b u i l t - i n biases and preferences, developing the necessary s k i l l s , and f a m i l i a r i z i n g the future inspector with the procedure, are required to reach uniform and consistent grading. Attention should be paid to the grading of marginal l o t s where the p o t e n t i a l for m i s c l a s s i f i c a t i o n could be higher. Errors i n the assignment of grade and release of questionable products may i n i t i a t e strong consumer complaints, r e s u l t in substantial monetary loss by a p r i v a t e company, and lead to plant closure with 3 consequent loss of employment. The 1985 incident of Starkist canned tuna, where "tainted" products (oxidized but safe for consumption) were released for r e t a i l and required a co s t l y r e c a l l , i s a good reminder. The search for chemical compounds useful to indicate spoilage has been the subject of many inve s t i g a t i o n s . In the past, research has mainly focussed on (a) amines such as t o t a l v o l a t i l e bases (Botta et a l . 1984), trimethylamine/dimethylamine (Hebard et a l . , 1982), ammonia (Leblanc and G i l l , 1984), and biogenic amines (Mietz and Karmas, 1978), (b) nucleotides such as hypoxanthine (Jahns et a l . , 1976), and the K-value (Ehira and Uchiyama, 1987), and (c) ethanol (Hollingworth et a l . , 1986). Although some methods perform better than others, none were found to accurately r e f l e c t the quality of a l l f i s h products. This can be att r i b u t e d to the d i v e r s i t y with regard to composition, mode of spoilage, and processing of d i f f e r e n t products. Despite constant improvements of the chemical methods, only sensory evaluation presently meets the p r a c t i c a l requirements, and therefore constitutes the major routine procedures. Fish i s one of our most perishable protein resources. When i t sp o i l s , unpleasant odours and flavours are released. Seafood decomposition i s a complex process in v o l v i n g a u t o l y t i c degradation, chemical oxidation, and b a c t e r i a l a c t i v i t y . Since the decomposition process could follow any combination of pathways, measurement of several indices may be e s s e n t i a l . Most breakdown products are smaller than t h e i r precursors, have higher vapour pressures, and are more l i k e l y to contribute to the t o t a l flavour of the product. An instrumental method that measures a number of v o l a t i l e compounds simultaneously and correlates well with sensory analysis would be highly desirable. It should also be p a r t i c u l a r l y e f f e c t i v e at the e a r l y stages of decomposition, where c o n t r o v e r s i a l situations are more l i k e l y to a r i s e . Gas chromatographs are contemporary instruments extensively used for v o l a t i l e analysis. A method applied to canned salmon could be developed with such an instrument. M u l t i v a r i a t e s t a t i s t i c s are foreseen as necessary to interpret the generated complex patterns of peaks from canned salmon of various q u a l i t y and re l a t e them to concurrent sensory assessments. Furthermore, i t would be of interest to i d e n t i f y the various v o l a t i l e compounds encountered i n canned salmon and t h i s could be accomplished by gas chromatography-mass spectrometry (GC-MS). 5 A . Q u a l i t y of seafood produc ts In r e l a t i o n to f i s h e r y products, the term 'quality' sometimes leads to confusion as i t s meaning can be defined i n many ways. Very often, i t i s synonymous with freshness, or degree of spoilage. However, species of salmon and stage of sexual maturity a f f e c t the appearance and other sensory c h a r a c t e r i s t i c s perceived by the consumer, and a f f e c t acceptance as well as market value. There are therefore a number of factors encompassing q u a l i t y i n f i s h e r y products and the extent to which they affect that quality v a r i e s . The remainder of t h i s introduction takes into consideration some background information covering the e f f e c t s of various important q u a l i t y factors, on odour i n p a r t i c u l a r . 1. Species Differences in the genetic material between species of salmon may be the cause of major influences on q u a l i t y . For example, pink salmon (Oncorhynchus qorbusha) possesses a more d e l i c a t e f l e s h than other species. Sockeye salmon (0^ nerka) i s noted for i t s d i s t i n c t i v e red colour not matched by pink or chum (0_;_ keta) . I n t r i n s i c variations i n composition of protein, f a t , pigment, low molecular weight compounds, etc., result in the differences seen i n colour, microstructure and texture, and flavour (Wheaton and Lawson, 1985) . 6 2. Diets What a f i s h eats has a s i g n i f i c a n t e f f e c t on the colour and flavour of i t s f l e s h . Fish, phytoplankton, zooplankton, invertebrates and other organisms constitute the main diet of wild salmon. The carotenoid pigments responsible for the orange coloration of salmon f l e s h are derived from i t s food and consist primarily of astaxanthin (Simpson, 1982) . For aquaculture salmon, di e t s that are d e f i c i e n t i n these carotenoids result i n f l e s h lacking orange coloration (Chen et. a l . , 1984). High degree of pigmentation i s associated with consumer acceptance and often with r i c h e r flavour (Skrede and Storebakken, 1986). A common flavour defect described as blackberry, weedy, d i e s e l , or s u l f i d e flavour has been known to develop i n cod and other gadoids, mackerel, and salmon (e.g., chum). This has been at t r i b u t e d to the presence of dimethylsulfide, or DMS (Connell, 1980) . It generally occurs i n f i s h feeding on a planktonic molluscs known as pteropods. The pteropods contain dimethyl-B-propiothetin, which may be converted to DMS i n the f i s h . A low concentration of t h i s compound may not be detected i n fresh f i l l e t s on i c e . The odour becomes more pronounced i n f l e s h such as salmon that i s canned. The location of the f i s h i n g ground plays an i n d i r e c t role in the q u a l i t y of the f i s h e r y product (Jones, 1969) . Flavour can vary from one f i s h i n g ground to the next, and can also vary from one season to the next. Winds, tides, water conditions, and migratory patterns influence the q u a l i t y of a f i s h before harvest. These factors have repercussions on the type and abundance of food organisms a v a i l a b l e , which could a f f e c t the ph y s i o l o g i c a l condition of the f i s h . 3. Spawning Salmon are anadromous f i s h that begin l i f e i n fresh water, migrate to the sea for a period of feeding and development, and return to freshwater to spawn. They do not take food during the spawning migration; feeding may cease while s t i l l i n the ocean> or at the time they enter the brackish water at the mouth of the r i v e r s (Childerhose and Trim, 1981). Their muscles contain minute amounts of carbohydrate and consist mainly of protein and f a t . During the maturation process, the energy and nutrients needed for physical a c t i v i t y and for gonadal growth are taken from the viscera and the f l e s h . As the depletion of fat and protein i n muscle occurs, the water content increases, probably mostly i n the i n t e r c e l l u l a r space (Greene, 1926; Love, 1980; Aksnes et a l . , 1986). Although the fat content decreases, the f a t t y a c i d composition i s not strongly affected. However, the increase i n the hydroxyproline i n protein indicates that the metabolism of protein from the muscle during maturation takes place at the expense of non-collagenous and c e l l u l a r proteins. These occurrences have been documented by Lovern (1934),, Kaneko et a l . (1966), Love (1970), and Aksnes et a l . (1986). After "Spawning, some muscle fibres have been found to be i n the process of d i s i n t e g r a t i o n . 8 S t r i a t i o n disappeared and the myoplasm eventually presented the appearance of a structureless but swollen c o l l o i d (Greene, 1926) . The sensory a t t r i b u t e s of the cooked product are a f f e c t e d by the maturation process. The f i r s t evident signs of maturation are usually the change i n the skin pigmentation from s i l v e r y to various shades of brown, green, or red depending on the species. The colour of the f l e s h i s also affected as red colours fade due to pigment m o b i l i z a t i o n . The depletion of f a t and protein correlates with the gradual disappearance of odour and flavour of the cooked product. During maturation of A t l a n t i c salmon, the odour of cooked samples becomes less pronounced and i s often c l a s s i f i e d as neutral. The o r i g i n a l flavour gradually changes and the maturing salmon i s eventually perceived as tasteless (Aksnes et a l . , 1986). In the case of chum salmon, jthe so-called "late-run odours" appear. The texture of immature f i s h i s evaluated as filamentous and f i r m while mature f i s h have a more watery, d i s t i n c t i v e l y soft and tough consistency. It i s possibly explained by the higher water content and the higher r e l a t i v e amount of collagenous ti s s u e (Aksnes et a l . , 1986) . In extreme cases, the depletion of p r o t e i n results in a gelatinous state which renders the f l e s h useless. 9 B. Postmortem d e t e r i o r a t i v e factors Deterioration of fresh f i s h begins as soon as the animal d i e s . It i s a complex s i t u a t i o n f o r which no sing l e factor i s responsible, but, rather, i s a combination of several i n t e r r e l a t e d processes. Eventually undesirable odours and o f f - f l a v o u r s develop as well as softening of the flesh and loss of c e l l u l a r f l u i d containing various nutrients. There are three basic modes of de t e r i o r a t i o n i n f i s h : microbial, enzymatic, and chemical. 1. Enzymatic degradation When captured or harvested before s t a r t i n g the spawning migration, salmon usually contain food i n t h e i r gut as well as digestive enzymes and bacteria. Enzymes present i n the f l e s h and stomach remain active a f t e r death of the f i s h , and are p a r t i c u l a r l y involved i n flavour changes that take place during the f i r s t few days of storage before b a c t e r i a l spoilage becomes s i g n i f i c a n t . In the postmortem muscle, one of the f i r s t metabolites to appear i s l a c t i c acid. Lactic a c i d accumulates because of the g l y c o l y t i c conversion of storage glycogen i n the f i s h muscle a f t e r the cessation of r e s p i r a t i o n . Lactic acid build-up can cause a drop i n pH, r e s u l t i n g i n the l i b e r a t i o n and activation of inherent acid c e l l -proteases, cathepsins (Connell, 1980). These proteases contribute to the weakening and softening of the flesh. Hemoglobin, p a r t i c u l a r l y concentrated i n the kidney, can be released and may migrate toward the body c a v i t y to cause red d i s c o l o r a t i o n c a l l e d belly-burn (Department of F i s h e r i e s and Oceans, 1989). Spoilage products released from the i n t e s t i n e may d i f f u s e throughout the f l e s h of the f i s h , producing offensive odours and d i s c o l o r a t i o n (Wheaton and Lawson, 1985) . In the muscle of l i v e animals, ATP predominates among the nucleotides under normal conditions. More than 90% of the nucleotides i n the muscle of f i s h and s h e l l f i s h are accounted for by purine derivatives (Seki, 1971) . From the food chemistry point of view, some nucleotide degradation products serve as important umami-producing fa c t o r s . Five prime nucleotides such as inosine 5'-monophosphate (IMP) and guanosine 5'-monophosphate (GMP) show a d i s t i n c t taste-enhancing e f f e c t i n combination with glutamic acid. However, during the process of nucleotide degradation, ammonia i s produced and may contribute to off-jflavours (Gunnar, 1982) . 2. M i c r o b i a l spoilage Aided by enzymatic a c t i v i t y , microbial spoilage i s by far the main mode of spoilage of c h i l l e d f i s h and s h e l l f i s h . Large numbers of bacteria are normally present i n the surface slime, on the g i l l s , and i n the i n t e s t i n e s of the l i v e f i s h . However, b a c t e r i a l spoilage does not begin u n t i l the passage of r i g o r mortis when c e l l u l a r f l u i d s are released from the muscle f i b r e s . Rate of microbial growth depends on the numbers and types of microorganisms present and the temperature at which the f i s h i s held. The b a c t e r i a l f l o r a of f i s h i s influenced by a number of factors such as season and environment. The f l o r a i s a r e f l e c t i o n of the f l o r a i n the water in which the f i s h i s caught. The surface f l o r a may also r e f l e c t post-mortem handling and contamination on board a vessel, during unloading, or i n the f i s h plant. While the f i s h are c h i l l e d , the psychrotrophic Pseudomonas species predominates, followed by Achromobacter and Flavobacterium species. At higher temperatures, the genera Micrococcus and B a c i l l u s appear to increase i n numbers (Frazier and Westhoff, 1978) . The primary substrate f o r b a c t e r i a l growth and the main source of spoilage products i s the soluble material i n the muscle. Since free carbohydrate i s very low i n f i s h , Pseudomonas r a p i d l y u t i l i z e the nonprotein nitrogen (NPN) f r a c t i o n of muscle, and t h i s i s a major reason for t h e i r rapid domination of the microflora during spoilage. The major components of NPN i n f i s h muscle are peptides, amino acids, nucleotides, trimethylamine oxide, urea, taurine, and related purine-based compounds. The free amino acids of A t l a n t i c salmon (Salmo salar) include moderate amounts of taurine, glutamic acid, glycine, and alanine (Cowey et a l . , 1962). According to Krzynowek and Murphy (1987) , a l l species of P a c i f i c salmon have a s i m i l a r amino a c i d composition and contain r e l a t i v e l y high amounts of leucine (1 .6g/100g), lysine (1.8g/100g), as p a r t i c a c i d (2.0g/100g), and glutamic a c i d ( 3 . 0 g / 1 0 0 g ) . With t h e i r decarboxylation capacity, bacteria can convert some of these amino acids to v o l a t i l e bases such as the diamines, cadaverine and putrescine, as well as histamine. Histamine, a very potent c a p i l l a r y d i l a t o r i n man, i s the major cause of scombroid food poisoning (Eitenmiller et a l . , 1982). In some species of the gadoid family, trimethylamine oxide (TMAO) i s found in r e l a t i v e l y high amounts. The odourless compounds TMAO and l a c t i c acid can be metabolized by b a c t e r i a l action to y i e l d trimethylamine (TMA) and acetic a c i d (Hebard et a l . , 1982). TMA i s characterized by an ammoniacal odour, but i n combination with other compounds, such as f a t , may give a " f i s h y " odour. Also, ammonia i s u l t i m a t e l y produced from the breakdown of these nitrogen-containing compounds. Some pro t e o l y s i s seems to occur in the e a r l y stages of spoilage, but there i s evidence that protease production by bacteria i s i n i t i a l l y repressed. This appears to be due to the presence of free amino acids at a high l e v e l . Proteolysis becomes more vigorous i n the l a t e r stages of spoilage as the amino acids are u t i l i z e d (Chung, 1968) . Various products can be formed d i r e c t l y by enzymatic deamination of amino acids. The end-products are categorized as alcohols, ketones, f a t t y acids, aldehydes, s u l f i d e s , t h i o l s / mercaptans, etc. These products, i n d i c a t i v e of putrefaction, increase i n concentration as the f i s h become unacceptable. A c o r r e l a t i o n has been made between b a c t e r i a l counts and the occurrence of v o l a t i l e s u l f u r compound producing b a c t e r i a . When the s u l f i d e producers reach 40% of the 6 2 population (often corresponding to a b a c t e r i a l count exceeding 10 /cm ) , overt spoilage occurs (Liston, 1982). Microorganisms may also i n i t i a t e the hydrolysis of t r i a c y l g l y c e r o l s . The resulting free f a t t y acids are p a r t i c u l a r l y susceptible to oxidation which proceeds v i a hydroperoxides. Most microorganisms can degrade peroxides and the formation of secondary fat oxidation products by microorganisms i s very complex (Alford et a l . , 1971) . This leads to numerous stable end products, such as saturated and unsaturated aldehydes, ketones, dicarbonyl compounds, alcohols, alkanes, alkenes, and methyl furans (Grosch, 1984). The carbonyl compounds are odorous even i n trace concentrations and cause the rancid aroma of oxidized f a t . 3 . Chemical s p o i l a g e As with many other seafoods, monounsaturated and polyunsaturated fa t t y acids are more abundant i n P a c i f i c salmon than saturated f a t t y acids (Hearn et a l . , 1987) . Among the wide array of f a t t y acids, p a l m i t i c (16 :0) , o l e i c ( 1 8 : l n 9 ) , and docosahexaenoic (22:6n3) acids dominate i n fresh wild P a c i f i c salmon and remain dominant a f t e r cooking or canning (Ackman and McLeod, 1988; Barber et a l . , 1987; Barber et a l . , 1988; Braddock and Dugan, 1972). Fish normally have a much higher degree of l i p i d unsaturation than most other foods and, therefore, are p a r t i c u l a r l y prone to oxidative d e t e r i o r a t i o n and r a n c i d i t y . This i s the case f o r salmon, mackerel, tuna, and herring. Microbial enzymes may be involved i n the oxidation of fats and o i l s , but autoxidation i s more common during frozen storage. Freezing temperature, oxygen concentration, r e l a t i v e humidity, and c a t a l y s t concentration are important f a c t o r s i n l i p i d oxidation (Powrie, 1984) . These factors determine the rate at which oxygen reaches the muscle surface and ultimately reacts with the l i p i d s . However, endogenous enzyme systems, metallo- and hemoproteins, and metal ions i n the tiss u e play a role i n oxidation by acting as pro-oxidants. Hardy et a l . (1979) found that l i p o l y s i s was responsible f o r the major changes i n the l i p i d components during the frozen storage of cod. Oxidation was slow and occured p r i m a r i l y i n the phospholipid f r a c t i o n . In the i s o l a t e d microsomal membrane f r a c t i o n from f i s h s k e l e t a l muscle, enzymic oxidation of l i p i d components occured at temperatures as low as -20°C (Apgar and Hultin, 1982) . Lipases and phospholipases release free f a t t y acids from the l i p i d s and these free f a t t y acids may then undergo oxidation producing lower molecular weight compounds. Pretreatment with microwave heating i s another factor found to reduce the storage l i f e of f i s h by cat a l y z i n g the development of oxidative r a n c i d i t y (Ke et a l . , 1978). L i p i d oxidation r e s u l t s p r i n c i p a l l y i n undesirable flavours often described as "painty". In addition, there may be some e f f e c t s on the n u t r i t i v e components of the f i s h t i s s u e due to degradation of oxidizable nutrients such as e s s e n t i a l f a t t y acids, some amino acids, and some vitamins. Oxidation of heme pigments can y i e l d d i s c o l o r a t i o n , and texture changes may occur due to protein c r o s s - l i n k i n g (Nawar, 1985) . 15 C. Formation of flavour v o l a t i l e s from muscle food Fresh raw meat and f i s h have l i t t l e odour although i t i s usually possible to d i s t i n g u i s h between animal species by s n i f f i n g . Recent investigations using various chromatographic and biochemical techniques have i d e n t i f i e d v o l a t i l e components of several, f i s h species. Raw freshwater and marine f i s h possess mild flavours that r e s u l t from the conversions of polyunsaturated f a t t y acids by lipoxygenases (Josephson et a l . , 1984a). Josephson (1987) found that, during each of the stages of l i f e cycle, P a c i f i c salmon (Oncorhvnchus spp.) possessed 8-carbon alcohols and carbonyls which contributed d i s t i n c t p l a n t - l i k e aromas to the f i s h . Spawning condition salmon i n freshwater environments a d d i t i o n a l l y possessed a group of 9-carbon alcohols and carbonyls that added cucumber-or melon-like aroma notes. During storage, the i n i t i a l changes in the aroma of freshly caught f i s h usually involve a s h i f t from these fresh, planty aroma notes to a neutral, or flat-sweet odour and eventually detectable amounts of malodorous spoilage compounds are produced. In general, meat and f i s h must be cooked i n some fashion i n order to develop desirable odour and flavour. During cooking or canning of salmon, flavour compounds are formed which replace the fresh f i s h compounds, to y i e l d the c h a r a c t e r i s t i c flavour of cooked salmon. Although l i t t l e information has been published about the flavours of cooked seafoods, v o l a t i l e carbonyls have been suggested as s i g n i f i c a n t contributors to cooked f i s h flavours (Pokorny, 1980) . In general, meat and f i s h aroma i s not the r e s u l t of one constituent but the sum of the sensory e f f e c t s of a multitude of v o l a t i l e s a r i s i n g from the thermal process. However, cer t a i n groups of compounds appear to be important contributors to meat flavours. Chang and Peterson (1977) suggested that lactones, furanoid compounds, a c y c l i c s u l f u r - c o n t a i n i n g compounds and heterocyclic compounds containing S, N, and O may have a large flavour impact. It i s recognized that the low molecular weight f r a c t i o n a r i s i n g from the degradation of sugars, proteins, and l i p i d s are responsible f o r the development of meat and f i s h flavours upon heating. The various reactions occurring i n the process depend on the type and concentration of the non-volatile precursors, the heating temperature, the pH of the medium, and the water a c t i v i t y of the meat. 17 I I . STATIC HEADSPACE GAS CHROMATOGRAPHIC METHOD TO ANALYZE VOLATILES IN CANNED SALMON A. Introduction Food v o l a t i l e s , usually found i n small amounts, constitute an in t e g r a l part of the flavours perceived by consumers. Gas chromatography (GC) i s a widely employed a n a l y t i c a l technique to analyze v o l a t i l e compounds, and a number of methods have been developed f o r sample preparation. One way of analyzing the v o l a t i l e constituents i s by d i r e c t i n j e c t i o n of the sample i t s e l f or of the sample headspace gases. More often, food v o l a t i l e s must be i s o l a t e d and concentrated i n some manner before GC analysis. Samples can be subjected to extraction, adsorption, d i s t i l l a t i o n , cryogenic procedures, or a combination of the above. A description of these methods as well as t h e i r advantages and disadvantages were recently reviewed by Heath and Reineccius (1986) and Parliment (1986) . Equilibrium or s t a t i c headspace analysis (SHA) involves the chromatographic separation of a predetermined volume of vapour headspace above a sample held i n a closed v i a l . V o l a t i l e s of many food commodities including macadamia nuts (Crain and Tang, 1975), cooked Brassicaceous vegetables (Maruyama, 1970), c i t r u s juice (Davis and Chace, 1969), dehydrated potatoes (Sapers et a l . , 1970), beer (Hoff and Herwig, 1976), peanuts (Young and Hovis, 1990), wine (Noble et a l . , 1980), and spices (Pesek et a l . , 1985) have been analyzed by t h i s technique. This method has also been applied to the determination of dimethyl sulphide i n cod 18 (Sipos and Ackman, 1964), amines in hake, sole, cod, rockfish, perch and lingcod (Miller et a l . , 1972a), v o l a t i l e halocarbons i n e e l , carp, s t r i p e d bass, and spot f i s h (Entz and H o l l i f i e l d , 1982) , and ethanol i n canned salmon (Hollingworth et a l . , 1986). SHA has po t e n t i a l to be used for q u a l i t y control purposes due to i t s simple and rapid operation. However, detection i s l i m i t e d to abundant compounds of low b o i l i n g point. One way to improve s e n s i t i v i t y i s to raise the temperature of the sample. Furthermore, the i n j e c t i o n procedure can be automated by a headspace sampler where the amount of headspace volume taken i s constant, e l i m i n a t i n g errors associated with manual handling. The main objective of t h i s research was to develop a method based on s t a t i c headspace gas chromatography (SHGC) that would simultaneously measure a large number of v o l a t i l e compounds from canned salmon i n a short period of time. More s p e c i f i c a l l y , t h i s i n v e s t i g a t i o n was aimed at: 1) studying some sample preparation and incubation r e l a t e d -factors which influenced chromatographic s e n s i t i v i t y ; 2) optimizing factors involved i n i n i t i a l sample introduction and separation, and 3) i d e n t i f y i n g the headspace v o l a t i l e s separated by the SHGC method. 19 B. Materials and methods 1. C o l l e c t i o n and preparation of samples Ocean caught pink salmon (Oncorhvnchus qorbuscha) were c o l l e c t e d from the f i s h pumped out of commercial boats at a l o c a l processing plant. After washing and butchering the f i s h , approx. 215g salmon steaks and 2g sodium chloride were put i n 307 x 115 two-piece cans. The cans were then vacuum-sealed (10 in Hg), processed i n a ten busse hor i z o n t a l batch s t i l l steam r e t o r t , and cooled with chlorinated water. The i n s t a l l a t i o n and practice of the commercial plant followed e s t a b l i s h e d Canadian f i s h e r i e s regulations ensuring a minimum l e t h a l i t y (F Q) of 5.6 min. Cans were treated as independent observations from a random sampling. Before GC analysis, each can was opened and the l i q u o r was jdrained for approx. 2 min. The "white" muscles of pink salmon were gently flaked (3-5 mm i n size) with a spatula c a r e f u l l y avoiding dark meat, skin, and bones. The selected amount of f i s h flakes was tr a n s f e r r e d to 20 mL headspace v i a l s (Hewlett Packard, Avondale, PA) . Aluminum caps and teflon-faced s i l i c o n e septums were then crimped on the v i a l s . When a l l the factors described above were optimized, a 1.0 mL sol u t i o n of 3-hexanol (Aldrich Chemical Co., Milwaukee, WI) at a concentration of 84.8 mg/L, i n deionized d i s t i l l e d water, was added to the salmon flakes just before s e a l i n g and served as a reference compound. 20 2. S t a t i c headspace gas chromatography (SHGC) The transfer l i n e of an HP 19395A headspace sampler (Hewlett Packard, Avondale, PA) was attached to the i n j e c t o r port of a HP 5890 gas chromatograph equipped with a flame i o n i z a t i o n detector (FID). Chromatograms were recorded with an HP 3396A integrator. V o l a t i l e separation was accomplished with an ULTRA 2 (cross-linked 5% phenyl methyl silicone) fused s i l i c a c a p i l l a r y column (0.52(lm f i l m thickness x 0.32mm i . d . x 25m length, Hewlett Packard Co., Avondale, PA). The c a r r i e r gas was helium. Some working conditions that remained constant for a l l runs were as follows: i n j e c t o r temperature, 240°C; detector temperature, 250°C; septum purge vent, 2.5 mL/min. The gas flowrates of the FID detector were 33 mL/min for hydrogen, 375 mL/min for a i r , and 30 mL/minor make-up helium. The GC oven was programmed to spend 5 min at the i n i t i a l temperature and to reach 175°C at a rate of 10°C/min. The temperature of the valve, the 3 mL loop, and the transfer l i n e of the headspace sampler was set 5°C above the incubation bath temperature. Other SHA conditions were: v e n t / f i l l loop time, 2 s; i n j e c t i o n time, 30 s. 3. Examination of the GC factors The following l i s t presents the factors and the ranges at which they were studied: temperature of v i a l incubation (TPI), 35 - 125°C; time of v i a l incubation (TMI), 0.25 - 5.5 h; amount of meat in v i a l (AMV), 2.5 -21 12.5 g; i n i t i a l oven temperature (IOT), 35 - 100°C; column headpressure (CHP), 30 - 120 kPa; t o t a l helium flowrate (TFR), 2 - 200 mL/min. The f i r s t three factors were investigated independently while the la s t three were simultaneously optimized using the random-centroid optimization program (RCO) written by Nakai (1989) . The flowchart of the optimization program i s shown i n Figure 1. It i n i t i a l l y consisted of cycles of eight to ten randomly selected experiments for 3-5 factors and a subsequent centroid search with two to three experiments. A mapping subroutine (Nakai et a l . , 1984) was then used to v i s u a l i z e the response surface and guide the reassignment of narrower search ranges. When the approximate optimum areas were l o c a l i z e d , the cycle was not repeated. Instead, a simultaneous s h i f t procedure with up to f i v e experiments was car r i e d out to complete the optimization process. The objective function used was based on Kaiser's concept of peak separation (Kaiser, 1960b) summed over a l l adjacent pairs of peaks. Symmetry of peaks were assumed, therefore f a c i l i t a t i n g the estimation of res o l u t i o n . Peaks with area counts below 100 were not considered. 4. Gas chromatography-mass spectrometry (GC-MS) Tentative v o l a t i l e i d e n t i f i c a t i o n was performed by connecting the headspace sampler to a Hewlett Packard 5985B GC-MS system. The ULTRA 2 c a p i l l a r y column was d i r e c t l y i n terfaced to the ion source. The oven temperature was programmed for 20°C for 5 min and then ramped to 175°C at 10°C/min. The mass spectrometer was operated i n the electron impact mode 22 Start Random search Centroid search Figure 1. Flowchart of the random centroid optimization program. with the following conditions: ion source temperature, 200°C; i o n i z i n g energy, 70 eV; scan range, 34-350 amu at 1 A/D measurement. The mass spectra were acquired with the data system (Rev G) and subsequent data reduction i d e n t i f i e d the peaks with base peak p r o b a b i l i t y matching using the l i b r a r y EPA-NIH Mass Spectral Database (Eight Peak Index Mass Spectra). Confirmation of several compounds was accomplished by comparing t h e i r retention times (within 1.5 s range) with those of reference standards (Aldrich Chem. Co., Milwaukee, WI). 24 C. Results and d i s c u s s i o n Figure 2 shows the e f f e c t of flaked f i s h weight i n v i a l s on the t o t a l area of the chromatograms after 1 h incubation at 75°C. The t o t a l area increased i n a s l i g h t l y c u r v i l i n e a r fashion with the amount of meat between 2.5 g and 10 g. The means of t o t a l area {ni = 5) from t h i s l i n e were l i n e a r l y r e l a t e d to t h e i r standard deviations giving a constant c o e f f i c i e n t of v a r i a t i o n of 8.4% (range 8.1-8.9). This r e l a t i o n s h i p r e f l e c t s increased v a r i a b i l i t y i n flake compactness (or flake density) with f i s h weight i n v i a l s , which in turn, t r a n s l a t e d into increased v a r i a b i l i t y i n flake surfaces exposed to the headspace and therefore i n inte r n a l pressure. A further increment from 10 g to 12.5 g d i d not increase the t o t a l area to a major extent while the c o e f f i c i e n t of v a r i a t i o n was near 10%. The influence of incubation temperature on the chromatographic integrated area i s i l l u s t r a t e d i n Figure 3. As the v i a l s were incubated at higher temperatures, both o v e r a l l peak i n t e n s i t i e s and number of peaks augmented and r e s u l t e d i n a quadratic increase i n t o t a l area. When samples were subjected to incubation temperatures above 110°C for more than 1 h, septums tended to bulge due to i n t e r n a l pressure and leaks were encountered. Figure 4 shows that the t o t a l chromatographic area i n i t i a l l y followed a l i n e a r r e l a t i o n s h i p with incubation time ( f i s h weight, 10 g; incubation temp., 75°C) . After a period of 3 1/2 - 4 h, the t o t a l area reached a plateau i n d i c a t i n g that the equilibrium state between the l i q u i d coating on the f i s h flakes and the headspace was established. In simple 25 2 4 6 8 10 12 14 Weight of fish (g) Figure 2. E f f e c t of f i s h weight on t o t a l chromatographic area (temperature of v i a l incubation, 75°C; time of v i a l incubation, 1 h; column headpressure, 60 kPa; s p l i t r a t i o , 50:1). 2 8 0 3 0 5 0 7 0 9 0 1 1 0 1 3 0 Temperature ( °C) Figure 3. E f f e c t of incubation temperature on t o t a l chromatographic area (time of v i a l incubation, 1 h; f i s h weight, lOg; column head-pressure, 60 kPa; s p l i t r a t i o , 50:1). 2 5 0 D I i I i I i I i I i I i I 0 1 2 3 4 5 6 Time of incubation (h) Figure 4. E f f e c t of incubation time on t o t a l chromatographic area (temperature of v i a l incubation, 75°C; f i s h weight, lOg; column headpressure, 60 kPa; s p l i t r a t i o , 50:1). 28 situations such as d i l u t e aqueous solutions of organic v o l a t i l e s at constant temperature (25°C) and pressure (760 mm), equilibrium between gas and l i q u i d phases has usually been attained within 30 min (Buttery et a l . , 1971) . However, the factors c o n t r o l l i n g the equilibrium in food systems (e.g., salmon flesh) that contain a large number of embodied aqueous and organic constituents are very complex. These factors, i n addition to high incubation temperature of salmon flakes which increased the amount of high b o i l i n g point v o l a t i l e s i n the headspace, seemed to increase the equilibrium time to 4 h. A period of 1 h was however considered more p r a c t i c a l and increased s e n s i t i v i t y to a s u f f i c i e n t extent. This assessment was based on the o r i g i n a l incubation temperature of 75 °C and was assumed to hold for the optimized temperature of 105°C. Results from RCO on the i n i t i a l oven temperature (IOT) , column headpressure (CHP), and t o t a l flowrate (TFR) are shown i n Figures 5, 6, and 7. After repeating the sequence of random simplex and centroid search followed by adjusting the search areas using the mapping procedure, the simultaneous s h i f t process was c a r r i e d out for f i n a l convergence. The li n e s drawn i n these figures were derived from f i t t e d quadratic equations to provide guidance toward l o c a l i z a t i o n of the optima. Among the 25 vertices explored, vertex 24 gave the best separation response (88.0) which corresponded to 35°C, 95.5 kPa, and 44 mL/min for IOT, CHP, and TRF respectively. The lowest IOT (35°C) was necessary since many low-boiling point compounds needed to be separated. The- s p l i t r a t i o 11:1 was calculated from the optimal GC conditions and constituted the best compromise between the s e n s i t i v i t y required for detection of h i g h - b o i l i n g Figure 5. Optimization mapping results of peak separation as a function of i n i t i a l oven temperature (temperature of v i a l incubation, 105°C; time of v i a l incubation, 1 h; f i s h weight, lOg). 100 2£ 40 -CO <D Q . 20 -0 I 1 1 1 1 1 30 49 68 87 106 125 Column headpressure (kPa) Figure 6. Optimization mapping results of peak separation as a function of column headpressure (temperature of v i a l incubation, 105°C; time of v i a l incubation, 1 h; f i s h weight, lOg). o 100 0 I 1 1 1 1 1 1 40.8 80.6 120.4 160,2 200 Total flowrate (mL/min) Figure 7. Optimization mapping results of peak separation as a function of t o t a l flowrate (tempearture of v i a l incubation, 105°C; time of v i a l incubation, 1 h; f i s h weight, lOg). 32 point components and r e s o l u t i o n of e a r l y e l u t i n g v o l a t i l e s . Figure 8 i s a t y p i c a l chromatogram obtained under optimized conditions for 10 g f i s h and 1 h incubation at 105°C. Eighty peaks were detected with t h i s method, using a c a p i l l a r y column and a flame i o n i z a t i o n detector. For sake of comparison, a d i r e c t i n j e c t i o n of a headspace volume drawn from a v i a l containing a l i q u o r sample of canned salmon and equi l i b r a t e d at room temperature would allow the detection of approx. 6 peaks on a packed column (Hollingworth and Throm, 1983) . In contrast, the SHGC method considerably enlarged the spectrum of detectable v o l a t i l e compounds. It was probable that some of the detected v o l a t i l e s were a result of thermal and oxidative degradation due to incubation, even though canned salmon i s a product which has undergone a s t e r i l i z i n g heat treatment. The i d e a l s i t u a t i o n would be to perform the optimization on the integrator response for known compounds which respond only to the j tested parameters. In applications where the food v o l a t i l e s of i n t e r e s t were thermally l a b i l e , other means which could overcome t h i s d i f f i c u l t y (e.g., s a l t i n g out effect) should be used i n the optimization. However, the usefulness of the SHGC method could c e r t a i n l y be extended to the analysis of v o l a t i l e s from various thermally processed food products. A d i s t i n c t advantage was that insoluble and c e l l u l a r materials such as meat could be analyzed with minimum sample preparation and contamination. Another advantage concerns the high sample throughput. Apart from incubation time of v i a l s , each sample analysis required approx. 30 min, o f f e r i n g the p o s s i b i l i t y of analyzing 13 to 15 samples a day. 2 3 4 6 8 10 12 14 Uncorrected retention time (min) Figure 8. Chromatogram of volTtTiles from canned pink salmon (temperature of v i a l incubation, 105°C; time of v i a l incubation, 1 h; f i s h weight, lOg; i n i t i a l oven temperature, 35°C; column headpressure, 95 kPa; t o t a l flowrate, 44 mL/min). Table 1 l i s t s the name of the compounds associated with the peak number found on the chromatogram of F i g . 8, the corresponding uncorrected retention time and percent peak area of a t y p i c a l run, and the method used to i d e n t i f y the v o l a t i l e s . Tentative i d e n t i f i c a t i o n was achieved by comparing mass spectrometry data with that of EPA/NIH l i b r a r y . Retention time, used as a f i r s t step confirmation, was checked with that of authentic standards. Compounds of various classes were found: eight aldehydes; six alkanes; two alkenes; f i v e aromatic compounds; two ketones; three sulfur-containing compounds; one alcohol; one acid, and s i x miscellaneous compounds. V o l a t i l e s that produced large detector responses were hydrogen s u l f i d e , acetaldehyde, methanethiol, butane, t h i o b i s -methane, butanal, l-penten-3-ol, heptane, 1,5-dimethyl-cyclopentene, methyl-benzene, 3-ethyl-2-methylpentane, and 4-ethyl-benzenemethanol. In addition, Table 1 compiles several references where more than half of the v o l a t i l e s i d e n t i f i e d i n t h i s study were also found i n other seafood products. Ethanol was one of the compound cross-referenced and has been used i n the past as an i n d i c a t o r of decomposition i n canned salmon (Hollingworth et a l . , 1986) . In theory, the SHGC method could be used for the determination of ethanol. The o r i g i n of the remaining headspace v o l a t i l e s as well as t h e i r contribution to perceived flavour i s a speculative matter at t h i s point and should be the object of further studies. 35 Table 1. Identification of volatile compounds detected in canned pink salmon. Peak Compound name Uncorrected Peak area IDa Reference no. retention time (%) (min) 1 hydrogen sulfide 0.905 29.013 MS 3,6, 10 2 acetaldehyde 0.935 2.817 MS, RT 5,6, 10 3 methanethiol 1.003 15.003 MS 3,6 4 ethanol 1.135 0.962 MS, RT 2,4, 5,6, 10 5 butane 1.226 12.468 MS 6 3-methyl-l-butene 1.292 0.985 MS, RT 7 thiobis-methane 1.330 1.116 MS, RT 3,6, 8,9, 10 8 2-methyl-propanal 1.530 0.692 MS, RT 9 hexane 1.773 0.779 MS, RT 10 butanal 1.815 1.535 MS, RT 5,10 11 2-methyl-furan 1.869 0.707 MS 12 3-methyl-butanal 2.453 0.175 MS 5,6, 8,9 13 benzene 2.501 0.622 MS, RT 1,7, 10 14 2-methyl-butanal 2.590 0.287 MS, RT 15 2,2-dimethyl-propanal 2.656 0.543 MS 16 l-penten-3-ol 3.000 1.952 MS, RT 4,5, 8 17 heptane 3.305 1.330 MS, RT 6 18 1,5-dimethyl-cyclopentene 3.373 11.223 MS 19 unknown 1 4.611 0.470 -20 acetic acid 5.104 0.340 MS, RT 6 21 methyl-benzene 5.414 1.881 MS, RT 1,6, 7,8, 10 22 3-hexanone |6.205 0.200 MS, RT 8 23 unknown 2 1 6.322 0.032 - 8 24 3-hexanol (standard) 6.522 5.372 -25 3-ethyl-2-methyl-pentane 6.653 1.528 MS 26 ethylidene-cyclohexane 7.093 0.157 MS 27 unknown 3 7.663 0.125 -28 nonane 9.475 0.373 MS, RT 9 29 1-propenyl cyclohexane 10.572 0.021 MS 30 benzaldehyde 10.614 0.125 MS, RT 5,6, 8,9 31 7-octen-4-ol 10.875 0.242 MS 32 2-pentyl-furan 11.550 0.247 MS 8 33 decane 11.673 0.013 MS, RT 9 34 4-ethyl-benzenemethanol 11.750 1.869 MS 35 octanal 11.944 0.257 MS, RT 36 3-ethyl-l,4-hexadiene 12.295 0.105 MS 37 2-nonanone 13.097 0.162 MS, RT 5,6, 8,9 38 nonanal 13.700 0.270 MS, RT 5,6, 8,9 a MS, tentatively identified by mass spectrometry; RT, retention time consistent with that of authentic compounds. b 1, Easley et al. (1981); 2, Hollingworth et al. (1986); 3, Hughe3 (1964); 4, Human and Khayat (1981); 5, Kubota et al. (1982); 6, McGill et a l . (1977); 7, Reinert et al. (1983); 8, Tanchotikul and Hsieh (1989); 9, Vejaphan et al. (1988); 10, Wong et al. (1967). 36 I I I . CANNED SALMON QUALITY CLASSIFICATION BY MULTIVARIATE ANALYSIS OF GAS CHROMATOGRAPHIC DATA A. Introduction Modern a n a l y t i c a l techniques provide the p o t e n t i a l to obtain large amounts of information from every experimental u n i t . On one hand, the data may originate from d i f f e r e n t methods measuring various parameters of in t e r e s t . On the other hand, multiparameter methods, such as g a s - l i q u i d chromatography and sensory evaluation, can generate multidimensional bodies of data i n a single analysis. Given the large number of factors contributing to the changes i n the food properties being studied, multivariate analysis (MVA) constitutes a powerful t o o l to a s s i s t i n int e r p r e t a t i o n of complex datasets. MVA includes several procedures, a l l designed to describe the relationships that e x i s t i n a mul t i v a r i a t e dataset, by i s o l a t i n g and id e n t i f y i n g redundancies. These procedures o f f e r advantages over univariate s t a t i s t i c a l methods i n being able to deal with many variables simultaneously and thereby uncover r e l a t i o n s h i p s that could not be observed when examining the various random fa c t o r s one by one. Although these multivariate techniques are s i m i l a r i n many ways, there are important differences with regard to the questions they are designed to address and the in t e r p r e t a t i o n of t h e i r r e s u l t s . Multivariate techniques can be div i d e d i n two groups, dependence and interdependence methods (Dillon and Goldstein, 1984) . The f i r s t type relates to the techniques dealing with the association between a set of c r i t e r i o n (predictant) variables and a set of predictor variables, i . e . , l i n e a r discriminant analysis and canonical c o r r e l a t i o n a n a l y s i s . The second type i s less p r e d i c t i v e i n nature and centres on the mutual association across a set of given variables, i . e . , p r i n c i p a l component analysis, factor analysis, c l u s t e r analysis, and multidimensional s c a l i n g . P r i n c i p a l component analysis (PCA) derives l i n e a r combinations (pr i n c i p a l components) of the o r i g i n a l variables, e.g., Z = WX (1) where Z i s a matrix of PC, X a matrix of o r i g i n a l variables, and W a matrix of weights which have been computed to maximize the r a t i o of the variance of PC to the t o t a l v a r i a t i o n . The successive l i n e a r combinations are extracted to produce uncorrelated p r i n c i p a l components (PC) that consecutively account for smaller amounts of the t o t a l variance. The complexity of the Lata can be reduced by representing the numerous variables in a few selected components which comprise a large proportion of the o r i g i n a l variance. Based on the uniqueness and exact so l u t i o n of PCA, scores of selected PC can be used i n l a t e r analyses i n place of the o r i g i n a l responses. The i n t e r p r e t a t i o n of sensory c h a r a c t e r i s t i c s has been f a c i l i t a t e d by using PCA i n the study of wine (Heyman and Noble, 1989) , beer (Glapperton and Piggott, 1979) , f i s h (Quarmby and Ratkowshy, 1988) , frozen peas (Sanford et a'l. , 1988) , and selected food items (Syarief et a l . , 1985) . PCA was applied to data obtained from a v a r i e t y of p h y s i c a l , 38 chemical, biochemical, and other instrumental methods for analysis of rum (Herranz et a l . , 1990), durum wheat (Autran et a l . , 1986), whisky (Headley and Hardy, 1989), orange ju i c e (Aries et a l . , 1986), cheese (Kwak et a l . , 1989), and wine (Moret et a l . , 1986). The common f a c t o r - a n a l y t i c model (CFA), or less p r e c i s e l y factor analysis, expresses the o r i g i n a l variables i n terms of newly generated common factors and a unique factor, e.g., X = WF + E (2) where X i s a matrix of o r i g i n a l variables, W a matrix of weights associating the common factors to the o r i g i n a l variables, F a matrix of latent common factors, and E a matrix of unique fac t o r s . Its purpose consists of find i n g the part of the t o t a l variance that a p a r t i c u l a r variable shares with the other v a r i a b l e s . In other words, CFA attempts to explain the co r r e l a t i o n s or covariances among a set of variables i n terms of a l i m i t e d number of unobservable, latent v a r i a b l e s . Factor scores, which give the coordinates of each observation i n the multidimensional space of the common factors, cannot be computed d i r e c t l y . They must instead be estimated. Because of t h i s indeterminacy problem, the non-uniqueness of factor scores sometimes arouse c r i t i c i s m s of t h e i r use for subsequent analyses. CFA has been used to determine relationships among instrumental methods and sensory q u a l i t i e s of cheese (Vangtal and Hammond, 1986), green beans (Powers et a l . , 1977), peanut beverage (Rubico et a l . , 1988), dry beans (Hosfield et a l . , 1984), and tomatoes (Resurreccion and Shewfelt, 1985). It was also applied i n the study of frozen cod (Leblanc et a l . , 1988), cooked beef (Gait and MacLeod, 1983), and chicken p a t t i e s (Lyon, 1988) . Rotation i s a post hoc step which could be performed on both PCA and CFA solutions i n order to improve the i n t e r p r e t a b i l i t y of the r e s u l t s . Two types of procedures for r o t a t i n g axes are encountered, orthogonal and oblique. Orthogonal rotations r e s t r i c t the factors to be mutually independent (uncorrelated) by preserving t h e i r perpendicular orientation, and oblique rotations allow the resultant factors to adopt various angles and therefore be correlated. Rotating a set of factors does not change the s t a t i s t i c a l explanatory power of the f a c t o r s . However, a consequence of correlated factors as a r e s u l t of using oblique rotation i s that no single unambiguous measure of the importance of a factor can explain a variable. Linear discriminant analysis (LDA) i s a multivariate approach which handles the case of a q u a l i t a t i v e , or c a t e g o r i c a l dependent variable and a set of independent vari a b l e s . This technique aims at c o r r e c t l y c l a s s i f y i n g observations into mutually exclusive groups known a p r i o r i . Original independent variables are l i n e a r l y assembled to form discriminant functions with the objective of minimizing the m i s c l a s s i f i c a t i o n rate. This i s accomplished under the s p e c i f i c a t i o n that the between-group variance of the weighted predictor variable combinations r e l a t i v e to the within-group variance i s maximal. LDA r e l i e s upon ce r t a i n conditional assumptions. In p a r t i c u l a r , the independent variables must have 40 multivariate normal d i s t r i b u t i o n s and t h e i r variance-covariance matrix i n each group should be considered homogeneous. Discriminant analysis (DA) of GC v o l a t i l e patterns have often been used to segregate d i f f e r e n t groups within various food commodities including wine (Montedero and B e r t u c c i o l i , 1983; Marais et a l . , 1982; Moret et a l . , 1986), beer (Helbert and Hoff, 1974; Hoff et a l . , 1975; Lindsay, 1977), milk (Leland et a l . , 1987; Smeyers-Verbeke et a l . , 1977), coffee and potato chips (Powers and Keith, 1968), cola beverage (Young et a l . , 1970), grape brandy (Schreier and Reiner, 1979), soy sauce (Aishima, 1979a; Aishima, 1979b; Aishima, 1983), cheese (Aishima and Nakai, 1987), grapefruit juice (Pino et a l . , 1986), and corn (Dravnieks et a l . , 1973; Dravnieks and Watson, 1973). Interpretation of data from other instrumental methods have also b e n e f i t t e d from the use of DA, e.g., rheological c h a r a c t e r i s t i c s of cheese (Amantea et a l . , 1986), heavy metal c h a r a c t e r i s t i c s of cereals (Stryjewska et a l . , 1987), high pressure l i q u i d chromatography of cheese extract (Pham and Nakai, 1984; Mohler-Smith and Nakai, 1990), milk c l o t t i n g a c t i v i t y of p r o t e o l y t i c enzymes (Aishima et a l . , 1986), meat protein f u n c t i o n a l i t y (Li-Chan et a l . , 1987), and p y r o l y s i s mass spectrometry of orange juice (Aries et a l . , 1986). Discriminative relationships among sensory and instrumental data have been examined in canned blueberries (Powers et a l . , 1978), butter (Woo and Lindsay, 1983), and beer (Brown and Clapperton, 1978). The di s c r i m i n a t i v e t e x t u r a l characterization of various ph y s i c a l properties was demonstrated i n cooked f i s h by Hatae et a l . (1984) . S i m i l a r l y , sensory properties of 18 common A t l a n t i c species of f i s h were assessed for t h e i r usefulness i n 41 species discrimination (Sawyer et a l . , 1988). The SHGC method developed i n Chapter II produced a large array of responses each time a sample of canned salmon was submitted for v o l a t i l e analysis. The pattern of v o l a t i l e compounds may r e f l e c t a number of qua l i t y a t t r i b u t e s including species of P a c i f i c salmon, stage of sexual maturity, and l e v e l of freshness. Therefore, the objectives of the second part of t h i s research were: (1) to apply multivariate s t a t i s t i c s to data from the SHGC method for c l a s s i f i c a t i o n of canned P a c i f i c salmon i n terms of species, stages of sexual maturity of chum salmon, and r e f r i g e r a t e d decomposition of pink salmon, and (2) to study the re l a t i o n s h i p s between the results from sensory evaluation and gas chromatography of fresh and canned pink salmon during r e f r i g e r a t e d storage. 42 B. Materials and methods 1. C o l l e c t i o n and canning of salmon P a c i f i c salmon of selected species were c o l l e c t e d d i r e c t l y from commercial boats at a l o c a l f i s h processor i n Vancouver. A l l f i s h were kept refrigerated at approximately 5°C for 2-3 h i f they could not be cleaned and eviscerated immediately. Steaks of approx. 3.8 cm i n width were cut and 215 g salmon f l e s h was placed i n 307 x 115 two-piece cans with 2 g sodium chl o r i d e . The cans were then vacuum-sealed, retorted, and cooled according to pr a c t i c e of the commercial plant as described i n Chapter I I . 2. Investigated treatments Table 2 shows the number of P a c i f i c salmon that were cleaned and processed for the three treatments under i n v e s t i g a t i o n . These treatments were studied over a two year sampling period to account for possible year to year v a r i a t i o n i n the f i s h and analyses. The f i r s t treatment of interest covered four d i f f e r e n t species of P a c i f i c salmon, namely, pink (Oncorhynchus gorbuscha Walbaum), sockeye (O^ nerka Walbaum), coho (0.  kisutch Walbaum), and chum (0^ keta Walbaum). For a l l treatments but the second year of the species study, two samples per f i s h were analyzed to account for w i t h i n - f i s h v a r i a t i o n . A l l observations were however considered independent of each other i n the s t a t i s t i c a l analyses. 43 Table 2. Number of P a c i f i c salmon processed with respect to the species, stages of sexual maturity, and r e f r i g e r a t e d decomposition studies f o r two sampling years. Treatment Year" Species of P a c i f i c salmon pink sockeye coho chum 6 (2) 4 (2) 6 (2) 4 (2) 6 (1) 6 (1) 6 (1) 6 (1) Sexual maturity of chum salmon s i l v e r - b r i g h t (stage 1) semi-bright (stage 2) commercial dark (stage 3) 6 (2) 12 (2) 6 (2) 11 (2) 6 (2) 6 (2) Refrigerated storage of pink salmon 42 (2) 54 (2) a The d i g i t i n parenthesis represents the number of cans per f i s h analyzed by SHGC. The second treatment focussed on the aspect of sexual maturity of chum salmon. Fish were c l a s s i f i e d i n three d i f f e r e n t stages of maturity based on external appearance using the Color Evaluation Guide for P a c i f i c Salmon (Alaska Seafood Marketing I n s t i t u t e , Juneau, AK) . F i s h corresponding to i l l u s t r a t i o n s A, B, and C were c l a s s i f i e d as stage 1 ( s i l v e r - b r i g h t ) , D,E, and F as stage 2 (semi-bright), and G, H, and I as stage 3 (dark). The t h i r d treatment was r e f r i g e r a t i o n storage time of pink salmon. Fish with good external appearance were taken from unloading commercial boats and stored between layers of ice i n a cold room at 2-5°C. The f i s h were not eviscerated before storage i n i c e . About every two days f o r up to 13 and 21 days of r e f r i g e r a t e d storage for years 1 and 2, r e s p e c t i v e l y , s i x pink salmon were randomly sampled; the remaining f i s h held i n the bogey were re-iced i f needed. The two day sampling period was increased towards the end of storage to 3 or 4 days i n order to extend the decomposition process. 3. Sensory assessment o f raw and canned salmon The quality of each f i s h selected was assessed before canning. Chum salmon of d i f f e r e n t stages of maturity were pre-categorized by commercial plant personnel and t h e i r c l a s s i f i c a t i o n based on the Colour Evaluation Guide for P a c i f i c Salmon were subsequently determined. Pink salmon subjected to the r e f r i g e r a t e d storage treatment were assessed by the author with the help of experienced personnel from a l o c a l f i s h processor using the grading guide shown in Table 3. The procedure involved evaluating condition of the eyes, the g i l l s , and the f l e s h texture for the external c h a r a c t e r i s t i c s . Once the f i s h was opened and eviscerated, the i n t e g r i t y and odour of the b e l l y cavity were then the main c r i t e r i a examined. Canned samples were evaluated by a tra i n e d and experienced Federal Fish Inspector using the 1986 Canned P a c i f i c Salmon Grade Standard (Department of Fisheries and Oceans, Vancouver, BC) . The canned pink salmon from the r e f r i g e r a t e d storage studies were randomized amongst the cans of the other treatments. The vacuum of each can was measured with a Seaman Vacuum Gauge and most values were found to be between 11-12 i n Hg. The cans were then opened and drained for 2-5 min. The colour grade was obtained by v i s u a l comparison with coloured p o r c e l a i n enamel t i l e s (Department of Fisheries and Oceans, Vancouver, BC). Appearance of the canned salmon was examined for attributes such as watermarking (skin with uncharacteristic colour), belly-burn (reddened f l e s h ) , and curd. The f i n a l grade of the products was however based s o l e l y on odour and taste. The important flavour defects were characterized as decomposed, rancid, late (advanced sexual maturity), and overheating. If possible, the decomposed flavours were further q u a l i f i e d as f r u i t y , sour, faecal, cheesy, ammoniacal, or by use of other relevant d e s c r i p t o r s . Every c r i t e r i o n assessed was rated to describe i t s i n t e n s i t y , and samples were subsequently assigned scores integrating the ratings. F i n a l l y , scores were translated into grades corresponding to one of the following categories of q u a l i t y : A, B, and reject (Table 4). Table 3. Grading guide f o r whole raw P a c i f i c salmon. C r i t e r i a Grade A Grade B Reject External c h a r a c t e r i s t i c s Eyes clear , bright, convex s l i g h t l y sunken and d u l l d u l l , sunken, and cloudy G i l l s colour G i l l s odour bright red fresh, seaweed, or s h e l l f i s h odour pink neutral, s l i g h t but d e f i n i t e sour, faecal, or p u t r i d odour brown, grey strong sour, p u t r i d , p u t r i d or faecal odour Texture firm and r e s i l i e n t , f l e s h springs back when thumb depression i s released moderately soft, thumb indentations may slowly f i l l out very soft, thumb indentations may remain i n f l e s h Internal c h a r a c t e r i s t i c s B e l l y c a v i t y transparent, i n t a c t p e r i t o n e a l l i n i n g Odour of b e l l y fresh and character-i s t i c odour moderate reddening, some r i b s may protrude s l i g h t but d e f i n i t e sour, faecal, or pu t r i d odour extensive reddening, li q u e f a c t i o n of b e l l y walls strong sour, faecal, or p u t r i d odour 0\ 47 Table 4. Grading guide f o r canned P a c i f i c salmon. C r i t e r i a Grade A Grade B Reject Colour 0-2 3-4 Appearance 3 0-1 2-3 -b c Odour and taste ' 0-2 3 4-9 a Cross-packing, belly-burn, watermarking Descriptive terms: decomposed ( f r u i t y , vegetable, sour, faecal, putrid) , rancid, l a t e , overheating, contamination, etc. ° Grade assignment: Ratinq Score Grade 1 0, A (1+1) l' | A (1+1+1) etc. 2 A 2 2 A 3 4 B 4 8 R 5 16 R 6 32 R 7 64 R 8 128 R 9 256 R 48 4. S t a t i c headspace gas chromatography (SHGC) GC analyses were performed on canned salmon that had been stored at room temperature for 3 to 12 months. Sample preparation f o r gas chromatography was done i n the following manner. Each can was opened and the l i q u o r was drained by t i l t i n g the can and i t s l i d for 2 min. Af t e r f l a k i n g the white muscle 3-5 mm i n si z e with a spatula, 10 g were transferred i n 20 mL headspace v i a l s (Hewlett Packard, Avondale, PA) avoiding dark meat, skin, and bones. One m i l l i l i t r e of 3-hexanol working solu t i o n (84.8 ppm) was added to the v i a l s , which were then sealed with Teflon-faced s i l i c o n e septums and aluminum caps. Some of the raw salmon from the r e f r i g e r a t e d storage of year 2 was also kept for SHGC analysis. Samples were inserted i n p l a s t i c freezer bags (Baggies, Colgate-Palmolive Canada, Toronto, Canada), and a f t e r evacuating the a i r with a straw, the bags were closed with twist t i e s before freezing at -20°C for approx. 1 month. As needed, the frozen samples of raw salmon were thawed i n water (20°C) for 3-4 hours. The thawed white muscles were then cut i n pieces of 2-5 g with clean s c i s s o r s and 12.5 g was put i n headspace v i a l s . Aliquots of 3-hexanol standard (84.8 ppm) were then added to the v i a l s before sealing as described e a r l i e r for canned salmon. The- SHGC method used was derived from Chapter I I . A HP 5890 gas chromatograph (Hewlett Packard, Avondale, PA) equipped with a flame i o n i z a t i o n detector (FID) was connected to a HP 19395A headspace sampler. A HP 3396A integrator recorded the chromatograms and also transmitted the output to an IBM-compatible personal computer through a RS-232-C cable. The signal was handled by a program c a l l e d FILE SERVER which allowed the host computer to serve as a remote external disk drive. Separation of v o l a t i l e s was made on a HP ULTRA 2 fused s i l i c a c a p i l l a r y column (25m length x 0.32mm i . d . x 0.52um f i l m thickness). The operating conditions of the GC system are shown i n Table 5. 5. S t a t i c headspace gas chromatography-mass spectrometry (SHGC-MS) Tentative v o l a t i l e i d e n t i f i c a t i o n was performed by connecting the headspace sampler to a Hewlett Packard 5985B GC-MS system. The conditions of operation and separation as well as mass spectra a c q u i s i t i o n and confirmation by retention time comparison with reference standards were executed as described i n Chapter I I . 6. Preparation of standard s o l u t i o n s Approx. 1 mL of 3-hexanol (Aldrich Chemical Co., Milwaukee, WI) was weighed i n 1 L volumetric f l a s k before d i l u t i n g to volume with deionized d i s t i l l e d water (DDE). Ten m i l l i l i t r e s of t h i s stock s o l u t i o n was pipetted into a 100 mL v o l . f l a s k . The flask was brought up to volume with DDE and served as working s o l u t i o n (84.8 ppm). 50 Table 5. Operating conditions for the s t a t i c headspace gas chromatographic method used to analyze v o l a t i l e s i n canned P a c i f i c salmon. Headspace sample conditions incubation temperature of v i a l s incubation time of v i a l s loop si z e valve temperature v e n t / f i l l loop time i n j e c t i o n time 105°C 1 h 3 mL 110°C 2 s 30 s GC conditions c a r r i e r gas in j e c t o r temperature detector temperature column headpressure s p l i t vent septum purge vent helium 240°C 250°C 95 kPa 40 mL/min 2.5 mL/min Detector flowrates hydrogen a i r make-up gas (helium) 33 mL/min 375 mL/mih 30 mL/min Oven temperature program i n i t i a l oven temperature time at i n i t i a l temperature ramp rate f i n a l temperature 35°C 5 min 10°C/min 175°C Integrator conditions i n i t i a l settings attenuation chart speed threshold peak width 0 2 -1 0.03 51 7. Data handling and s t a t i s t i c a l a n alysis A computer program was written i n QuickBasic 4.5 (Microsoft Corporation) to preprocess the chromatographic data. Raw data were f i r s t extracted from the stored d a t a f i l e s of the integrator reports and f i l t e r e d to compile chromatographic area within the desired retention time i n t e r v a l s . Standardization was done by d i v i d i n g each f i l t e r e d chromatographic area with the peak area of the 3-hexanol standard and multiplying by 10000. A l l data were then accumulated i n a spreadsheet format. A l l duplicate and r e p l i c a t e analyses of both sampling years, which represented within f i s h , between f i s h , and year variations, respectively, were gathered and combined f or the four species of P a c i f i c salmon, three sexual maturity stages of keta salmon, and three decomposition l e v e l s of pink salmon with the goal of d i f f e r e n t i a t i n g these s p e c i f i c groups. The S t a t i s t i c a l Analysis System (SAS, 1989) was used f o r multivariate and regression analyses. P r i n c i p a l component analysis (PCA) and common factor analysis (CFA) with and without r o t a t i o n were performed using the FACTOR procedure. Computations of l i n e a r , quadratic, and non-parametric discriminant analyses were c a r r i e d out with the DISCRIM procedure. The REG procedure was applied to f i t least-squares estimates to l i n e a r and multiple regression models. Estimated l i n e a r models were also obtained with the CATMOD procedure when the models involved categorical v a r i a b l e s . 52 SYSTAT (Wilkinson, 1990) was another s t a t i s t i c a l package used f o r data analysis. The o r i g i n a l spreadsheets of standardized data were f i r s t converted to compatible workfiles with the IMPORT command of the DATA module. The NPAR module was then used to compute the Kolmogorov-Smirnov (KS) s t a t i s t i c with the L i l l i e f o r s option f o r t e s t i n g normality. B a r t l e t t ' s test for homogeneity of variance was computed using the STATS module. 53 C. Results and d i s c u s s i o n 1. M V A of v o l a t i l e s from canned salmon of d i f f e r e n t species For every sample of canned salmon analyzed, numerous v o l a t i l e s were separated and constituted the chromatograms. A v i s u a l examination of the SHGC chromatograms indi c a t e d l i t t l e q u a l i t a t i v e differences i n the patterns of v o l a t i l e s between the species of P a c i f i c salmon studied. The use of multivariate s t a t i s t i c a l methods was d i c t a t e d by the complex array of responses obtained from each can. The i n i t i a l step of data analysis involved the s e l e c t i o n of peaks to b u i l d up complete matrices. Small peaks having actual area counts below 100 were not detected and/or integrated reproducibly and therefore were not considered. Forty-four peak responses were selected on the basis of consistent detection and integration of the same treatments, and most of them had a c o e f f i c i e n t of v a r i a t i o n between 4 and 12%. The 44 selected peaks are shown i n Figure 9 and t h e i r i d e n t i t y i n Table 6. On occasion, the f i r s t two peaks of some chromatograms were not adequately separated. For t h i s reason, they were combined to provide one response and formed peak 1. The f i r s t m u l t ivariate analysis performed was p r i n c i p a l component analysis (PCA). In theory, the number of p r i n c i p a l components that could . be extracted from the data set of salmon species equals the rank of the c o r r e l a t i o n matrix. Of the 44 possible components, the f i r s t nine components followed Kaiser's rule of thumb (Kaiser, 196.0a) which recommends to r e t a i n those components having eigenvalues (latent roots) 2 4 17 22 Std Uncorrected retention time (min) Figure 9. Chromatogram of v o l a t i l e s from canned pink salmon selected to carry out multivariate s t a t i s t i c a l analyses. 55 Table 6. I d e n t i f i c a t i o n of v o l a t i l e compounds used in multivariate analysis of canned P a c i f i c salmon. Peak no. Compound name ID 4 1 hydrogen s u l f i d e MS 1 acetaldehyde MS, RT 2 methane t h i o l MS 3 ethanol MS, RT 4 butane MS 5 3-methyl-l-butene MS, RT 6 dimethyl s u l f i d e MS, RT 7 2-methyl propanal MS, RT 8 hexane MS, RT 9 butanal MS, RT 10 2-methyl furan MS 11 benzene MS, RT 12 2-methyl butanal MS, RT 13 2,2-dimethyl propanal MS 14 unknown 1 -15 l-penten-3-ol MS, RT 16 heptane MS, RT 17 1,5-dimethyl cyclopentene MS 18 3-methyl-butanol MS, RT 19 2-methyl-2-butenal MS, RT 20 unknown 2 -21 acetic a c i d MS, RT 22 toluene MS, RT 22 3-hexanone MS, RT 23 unknown 3 -24 unknown 4 25 unknown 5 ^ MS 26 3-ethyl-2-methyl-pentane 27 ethylidene cyclohexane MS 28 unknown 6 -29 unknown 7 -30 nonane MS, RT 31 unknown 8 -32 benzaldehyde MS, RT 33 7-octen-4-ol MS 34 unknown 9 -35 unknown 10 -36 unknown 11 -37 unknown 12 -38 2-pentyl furan MS 39 4-ethyl benzene methanol MS 40 3-ethyl-l,4-hexadiene MS 41 nonanal MS, RT 42 unknown 13 -43 unknown 14 -44 unknown 15 -* MS, t e n t a t i v e l y i d e n t i f i e d by mass spectrometry; RT, retention time consistent with that of authentic compounds. 56 greater than one (Table 7) . But since the tenth component had a value approaching one, i t was also included i n the subsequent analyses. Interpretation of p r i n c i p a l components was f a c i l i t a t e d by computing the component loadings. A component loading gives the product-moment co r r e l a t i o n of each variable and the respective component. Table 7 presents these correlations for the peak area variables from the salmon species data. Based on those variable loadings highest on a given factor, the f i r s t p r i n c i p a l component (PCI) was dominated by peaks 1 (hydrogen s u l f i d e and acetaldehyde), 5 (3-methyl-l-butene), 16 (heptane), 17 (1,5-dimethyl cyclopentene), 29 (unknown 7), 30 (nonane), 31 (unknown 8), 32 (benzaldehyde), 37 (unknown 12), and 38 (2-pentyl furan). Peaks 9 (butanal) and 34 (unknown 9) had high negative c o r r e l a t i o n s with p r i n c i p a l component 2 (PC2) while peak 14 (unknown 1) had a dominant p o s i t i v e c o r r e l a t i o n . Because of the b i p o l a r i t y i n algebraic signs, PC2 contrasted the group of peaks 9 and 34 with peak 14. Similar interpretations could be done for the remaining of the components. Nearly 40 percent of the variance was explained by the f i r s t p r i n c i p a l component. Together, the ten p r i n c i p a l components i n Table 7 accounted for over 87 percent of the t o t a l variance. PCA e f f e c t i v e l y reduced the dimensionality of the data set i n d i c a t i n g a high degree of in t e r c o r r e l a t i o n s among the o r i g i n a l v a r i a b l e s . By d e f i n i t i o n , PCA transforms the peak area factors into new latent variables (PC's) uncorrelated to each other. Therefore a m u l t i c o l i n e a r i t y problem due to high i n t e r c o r r e l a t i o n s has been a l l e v i a t e d which otherwise would have 57 Table 7. Correlation (loadings) of ga3 chromatographic peak variables from canned Pacific salmon (chum, coho, pink, sockeye) with the f i r s t ten principal' components. Peak Principal component 1 2 3 4 5 6 7 8 9 10 1 0. ,847 0. 125 0. ,296 -0. 119 -0. 234 0. 006 -0. 044 0. 089 0. 203 -0. 051 2 0. 257 -0. 354 0. 661 -0. 423 0. 009 0. 211 0. 142 -0. 166 0. 063 0. 049 3 -0. ,373 0. 327 -0. 591 0. 000 0. 074 0. 339 -0. 171 -0. 028 0. 050 0. 175 4 0. 526 -0. 233 0. 637 0. 041 -0. 078 -0. 010 -0. 032 -0. 229 0. 150 -0. 168 5 0. ,810 0. 371 0. 145 0. ,062 -0. 094 0. 003 0. 057 0. 144 0. ,193 -0. 162 6 0. 682 0. 312 0. ,113 0. 278 -0. 151 -0. 119 -0 . 144 -0. 102 -0. 340 -0. 089 7 0. ,539 -0. 460 0. ,309 -0. 485 -0. ,072 0. 288 0. 078 0. 039 -0. 075 -0. .051 8 0. ,500 0. 344 -0. ,307 -0. 017 -0. 552 0. 159 0. 087 -0. 236 0 . 193 0. 002 9 0. ,405 -0. ,740 0. ,363 0. ,042 0. ,057 0. 167 -0. 206 -0. 048 -0. ,081 -0. ,045 10 0. ,744 0. 321 0. ,093 0. 130 0. 017 0. 003 0. 040 0. ,039 -0. 250 -0. ,150 11 0. ,677 -0. ,316 -0. ,138 -0. ,388 -0. 312 0. 110 0. ,050 0. ,170 -0. ,246 0. ,054 12 0. ,195 -0. ,231 0. ,007 -0. 520 -0. 195 -0. ,210 -0. ,505 0. 031 0. 196 0. ,044 13 0. .454 -0. ,428 -0. .339 -0. ,513 -0. ,182 0. ,134 0. ,062 0. ,116 -0. ,214 0. ,113 14 0. .273 0. .668 0. .348 -0. ,181 -0. ,107 -0. 006 0. ,084 0. ,022 0. ,033 0. .140 15 0. .783 -0. ,112 -0. .023 0. ,313 -0. .448 0. ,031 0. ,052 -0. ,074 0. ,037 0. ,173 16 0. .800 -0. ,422 -0. .085 0. ,191 -0. ,147 -0. ,094 -0. ,017 0. ,251 0. ,052 0. ,045 17 0. .802 -0. .053 -0. .515 0. .003 0. .022 0. ,114 -0. .039 -0. .025 0. .098 -0. .043 18 0. .143 0. .476 -0. .368 -0. .344 -0. ,099 0. .119 0. .234 0. .261 -0. .076 -0. .448 19 0. .193 0. .358 0. .230 0. .241 0. .080 0. .588 -0. .483 0. .102 -0. .069 0. .049 20 0. .505 -0, ,562 0. .316 0. ,186 0. .147 -0. .054 -0, .053 -0, .078 0. .306 -0, .240 21 0. .481 -0. .054 0, .240 0. .207 -0. .552 0. .061 0. .201 -0. .250 -0. .170 0. .097 22 0. .003 -0. .460 -0, .679 -0, .055 0, .183 0, .088 0, .044 -0, .265 -0, .078 -0, .025 23 j 0, .598 0. .093 0, .262 0. .066 0, .347 -0. ,129 -0. .302 -0, .126 -0. .412 0, .128 24 0, .589 -0, .419 -0 .075 0, .247 -0, .364 0. .004 -0, .017 0.285 0, .012 0, .221 25 ' 0. .708 0, .338 -0, .119 0, .169 -0, .130 -0. .114 0. .118 -0, .202 0, .291 0, .143 26 0, .551 -0, .032 0 .058 0 .317 -0, .146 -0, .051 -0, .082 -0, .031 -0, .208 -0, .491 27 0, .788 0, .081 0, .128 0, .099 0, .311 -0. .187 -0, .018 0, .213 0, .015 0, .252 28 0 .785 0, .320 0 .138 -0, .047 0 .287 -0, .183 0, .016 0, .184 0 .004 0, .117 29 0, .862 0, .014 -0 .125 -0, .051 0, .315 -0, .137 -0, .011 0, .086 0, .075 -0 .026 30 0 .821 -0 .200 -0 .431 0, .115 0 .084 -0, .058 0 .001 0 .041 0, .200 0 .028 31 0, .841 0, .156 -0, .274 0, .173 -0, .209 0, .109 -0, .033 -0, .119 -0, .162 -0, .001 32 0 .919 0 .066 0 .038 0 .129 0 .160 -0, .041 -0 .025 0 .100 -0 .037 0 .029 33 0 .602 -0, .488 0 .356 -0, .271 0, .179 0, .056 0 .091 -0 .011 0, .057 -0 .036 34 0 .056 -0 .871 -0 .125 0 .252 0 .174 -0, .042 0 .058 0 .046 -0 .058 -0 .045 35 0 .466 0 .318 -0 .089 -0, .347 0, .198 0, .023 -0 .130 -0 .588 -0 .070 0 .043 36 0 .783 0 .383 0 .047 -0 .213 0 .092 0, .001 -0 .033 -0 .050 0 .094 0 .122 37 0 .888 0 .226 0 .120 -0 .043 0 .014 -0 .024 -0 .055 0 .042 -0 .050 0 .120 38 0 .879 0 .263 -0 .083 -0 .220 0 .177 -0 .029 0 .112 0 .080 -0 .090 -0 .003 39 0 .679 -0 .372 -0 .520 -0, .031 0 .259 0 .102 0 .039 -0 .085 0 .023 -0 .041 40 0 .336 -0 .011 0 .227 0 .181 0 .396 0 .047 0 .610 -0 .178 -0 .078 0 .131 41 0 .556 -0 .599 -0 .489 0, .180 0 .074 0 .029 -0 .006 -0 .096 -0 .003 0 .069 42 0 .765 0 .423 -0 .251 -0 .183 0 .226 -0 .015 0 .115 0 .067 -0 .010 -0 .118 43 0 .662 0, .174 -0 .166 0 .072 0 .241 0 .177 -0 .206 -0 .134 0 .239 -0 .183 44 -0. .048 0 .123 0 .258 0 .290 0 .256 0 .714 0 .148 0 .189 0 .097 0 .062 nt root 17 .178 5 .945 • 4 .276 2 .448 2 .383 1 .480 1 .351 1 .228 1 .109 0 .982 ance (%) 39 .042 13 .512 9 .719 5 .564 5 .415 3 .363 3 .070 2 .790 2 .521 2 .233 58 occurred during the use of the o r i g i n a l variables i n further s t a t i s t i c a l procedures. Figure 10 shows a plot of the scores of the d i f f e r e n t salmon species for the f i r s t two p r i n c i p a l components. There was an i n i t i a l attempt to uncover differences between the species based on PCI. On the other axis, PC2 c l e a r l y contrasted the salmon species between the two sampling years. At least two major sources of v a r i a t i o n could be put forward to explain t h i s demarcation on PC2. F i r s t l y , salmon of d i f f e r e n t years may produce variations in v o l a t i l e patterns due to t h e i r differences in feed sources, water temperature changes, and other possible environmental f a c t o r s . However, not a l l salmon of a given species were from the same runs, as they were sampled on d i f f e r e n t days during the f i s h i n g seasons. Since salmon of d i f f e r e n t runs, but c o l l e c t e d the same year, responded s i m i l a r l y on PC2, b i o l o g i c a l explanations do not provide s a t i s f a c t o r y answers. The two year-groups were analyzed approximately one year apart. Although the analyses were performed on the same GC column and on the same instrument, s l i g h t changes i n factors such as detector s e n s i t i v i t y , column phase, and gas flow rates could contribute to v a r i a t i o n s i n the GC responses i n a consistent manner between years. A c a r e f u l inspection of peaks 9, 14, and 34 which received high loadings on PC2 revealed that they had decreased i n resolution between year 1 and year 2. To a large extent, PC2 may r e f l e c t the loss in chromatographic separation or increased measurement error for these peaks. CM O "ca Q. 15 c Q_ 1 i-0 --1 --2 h -3 - 3 - 2 - 1 0 1 2 3 Principal component 1 Figure 10. Plot of the f i r s t two p r i n c i p a l component scores for the salmon species (A, pink-year 1; B, coho-year 1; C, chum-year 1; D, sockeye-year 1; E, pink-year 2; F, sockeye-year 2; G, coho-year 2; H, chum-year 2 ) . The data for both years were not subsequently analyzed separately as the number of observations was i n s u f f i c i e n t , and they therefore remained i n a pooled format. Scores of the ten uncorrelated p r i n c i p a l components were used to carry out discriminant analyses. Homogeneity of within-group variance-covariance was the f i r s t underlying assumption examined. Table 8 presents both univariate and multivariate s t a t i s t i c s to test the n u l l hypothesis regarding t h i s aspect. When variances were taken separately for each v a r i a b l e , four PC's were s i g n i f i c a n t at 0.05 l e v e l and one of them, namely PC6, was s i g n i f i c a n t at 0.01 l e v e l . The multivariate test based on B a r t l e t t ' s modification of the l i k e l i h o o d r a t i o was also found to be highly s i g n i f i c a n t (P<0.0001). This indicates that the assumption of homogeneity of dispersion matrices between groups was not multivariately respected. The next assumption examined was normality. Although multivariate normality was not tested, c e r t a i n necessary conditions (e.g., univariate normality) can be v e r i f i e d i n order for i t to not be rejected. In t h i s context, Kolmogorov-Smirnov t e s t s based on L i l l i e f o r s p r o b a b i l i t y d i s t r i b u t i o n s where means and standard deviations are not a p r i o r i assumed were c a r r i e d out on the salmon species for each p r i n c i p a l component. Results presented in Table 9 shows that sockeye was univariate normal i n a l l but one p r i n c i p a l component. On the other hand, coho groups behaved i n a non-normal fashion i n four instances of which three came out s i g n i f i c a n t at the 0.01 l e v e l . Chum and pink salmon both were s i g n i f i c a n t at the 0.05 l e v e l i n two occasions. 61 Table 8. B a r t l e t t ' s t e s t s for homogeneity of within group variance-covariance between canned P a c i f i c salmon (chum, coho, pink, sockeye) f o r the f i r s t ten p r i n c i p a l components. Variable df Chi-square Univariate PCI 3 2.983 PC2 3 8.874 * PC3 3 5.230 PC4 3 7.717 * PC5 3 9.090 * PC6 3 14.800 ** PC7 3 7.533 PC8 3 I 9.085 * P.C9 3 ] 0.87 4 PC10 3 4.440 Mul t i v a r i a t e PC1-10 165 520.476 *** * S i g n i f i c a n t d i f f e r e n c e at 0.05. ** S i g n i f i c a n t d i f f e r e n c e at 0.01. *** S i g n i f i c a n t d i f f e r e n c e at 0.0001. 62 Table 9. Kolmogorov-Smirnov normality t e s t of canned P a c i f i c salmon (crnam, coho, pink, sockeye) within p r i n c i p a l components. Variable Maximum difference Coho Chum Pink Sockeye (n = 18) (n= = 14) (n= =17) (n=13) PCI 0. 294 ** 0. 165 0. 145 0.198 PC2 0. 165 0 . 189 0. 174 0.210 PC3 0. 175 0. 258 * 0. 176 0.207 PC4 0. 124 i 0. 254 * 0. 141 0.239 * PC5 0. 105 0. 203 0. 118 0.184 PC6 0. 230 * 0. 182 0. 141 0.219 PC7 0. 321 ** 0. 222 0. 230 * 0.117 PC8 0. 129 0 . 126 0. 215 * 0 .104 PC9 0 . 158 0. 149 0. 152 0.157 PC10 0. 293 ** 0. 176 0. 108 0 .184 * S i g n i f i c a n t difference at 0.05 l e v e l based on L i l l i e f o r s p r o b a b i l i t y . ** S i g n i f i c a n t difference at 0.01 l e v e l based on L i l l i e f o r s p r o b a b i l i t y . 63 As the normality assumption was v i o l a t e d a number of times and the group dispersion structures were unequal, the conclusions that can be extracted from the t e s t s of s i g n i f i c a n c e and estimated c l a s s i f i c a t i o n error rates which r e l y on parametric c l a s s i f i c a t i o n c r i t e r i a may be biased. Caution should be exercised i n the i n t e r p r e t a t i o n of the sign i f i c a n c e tests based on the parametric d i s t r i b u t i o n s since f a i l u r e to meet the assumptions influences the estimates of variance and a f f e c t s the p r o b a b i l i t y of an hypothesis to be true or f a l s e . Linear discriminant analysis (LDA) was performed with the p r i o r p r o b a b i l i t i e s set proportional to the unequal sample s i z e s . Table 10 summarizes both univariate and multivariate test s t a t i s t i c s . Six of the ten p r i n c i p a l components were s i g n i f i c a n t at or beyond the 5% l e v e l , namely, PCI, PC2, PC3, PC4, PC6, and PC7. These p r i n c i p a l components were important elements of v a r i a t i o n between the salmon species, e s p e c i a l l y PC3, PCI, and PC4, based on t h e i r F r a t i o s . Several v o l a t i l e compounds had high correlations with the important discriminators PCI, PC3, and PC4. Methane t h i o l , ethanol, butane, toluene, 4-ethyl benzenemethanol, and nonanal gave large loadings to PC3 while 2-methyl propanal, 2-methyl butanal, and 2,2-dimethyl propanal provided loadings of substantial magnitude toward PC4. As mentioned e a r l i e r , at least ten v o l a t i l e s dominated PCI (see page 56) . Although PC3 and PCI regrouped large contributors having various functional groups, three aldehydes strongly correlated with PC4. 64 Table 10. Univariate and multivariate t e s t s t a t i s t i c s of discriminant analysis on the f i r s t ten p r i n c i p a l components from the v o l a t i l e s of canned P a c i f i c salmon (chum, coho, pink, sockeye). Variable num df,den df Univariate PCI PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9 PC10 3,58 3, 58 3, 58 3, 58 3, 58 3,58 3,58 3, 58 3,58 3, 58 12.7976 *** 4.0722 * 43.0986 *** 10.1754 *** ,0423 .4860 ** ,5366 * .6748 ,8462 . 9877 Multivariate Wilk's lambda = 0.0230 30,144.5 P i l l a i trace = 2.1258 30,153 Hotelling-Lawley trace = 7.8011 30,143 Roy's largest root = 3.5936 10,51 12.591 *** 12.402 *** 12.395 *** 18.327 *** Canonical LR LR2 = LR, = 0.0230 0.1058 0.3728 30,144.5 18,100 8,51 12.5914 *** 11.5280 *** 10.7248 *** * S i g n i f i c a n t difference at 0.05 l e v e l . ** S i g n i f i c a n t difference at 0.01 l e v e l . *** S i g n i f i c a n t difference at 0.0001 l e v e l . LR stands for Likelihood r a t i o . The multivariate tests including Wilk's lambda, P i l l a i ' s trace, Hotelling-Lawley's trace, and Roy's greatest root were computed using the eigenvalues extracted from Wishart matrices (matrices of corrected sums of squares and products of assumed multinormal v a r i a t e s ) , and constitute various ways to test the hypothesis that the c l a s s means were equal i n the population (Morrison, 1976). A l l highly s i g n i f i c a n t (P < 0.0001), these tests c l e a r l y revealed underlying between-group differences involving the ten variables j o i n t l y . A plot of estimated density d i s t r i b u t i o n s for PC3, the largest contributor toward group separation, i s shown in Figure 11. In t h i s context, normality and homoscedasticity were modalities assumed to be respected. Among the r e s u l t i n g curves equivalent i n form, only two groups were well separated i n the PC3 dimension. The curve describing the sockeye group was located i n the negative range of PC3 and intersected only marginally with the chum salmon curve s i t u a t e d i n the p o s i t i v e end of the scale. Canonical functions were derived to f i n d uncorrelated composites of the 10 PC's that best summarize the d i f f e r e n c e s among the salmon species. The approximate F s t a t i s t i c based on Rao's approximation of the l i k e l i h o o d r a t i o d i s t r i b u t i o n tested the hypotheses that the canonical c o r r e l a t i o n associated with a p a r t i c u l a r discriminant function and a l l succeeding smaller ones were not d i f f e r e n t than zero (Rao, 1973) . Canonical analysis indicated that a l l three canonical variates (CV) were highly s t a t i s t i c a l l y s i g n i f i c a n t at the 0.0001 l e v e l (Table 10). In other words, CV1, CV2, and CV3 were a l l necessary to adequately describe differences between the four salmon species. The standardized c o e f f i c i e n t s or weights for the 66 0.8 - 3 - 2 - 1 0 1 2 3 Principal component 3 Figure 11. Estimated density d i s t r i b u t i o n s of canned P a c i f i c salmon f o r p r i n c i p a l component 3 based on l i n e a r discriminant analysis ( V / sockeye; • , pink; Q / chum; , coho). canonical variates are presented i n Table 11. PC3 with a c o e f f i c i e n t of 1.1161 completely dominated the f i r s t canonical variate CV1. The largest c o e f f i c i e n t of 1.0318 on CV2 was associated with PC4. PC7 and PC6 were also attributed r e l a t i v e l y large c o e f f i c i e n t s on CV2. Due to opposite signs found on the weights of importance, CV2 was e s s e n t i a l l y a contrast between PC4 and PC7 with PC6. The f i r s t two p r i n c i p a l components contributed largely toward CV3. In a l l , the large c o e f f i c i e n t s given to PCI, PC2, PC3, PC4, PC6, and PC7 i n the three discriminant functions corroborated the r e s u l t s of the univariate s t a t i s t i c s i n Table 10. The means of the four groups of salmon species are p l o t t e d i n Figure 12 in a three dimensional canonical discriminant function space. The p l o t s on the 3 facets of the graph are two-dimensional image perspectives brought onto the background planes at right angles. The centroids of the four groups were not equally distant from one another i n the canonical function space, but appeared e i t h e r near or far from each other depending on the observed plane. Graphically, a l l three canonical v a r i a t e s were needed to separate the groups e f f e c t i v e l y . CV1 p r i m a r i l y separated sockeye and coho from pink and chum, while CV2 e s s e n t i a l l y separated coho from the remaining groups. CV3 mainly segregated coho and pink from sockeye and chum. In addition to the depiction of distances between group means, Figure 12 shows isodensity contours of the b i v a r i a t e normal d i s t r i b u t i o n of the canonical v a r i a t e s . Because standardized discriminant function c o e f f i c i e n t s were used to compute scores whose pooled-group covariance matrix was an i d e n t i t y , the isodensity e l l i p s e s about the centroids are c i r c l e s . The c i r c l e s surrounding each centroid represent 68 Table 11. Standardized canonical variate c o e f f i c i e n t s for species dis c r i m i n a t i o n of canned P a c i f i c salmon (chum, coho, pink, sockeye). Dependent va r i a b l e Canonical v a r i a t e 1 2 3 PCI 0. .3737 -0. .0739 0. . 9584 PC2 -0. .3046 0. .4552 0. .6058 PC3 1. .1161 -0, .0774 -0. .0414 PC4 0. . 1530 1. .0318 -0, .1550 PC5 -0, .3106 -0. .2167 0, .3107 PC6 -0, .0001 -0, .6570 -0, .5279 PC7 0. .1366 0. .7845 -0. .1030 PC8 0. .1389 0. .2123 -0. .2935 PC9 -0. .1904 -0. .3118 0. .2469 PC10 -0. .0928 0, .3946 -0. .2370 69 the b i v a r i a t e dispersion of each group at a selected p r o b a b i l i t y l e v e l of 0.50,. that i s , the c i r c l e s are expected to contain approximately 50 percent of the observations i n each group. Some intersections between c i r c l e s indicated that groups were overlapping i n the two dimensional discriminant space. However, the overlaps appeared to be minimal when considered among the three canonical dimensions, and therefore, m i s c l a s s i f i c a t i o n errors were expected to remain low. The Mahalanobis distances were computed using the canonical scores and the p r o b a b i l i t i e s of group membership were then c a l c u l a t e d from these distances. The closer an observation i s to a p a r t i c u l a r group's l o c a t i o n or centroid i n the discriminant space, the more l i k e l y i t i s that i t belongs to the group. The res u b s t i t u t i o n method, which simply y i e l d s the direct sample proportions of m i s c l a s s i f i e d observations ( H i l l s , 1966) was used to cross-tabulate the actual and predicted group membership of salmon species (Table 12). Nearly a l l samples were c o r r e c t l y a l l o c a t e d to t h e i r respective species, except one pink salmon was categorized as coho. The apparent error rate generated was therefore 1.6%. In view of the heterogeneity of variance-covariance previously tested, quadratic discriminant analysis was performed where the i n d i v i d u a l within-group covariance matrices were not pooled i n c a l c u l a t i n g the generalized squared distances. Estimated densities from the quadratic functions were p l o t t e d against PC3 i n Figure 13. The sockeye and chum salmon curves had wider spread d i s t r i b u t i o n s than those of coho and pink but remained r e l a t i v e l y well detached from one another. Computations of 71 Table 12. C l a s s i f i c a t i o n matrix f o r act u a l and predicted group membership of canned P a c i f i c salmon (chum, coho, pink, sockeye) by l i n e a r discriminant analysis using the resubstitution method. Actual group Predicted group Pink Sockeye Coho Chum Pink 16 0 1 0 Sockeye 0 13 0 0 Coho 0 0 18 0 Chum 0 0 0 14 72 1.0 1 1 1 F - 2 - 1 0 1 2 3 Principal component 3 Figure 13. Estimated density d i s t r i b u t i o n s of canned P a c i f i c salmon for p r i n c i p a l component 3 based on quadratic discriminant analysis ( V / sockeye; • , pink; O, chum; -7V, coho) . the quadratic rule on the ten p r i n c i p a l components using the resubstitution method resu l t e d i n the assignment of a l l observations i n t h e i r respective group, hence leading to a 100% of correct c l a s s i f i c a t i o n . The influence of unequal dispersions on m u l t i v a r i a t e s t a t i s t i c a l tests has been studied i n the past. Evidence ind i c a t e d that the e f f e c t s of heterogeneity of dispersions on the tests of s i g n i f i c a n c e depends on both the number of v a r i a b l e s and the group sample s i z e s ( D i l l o n and Goldstein, 1984). Under conditions of heteroscedasticity, s e n s i t i v i t y to rejecting the n u l l hypothesis (mean vectors being equal) increases with increasing number of variables or with increasing disproportion between sample sizes. Although the apparent error rate of the l i n e a r and quadratic discriminant functions did not appear to be d r a s t i c a l l y d i f ferent for the salmon species data, unequal dis p e r s i o n matrices i s a factor known to a f f e c t the c l a s s i f i c a t i o n rule (Marks and Dunn, 1974) . Other factors i n f l u e n c i n g c l a s s i f i c a t i o n include the separation among groups, the number of v a r i a b l e s , and sample s i z e s . Agreement between results obtained from l i n e a r and quadratic equations diminishes as the differences in group dis p e r s i o n increase, the group means become closer, the number of variables increases, or the sample sizes decrease. When the data within each group are not assumed to possess any specified d i s t r i b u t i o n or are assumed to have d i s t r i b u t i o n s d i f f e r e n t from multivariate normal, nonparametric methods - can be considered as alternatives. The nonparametric approach does not constrain the s t a t i s t i c a l d i s t r i b u t i o n i n question to f a l l i n a given parametric family with r i g i d predetermined assumptions of homoscedasticity and normality, but s t i l l r e l i e s on independently c o l l e c t e d observations o r i g i n a t i n g from i d e n t i c a l d i s t r i b u t i o n s . Nonparametric discriminant methods are based on nonparametric estimates of group-specific p r o b a b i l i t y d e n s i t i e s . For the following analyses, the kernel estimator was chosen among the nonparametric methods a v a i l a b l e . The kernel method uses a f i x e d p-dimensional volume by s p e c i f y i n g a radius, r, and estimate the group densities at each observation vector (Hand, 1982). The parameter r also has the property of determining the degree of i r r e g u l a r i t y or smoothness in the estimation of the density function. Small values of t h i s smoothing parameter produce jagged contours while large values obscure and broaden the curve d e f i n i t i o n of kernel estimates. Although various methods for se l e c t i n g the smoothing parameter have been suggested, there i s as yet no simple solution to t h i s problem. One way to choose the smoothing parameter i s to plot c l a s s i f i c a t i o n error rates with d i f f e r e n t r values and select the estimate which minimizes the m i s c l a s s i f i c a t i o n rate (Hand, 1982). Over the years, a number of kernel functions have been developed to estimate the d e n s i t i e s . Silverman (1986) compared the e f f i c i e n c i e s of some established kernels and found very l i t t l e d i f f e r e n c e between them as a l l obtained e f f i c i e n c y values close to one. The Epanechnikov kernel was applied to the p r i n c i p a l components extracted from the salmon species dataset i n t h i s study. In Figure 14, the behaviour of the error count estimate was followed over a range of the smoothing parameter that spanned from 2 to 18. Based on the re s u b s t i t u t i o n method (straightforward 75 Figure 14. Comparison of two error count estimation methods against the smoothing parameter of the Epanechnikov kernel c l a s s i f i e r on p r i n c i p a l component scores of canned P a c i f i c salmon (chum, coho, pink, sockeye) v o l a t i l e s . r e c l a s s i f i c a t i o n ) to c a l c u l a t e the er r o r rate, the three f i r s t r values generated no c l a s s i f i c a t i o n e r r o r . Although simple in nature, the apparent error rate computed by r e s u b s t i t u t i o n has been c r i t i c i z e d i n the past as being over-optimistic (Hand, 1982) . The bias originates from the fact that the c l a s s i f i e r has been designed to optimize some function of the dataset and therefore performs better on the o r i g i n a l observations than on other s i m i l a r l y generated datasets. Another method for error rate estimation introduced by Lachenbruch (1967) i s the cross-tabulation or U-method. Also s e l f - e x p l a n a t o r i l y r e f e r r e d to as the "leave-one-out" method, i t consists of holding out one observation at a time, estimating the discriminant function on n - l observations, and c l a s s i f y i n g the held-out observation. This process i s repeated for each observation, and achieves a nearly unbiased estimate but with a r e l a t i v e l y large variance (Glick, 1978) . When cro s s - v a l i d a t i o n was applied to evaluate the kernel performance, i t was found that the three f i r s t smoothing parameters mentioned e a r l i e r (r = 2,3,4) yielded much larger e r r o r counts as opposed to those from re s u b s t i t u t i o n (Figure 14). On the other hand, the next three r values (r = 5,6,7) gave the lowest error rate of 0.0161 which was also the same as that obtained with r e s u b s t i t u t i o n . In fact, the er r o r rates associated with the smoothing parameters ranging from 5 to 18 were i d e n t i c a l for both estimation methods. The estimated density d i s t r i b u t i o n s computed using the Epanechnikov kerne-1 with r=6 f o r p r i n c i p a l component 3 are shown i n Figure 15. Compared to the previous r e s u l t s based on parametric discriminant analyses, the curves possess s i m i l a r mean positions, a high degree of 77 -6 -4 -2 0 2 4 6 Principal component 3 Figure 15. Estimated density d i s t r i b u t i o n s of canned P a c i f i c salmon for p r i n c i p a l component 3 based on non-parametric discriminant analysis ( V/ sockeye; • , pink; Q ' chum; -fc, coho). 78 symmetry, but wider spread consequently causing more overlapping between a l l salmon groups. The species averages i n terms of estimated density had the same ranking order (sockeye, chum, coho, and pink) for the quadratic and kernel methods. However, the Epanechnikov kernel produced rounder contours of the b e l l shape curves. In order to make an appropriate comparison between the above three discriminant analyses, er r o r count estimates were c a l c u l a t e d by both resubstitution and c r o s s - v a l i d a t i o n for the salmon species (Table 13) . The c r o s s - v a l i d a t i o n error estimate for l i n e a r discriminant analysis was 11.29% while i t was 1.61% when computed using r e s u b s t i t u t i o n . There was an even more d r a s t i c difference between c r o s s - v a l i d a t i o n and resubstitution for quadratic discriminant analysis (19.3% and 0.0%, respectively). No d i f f e r e n c e was found for the Epanechnikov kernel method as was shown i n Figure J ^ * Results c l e a r l y indicated the danger of relying on the r e s u b s t i t u t i o n method for error rate c a l c u l a t i o n as i t generally provides a biased l e v e l of confidence. The rank i n increasing order on the basis of c r o s s - v a l i d a t i o n error count estimate for the discriminant analysis methods was kernel, l i n e a r , and quadratic. This sequence i s i n accord with that of Van Ness and Simpson (1976) who studied the r e l a t i o n s h i p between dimensionality, error rate, and sample size for d i f f e r e n t types of c l a s s i f i e r . Besides the absence of the r e s t r i c t i v e parametric assumptions, they also a t t r i b u t e d the s u p e r i o r i t y of the kernel- over l i n e a r discriminant methods to the s t a b i l i t y of the kernel estimator i n high dimensionality. The l i n e a r estimator performed r e l a t i v e l y well despite the non-normality and heteroscedasticity of the 79 Table 13. Comparison of error count estimation methods for d i f f e r e n t discriminant analysis (DA) applied to the f i r s t ten p r i n c i p a l components from canned P a c i f i c salmon (chum, coho, pink, sockeye) v o l a t i l e s . DA func t i o n 3 Error count estimate b Coho Chum Pink Sockeye % Total (n=18) (n=14) (n=17) (n=13) (n=62) Resubstitution l i n e a r 0 .000 0.000 0.059 0.000 1.6 (0) (0) (1) (0) quadratic 0.000 0.000 0 .000 0.000 0.0 (0) (0) (0) (0) nonparametric 0.000 0.000 0.059 0.000 1.6 (0) (0) (1) (0) Cross-validation l i n e a r 0.167 0.000 0.176 0.077 11.3 (3) (0) (3) (1) quadratic 0.056 0.143 0.118 0.538 19.4 (1) (2) (2) (7) nonparametric 0.000 0.000 0.059 0.000 1.6 (0) (0) (1) (0) Pr i o r prob. 0.290 0.226 0 .274 0.210 The nonparametric method consists of the Epanechnikov kernel with r=6. The numbers i n parentheses are the m i s c l a s s i f i c a t i o n s associated with the error count estimates. 80 species data. Lachenbruch et a l . (1973) investigated the performance of lin e a r discriminant function applied to some multivariate non-normal d i s t r i b u t i o n s . They concluded that results i n such situ a t i o n s can be misleading though the o v e r a l l c l a s s i f i c a t i o n error rates were usually not severely affected. In p a r t i c u l a r , the i n d i v i d u a l group error rates were di s t o r t e d i n that error rates were generally much larger than the optimal value for one population group, and much smaller f o r the other. Attempts to adjust f o r unequal dispersion matrices by use of the quadratic rule proved to be i n e f f e c t i v e . Sockeye salmon e s p e c i a l l y received a high rate of i n c o r r e c t c l a s s i f i c a t i o n . The poor performance of the quadratic method could be a t t r i b u t e d to sample covariance matrices being'' poor estimates i n presence of large dimensionality/sample size r a t i o s (Van Ness and Simpson, 1976). Therefore, larger sample sizes could improve the performance of the quadratic functions. 2 . MVA o f v o l a t i l e s from canned chum salmon a t t h r e e 3tages o f s e x u a l m a t u r i t y Data on the forty-four GC peaks i d e n t i f i e d e a r l i e r were gathered from samples of adult chum salmon i n three d i f f e r e n t ranges of maturity during t h e i r spawning migration. Only quantitative differences i n the GC patterns could be found between the samples as no appearance or disappearance of peaks were observed. This s i t u a t i o n was comparable to the previous species study and therefore a s i m i l a r multivariate approach was undertaken to analyze and extract information about the structure of the dataset. 81 Pr i n c i p a l component analysis was c a r r i e d out on the c o r r e l a t i o n matrix. Table 14 shows the f a c t o r loadings of the f i r s t eight PC's that had a latent root higher than 1.0. Thirteen peaks were all o c a t e d loadings of 0.8 and above on PCI. Of these t h i r t e e n peaks, eight of them had correspondingly high and p o s i t i v e loadings on the f i r s t p r i n c i p a l component of the species dataset (Table 7). They were peaks 1 (hydrogen s u l f i d e and acetaldehyde), 5 (3-methyl-l-butene), 16 (heptane), 17 (1,5-dimethyl cyclopentene), 29 (unknown 7), 30 (nonane), 32 (benzaldehyde), and 38 (2-pentyl furan) . PC2 highly correlated with three peaks: 14 (unknown 1) , 22 (3-hexanone) and 34 (unknown 9) . Again, two of these peaks (14 and 34) loaded high c o e f f i c i e n t s on PC2 i n the analysis of the salmon species data (Table 7) . Compared to the e a r l i e r study, peak 14 also contrasted with peak 34 based on t h e i r opposite algebraic signs but in a reverse manner. Forty-one percent of the t o t a l variance was explained by PCI and d i d not d i f f e r markedly from the respective value of 39.0% found i n the e a r l i e r analysis. The rate at which the variance explained by the succeeding p r i n c i p a l components decreased, was also a common aspect between the species and sexual maturity studies. As a r e s u l t , the variance cumulated over the f i r s t eight components was 84.2%, and corresponded well with the fig u r e of 87.2% obtained previously. Notwithstanding the p a r a l l e l drawn between the two studies, the s i m i l a r i t i e s i n component loadings were found mainly with the two f i r s t PC's and the remaining p r i n c i p a l components showed more s p e c i f i c i t y to t h e i r respective dataset. 82 T a b l e 14 . C o r r e l a t i o n ( l o a d i n g s ) o f g a s c h r o m a t o g r a p h i c p e a k v a r i a b l e s f r o m c a n n e d chum s a l m o n t e s t e d a t t h r e e s e x u a l m a t u r i t y s t a g e s w i t h t h e f i r s t e i g h t p r i n c i p a l c o m p o n e n t s . P e a k P r i n c i p a l c o m p o n e n t 1 2 3 4 5 6 7 8 1 0. ,829 -0. 292 -0. 052 -0. 299 -0. 086 -0. 063 0. 113 -0. 170 2 0. ,596 0. 394 0. 083 -0. 548 -0. ,260 -0. 126 -0. 049 0. 055 3 -0. .353 0. 084 -0. 049 0. 341 0. ,486 0. ,478 0. 218 0. 015 4 0. ,779 0. 068 -0. 027 -0. 532 0. ,067 -0. 053 -0. 031 0. 046 5 0. ,808 -0. 427 -0. 089 -0. 010 -0. 088 -0. 033 0. 254 -0. 078 6 0. ,245 0. 269 -0. 315 0. 214 -0. 562 0. 214 -0. 069 0. 257 7 0. ,341 0. 382 0. ,725 -0. ,301 -0. ,056 -0. .044 0. ,180 0. 124 8 0. ,479 -0. 265 0. 116 -0. ,186 0. ,265 0. 581 0. ,098 0. 023 9 0. .790 0. 316 0. ,106 -0. ,430 -0. ,085 -0. ,002 0. .101 -0. ,067 10 0. .641 0. ,325 0. ,177 0. ,126 -0. ,001 0. ,092 0. .289 0. 229 11 0. ,073 0. 131 0. ,909 -0. ,044 0. ,159 -0. ,095 0. ,090 0. ,148 12 0. .219 0. .573 -0. ,100 0. ,155 -0. ,501 0. ,327 0. .140 -0. ,147 13 0. .012 0. ,527 0. .483 0. ,212 0. ,549 -0. ,032 0. ,084 0. ,141 14 0. ,290 -0. 672 0. 349 -0. ,351 -0. ,237 0. ,191 0. .084 0. ,155 15 0. .767 -0. .412 -0. ,318 -0. ,171 0. .132 0. ,154 -0. .029 0. .079 16 0. .937 0. ,030 -0. .101 -0. .117 -0. .043 0. ,013 0. .188 -0. .067 17 0. .819 0. .148 0. .015 0. .363 0. .247 -0. .009 0. .038 -0. .095 18 0. .112 0. ,097 0. ,154 0. .173 -0. .217 0. .022 0. .005 0. .131 19 0. .025 -0. .038 0. .544 -0. .067 -0, .124 -0. .026 0. .162 0. .281 20 0. .656 0. ,091 -0. .309 -0. ,526 -0, .061 -0, .027 0, .149 0, .003 21 0, .465 0. .394 -0. .027 -0. .164 0. .492 -0. .001 -0. .310 0. .030 22 0. .076 0. .717 -0, .022 0. .164 -0, .270 0, .467 -0, .003 -0, .121 23 0, .411 0, .515 -0, .116 -0. .013 0, .297 -0. .078 0 .195 0. .039 24 0. .696 0. .180 -0, .388 -0. .237 0. .322 0. .023 -0, .083 0. .158 25 0. .666 -0, .274 -0, .354 -0, .126 0, .236 -0, .044 0 .027 0 .220 26 0.798 0, .339 -0, .147 -0, .023 0 .004 -0, .027 -0 .107 -0, .093 27 0. .886 0, .016 -0, .087 0, .224 -0, .174 -0. .106 0, .196 -0, .109 28 0. .916 0. .033 -0 .065 0, .228 -0, .115 -0, .125 0 .176 -0 .110 29 0 .840 0, .191 -0, .194 0, .310 -0 .054 -0, .129 0 .074 -0, .113 30 0 .889 0 .051 -0 .003 0 .154 0 .228 0 .029 . 0 .069 -0 .168 31 0.763 -0, .071 0, .164 -0, .149 0 .347 0, .190 -0 .253 -0 .050 32 0 .911 -0 .200 0 .086 0 .067 -0 .183 -0 .049 -0 .023 -0 .042 33 0 .588 0 .534 0 .156 -0 .040 -0 .123 -0 .093 -0 .342 0 .111 34 0 .299 0, .775 -0, .327 -0, .029 -0 .089 -0 .075 -0 .073 -0 .069 35 0 .845 -0 .050 -0 .225 0 .024 0 .013 0 .217 -0 .147 0 .197 36 0 .709 -0 .504 -0 .083 0 .352 -0 .014 0 .010 -0 .116 0 .176 37 0 .711 -0, .412 -0 .139 0 .338 -0 .155 0 .040 -0 .188 0 .225 38 0 .808 -0 .190 0 .189 0 .458 -0 .084 -0 .083 -0 .098 0 .038 39 0 .739 0 .379 0 .171 0 .382 0 .200 -0 .137 -0 .062 -0 .096 40 0 .131 0, .277 0 .417 -0, .056 -0 .305 0 .188 -0 .340 0 .013 41 0 .836 -0 .320 0 .296 -0 .059 -0 .010 0 .099 -0 .155 -0 .029 42 0 .610 -0 .567 0 .359 0 .138 -0 .055 0 .003 -0 .041 -0 .170 43 0 .734 -0 .184 0 .387 0 .371 -0 .022 -0 .122 -0 .064 -0 .103 44 -0 .039 -0 .197 0 .535 -0 .270 0 .064 0 .176 -0 .146 -0 .565 n t r o o t 18 .060 '5 .522 4 .461 3 .051 2 .510 1 .315 1 .066 1 .059 a n c e (%) 41 .046 12.55 10 .139 6 .933 5 .705 2 .989 2 .423 2 .406 83 The scores of the second p r i n c i p a l component were pl o t t e d against those from PCI i n Figure 16. No d e f i n i t e group separation could be distinguished along the PCI axis. Beside an expression of general va r i a t i o n common to a l l groups, t h i s dimension d i d not suggest any evident interpretation. On the other hand, PC2 d i s t i n c t l y segregated samples from year 1 and 2. This p r i n c i p a l component, car r y i n g high loadings for peaks 11, 14, and 34, r e f l e c t e d a form of year to year v a r i a b i l i t y . As expressed i n the species study, s l i g h t but notable decreases i n peak resolution were also observed i n the three mentioned GC areas. Although not disproving the explanation inherent to the b i o l o g i c a l nature of the chum salmon samples themselves, the separation of the year-groups along PC2 was l i k e l y due to chromatographic factors since a comparable s i t u a t i o n emerged in a l l species (Figure 10). Linear discriminant analysis was performed on the f i r s t eight p r i n c i p a l components extracted from the chum salmon data. P r i o r p r o b a b i l i t i e s were s p e c i f i e d to be proportional to the sample siz e s . Five out of eight p r i n c i p a l components were found s i g n i f i c a n t at the 5% l e v e l or below as indicated i n the upper portion of Table 15 dealing with the univariate s t a t i s t i c s . The four largest contributors toward group discrimination were PC3, PC5, PC6, and PC4 i n decreasing order. Although they account for less variance of the dataset than the i n i t i a l p r i n c i p a l components, PC3-PC6 were selected over PC1-PC2 for group discrimination. This s i t u a t i o n underlined the danger of r e j e c t i n g components accounting for small amount of v a r i a t i o n when they are to be used i n subsequent data CM -4—< c o CL 1 -0 -E -1 -2 -3 fib1 B B O A A C <j£c B 0 A \ C c | A B BA B A B B B Year 1 B B A B Year 2 F D D F D D 9 D^F F D E & D E F C E E E D -2 -1 0 1 Principal component 1 Figure 16. Plot of the f i r s t two p r i n c i p a l component scores for the canned chum salmon tested at three sexual maturity stages (A, si l v e r - b r i g h t / y e a r 1; B, semi-bright/year 1; C, dark/year 1; D, s i l v e r - b r i g h t / y e a r 2; E, semi-bright/year 2; F, dark/year 2). 00 85 Table 15. Univariate and multivariate t e s t s t a t i s t i c s of discriminant analysis on the f i r s t eight p r i n c i p a l components from v o l a t i l e s of canned chum salmon tested at three sexual maturity stages. Variable num df,den df Univariate PCI 2, 91 3. 4606 * PC2 2, 91 1. 1410 PC3 2, 91 31. 9029 *** PC4 2, 91 13. 6433 * * PC5 2, 91 23. 1504 *** PC6 2, 91 16. 6785 *** PC7 2, 91 0. 1749 PC8 2, 91 0. 8634 Multivariate Wilk's lambda = 0.0726 16,168 P i l l a i trace = 1.3659 16,170 Hotelling-Lawley trace = 6.7338 16,166 Roy's largest root = 5.6683 80,85 28.4686 *** 22.8872 *** 34.9318 *** 60.2256 *** Canonical LR1 = 0.0726 LR, = 0.4841 16,168 7, 85 28.4686 *** 12.9388 *** * S i g n i f i c a n t difference at 0.05 l e v e l . ** S i g n i f i c a n t difference at 0.01 l e v e l . *** S i g n i f i c a n t difference at 0.0001 l e v e l , LR stands for Likelihood r a t i o . 86 treatments. Sometimes only the largest p r i n c i p a l components are retained as predictors. However, as Dunteman (1984) pointed out, "There i s no compelling reason to suspect that larger components are better predictors of some dependent variable than smaller components." PC3 and PC5, the p r i n c i p a l components with high F r a t i o s , received high loadings from 2-methyl propanal and benzene, and from ethanol, dimethyl s u l f i d e , 2-methyl butanal, 2,2-dimethyl propanal, and a c e t i c acid, respectively. Ethanol also had a high loading with PC6 along with hexane and toluene while PC4 gathered c o r r e l a t i o n s above 0.5 from methane t h i o l , butane, and unknown compound 2. The l a s t three v o l a t i l e s as well as two others from PC5 possessed negative loading values. Among these v o l a t i l e s , methane t h i o l and dimethyl s u l f i d e are compounds that could be generated from sulfur-containing amino acids. As spawning migration progresses, salmon are using t h e i r non-collagenous and c e l l u l a r protein supplies and are not replenishing them (Love, 1970; Aksnes et a l . , 1986). The u t i l i z a t i o n of amino acids for metabolic processes could therefore decrease the content of sulfur-containing amino acids a v a i l a b l e for the production of methane t h i o l and dimethyl s u l f i d e . Ethanol, used i n the past as an indicator of decomposition, was also influenced by the changes of sexual maturity. As a consequence, sexual maturity of chum salmon i s a f a c t o r that could bias the r e s u l t of q u a l i t y determination when established on the basis of ethanol q u a n t i f i c a t i o n only. The other species of P a c i f i c salmon should be investigated with t h i s regard i n order to ascertain the q u a l i t y c l a s s i f i c a t i o n of canned salmon of progressing sexual maturity. 87 In terms of m u l t i v a r i a t e s t a t i s t i c s of the l i n e a r discriminant analysis, a l l tests were s i g n i f i c a n t at P<0.0001 (Table 15). They indicated that the p r i n c i p a l components taken together affirmed the presence of between-group d i f f e r e n c e s . Canonical analysis shown at the bottom part of Table 15 revealed that the two possible canonical functions were necessary to capture the differences among the three maturity stages of chum salmon as judged by the high levels of s i g n i f i c a n c e computed (P<0.0001). Table 16 presents the standardised c o e f f i c i e n t s a l l o c a t e d to each p r i n c i p a l component f o r both canonical v a r i a t e s . Three of the important factors explaining the between-group variables, found i n Table 15, i . e . , PC3, PC4, and PC5, occupied predominant posi t i o n s having large c o e f f i c i e n t s for CV1. On the other hand, the second canonical variate was monopolized by PC6. The graphical representation of the scores of the two canonical functions i s shown i n Figure 17. Scores of the three ] maturity stages, namely s i l v e r - b r i g h t , semi-bright, and dark, were d i s t i n c t l y regrouped. The centroids of each group were not p l o t t e d but the density c i r c l e s about the centroids of the b i v a r i a t e dispersions were drawn at the p r o b a b i l i t y l e v e l of 0.50. Group memberships were computed from the Mahalanobis distances of the canonical scores. Two samples i n each of the s i l v e r - b r i g h t and semi-bright maturity stages were m i s c l a s s i f i e d using the r e s u b s t i t u t i o n method to c a l c u l a t e the e r r o r count estimate. Thus, t h i s l i n e a r discriminant analysis y i e l d e d an apparent error rate of 3.2%. 88 Table 16. Standardized canonical variate c o e f f i c i e n t s for discrimination of canned chum salmon tested at three sexual maturity stages. Dependent variable Canonical variate 1 2 PCI -0.4833 -0.3971 PC2 0.4282 -0.0476 PC3 1.5820 , 0.5809 PC4 1.0707 -0.5645 PC5 1.5627 -0.2735 PC6 -0.0487 1.0246 PC7 0.0973 0.1009 PC8 0.0703 -0.2654 Figure 17. Plot of the two canonical variate scores for canned chum salmon tested at three sexual maturity stages (S, s i l v e r bright; B, semi-bright; D, commercial dark). CO Figure 17 can also be i n t e r p r e t e d from a d i f f e r e n t point of view. In addition to the three d i s t i n c t i v e areas where the groups had dispersed, a U-shape pattern also emerged: s i l v e r - b r i g h t at the bottom-right, semi-bright at the top-centre, and dark at the bottom-left. During the spawning migration of salmon, which begins i n the ocean and ends at the spawning beds, gradual p h y s i o l o g i c a l changes occur. This process of sexual maturation proceeds i n a continuous manner. However, categorization of salmon i n d i f f e r e n t stages based on physical appearance i s currently done by processors to e s t a b l i s h q u a l i t y grading systems. In l i g h t of t h i s , the r e s u l t s of Figure 17 can also be viewed as an uninterrupted sequence from the s i l v e r - b r i g h t stage to the commercial dark stage behaving according to a quadratic function with a maximum inve r s i o n point at the semi-bright stage. Before carrying out discriminant analysis, B a r t l e t t ' s t e s t s f o r homogeneity of variance were computed on the f i r s t ten p r i n c i p a l components of chum maturity data and r e s u l t s are found i n Table 17. Five p r i n c i p a l components out of eight d i d not show any s t a t i s t i c a l sign of heteroscedasticity across the maturity stages. However, the chi-square values associated with PC3 presented a s i g n i f i c a n t difference at the 0.0001 l e v e l while PC2 and PC8 were s i g n i f i c a n t at the 1% p r o b a b i l i t y . The multivariate test was also highly s i g n i f i c a n t (P<0.0001) i n d i c a t i n g that the p r i n c i p a l components, when considered i n a joi n t fashion, were not able to s a t i s f y the parametric assumption of homogeneity of variance. 91 Table 17. B a r t l e t t ' s tests for homogeneity of within group variance-covariance between canned chum salmon tested at three sexual maturity stages f o r the f i r s t eight p r i n c i p a l components. Variable df Chi-square Univariate PCI 2 0.290 PC2 2 10.872 ** PC3 2 30.027 *** PC4 2 0.020 PC5 2 5.460 PC6 2 2.769 PC7 , 2 3.426 PC8 2 10.724 ** Mult i v a r i a t e PC1-10 72 302.937 *** * S i g n i f i c a n t difference at 0.05. ** S i g n i f i c a n t difference at 0.01. *** S i g n i f i c a n t difference at 0.0001. 92 Results of the Kolmogorov-Smirnov normality t e s t s on the maturity data are shown i n Table 18. Scores of four out of eight p r i n c i p a l components departed from normality for s i l v e r - b r i g h t chum but only PC2 and PC3 were in a s i m i l a r s i t u a t i o n f or semi-bright chum. In addition, the normality assumption was concurrently respected by a l l p r i n c i p a l components for dark chum. In order to respond to the heterogeneity of variance-covariance found between the chum salmon maturity stages, quadratic discriminant analysis was c a r r i e d out on the same eight p r i n c i p a l components. The few cases of non-conformity to the assumption of normality also prompted the re-analysis of the sexual maturity data with non-parametric discriminant analysis using the Epanechnikov kernel. Figure 18 compared the e r r o r count estimate methods, res u b s t i t u t i o n and c r o s s - v a l i d a t i o n , with regard to the smoothing parameter of the kernel estimator. In accordance to what was found i n the section on the analysis of the species data, the resubstitution method underestimated the error rates f o r small values of r but eventually reached perfect agreement as the smoothing parameter increased. The lowest error rate estimate along the c r o s s - v a l i d a t i o n curves was associated with a r of 5. At that point, however, there was a discrepancy between the two error estimation methods. A l b e i t a s l i g h t difference, i t i n d i c a t e d that the error rates estimated by the two methods may not always be concordant at the optimal c r o s s - v a l i d a t i v e conditions. The error count r e s u l t s computed using both estimation methods for a l l three discriminant analyses were compiled i n Table 19. Based on 93 Table 18. Kolmogorov-Smirnov normality t e s t of canned chum salmon tested at three sexual maturity stages within the f i r s t eight p r i n c i p a l components. Variable Maximum difference S i l v e r - b r i g h t (n=34) Semi-bright (n=36) Dark (n=24) PCI 0.119 0.117 0.102 PC2 0.209 * * 0.190 ** 0.144 PC3 0.160 * 0.154 ** 0.157 PC4 0.177 * * 0.087 0.085 PC5 0.102 0.105 0.163 | PC6 0.160 * 0.114 0.104 PC7 0.105 0.080 0.147 PC8 0.161 0.087 0.169 * S i g n i f i c a n t difference at 0.05 l e v e l based on L i l l i e f o r s p r o b a b i l i t y . ** S i g n i f i c a n t difference at 0.01 l e v e l based on L i l l i e f o r s p r o b a b i l i t y . 94 0.7 0 5 10 15 2 0 r Figure 18. Comparison of error count estimate method against the smoothing parameter of the Epanechnikov kernel c l a s s i f i e r on p r i n c i p a l component scores of canned chum salmon tested at three sexual maturity stages. 95 Table 19. Comparison of er r o r count estimation methods for d i f f e r e n t discriminant a n a l y s i s (DA) functions applied to the f i r s t eight p r i n c i p a l components from canned chum salmon tested at three sexual maturity stages. DA func t i o n 3 Error count estimate b S i l v e r - b r i g h t Semi-bright Dark % Total (n=34) (n=36) (n=24) (n=94) Resubstitution l i n e a r 0.059 0.056 0.000 4.3 (2) (2) (0) quadratic 0.059 0.000 0.000 2.1 (2) (0) (0) nonparametric 0.088 0.000 0.000 3.2 (3) (0) (0) Cross-validation l i n e a r 0.059 0.083 0.000 5.3 (2) (3) (0) quadratic 0.059 0.056 0.000 4.3 (2) (2) (0) nonparametric 0.118 0.000 0.000 4.3 (4) (0) (0) P r i o r prob. 0.362 0.383 0.255 a The nonparametric method consists of the Epanechnikov kernel with r=5. The numbers i n parentheses are the m i s c l a s s i f i c a t i o n s associated with the erro r count estimates. resubstitution, the quadratic and kernel analyses improved the error rates compared to those obtained by the l i n e a r functions, but the differences were small. A l l t o t a l e r r o r rates c a l c u l a t e d with the c r o s s - v a l i d a t i o n method were larger than that of r e s u b s t i t u t i o n thereby bringing again forward the bias of the l a t t e r method. Yet the two sets of estimated error rates were not as disparate as those found with the species data (Table 13) . In terms of c r o s s - v a l i d a t i o n , the t o t a l error rate was lower for the non-parametric kernel estimator and the quadratic than for the l i n e a r discriminant functions. The error counts nevertheless remained i n the 4 to 5% region. These r e s u l t s therefore suggested that even though the parametric assumptions d i d not hold true, the m i s c l a s s i f i c a t i o n rates could s t i l l be kept low. Besides dark chum and two instances for semi-bright chum, most cases of non-normality were found for s i l v e r - b r i g h t chum salmon data (Table 18) . Since PC2 and PC8 were not involved i n group separation for l i n e a r discriminant analysis as indicated by a nonsignificant univariate F r a t i o (Table 15), the strong non-normality of these components d i d not bear as much negative influence on the membership c l a s s i f i c a t i o n r e s u l t s as PC3, PC4 and PC6 seemingly could have. Hence, membership c l a s s i f i c a t i o n could have r e f l e c t e d a c e r t a i n s t a b i l i t y or robustness to a few non-normal anomalies. In addition, PC3 being the only major discriminator a f f l i c t e d by heterogeneity of variance was expected to negatively a f f e c t the r e s u l t s of LDA (Table 17) . On one hand, an explanation among others already discussed i n the section on the salmon species data refers to the distance between group means. As the distances 97 between group centroids and t h e i r overlaps i n the multivariate space increases, the e f f e c t s of heteroscedasticity on m i s c l a s s i f i c a t i o n rates diminishes (D i l l o n and Goldstein, 1984) . In these circumstances, l i n e a r and quadratic discriminant functions may appear to be equally e f f e c t i v e . On the other hand, the two samples of s i l v e r - b r i g h t chum that were categorized as dark chum (Figure 17) could have been o u t l i e r s with c h a r a c t e r i s t i c s uncommon to t h e i r group. If they were removed from the analysis, the number of rejections of the parametric assumptions could l i k e l y be decreased and thereby help to account for the small differences between discriminant functions. 3. MVA of volatiles from canned pink salmon during refrigerated decomposition Unlike the previous SHGC chromatograms obtained for the species and sexual maturity studies, the patterns of v o l a t i l e s from canned pink salmon during the storage periods possessed q u a l i t a t i v e as well as quantitative differences. Figure 19 shows t y p i c a l chromatograms from canned pink salmon at day 0, day 8, and day 13 of year 1 which r e s p e c t i v e l y corresponded to grade A, B, and reject, as determined p r i o r to canning. Peaks 3 (ethanol), 18 (3-methyl-l-butanol), and 19 (2-methyl-2-butenal) have been l a b e l l e d and were some of the v o l a t i l e s that storage time influenced. Their peak sizes increased with time, and the sensory grade associated with these t y p i c a l samples sequentially progressed from grade A to grade r e j e c t . The peak areas of these three v o l a t i l e compounds have been translated into actual concentrations using standard curves. The ranges of concentrations within which a l l measurements f e l l , are presented 98 3 • i Jiii 1 • A OJLTVJI—• • 1 u 1 *v— B ..Ll. 1 A > ie 10 / 1 1 1 1 Cj-,i i — L _ 0 1 2 3 S 7 9 11 13 13 17 19 Uncorrected retention time (min) Figure 19. Chromatogram of v o l a t i l e s from canned pink salmon during the refr i g e r a t e d storage of year 1. (A, day 0/grade A; B, day 8/ grade B; C, day 13/grade r e j e c t ) . for each grade l e v e l i n Table 20. With changing grade l e v e l s , both the increase i n concentration and the spread of the ranges appeared to augment in a non-linear manner. Peak 18 (3-methyl-l-butanol) was the only one, among 44 analyzed v o l a t i l e s , that was not detected i n grade A canned pink salmon. Regarding the comparison of the obtained values for ethanol with those of Hollingworth et a l . (1986), a 1000-fold factor i n scale could be observed. The main explanation l i e s i n the sample preparation and analysis. Results of t h i s study were obtained using an automated method sampling headspace v o l a t i l e s above heated salmon flakes while Hollingworth et a l . (1986) manually i n j e c t e d headspaces of l i q u i d samples e q u i l i b r a t e d at room temperature. Common factor analysis (CFA) with varimax rotation was the f i r s t multivariate method performed on the 44 v o l a t i l e compounds analyzed throughout the re f r i g e r a t e d storage study. As Table 21 indicates, nine factors were allocated eigenvalues of 1.0 and over, explaining nearly 88% of the common variance. The v o l a t i l e s outlined i n Figure 19, i . e . , peaks 3, 18, and 19, a l l received loadings above 0.9 for factor 3. The standardized peak area of these three v o l a t i l e s were p l o t t e d i n Figure 20 for both years of the r e f r i g e r a t e d storage study. Constant values over a period of time, followed by an exponential increase was a pattern common to a l l peaks. Peak 18 (3-methyl-l-butanol) however was d i s t i n c t i n that i t was undetected up to a c e r t a i n point i n time, day 8 for year 1 and day 13 for year 2. By the time peak 18 appeared on the chromatogram, the -fish from which the samples o r i g i n a t e d had consistently been c l a s s i f i e d as grade B. 100 Table 20. Concentration ranges of three v o l a t i l e compounds from canned pink salmon of d i f f e r e n t q u a l i t y grades. 3 b ~ 3 Grade Concentration (xlO ppm) Ethanol 0 3-methyl-l-butanol d 2-methyl-2-butenal e A 2. .6 - 37.0 - 0. .0 - 5.0 B 31. .4 - 210.3 1.7 - 13.7 3. .6 - 16.1 reject 263, .7 - 878.9 18.7 - 128.0 19. .3 - 138.7 a Measurements obtained using a headspace sampler with a 3 mL loop. b Grade of canned salmon based on assessment p r i o r to canning. c FID response = 2920.2 x [ethanol (ng/mL)], r 2 = 0.997, n = 22, S e r r = 24.2. d FID response = 1097.0 x [3-ME-l-butanol (ng/mL)], r 2 = 0.973, n = 15, S e r r = 28.2. e FID response = 364.9 x [2-ME-2-butenal (ng/mL)], r 2 = 0.984, n = 21, S e r r = 6.2. 101 T a b l e 2 1 . L o a d i n g s o f t h e f i r s t n i n e v a r i m a x r o t a t e d f a c t o r s f r o m f a c t o r a n a l y s i s o f c a n n e d p i n k s a l m o n v c l a t i l e s o f t h e r e f r i g e r a t e d s t o r a g e s t u d y . P e a k F a c t o r s 1 2 3 4 5 6 7 8 9 1 0. .372 0. ,674 0. ,080 0. ,096 0. ,019 0. 182 0. 105 -0. ,143 -0. 131 2 0, .232 -0. .032 -0. ,368 0. .134 0. ,242 -0. 072 -0. .195 0. ,581 0. 184 3 0. .145 0, .175 0. .904 0, .255 0. .086 -0. 027 0. ,057 -0. .135 0. ,005 4 0. .530 0. .466 -0. ,041 0. .248 0. ,430 0. 180 0. ,013 0. ,097 0. ,022 5 0. .297 0. .826 0. .243 0, .258 -0, .039 0. 111 0. ,181 -0. .065 -0. ,016 6 0. ,141 0. ,122 -0. .066 -0. .097 -0. ,191 -0. 087 0. ,123 -0. ,015 0. 003 7 0. ,285 0. .270 0. ,156 0, .868 0. ,069 0. 007 -0. ,035 0. .080 0. ,085 8 -0. ,053 0. ,115 -0. ,050 0. .024 0. .157 0. ,898 0. ,029 -0. .007 -0. .031 9 0. .582 0. .387 -0. .022 0. .387 0. .301 -0. ,399 -0. ,079 0. .219 0. ,029 10 0. .585 0. .222 0, .264 0, .346 0. .306 -0. ,059 -0. .030 0. .220 0. .029 11 0. .168 0. ,173 0. .306 0. .876 0. .187 -0. ,015 0. ,013 -0. .110 0. ,048 12 0. .013 0. .057 -0. .002 -0, .052 0. .163 -0. .045 0. .044 0. .192 0. .132 13 0. .176 0. .155 0. .345 0. .862 0. .137 -0. ,029 0. ,004 0. .007 0. ,039 14 0. .039 0. .144 0. .205 0, .152 -0. .285 -0. ,071 0. .546 -0. ,074 -0. .198 15 0. .360 0. .374 0. .217 0, .040 0. .542 0. ,166 0, .501 -0, .153 0. .026 16 0, .558 0. .522 -0, .008 0, .143 0. .277 0. ,083 -0. .071 0. ,111 -0. .018 17 0. .796 0, .084 -0. .092 0, .143 0. .359 0. .043 0. .028 0, .244 -0. ,013 16 0. .059 0. .111 0. ,963 0. .171 0. .010 -0. ,002 0. .110 -0, ,071 -0. .047 19 0. .031 0. .108 0, .956 0, .197 -0. .016 -0. .001 0. .105 -0, .065 -0. .059 20 0. .549 0. .109 0. .182 0. .077 0. .030 0. .123 0, .209 -0, .012 -0. .115 21 0. .049 -0. .003 -0, .022 0, .209 0, .822 0. .024 0. .045 0, .184 0. .226 22 -0, .059 -0, .128 -0, .148 -0, .081 0, .122 -0. .124 -0.247 0 .700 0. .081 23 0. ,566 -0. .163 -0, .038 -0, .103 0. .078 -0. .002 7°' .041 0, • 111 -0. .057 24 0. .488 0.217 0, .196 0, .128 0, .650 0. .086 0. .250 0 .000 0, .084 25 0. ,i54 0. .060 0 .121 -0, .126 0, .172 0. .146 ' 0. 783 -0 .196 0. .050 26 0, .472 0. .020 0, .019 0, .207 0, .636 0. .011 -0, .163 0 .280 0, .080 27 0, .421 0, .761 0, .256 0 .297 0 .076 -0, .014 0 .097 -0 .044 0 .063 28 0. .549 0. .704 0, .268 0, .277 0. .074 0. .002 0, .086 -0, .023 0. .040 29 0. .775 0, .271 0, .091 0 .125 0, .073 0, .195 -0 .118 -0 .120 -0 .113 30 0. .570 0. .467 0, .004 0, .006 -0, .068 0.227 0, .116 -0, .076 0 .003 31 0. .614 0, .061 0 .213 0, .233 0, .477 -0, .260 -0 .023 -0 .004 -0, .013 32 0. .858 0. .288 0, .184 0, .182 0, .129 0 .003 0 .046 -0 .047 0, .051 33 0. .645 0, .077 -0 .028 0 .435 0, .266 -0, .021 -0, .149 0, .280 0, .230 34 0, .074 -0, .016 -0 .069 0 .112 0 .143 -0, .036 -0 .013 0 .134 0 .726 35 0, .358 0, .063 0 .031 -0 .060 -0 .164 0, .794 0 .106 -0 .118 -0, .023 36 0. .657 0, .238 0 .085 -0 .010 -0 .221 0 .072 0 .462 -0 .152 0 .067 37 0. .797 0. .231 0, .136 0 .104 0, .122 0, .071 0, .273 -0 .168 0, .241 38 0. .936 0, .168 0 .070 0 .166 0 .026 -0 .065 0 .100 -0 .030 0 .130 39 0. .856 0. .064 -0 .041 0 .169 0, .297 0 .003 -0 .059 0 .206 0 .019 40 0. .181 0. .140 0, .021 -0 .074 -0 .472 0, .353 0 .257 -0 .205 0 .180 41 0. .775 0, .288 0 .205 0 .180 0 .111 0 .210 0 .147 -0 .098 -0 .148 42 0. .880 0. .171 -0, .041 -0 .010 -0, .112 0, .093 0 .166 -0 .085 -0 .008 43 0. .858 0, .180 0 .005 0 .032 -0 .159 0 .047 0 .123 -0 .113 -0 .006 44 0. .191 -0. .017 0, .123 -0 .042 -0 .273 0 .015 0 .046 -0 .333 -0 .211 L a t e n t r o o t 11.543 4.083 3.707 3.624 3.602 2.148 2.015 1.718 0.999 V a r i a n c e (%) 30.371 10.743 9.755 9.535 9.476 5.652 5.303 4.520 2.630 102 Year 1 Peak 3 m 250000 V u m M 200000 4 ^ 150000 Tj V 5 icoooo u •S 60000 c W 0 a V u 9 M « i •a o N 7000 6000 6000 4000 3000 2000 - i r Peak 18 1000 e m *> Vi o Peak 1 9 9 2600 CI u m M 2000 4 t l 1600 •a •> 5 1000 TJ u m TJ 600 e « 01 0 s . 8 8 g I 8 8 „ -i 1— Peak 3 Year 2 — i 1 1 r 8 - i 1 1 r Peak 18 o o o o o o -| 1 1 r Peak 1 9 I 400000 300000 200000 100000 26000 20000 15000 10000 6000 Ho 8000 ecoo 4CO0 2C00 Ho to 0-» a. n a-TJ A » K Co 1 n m to rt 8 a H-ta tt 0. t* 8 ft* H 10000 to a a N tt tt 01 1 tt » 6 10 Tiae (day) 16 6 10 16 20 26 Tine (day) Figure 20. Plots of v o l a t i l e s from canned pink salmon of the r e f r i g e r a t e d storage study with high loadings for fa c t o r 3. 103 Upon further i n v e s t i g a t i n g the r e s u l t s of CFA i n conjunction with refrigerated storage time, factor 4 also appeared to be influenced by storage. Based on Table 21, three v o l a t i l e compounds, peak 7 (2-methyl propanal), peak 11 (benzene), and peak 13 (2,2-dimethyl propanal), were allocated high loadings by factor 4. As was expected of variables that were correlated to each other, peaks 7, 11, and 13 had s i m i l a r trends (Figure 21) ; they increased i n concentration near the beginning of storage, and l e v e l l e d o f f towards the end of the experiments. In addition, i t i s worth noting that the spread i n peak area increased with time for a l l s i x v o l a t i l e s just discussed (Fig. 20 and 21). P r i n c i p a l component analysis was also c a r r i e d out with a varimax rotation on the same data matrix gathered from canned salmon of the refrigerated storage study. Figure 22 shows the 44 loadings as vectors i n the subspace spanned by PC4 and PCS. Loadings with a b s o l u t e values higher than 0.2 have been l a b e l l e d . Peaks 3, 18, and 19, and peaks 7, 11, and 13 were found i n the v i c i n i t y of each other and each group of peaks dominated PC4 and PC5, respectively. Figures 23 and 24 are p l o t s of scores of PC4 and PC5 during the r e f r i g e r a t e d storage for years 1 and 2. The trends found i n these figures were c l e a r l y comparable to the sets of graphs in F i g . 20 and 21. PC4 and PC5 e s s e n t i a l l y contained the same information as factors 3 and 4. The results of CFA and PCA were a l i k e i n terms of peak grouping and confirm the presence of i n t e r c o r r e l a t i o n s among several of the variables i n the o r i g i n a l dataset. Although PCA could be viewed as another type of 104 Year 1 Year 2 P e a k 7 m 18000 « u m M 9 10000 TJ 0 N Tt TJ 6000 M 9 TS G 9 4J I S -T r P e a k 1 1 4 20000 0 U 9 "2 1*000 I 9 0 <X *S 10000 I N I •H <9 6000 I % 9 •P w m 16000 0 u 9 2 10000 p. •o 0 N Tj 6000 4 TJ e <o w o: i ! P e a k 1 3 I I I - | 1 r P e a k 7 O o 1 1 1 % 8 8 16000 JJ lb 10000 £ N f» p. t j A 6000 -i 1 1 r P e a k 1 1 .8 16000 10000 6000 o 20000 W 9 0-n p. -1 1 T P e a k 1 3 I . I N A p. >T ft) 1 A 16000 C A rt-to 9 » 10000 P. H-N A P. 6000 A » •1 A to 6 10 Tine (day) 16 5 10 16 20 Tine (day) 26 Figure 21. Plots of v o l a t i l e s from canned pink salmon of the r e f r i g e r a t e d storage study with high loadings for factor 4. 105 -1.0 -0.5 0.0 0.5 1.0 1.5 Principal component 4 Figure 22. Projection of gas chromatographic peak variable loadings on p r i n c i p a l components 4 and 5 for canned pink salmon of the re f r i g e r a t e d storage study. 0 5 10 15 20 25 Time (days) 5 B CD o Figure 23. Plot of the scores of p r i n c i p a l component 4 over r e f r i g e r a t e d storage time f o r canned pink salmon. o 4 3 5 c O i |_ Q. 1 -Q. •E -1 OL -2 --3 -Year 1 o o o o o o o 8 o Year 2 o o o 8 8 o 5 10 Time (days) 15 10 15 Time (days) o 8 20 25 3 2 5 1 a 0 1 -1 2-Cn -2 Figure 24. Plot of the scores of p r i n c i p a l component 5 over r e f r i g e r a t e d storage time f o r canned pink salmon. o 108 factor analysis, there are t h e o r e t i c a l differences between them. Nevertheless, these two techniques often y i e l d solutions that are very s i m i l a r (Dillon and Goldstein, 1984) . When PC4 and PC5 were pl o t t e d against each other, segregation of the 3 q u a l i t y levels became apparent (Fig. 25). PC4 separated grade rej e c t , found on the right side of the graph, from grade A and grade B located on the l e f t . Referring back to F i g . 23, score values of PC4 remained r e l a t i v e l y constant for some time but began to increase toward the end of the storage periods. These late changes were cor r e l a t e d with the increase i n concentration of v o l a t i l e s that received e s p e c i a l l y high loadings on PC4 (peaks 3, 18, and 19) and coincided with a s h i f t of q u a l i t y grade r e j e c t . S i m i l a r l y , PC5 mainly served to v e r t i c a l l y separate grades A and B. The increasing scores of PC5 from the s t a r t of storage up to a maximum point (Fig. 24) provided valuable information to discern samples of grade A from grade B (Fig. 25). Table 22 regroups the r e s u l t s of univariate and multivariate B a r t l e t t ' s tests f or homogeneity of variance on peak variables and p r i n c i p a l components involved i n the above discriminant analyses. Important discriminating variables such as PC4, PC5, peaks 3 and 18 had s i g n i f i c a n t Chi-square s t a t i s t i c s at P<0.05. The multivariate p r o b a b i l i t i e s that the peaks and p r i n c i p a l components had equal variance-covariance matrices for a l l q u a l i t y grades were very low (P<0.0001) . Kolmogorov-Smirnov normality tests were also performed on the peaks and p r i n c i p a l components (Table 23) . Both sets of variables had instances - 2 - 1 0 1 2 3 4 Principal component 4 - 1 0 1 2 3 4 Principal component 4 Figure 25. Plot of the scores of p r i n c i p a l component 5 against 4 for canned pink salmon of the r e f r i g e r a t e d storage study. o 110 Table 22. B a r t l e t t ' s t e s t s f o r homogeneity of within group variance-covariance between q u a l i t y grades of canned pink salmon f or selected gas chromatographic peaks variables and the f i r s t ten p r i n c i p a l components. Variable df Chi-square Univariate PCI PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9 PC10 2 2 2 2 2 2 2 2 2 2 9.2574 ** 16.8106 ** 28.1296 *** 175.5999 *** 19.0639 ** 2.3869 7.8339 * 21.9451 *** 3.4393 8.2366 * Peak 3 Peak 7 Peak 18 503.4706 *** 5.6272 1504.9712 *** Multivariate PC1-PC10 Peak 3,7,18 110 12 635.3260 *** 1867.8104 *** * S i g n i f i c a n t difference at 0.05. ** S i g n i f i c a n t d i f f e r e n c e at 0.01. *** S i g n i f i c a n t d i f f e r e n c e at 0.0001. I l l Table 23. Kolmogorov-Smirnov normality test of q u a l i t y grades of canned pink salmon for selected gas chromatographic peaks and the f i r s t p r i n c i p a l components. Variable Maximum dif f e r e n c e Grade A Grade B Reject (n=107) (n=48) (n=34) PCI 0.0710 0.1380 * 0.1667 * PC2 0.0843 0.1004 0.1116 PC3 0.0620 0.1134 0.1295 ** PC4 0.0822 0.0694 0.2108 * PC5 0.0645 0.1320 * 0.1562 PC6 0.0754 0.0923 0.1238 PC7 0.1981 *** 0.1773 ** 0.1281 PC8 0.li36 ** 0.0889 0.1535 * PC9 0.0768 0.0987 0.1254 PC10 0.1341 ** 0.0864 0.1455 Peak 3 0.104 ** 0.161 ** 0.184 ** Peak 7 0.064 0.094 0.126 Peak 18 0.529 *** 0.174 ** 0.281 *** * S i g n i f i c a n t d i f f e r e n c e at 0.05 l e v e l based on L i l l i e f o r s p r o b a b i l i t y . ** S i g n i f i c a n t d i f f e r e n c e at 0.01 l e v e l based on L i l l i e f o r s p r o b a b i l i t y . *** S i g n i f i c a n t d i f f e r e n c e at 0.0001 l e v e l based on L i l l i e f o r s p r o b a b i l i t y . 112 where the assumption of normality was not respected in grade A, B, and r e j e c t . However, while peaks 3 and 18 showed non-normal behaviour i n a l l grades, the d i s t r i b u t i o n s of PC4 and PC5 were s t a t i s t i c a l l y d i f f e r e n t from a normal one for only grade reject and grade B, respectively. The scores of the rotated p r i n c i p a l components of canned pink salmon from the storage study were submitted to l i n e a r discriminant analysis. The univariate tests on the ten PC's having eigenvalues above 1.0 showed that the f i r s t f i v e were s i g n i f i c a n t at e i t h e r the 0.0001 or 0.05 l e v e l s (Table 24). The two largest F r a t i o s were assigned to PC4 and PC5, and confirmed t h e i r substantial involvement i n grade categorization. The joi n t contribution of the PC's also proved to be highly s i g n i f i c a n t (P<0.0001) in discriminating the 3 q u a l i t y grades as analyzed by the multivariate t e s t s . Table 24 indicates the need for using both possible canonical functions for the computation of optimal distances between group centroids based on the canonical a n a l y s i s . The t o t a l e r r o r rate, calculated using the c r o s s - v a l i d a t i o n method, generated by the l i n e a r discriminant functions was 6.9%. Going back to the r e s u l t s of p r i n c i p a l component analysis, peak 18 (3-methyl-l-butanol) was previously shown to be strongly correlated to peaks 3 and 19, and a l l three were major contributors to the l i n e a r r e l a t i o n s h i p making up PC3. When the behaviour of these three peaks were compared, peak 18 was found to be the only compound that remained undetected for the i n i t i a l period of r e f r i g e r a t i o n storage. This p e c u l i a r i t y was not taken into account when carrying out tandem PCA-LDA. 113 Table 24. Univariate and mult i v a r i a t e test s t a t i s t i c s of l i n e a r discriminant analysis on the f i r s t ten varimax rotated p r i n c i p a l components from canned pink salmon during the refrigerated storage study. Variable num df,den df Univariate PCI 2,186 3. .8697 * PC2 2,186 9. .9110 * * * PC3 2,186 4, .1739 * PC4 2,186 154. .0639 * * * PC5 2, 186 61. .4829 *** PC6 2,186 2. .5676 PC7 2, 186 0. .1830 PC8 2, 186 2. .5589 PC9 2,186 2, .6270 PC10 2, 186 1 .2413 Mul t i v a r i a t e Wilk's lambda = 0.07785 20,354 P i l l a i trace = 1.29703 20,356 Hotelling-Lawley trace = 7.02994 20,352 Roy's largest root = 6.26080 10,178 45.7376 *** 32.8421 *** 61.8635 *** 111.4423 *** Canonical LR, = 0.07785 LR2 = 0.56525 20,354 9,178 45.7376 *** 15.2118 *** * S i g n i f i c a n t difference at 0.05 l e v e l . ** S i g n i f i c a n t difference at 0.01 l e v e l . *** S i g n i f i c a n t difference at 0.0001 l e v e l , LR stands for Likelihood r a t i o . To investigate the p o t e n t i a l usefulness of t h i s information, a form of stepwise discriminant analysis was performed on selected peak variables (Table 25). Functions based only on peak 3 (ethanol) provided a cross-validated t o t a l e r r o r rate of 19.6%, the majority of the error made on c l a s s i f y i n g grade B to grade A samples. When peak 18 was combined with peak 3 for the analysis, the erro r rate for grade B was reduced from 75% to about 42%. This underlined the fact that, even though an increased amount of ethanol s i g n a l l e d a s h i f t from grade A to grade B with or without the presence of 3-methyl-l-butanol, peak 18 was consistently associated with samples of grade B. Peak 7, having the highest c o r r e l a t i o n with PC5, helped achieve a further decrease i n error rate to approx. 17% for grade B whether or not peak 19 was put i n the model. Peak 19 was redundant due to i n t e r - c o r r e l a t i o n s with peaks 3 and 18. Peaks 11 (benzene), 5 (3-methyl-l-butene), 21 (acetic a c i d ) , and 38 (2-pentyl furan), which were a l l given high loadings onto PC5, PC2, PC3, and PCI, respectively, d i d not improve the c l a s s i f i c a t i o n rate of 93% established by peaks 3, 18, and 7. In view of the heteroscedasticity and non-normality cases (Table 23), the i n v e s t i g a t i o n of quadratic and non-parametric discriminant analyses was j u s t i f i e d and pursued for the c l a s s i f i c a t i o n of canned salmon i n the 3 q u a l i t y grades. Error counts estimated by the cro s s - v a l i d a t i o n method were computed for the three types of discriminant functions and are reported i n Table 26. As previously discussed, the t o t a l error rates were comparable for the two LDA's. Out of the 3 q u a l i t y l e v e l s , grade B was the one where the l i n e a r functions performed the least s u c c e s s f u l l y (12.6% 115 Table 25. Cross-validated error count estimates of l i n e a r discriminant functions c a r r i e d out on selected gas chromatographic peaks from canned pink salmon of the r e f r i g e r a t e d storage study. Peak v a r i a b l e Error count estimate 3 Grade A Grade B Reject % Total (n=107) (n=48) (n=34) (n=189) 3 0. .000 (0) 0 .750 (36) 0. .029 (1) 19.6 3,18 0. .000 (0) 0 .417 (20) 0. .029 (1) 11.1 3,18,19,7 0. .037 (4) 0 .167 (8) 0. .029 (1) 6.9 3,18,7 0. .028 (3) 0 .167 (8) 0. .029 (1) 6.4 3,18,7,11 0. .028 (3) 0 .188 (9) 0, .029 (1) 6.9 3 ,18,7 ,5 0. .028 (3) 0 .188 (9) 0. .029 (1) 6.9 3,18,7,21 0. .028 (3) 0 .208 (10) 0. .029 (1) 7.4 3,18,7,38 0. .028 (3) 0 .167 (8) 0, .029 (1) 6.4 a The numbers i n parentheses are the m i s c l a s s i f i c a t i o n s associated with the er r o r count estimates. 116 Table 26. Comparison of cross-validated error count estimates f o r di f f e r e n t discriminant analysis functions (DA) of selected gas chromatographic peaks and the f i r s t ten p r i n c i p a l components from canned pink salmon of the r e f r i g e r a t e d storage study. DA f u n c t i o n 3 Error count estimate Grade A Grade B Reject % Total (n=107) (n=48) (n=34) (n=189) P r i n c i p a l components l i n e a r 0.047 (5) 0.125 (6) 0.059 (2) 6.9 quadratic 0.028 (3) 0.083 (4) 0.000 (0) 3.7 non-parametric (r=4) 0.019 (2) 0.104 (5) 0.000 (0) 3.7 Three peaks (3, 7, 18) li n e a r 0.028 (3) 0.167 (8) 0.029 (1) 6.4 quadratic 0.009 (1) 0.062 (3) 0.000 (0) 2.1 non-parametric (r=3) 0.009 (1) 0.042 (2) 0.000 (0) 1.6 Pri o r prob. 0.566 0.254 0.180 a The nonparametric method consists of the Epanechnikov kernel. b The numbers i n parentheses are the m i s c l a s s i f i c a t i o n s associated with the error count estimates. and 16.7%). The quadratic and non-parametric kernel methods were found to be equally e f f e c t i v e i n bringing down the error count of a l l grades; but grade B samples benefitted most from t h e i r implementation. Although rates of correct c l a s s i f i c a t i o n were heightened to l e v e l s between 96% and 98%, non-parametric discriminant functions should be regarded as more stable due to the considerations encompassing the non-normal behaviour of some of the variables involved. Furthermore, the use of the non-parametric Epanechnikov kernel DA based on the three peak variables had the advantage of computing the functions d i r e c t l y on the peak areas themselves. In t h i s manner, p r i n c i p a l component analysis was by-passed, the number of v a r i a b l e s handled was considerably reduced, thereby lessening the amount of computation. 4. Performance of sensory evaluation and M V A The pink salmon that had been subjected to r e f r i g e r a t e d storage were graded before canning. The grades obtained p r i o r to canning served as reference grades i n the m u l t i v a r i a t e discriminant techniques used on the SHGC chromatograms of v o l a t i l e s . Cans from each of the pink salmon were also submitted to a trai n e d federal inspector for sensory q u a l i t y assessment. Figure 26 presents the sensory c l a s s i f i c a t i o n of canned salmon plotted against the grades determined before canning. While the post-canning results f o r grade A were i n perfect agreement with pre-canning q u a l i t y assessment, some divergence was observed for a number of grade reject samples. The largest discrepancy occured with grade B samples, of which a notable portion were c l a s s i f i e d as grade A. The 34-3 2 : 3 0 : c 28-| 2 6 i • 24-co 22 v 20 c c 18H o o 16 o o u 00 14-12 : 10' 8H 6-4-2 : " 0- JIJIIJTUTTUIRJI —. . . rTfk m M M fl n = 46 n = 20 n = 15 C o E o CO •o R £ c o o d> "O o o B A A B R Grade of salmon before canning Figure 26. Sensory c l a s s i f i c a t i o n before and af t e r canning pink salmon of the r e f r i g e r a t e d storage study. present post-canning grading scheme grants r e l a t i v e l y small ranges of scores for grade A and B compared to grade r e j e c t . Samples of grade B were expected to be i n greater numbers at the onset of decomposition. The experiments were c a r r i e d out at r e f r i g e r a t e d temperatures which were designed to allow time f o r t r a n s i t i o n from one grade to another. While grade B targeted samples with s l i g h t but d e f i n i t e l y persistent and perceptible odours of decomposition, samples grouped i n grade reject possessed a strong odour of decomposition as evidence of advanced decay. Figure 2 6 indicates the tendency of sensory evaluators to segregate samples of canned salmon into two categories, grade A and grade reject, and consequently underlined the d i f f i c u l t y of c o n s i s t e n t l y detecting early signs of decomposition. This problem could be l i n k e d to sensory fatigue due to the evaluation of numerous cans of salmon consecutively. The comparison of error rates between instrumental analysis and sensory evaluation of canned salmon i s shown i n Table 27. The data of categorical nature (correct or incorrect c l a s s i f i c a t i o n ) were analyzed with the SAS Catmod procedure which f i t s l i n e a r models to functions of categorical variables. Because the performance of the c l a s s i f i c a t i o n methods varied with grades, that i s the i n t e r a c t i o n was s i g n i f i c a n t (P<0.01), weighted least squares estimations were c a r r i e d out for each grade. The error rates for instrumental and sensory analyses were obtained d i r e c t l y from Table 26 and F i g . 26. Both non-parametric discriminant functions and sensory grading kept the -error rates for grade A to a minimum. While there was no error obtained by the discriminant analyses for grade r e j e c t , sensory grading had an e r r o r rate s i g n i f i c a n t l y 120 Table 27. Comparison of error rates f o r non-parametric discriminant functions (NPAR-DA) and sensory grading of canned pink salmon of the r e f r i g e r a t e d storage study. Method^ Error rate NPAR-DA Grade A Grade B Reject 5 PC's (r=4) 0.019 (2/107) 0.104 (5/48) 0.000 (0/34) 3 Peaks (r=3) 0.009 (1/107) 0.042 (2/48) 0.000 (0/34) Sensory grading 0.000 (0/46) 0.900 (18/20) 0.267*-(4/15) a _ f Error count estimates i n columns with the same l e t t e r s are not s i g n i f i c a n t l y d i f f e r e n t from each other (P<0.05). 9 The nonparametric method consists of the Epanechnikov kernel. The numbers i n parentheses are the . m i s c l a s s i f i c a t i o n r a t i o s associated with the error count estimates. 121 greater (26.7%). The rate for non-parametric DA based on 5 p r i n c i p a l components was s i g n i f i c a n t l y higher than that based on the 3 peak variables, but both rates were found to be between 4 to 10%. The most s t r i k i n g feature of Table 27 was the 90% error rate of m i s c l a s s i f i c a t i o n of grade B samples by sensory evaluation. Sensory grading of whole salmon r e l i e s on a larger number of observable variables than does sensory grading of canned salmon. S t i l l , odour i s among the c r i t e r i a used to evaluate a raw f i s h , and samples categorized i n grade B usually possessed a perceivable odour of decomposition. Spoilage, thermal processing, and t h e i r i n t e r a c t i o n can a f f e c t the aroma of a canned seafood i n various ways. In some instances, thermal processing causes an increase i n off-odour. For example, TMAO i s p a r t i a l l y degraded to TMA and DMA during the canning process of f i s h (Hebard et a l . , 1982) . M i c r o b i a l metabolism ofjineat components can change the composition of non-volatile precursors r e s u l t i n g i n a reduction or loss of the c h a r a c t e r i s t i c meat flavour. Decomposition products can also serve as reactants i n the complex chemical pool of flavour precursors, and be involved i n the formation of new v o l a t i l e s during heating. Pokorny (1980) reported that the reaction of proteins with rancid f i s h o i l s at elevated temperatures resulted i n the development of baked or roasted f i s h flavours. It i s therefore possible that components contributing to o f f -odours decrease i n concentration, or are masked by other thermally-generated v o l a t i l e s produced i n canned salmon but absent i n raw f i s h . This could happen p a r t i c u l a r l y during the i n i t i a l stages of spoilage. 122 In part, the success of the instrumental method may be due to the early detection of concentration changes to which sensory evaluation remains i n s e n s i t i v e . For example, the odour threshold for ethanol has been reported to vary between 100 ppm (Guadagni et a l . , 1963) and 900 ppm (Mulders, 1973) depending whether the method involved squeezable p l a s t i c flasks or smelling over open beakers. Hollingworth and Throm (1983) provided suggested ranges of ethanol concentration associated with q u a l i t y levels of canned salmon: sensory class 1 (grade A), 0-24 ppm ethanol; sensory class 2 (grade B), 25-74 ppm ethanol; and sensory cl a s s 3 (grade r e j e c t ) , 75 ppm ethanol and above. Therefore, changes of concentration below or around high threshold values s i m i l a r to those of ethanol could pass undetected by sensory evaluation. It i s very l i k e l y that the reverse i s also true. Experienced inspectors can r e l y on the perception of compounds with low sensory threshold values that are not detected by the SHGC method. But i t follows from Figure 2 6 and Table 27 that the accuracy with which these 'character-impact' odours were detected for grade B pink salmon i n pa r t i c u l a r , d i d not provide r e s u l t s that corroborated well with those obtained from sensory grading p r i o r to canning, whereas the instrumental SHGC method was successful i n t h i s regard. The nature of the sensory and SHGC-MVA re s u l t s accumulated thus f a r allowed further computations of regression equations i n order to investigate t h e i r r e l a t i o n s h i p s during the r e f r i g e r a t e d storage time. Plots of the sensory grading of canned salmon for years 1 and 2 are presented i n Figure 27. Because the data to be p l o t t e d were c a t e g o r i c a l , d i g i t s instead of points were used i n the graphs to represent the number Figure 27. Logarithmic r e l a t i o n s h i p s of sensory rating of canned salmon with r e f r i g e r a t e d storage time. D i g i t s represent the number of times a r a t i n g was encountered for each day of r e f r i g e r a t e d storage. 124 of times a rating was encountered for each selected day of r e f r i g e r a t e d storage. The changes in ratings were assumed to occur at an exponential rate. As v a r i a t i o n in ratings increased towards the end of the storage period, e s p e c i a l l y for year 2, a logarithmic transformation not only improved the f i t of the model but helped to diminish the heteroscedasticity i n the data. Although the data i s of c a t e g o r i c a l nature, the strength of the r e l a t i o n s h i p may be assessed i f the sensory ratings are taken to behave as continuous v a r i a b l e s . The c o r r e l a t i o n 2 index (r ) was used to measure the r e l a t i o n s h i p i n a way s i m i l a r to a c o r r e l a t i o n analysis context (Zar, 1984). For the curves of years 1 and 2, the c o r r e l a t i o n indices describing the amount of v a r i a b i l i t y i n logarithmically-transformed sensory r a t i n g of canned salmon accounted for by c o r r e l a t i o n with r e f r i g e r a t i o n storage were 0.55 and 0.60, respectively. Among the multivariate r e s u l t s , the scores of the two most discriminating p r i n c i p a l components, PC4 and PC5, served to calculate a 2 second degree polynomial q u a l i t y function (Z). A l l terms but PC4 were s i g n i f i c a n t i n the model (P<0.0001). This PC q u a l i t y function without the PC4 term i s p l o t t e d against r e f r i g e r a t e d storage time i n Figure 28. A c o e f f i c i e n t of determination of 0.80 (P<0.0001) was generated by the obtained l i n e a r i z e d function. Similar q u a l i t y functions based on peaks 3 (P3), 7 (P7), and 18 (P18) were computed for years 1 and 2 (Figure 29). Peaks 3 and 18 previously appeared to behave exponentially and therefore were l o g a r i t h m i c a l l y transformed before carrying out stepwise regressions by backward elimination. Three of the second degree terms of the Figure 28. L i n e a r i z e d r e l a t i o n s h i p of the polynomial q u a l i t y function using p r i n c i p a l components 4 (PC4) and 5 (PC5) over r e f r i g e r a t e d storage time. CO E CO 2S -20 -CO o c o o 1 , 1 ! Year 1 X - 0 6388 • 0.8964*t in« where Z - 142.1610 - 86.8602*Log(P3) + 28.4107*Log(P18) + 3.0932E-3«P7 + 12.9740»Uo9(P3))' -6 .6903«Log(P3)*Lo9(PlB) - 6.9656E-4«Lo9(P3)*P7 R ' - 0.90 S ere - 1.43 n - 83 _ 15 10 Z - 0.3467 + 0.9760*tlme where Z - 8S.5424 - 71.6338«Log(P3) + 21.7035'Log(P18) + 1.051E-2«P7 + 12.S435*(L09(P3))' + 1.3304»(Lo9(P18))* - 1.2E-7«P7' - 7.5277«Log(P3)«L09(P18) - 2.1202E-3*Log(P3)«P7 + 7.5034E-4*LO9(P18)*p7 R1 - 0.97 •+• S err « 1.20 - 106 Year 2 - 25 -0 CD B> 7C 10 15 Time (days) 10 16 Time (days) 20 - 15 - 10 - 6 - 0 CD Cl 20 25 Figure 29. Linearized r e l a t i o n s h i p of the polynomial q u a l i t y function using peaks 3, 7, and 18 over r e f r i g e r a t e d storage time. 127 polynomial equation for year 1 were found to be non-significant (P>0.05) but none of the terms i n the equation of year 2 were rejected. The highest c o e f f i c i e n t s of determination, 0.90 and 0.97, yet obtained from a l l above regressions were associated with these l a s t two equations for years 1 and 2 (P<0.0001), res p e c t i v e l y . In addition to better curve f i t t i n g s , the l i n e a r i z e d functions c a l c u l a t e d from SHGC data had lower c o e f f i c i e n t s of v a r i a t i o n , i . e . , 33.5%, 21.2%, and 13.8% for PC's and peaks quality functions of years 1 and 2, respectively, compared to 35.2% and 53.1% for the previous functions of years 1 and 2 applied on sensory data. A l l above functions were attempts to express the decomposition behaviour of canned salmon during r e f r i g e r a t e d storage periods. The d i f f e r i n g degrees of p r e c i s i o n embodied i n the various equations p a r t i a l l y depended on the nature of the methods themselves. They were also t i e d to many other factors that have not been considered but which influenced the rate of decomposition, e.g., the nature, d i v e r s i t y , and prominence of the microbial f l o r a . The intent of the regressions at t h i s point i s more des c r i p t i v e than p r e d i c t i v e . 5 . MVA of volatiles from raw pink salmon during refrigerated decomposition The SHGC method established i n Chapter II was optimized to analyze v o l a t i l e s from samples of canned salmon and provided important information that was used i n multivariate a n a l y s i s . The grading of raw salmon might also gain from s i m i l a r instrumentally-generated data treatments. The SHGC method was thus applied on samples of raw salmon from the r e f r i g e r a t e d storage study of year 2. Figure 30 presents t y p i c a l chromatograms of raw 128 J 1 J -4 / .18 KM 17 1» SO 39 39. ' » • A 1 I : 111 » JUlL jjttW B 1 1 1 i i i > 18 19 JlhJL -4—1  1 1 uJLi 1 1 1 1 JLuJ c i i 0 1 2 3 5 7 9 11 13 15 17 19 Uncorrected retention time (min) Figure 30. Chromatograms of v o l a t i l e s from raw pink salmon during the re f r i g e r a t e d storage of year 2. (A, day 0/grade A, day 10/grade B, day 21/grade r e j e c t ) . salmon after 0 day (A) , 10 days (B), and 21 days (C) of r e f r i g e r a t e d storage a f t e r sampling. The salmon, from which the samples were taken, were evaluated as grade A, B, or reject, r e s p e c t i v e l y . The peak areas from v o l a t i l e s of raw salmon were in general not as intense as with samples of canned salmon. The nature of the samples themselves was l i k e l y the reason. The pieces of raw salmon introduced i n headspace v i a l s received a thermal treatment for 1 h at an incubation temperature of 105°C. During the course of t h i s treatment, the m y o f i b r i l l a r proteins denatured to form matrices entrapping food components including v o l a t i l e s . In addition, raw salmon pieces d i d not provide as much surface exposure as canned salmon flakes from which v o l a t i l e s could escape and b u i l d up p a r t i a l pressures i n the v i a l headspace. Peaks 3 (ethanol), 18 (3-methyl-l-butanol), and 19 (2-methyl-2-butenal) were l a b e l l e d on the three graphs of F i g . 30 and t h e i r behaviours were found to bear a resemblance to those of canned salmon samples i n F i g . 19. A number of new v o l a t i l e s with r e l a t i v e l y low FID responses were detected, p a r t i c u l a r l y i n the second half of the chromatograms. Twenty two peaks having c o e f f i c i e n t s of v a r i a t i o n near 12% or less were selected for multivariate a n a l y s i s . These v o l a t i l e s are numerically indicated i n graph A of F i g . 30. On the basis of s i m i l a r retention time, a l l v o l a t i l e s except peak 39a were previously reported and are so l a b e l l e d i n Table 6. P r i n c i p a l component analysis with varimax rotation was performed on the standardized integrated areas of the 22 peaks; the r e s u l t i n g peak-p r i n c i p a l component c o r r e l a t i o n s are shown i n Table 28. Nine peaks had 130 Table 28. Loadings of the f i r s t four varimax rotated p r i n c i p a l components of v o l a t i l e compounds from raw pink salmon of the r e f r i g e r a t e d storage study (year 2 ) . Peak P r i n c i p a l components 1 2 3 4 1 0 .780 0 .236 0 .199 0.195 2 0 .692 -0 .221 0 .281 0.157 3 0 .098 0 .969 -0 .066 -0.027 5 0 .373 0 .100 0 .842 0.163 6 0 .308 0 .088 0 .206 0.918 7 0 .480 0 .701 0 .201 0.184 9 0 .883 0 .248 0 .244 0.129 10 0 .493 0 .302 0 .207 0.095 11 0 .351 0 .817 0 .163 0.138 13 0 .363 0 .776 0 .187 0.152 15 0 .833 0 .376 0 .214 0.009 16 0 .855 0 .221 0 .264 0.165 17 0 .909 0 .232 0 .199 0.095 18 0 .116 0 .983 0 .042 -0.003 19 0 . 148 0 .978 0 .050 0.030 20 0 .349 • 0 .092 0 .828 0.142 30 0 .698 0 .362 0 .234 0.241 31 0 .955 0 .108 0 .129 0.051 39 0 .900 0 .176 0 .177 0.076 39a 0 .922 0 .181 0 .073 0.165 41 0 .894 0 .165 0 .091 0.144 44 -0 .104 -0 .086 0 .013 -0.034 i t root 13 .061 3 . 657 1 .267 1.039 ance (%) 59 .369 16 .621 5 .758 4.721 loadings of approx. 0.8 and above for PCI: peak 1 (hydrogen s u l f i d e and acetaldehyde), peak 9 (butanal), peak 15 (l-penten-3-ol), peak 16 (heptane), peak 17 (1,5-dimethyl cyclopentene), peak 31 (unknown), peak 39 (4-ethyl benzenemethanol) , peak 39a (unknown), and peak 41 (nonanal) . PCI expressed correlated v a r i a t i o n s i n a d i v e r s i t y of v o l a t i l e compounds. Due to t h e i r loadings above 0.7, peak 3 (ethanol), peak 7 (2-methyl propanal), peak 11 (benzene), peak 13 (2,2-dimethyl propanal), peak 18 (3-methyl-l-butanol) , and peak 19 (2-methyl-2-butenal) were more strongly associated with PC2 than the other v o l a t i l e s detected. This l a s t p r i n c i p a l component regrouped the same v o l a t i l e compounds as the important discriminators, PC4 and PC5, from canned salmon. PC2 accounted f o r 16.6% of v a r i a t i o n while the f i r s t 4 PC's with latent roots above 1.0 explained 86.5% of the t o t a l variance. Scores of PC2 plotted against that of PCI are shown i n F i g . 31. Quality l e v e l s were segregated by PC2 while each grade was spread along the PCI dimension. Low values f o r PC2 were found i n grade A samples. Scores increased as grades changed from A to B, and eventually to r e j e c t . Samples of grade reject occupied a large part of the PC2 scale compared to samples from grades A and B, h i n t i n g at an heteroscedasticity problem. In any case, PC2 could be considered as the dimension representing the decomposition process during r e f r i g e r a t i o n . The changes in standardized area of the three selected peaks analyzed from raw pink salmon of the re f r i g e r a t e d storage study, previously used i n the discriminant analyses for canned salmon (peaks 3, 132 CN] •+-> C 0 C o QL E o o 16 Q. "o c 0_ 4 h 2 h 1 h 0 h -1 h -2 -1 0 Principal component 1 Figure 31. Plot of the scores of p r i n c i p a l component 1 against 2 for raw pink salmon of the r e f r i g e r a t e d storage study of year 2 (A, grade A; B, grade B; R, grade r e j e c t ) . 133 7, and 18), are shown i n F i g . 32. The non- l inear r e l a t i o n s h i p of peaks 3 and 18 was very s i m i l a r . Peak 18 (3-methyl- l -butanol) was not detected i n the f i s h at the beginning of the storage per iod . Peak 3 (ethanol) was, however, detected i n a l l samples. Peak 7 (2-methyl-2-butenal) had a d i f f e r e n t re la t ionsh ip to storage time than was observed with canned salmon. Instead of i n c r e a s i n g soon af ter the i n i t i a t i o n of the re f r igera ted storage, i t s concentrat ion remained r e l a t i v e l y low u n t i l the end, when i t s variance abrupt ly increased . The v i s u a l d i f f erence observed between peaks 3 and 18, and peak 7 (F ig . 32) was r e f l e c t e d i n a lower loading for peak 7 on PC2 (Table 28) . Subsequent m u l t i v a r i a t e analyses were c a r r i e d out on peaks 3, 7, and 18 only . The homogeneity of variance and normali ty assumptions were tested for the peak v a r i a b l e s . Univar ia te and m u l t i v a r i a t e B a r t l e t t ' s tes ts were a l l s i g n i f i c a n t (P<0.0001) and cases of non-normali ty , assessed by the Kolmogorov-Smirnov t e s t , were found for the three peaks (P<0.01) . Consequently, non-parametric DA funct ions , based on the Epanechnikov kerne l , were computed. As peak 7 showed some signs of weaker d i scr iminant power than peaks 3 and 18 ( F i g . 32 and Table 28), i t s involvement was invest igated by t e s t ing whether i t s presence improved the c r o s s - v a l i d a t e d error ra te . Functions obtained which r e l i e d on peaks 3 and 18, were as e f f ec t ive with or without peak 7, r e s u l t i n g i n a same t o t a l c l a s s i f i c a t i o n rate of 95.6%. C l a s s i f i c a t i o n rates for each q u a l i t y l e v e l determined by the non-parametric DA based on peaks 3 and 18 are shown i n Table 29. While a l l 53 samples of grade A were c o r r e c t l y c l a s s i f i e d by the funct ions , samples were m i s c l a s s i f i e d three times and one time for grades 134 600000 400000 300000 200000 100000 20000 15000 10000 16000 10000 6000 • 1 1 1 Peak 3 o — I § 8 8 0 0 _ 0 8 -o 8 § 0 1 ° g 1 ° 0 0 o 8 i e 1 • --Peak 7 --0 -- 0 -8 o 0 R 8 8 g o 8 • 8 8 i 8 1 i • 1 o 1 1 Peak 18 0 0 8 0 - 0 0 o 0 0 O ° | I e g o 0 0 1 O 0 • O S • 1 1 0 6 10 16 20 26 Time (days) Figure 32. Plots of v o l a t i l e s from raw pink salmon of the r e f r i g e r a t e d storage study of year 2 with high loadings for p r i n c i p a l component 2. 135 Table 29. C l a s s i f i c a t i o n by non-parametric discriminant analysis functions (NPAR-DA) of selected peaks (3 and 18) from the gas chromatographic analysis of raw pink salmon of the r e f r i g e r a t e d storage study. Assigned grade Predicted grade by NPAR-DAa Prior Prob. B Reject A (n=53) 1.000 (53) 0.000 (0) 0.000 (0) 0.582 B (n=16) 0.125 (2) 0.812 (13) 0.063 (1) 0.176 Reject (n=22) 0.0000 (0) 0.045 (1) 0.956 (21) 0.242 a The nonparametric method consists of the Epanechnikov kernel, r=3. 136 B and reject, respectively. These re s u l t s i n d i c a t e d that the NPAR-DA produced low error rates and demonstrated a c o m p a t i b i l i t y with the sensory grading p r i o r to canning. The d i f f e r e n t behaviour of peaks 7, 11, and 13, between raw and canned salmon, indicated that canning and the subsequent thermal processing could be an influencing f a c t o r . Ethanol and 3-methyl-l-butanol are known microbial c a t a b o l i c products. These short-chain alcohols are derived from the metabolic degradation of carbohydrates and amino acids, r e s p e c t i v e l y (Brock, 1979) . Ethanol has been suggested as p o t e n t i a l index of spoilage i n raw ( H i l l i g , 1958; Iida et a l . , 1981a, b; Human and Khayat, 1981; Kelleher and Z a l l , 1983; Ahamed and Matches, 1983) and canned f i s h (Holaday, 1939; Lerke and Huck, 1977; Crosgrove, 1978; Khayat, 1979; Iida et a l . , 1982; Tokunaga et a l . , 1982; Hollingworth et a l . , 1986). The concentration of 3-methyl-l-butanol has also been reported to increase during the course of spoilage j in f i s h ( Miller et a l . , 1973b; Kamiya and Ose, 1984; Ahamed and Matches, 1983) and mussels (Yasuhara, 1987). Production of ethanol i n sub s t a n t i a l amounts, followed by increases in 3-methyl-l-butanol, was previously observed i n re f r i g e r a t e d f i s h (Lindsay et a l . , 1987). Based on a study by Ahamed and Matches (1983) on b a c t e r i a l i s o l a t e s from king salmon (Oncorhynchus tshawytscha) and rainbow trout (Salmo i r r i d e u s ) , t e s t organisms appeared to prefer 5 and 6 carbon sugars as i n i t i a l energy sources and then began to u t i l i z e free- amino acids which are the precursors of various products such as alcohols. Indices of f i s h spoilage, such as some of the proposed discriminant functions, that r e l y on these two metabolic aspects, r e f l e c t the stages of decomposition that f i s h may undergo. Another advantage i n using alcohols as in d i c a t o r of microbial spoilage i n fresh and canned f i s h products i s t h e i r thermal s t a b i l i t y during processing. More research should therefore be c a r r i e d out to investigate the factors that a f f e c t the generation of these v o l a t i l e s . More work should be performed to increase the number of samples analyzed p a r t i c u l a r l y f o r grades B and r e j e c t thereby enhancing the accuracy of t h e i r population estimates i n multidimensional discriminant space. 138 IV. DYNAMIC HEADSPACE ANALYSIS OF VOLATILE FLAVOUR COMPONENTS IN CANNED SALMON A. Introduction V o l a t i l e aroma components are generally regarded as the most important parameters of food flavour q u a l i t y . Due to both economic importance and academic i n t e r e s t , research has been c a r r i e d out on the i d e n t i f i c a t i o n of v o l a t i l e flavour compounds i n various foods in c l u d i n g f i s h . Over the years, v o l a t i l e analysis by dynamic headspace techniques has received substantial attention from researchers. It i s based on a dynamic process i n which the sample v o l a t i l e s are tra n s f e r r e d with a stream of an inert gas to a trap and then desorbed by various means into a GC. The dynamically purged headspace v o l a t i l e s of tuna (Khayat, 1979; Human and Khayat, 1981) and herring (Hughes, 1964) were condensed by cryogenic trapping while adsorbent trapping was used for p i c k l e d smelt (Josephson et a l . , 1987), north sea f i s h o i l (Christensen et a l . , 1981), c r a y f i s h (Vejaphan et a l . , 1988; Tanchotikul and Hsieh, 1989), A t l a n t i c and P a c i f i c oysters (Josephson et a l . , 1985), prawns and sand-lobsters (Whitfield et a l . , 1982), and various species of f i s h (Easley et a l . , 1981; Reinert et a l . , 1983; Josephson et a l . , 1984a). Fresh P a c i f i c salmon, e i t h e r pan-fried, baked, or roasted, possesses a t y p i c a l flavour, d i s t i n c t i v e from other seafood, that i s recognized by consumers. Canned salmon releases, upon opening, an aroma which appears s i m i l a r but i s yet d i f f e r e n t from salmon processed by other means. The v o l a t i l e s responsible for these aromas have not been elucidated. 139 Josephson (1987) found that salmon carotenoids may e i t h e r serve as d i r e c t precursor compounds or modulate chemical reactions which convert f a t t y acids or other l i p i d precursors to salmon-loaf-like aroma compounds. The orange pigmentation of the f l e s h of adult wild P a c i f i c salmon i s due to carotenoids, e s p e c i a l l y astaxanthin, i n a free form. Kitahara (1983) showed that the l e v e l s of astaxanthin in muscles decreased markedly when salmon were engaged i n t h e i r spawning migration. It was transported to the skin and gonads via the blood serum. During anadromous migrations of salmon, the l i p i d s that depleted from the body stores were almost e n t i r e l y t r i a c y l g l y c e r o l s . The amount of t r i a c y l g l y c e r o l s i n dorsal muscle of chum salmon decreased from 2.1g/100g t i s s u e i n males and 3g/100g in females, to less than 0.2g/100g i n both sexes, between feeding i n the sea and spawning (Ando et a l . , 1985a; 1985b). The sexual maturation process also brings about a stimulation of mucus secretion, and a darkening and thickening of the skin r e s u l t i n g i n the scales becoming more deeply embedded. The biochemical regulation of mucus secretion i n la t e spawning-condition P a c i f i c salmon was further conceptually linked with increases of 8- and 9-carbon v o l a t i l e s which contributed d i s t i n c t cucumber or melon-like aromas (Josephson, 1987). Odour changes i n f i s h may also arise from spoilage as a consequence of microbial metabolism. Although numerous v o l a t i l e components have been i d e n t i f i e d i n spoiled f i s h , the d i r e c t association of s p e c i f i c compounds to p a r t i c u l a r odours remains a great challenge. Early e f f o r t s to rel a t e trimethylamine to f i s h y odour were reported by Davies and G i l l (1936) and Stansby (1962). F r u i t y odours were considered by M i l l e r et a l . (1973a) to be due to the presence of ethyl esters of acetate, butyrate, and hexanoate. Methyl-3-isopropyl pyrazine was determined to be responsible for the potato-like off-odour produced by Pseudomonas perolens (Miller et a l . , 1973c). The studies of Herbert and Shewan (1975, 1976) were instrumental in a t t r i b u t i n g the sulfurous odour to hydrogen s u l f i d e , methanethiol, and dimethyl s u l f i d e as well as uncovering t h e i r o r i g i n . The factors mentioned above, e.g., thermal processing, spawning migration, and s p o i l i n g during cold storage, have unknown e f f e c t s on the v o l a t i l e patterns of canned salmon. Therefore, the objectives of the t h i r d part of t h i s research were: (1) the separation and i d e n t i f i c a t i o n of v o l a t i l e compounds i n canned pink salmon of good q u a l i t y (grade A ) , canned pink salmon of advanced decomposition (grade r e j e c t ) , and canned chum salmon of advanced sexual maturity (spawning dark) by using sorbent trap sampling/GC/MS, and (2) the evaluation of the sensory c h a r a c t e r i s t i c s of cryofocussed GC e f f l u e n t s from these 3 d i f f e r e n t types of canned salmon. 141 B. Materials and methods 1. C o l l e c t i o n and canning of salmon Canned pink salmon (Oncorhynchus qorbuscha Walbaum) were taken among those processed for the second year of the r e f r i g e r a t e d decomposition study of Chapter I I I . Pink salmon of grade A and reject had undergone a cold storage of 2 and 17 days, r e s p e c t i v e l y . "Late-run" chum salmon (Oncorhynchus keta Walbaum) were obtained from the Chilliwack (B.C.) hatchery during the month of October. Very dark green or brown barr i n g of a thick skin with a heavy slime, eroded f i n s , hooked jaws, very t h i n b e l l y flaps and corpulence, pale greyish f l e s h colour, and strong " l a t e " odour of the f l e s h and skin were external signs i n d i c a t i v e of advanced sexual maturity. Approximately 215g of chum salmon f l e s h and 2g sodium chloride were put in 307 x 113 three-piece cans. The samples were then vacuum-sealed and thermally processed i n a FMC 500W Universal S t e r i l i z e r (F.M.C. Corporation, Santa Clara, CA) . A r e t o r t temperature of 120°C was used with a process time of 65 min. These conditions gave a l e t h a l i t y (F Q) of approx. 7.6 min based on data from C o l l i n s (1989) . 2. Tenax trap sampling/gas chromatography/mass spectrometry (TTS/GC/MS) Forty mL of drained l i q u i d from canned salmon were placed i n a headspace sampling tube of an HP 7675A Purge and Trap sampler. The samples were purged with p u r i f i e d helium (Zero Gas, Medigas, Vancouver, BC) at a rate of 20 mL/min while the sampling tube was immersed i n a water-bath held at 50°C. A f t e r a 30 min sparging time, the porous polymer TM Tenax TA (0.30g, 60-80 mesh, Hewlett Packard) sorbent trap (8.89cm x 0.64cm o.d.) was dry-purged f o r 1 h with helium. V o l a t i l e s were then thermally desorbed (185°C) and brought at a flowrate of 2 mL/min on an U l t r a 2 c a p i l l a r y column (30m x 0.32mm x 0.52 Mjti f i l m thickness) i n s t a l l e d i n a HP 5840A gas chromatograph/HP 5985B mass spectrometer. For v o l a t i l e separation from good q u a l i t y canned pink salmon, the oven temperature was programmed from 20°C to 220°C at a rate of 5°C/min with an i n i t i a l holding time of 1 min. Runs for samples of the two other treatments, e.g., canned pink salmon of poor q u a l i t y and canned late-run chum salmon, were programmed to hold the i n i t i a l temperature at 20°C for 1 min, increase to 120°C by 5°C/min, and subsequently reach 220°C at a rate of 10°C/min. Mass spectra a c q u i s i t i o n , tentative i d e n t i f i c a t i o n based on mass spe c t r a l l i b r a r y and confirmation by retention time comparison with reference standards was done as described i n Chapter I I . 3. Cryofocussing concentration sampling/gas chromatography /odour evaluation (CCS/GC/OE) In order to evaluate the sensory c h a r a c t e r i s t i c s of the GC e f f l u e n t , the v o l a t i l e s were concentrated by cryofocussing. The sample loop of a headspace sampler was replaced with a s t a i n l e s s s t e e l 4-way b a l l valve (Whitey Co., Highland Hts., OH) and a f i x t u r e assembled by coupling two can piercers (Alltech Associates Inc., Deerfield, IL) connected altogether 143 with s t a i n l e s s s t e e l tubing of 1.59 mm o.d. (Figure 33). A can was tightened i n the pie r c e r and immersed i n a water bath held at 50°C. After 30 min, the b a l l valve was opened and the v o l a t i l e s were swept with helium at a flowrate of 10 mL/min. Cryofocussing occurred i n the 40 cm section of the loop lowered into a bath of dry ice-acetone (-78°C). Thi r t y min l a t e r , the flowrate was reduced to 4 mL/min before the b a l l valve was switched to by-pass the can. The oven temperature program i l l u s t r a t e d i n Figure 34 was i n i t i a t e d , and the desorption process was started by immersing the loop i n an ice-water bath. The loop was subsequently dipped i n b o i l i n g water when the oven temperature reached -20°C. It remained i n the b o i l i n g water bath u n t i l an oven temperature of 0°C was obtained. The v o l a t i l e s t r a v e l l e d through the transfer l i n e (125°C) of the headspace sampler and reached the i n j e c t o r port. Two lengths of deactivated vitreous s i l i c a tubing (30 cm x 0.22 mm i.d.) were connected to the end of an U l t r a 2 c a p i l l a r y column (30m x 0.32mm x 0.52 m f i l m thickness) with a 1:1 r a t i o s p l i t t e r . One tube was dire c t e d to an FID maintained at 250°C while the second tubing l e d to a s n i f f i n g port heated at 175°C. The part of the ef f l u e n t for odour evaluation was mixed with a i r (300 mL/min) introduced v i a the gas supply of the second FID base and t r a v e l l e d through a Teflon tube (20 cm x 0.32 cm o.d. x 0.16 cm i.d.) connected with a s t a i n l e s s s t e e l Swagelok union (0.32 cm x .32 cm) to an adaptor f i t t i n g the FID e x i t . Two analysts having experience with canned salmon assessed the aroma of the GC eff l u e n t by noting the retention time and a des c r i p t i o n of the perceived sensation. 144 Figure 33. Can p i e r c e r f i x t u r e , valve, and s t a i n l e s s assembled to concentrate canned salmon cryofocussing (CCS). s t e e l tubing v o l a t i l e s by 145 2 2 0 / O °C/min 100 / ^ ° C / m i n 0 C/min - 2 0 - 6 0 i c e - w a t e r water ( O t ) ( 1 0 0 t ) ) Figure 34. Schematic representation of the v o l a t i l e desorption steps and the oven temperature program for cryofocussing concentration sampling/gas chromatography/odour evaluation (CCS/GC/OE). 146 C. Results and discussion Total ion chromatograms obtained f o r canned pink salmon of good quality, canned pink salmon of grade r e j e c t , and canned l a t e run chum salmon are shown in Figures 35, 36, and 37, respectively. V o l a t i l e s at the beginning of these chromatograms were not as well resolved as those e l u t i n g l a t e r on. With hindsight, the use of cryofocussing subsequent to the TTS procedure would have been b e n e f i c i a l , to increase the sharpness of the f i r s t compounds of lower b o i l i n g point. However, the information provided by the mass spectra was e f f i c a c i o u s l y deciphered at the data handling stage. Table 30 l i s t s the names of the compounds i d e n t i f i e d , the type of samples where the v o l a t i l e s were found, the means of i d e n t i f i c a t i o n , and the mass spectral data. I d e n t i f i c a t i o n s were made by comparison with l i b r a r y mass spectra and retention times of authentic compounds (MS,RT) or by deduction from the mass spectra alone (MS). The l a t t e r cases must be regarded as tentative i d e n t i f i c a t i o n s . There were instances where di f f e r e n t peaks bore the same chemical i d e n t i t y i n d i c a t i n g the d i f f i c u l t y i n d i s t i n g u i s h i n g isomers based on t h e i r mass spectra. Although a number of compounds were present i n only one type of sample, many v o l a t i l e components were found i n two or a l l three types of canned salmon. Among the 130 compounds that have been analyzed, the number of components _ detected for canned pink salmon of grade r e j e c t (83) were higher than that of canned pink salmon of grade A and canned la t e run chum salmon (60 and 66, r e s p e c t i v e l y ) . The classes of v o l a t i l e compounds included 147 Attention u> («ln> Figure 35. Total ion chromatogram obtained by gas chromatography/mass spectrometry (GC/MS) of headspace v o l a t i l e components from canned pink salmon of good q u a l i t y (grade A) concentrated using Tenax trap sampling (TTS). Compounds are i d e n t i f i e d by peak numbers shown i n Table 30. 148 Retention tXmt (alb) 1U /: \ 112 \120 • t t a n t l o o tl» Figure 36. Total ion chromatogram obtained by gas chromatography/mass spectrometry (GC-MS) of headspace v o l a t i l e components from canned pink salmon of advanced decomposition (grade reject) concentrated using Tenax trap sampling (TTS). Compounds are i d e n t i f i e d by peak numbers shown i n Table 30. 149 7.51*4 -, Figure 37. Total ion chromatogram obtained by gas chromatography/mass spectrometry (GC-MS) of headspace v o l a t i l e components from canned chum salmon of advanced sexual maturity (spawning dark) concentrated using Tenax trap sampling (TTS). Compounds are i d e n t i f i e d by peak numbers shown in Table 30. 150 Table 30. V o l a t i l e compounds t e n t a t i v e l y i d e n t i f i e d i n canned pink salmon of grades A and r e j e c t , and canned l a t e run chum salmon by Tenax trap sampling/ga3 chromatography/mass spectrometry (TTS/GC/MS). Peak Compound name Sample 3 ID Mass S p e c t r a l Data: no. mass t o charge r a t i o (abundance) 1 acetone G , P , L MS, RT 4 3 ( 1 0 0 ) ( 58( 5 1 ) , 41( 3 6 ) , 57 ( 3 4 ) , 56 ( 25) , 59 ( 24) , 42( 24) , 86( 8) 2 hexane G,P,L MS, RT 57(100) 41 ( 6 0 ) , 43( 5 8 ) , 56( 5 2 ) , 42( 32) , 29 ( 30) , 86( 11) , 39( 9) 3 2-butanone G,P,L MS, RT 43 (100) , 72 ( 21) 57( 8 ) , 42( 6 ) , 44 ( 5) , 82 ( 2) , 53( 1) , 41( 1) 4 cyclohexane G MS 56(100) 84( 9 3 ) , 41( 6 6 ) , 55( 3 8 ) , 69( 28) , 58 ( 23) , 57( 20) , 42( 12) 5 2-methyl-butanal P,L MS, RT 57(100) , 41( 9 2 ) , 58( 9 0 ) , 44( 2 9 ) , 86( 17) , 78 ( 14) , 84( 12) , 71( 11) 6 benzene G,P MS, RT 78 (100) 77 ( 2 0 ) , 51( 1 9 ) , 52( 1 6 ) , 50( 14) , 39 ( 11) , 79 ( 10) , 76 ( 4) 7 methoxy-ethane P,L MS 45 (100) , 29 ( 47) 60( 2 7 ) , 15( 2 4 ) , 27 ( 19) , 31 ( 19) , 59 ( 11) , 26 ( 7) 8 l-penten- 3-ol G MS, RT 57 (100) , 86( 20) 81( 14 ) , 59 ( 6 ) , 96( 4) , 53 ( 4) , 41{ 1) , 42 ( 1) 9 3-pentanone G,P,L MS, RT 57(100) , 86( 17) 43( 12 ) , 81( 9 ) , 56 ( 6) , 41 ( 4) , 55 ( 3) , 96( 3) 10 3-hydroxy-2-butanone P MS, RT 45(100) , 43 ( 76) , 59( 3 1 ) , 57( 2 5 ) , 81( 16) , 88 ( 12) , 42( 10) , 86( 8) 11 dimethyl d i s u l f i d e G MS, RT 94(100) , 79 ( 50) r 45 ( 2 8 ) , 80{ 1 5 ) , 46( 15) , 81 ( 12) , 47( 11) , 55( 10) 12 4-methyl-2-pentanone L MS, RT 43 (100) , 58 ( 56) , 57( 36) 41( 2 6 ) , 55( 23) , 85 ( 19) , 40( 15) , 1 0 0 ( 14] 13 3 ,3-dimethyl-2-butanone L MS, RT 57 (100) , 43 ( 72) , 41 ( 31) 58 ( 2 0 ) , 56( 17) , 59{ 14) , 1 0 0 ( 11) , 70 ( 10) 14 3-methyl-l-butanol P MS, RT 55(100) , 70 ( 89) , 41( 70) 42( 5 9 ) , 57 ( 58) , 56 ( 47) , 43( 46) , 45( 13) 15 methyl-benzene G,P,L MS, RT 91(100) , 92 ( 68) , 65( 15) 39 ( 10 ) , 63( 7) , 51 ( 5) , 93( 4) , 45( 4) 16 2-methyl-thiophene G, L MS 91(100) , 92( 57) , 65( 11) , 63( 7) 51 ( 6) , 50 ( 4) , 89( 4) , 94 ( 4 17 2-methyl-2,4-hexadiene G MS 81 (100) , 96( 49) , 79 ( 46) , 53( 21) 41( 18) , 67( 15) , 55 ( 15) , 54( 10 18 3-hexanone G MS, RT 57 (100) , 43( 79) r 71 ( 62) 100 ( 31) 44( 5] , 72 ( 4) , 42 ( 3] , 58 ( 2 19 unknown G MS 55 (100 ) , 84( 42) r 41 ( 36) 69( 27) 56( 25) , 42 ( 14) , 70 ( 4] , 58 ( 4 20 cyclopentanol P MS, RT 57(100) , 42 ( 35) , 41( 30) 55( 30) 70 ( 20 , 44 ( 20 , 43( 19 , 68 ( 7 151 Table 30. Volatile compounds tentatively identified in canned pink salmon of grades A and reject, and canned late run chum salmon by TTS/GC/MS (cont.d). Peak Compound name Sample3 IDb Mass Spectral Data: no. mass to charge ratio (abundance) 21 2-hexanone L MS,RT 22 unknown P MS 23 octane G,P MS,RT 24 tetrahydro-2,5-dimethyl- L MS furan 25 2-octene P MS 26 3-methyl-l,4-heptadiene G,P MS 27 3-ethyl-l,4-heptadiene G,P MS 28 unknown G MS 29 4-pyridinamine L MS 30 5-methyl-3-hexanone L MS 31 methyl-pyrazine P MS, RT 32 2,3-dimethyl-l,4- G MS hexadiene 33 3-ethyl-thiophene L MS, RT 34 (E,E)-l,3,6-octatriene G,P MS 35 ethyl-benzene L MS, RT 36 2,3,3-trimethyl-l,4- G MS pentadiene 37 dimethyl-benzene G,P,L MS 38 1,3-cyclopentanedione P MS 39 1,3-cyclooctadiene G,P MS 40 3-heptanone L MS, RT 43(100), 58 ( 67), 57 ( 27), 41( 18), 44( 14) ,100( 12) , 71{ 7) , 55( 6) 41 (100), 69 ( 83), 81( 56), 55( 54), 67( 47) , 84( 26) , 68( 22) , 56{ 17) 43(100), 56( 84), 57( 73), 44( 73), 41( 67) , 85( 31) , 71( 29) , 72( 19) 56(100) 44 ( 87), 57 ( 63), 41 ( 61), 43( 21) , 55 ( 19) , 72( 16) , 45( 14) 56 (100), 55 ( 94), 41( 83), 57( 66), 95( 58) , 70( 57) , 42( 34) , 59( 21) 81(100) , 67 ( 71) , 68( 63), 55 ( 30), 79( 28) ,110( 24) , 53( 22) , 41( 21) 81(100) , 67 ( 94), 68( 78), 79( 46), 55( 40) ,110( 36) , 53( 37) , 41( 36) 44(100), 55 ( 56), 69( 52), 43 ( 51), 41( 43) , 56( 40) , 42( 37) , 57( 31) 94(100) , 6 ( 47), 53( 17), 52( 9), 93 ( 8) , 95 ( 7) , 66( 6) , 41( 5) 57(100) , 44( 33) , 85( 24), 43( 23), 41( 10) ,114( 8) , 72 ( 6) , 45( 4) 94(100) , 67 ( 36) , 43( 19), 53 ( 8), 42 ( 8) , 45( 7) , 55( 5) , 41( 5) 95(100) 41 ( 50) 55( 45), 67( 44), 69( 29) , 57 ( 27) ,H0( 25) , 83( 21) 44(100) , 97 ( 49) ,112( 18) 84( 6), 59 ( 5) ,111( 4) , 71( 4) , 67( 4) 91(100) , 79( 92) , 77( 51) 93( 42), 108( 22) , 78 ( 13) , 66( 8) , 94( 5) 91(100) r 58 ( 37) r 43( 31), 106( 17), 44( 16) , 57( 9] , 73( 8) , 92 ( 8) 95(100) , 67 ( 49) ,110( 31) 55( 19), 41( 14) , 53 ( 8] , 81( 7) , 97 ( 5) 91(100) ,106( 43) ,105( 14) 77( 11), 95 ( 8] , 51 < 7) , 44( 7) , 79( 7) 98(100) , 41 ( 32) r 43( 27) r 42( 16), 70( 16) , 69 ( 15 , 53( 14) , 81( 13) 79(100) , 77 ( 40) ,108( 29) , 93 ( 15), 80( 15] , 44 < 15] , 66( 15] , 78( 15) 57(100) , 85( 21) , 41( 12) 72( 12), 43 ( 8] ,114( 7] , 58(- 4] , 95( 4) 152 Table 30. Volatile compounds tentatively identified in canned pink salmon of grades A and reject, and canned late run chum salmon by TTS/GC/MS (cont.d). Peak Compound name no. Sample ID Mass Spectral Data: ma33 to charge ratio (abundance) 41 2,5-diethyl furan P 42 2-heptanone G,P,L 43 nonane G,P 44 4-methyl-hexanal G,P,L 45 2,6-dimethyl-pyrazine P,L 46 2,3-dimethyl-pyrazine L 47 4-ethyl-phenol P 48 trimethyl-benzene P 49 4,5-dimethyl-thiazole L 50 unknown G,P 51 unknown P 52 unknown G,P 53 unknown G,P 54 unknown P 55 propyl-benzene G,P 56 unknown L 57 ethylmethyl-benzene G,P,L 58 2,4-dimethyl-hexanone L 59 dimethyl trisulfide G,P 60 3, 5, 5-trimethyl-2-hexene P,L MS 109(100), 124( 20), 104 ( 12), 56( 10), 110( 6) , 94( 5) , 57( 5) , 67 ( 5) MS, RT 43(100), 91( 87), 58( 73), 106( 38), 44( 31) , 71( 17) ,105( 14) , 77 ( 14] MS, RT 42(100) 57( 89), 85( 30), 56( 16), 71( 12) , 84( 10) , 41( 7) , 99( 6] MS 44(100) 70( 88), 43( 83), 55( 53), 57( 49) , 41( 47) , 41( 32) , 71( 23] MS, RT 108(100) , 42( 64) 43( 27), 71( 23) 57( 23) , 56( 8) , 41( 7) , 81 ( 7 MS, RT 67(100) r108( 47) , 44( 37), 41( 28) 59 ( 9) , 66( 8) , 52( 8) , 98 ( 6 MS, RT 107(100), 122 ( 38), 77( 15), 108( 10) 91 ( 5) , 39( 5) , 65( 3) , 53 ( 3 MS 57(100) 71( 42), 105( 36), 45( 33) 107( 26) , 41( 26) , 70( 25) ,108( 21 MS, RT 113(100) , 71( 59) , 85( 19), 86( 18) 45( 13) , 59( 8) , 58( 6) , 57 ( 6 MS 80(100) , 68( 63), 69( 59), 56 ( 53) 95( 51) , 70( 42) , 57 ( 31; , 97 ( 23 MS 64(100) , 41( 39), 57( 31), 71 ( 28) 111(!'26) ,113( 22) , 63( 22) , 95( 11 MS 95(100) ,126( 32), 43( 29), 57 ( 26) 41( 22) , 67 ( 18] , 65 ( 19) , 44 ( 18 MS 79(100) r 54( 75) , 81( 69) 55( 63) 72 ( 51) , 53( 50) , 82( 37) , 85( 37 MS 81(100) r 79 ( 92), 124( 70) 43 ( 44) 41( 34) , 54 ( 21) , 55( 20] , 67( 19 MS, RT 91(100) , 43( 31) , 57 { 27) r 44 ( 20) 120 ( 14] , 56 ( 12) , 41 ( 12] , 85 ( 11 MS 54(100) , 55( 64) , 71( 37) r 41( 30) 83 ( 24] , 56( 10] , 68( 10] , 42 ( 10 MS 105(100) ,120( 29) , 77( 11) ,101( 9) 91 ( 6] , 51 ( 5] , 57 ( 5] , 65 ( 4 MS 57(100) , 43( 37) , 55( 25) , 71( 23) 70( 21] , 41( 21] , 69( 12 , 85 ( 11 MS 126(100) , 79( 53) , 45( 25) F105( 24) U K 11] , 64( 11] , 47 ( 11 ,120( 9 MS 57(100) , 70( 45) , 55( 31) r 41( 24) 69 ( 23 , 43( 8 , 93 ( 7 , 42 ( 6 153 Table 30. Volatile compounds tentatively identified in canned pink salmon of grades A and reject, and canned late run chum salmon by TTS/GC/MS (cont.d). Peak Compound name no. Sample ID Mass Spectral Data: mass to charge ratio (abundance) 61 ethylmethyl-benzene G,P,L MS 62 unknown G,P MS 63 3-octanone L MS,RT 64 2-octanone L MS,RT 65 7-octen-2-one P MS 66 2-pentyl-furan P MS 67 trimethyl-benzene G MS 68 decane G,P,L MS,RT 69 unknown G,P,L MS 70 octanal L MS,RT 71 trijnethyl-pyrazine L MS, RT 72 (s)-2,3-dihydro-4-(l- P MS methylpropyl)-furan 73 unknown G,P,L MS 74 trimethyl-benzene G,P,L MS 75 (E)-2-hepten-l-ol G MS 76 unknown P,L MS 77 6-ethyl-2-methyl- G,P MS octane 78 1-decene L MS 79 trimethyl-octane G,P MS 80 1-phenyl-ethanone L MS 105(100) 91( 14 116(100) 42 ( 20 43(100) 71( 36 58(100) 41( 14 43(100) 55 ( 19 81(100) 105 ( 12 105(100) 91 ( 7 57(100) 85( 24 68(100) 79( 71 42(100) 57 ( 35 122(100) 54 ( 5 97(100) 43 ( 5 81(100) 41 ( 8 57(100) 43 ( 17 57(100) 93 ( 14 57(100) 41 ( 18 57(100) 70( 16 43(100) 41 ( 36 57(100) 70 ( 7 105(100) 43 ( 26 57 ( 24) , 55( 11 46( 65) , 43( 11 57( 59) , 41( 25 43( 91) , 81( 13 68( 48) , 69( 10 82( 22) , 53 ( 12 120 ( 48) , 81( 6 43 ( 81) , 56( 15 107( 88) , 81( 61 56( 46) , 44( 35 42( 77) , 52 ( 4 126( 13) , 98 ( 4 53( 16) , 55 ( 7 105( 89) , 71( 14 56( 55) , 69( 11 56( 30) , 45( 14 71 ( 60) , 41 ( 13 55( 79) , 82( 32 56( 25) , 58 ( 6 77 ( 87) , 51( 19 120( 24) , 70( 11 74( 31) , 73( 10 72( 47) , 55( 22 71( 21) , 57( 12 41( 27) , 71( 8 138 ( 14) , 41 ( 9 77 ( 8) , 79( 6 71( 38) , 81 ( 11 136( 78) , 77 ( 58 43( 44) , 55{ 32 81( 9) , 49( 2 45( 10) , 42 ( 4 110( 13) , 78( 6 120 ( 27) ,H9( 11 41( 21) , 67( 10 43( 29) , 55( 13 85( 26) , 55 ( 6 57( 67) , 97( 30 43( 15) , 85 ( 4 120 ( 48) , 85{ 17 44( 22), , 77( 11) 41 ( 22), , 45( 9) 99( 41), , 68( 18) 59( 18), , 85{ 8) 67 ( 19), , 97( 7) 43( 13), , 57( 9) 119( 8), ,106( 5) 41( 30), , 70( 11) 94( 74), , 53( 30) 84( 40), , 41 ( 32) 123 ( , 50 ( 44 ( , 41( 67 ( , 69 ( 56( 27), , 41( 8) 55 ( 17), , 54( 9) 44( 19), , 93( 12) 43( 19), , 58( 5) 70( 43), , 83( 27) 71( 14), ,113( 4) 57 ( 33), , 71( 13) 7) , 2) 5), 4) 8) , 5) 154 Table 30. Volatile compounds tentatively identified in canned pink salmon of grades A and reject, and canned late run chum salmon by TTS/GC/MS (cont.d). Peak Compound name Sample* ID Ma3S Spectral Data: no. mass to charge ratio (abundance) 81 unknown L MS 43(100) , 71( 72) , 95( 60) , 57( 36, 99( 26) ,124( 24) , 58( 24) , 69( 22) 82 trimethyl-octane G,P MS 57(100) , 56( 19) , 71( 18) , 43{ 9, 85( 8) , 41( 7) ,114( 4) , 57( 3) 83 dimethylethyl-benzene G,P MS 119(100) , 57 ( 67), 134( 26), 56( 24), 120( 17) , 71( 14) ,105( 11) ,124( 10) 84 trimethyl-octane G,P MS 57(100) 56( 44), 71( 36), 43( 31), 41( 11) , 70( 8) , 85( 6) , 55( 6) 85 trimethyl-octane G,P MS 71 (100) r 57( 83), 43( 61), 85{ 27), 70( 24) , 41( 15) , 56( 15) , 55( 13) 86 dimethylethyl-benzene G,P MS 119 (100) ,134( 29) , 91{ 8), 120( 7), 77 ( 6) ,115( 5) ,117( 4) ,105( 4) 87 unknown G,P MS 79(100) 81( 48), 67 ( 41), 54( 40), 58 ( 36) , 80( 32) , 96( 29) , 43 ( 29) 88 unknown G,P MS 79(100) 67( 55), 81 ( 52), 54 ( 50), 80 { 34) , 91( 26) , 77( 24) , 71( 23) 89 tetramethyl-pyrazine L MS, RT 136(100) , 54( 82) , 42 ( 42), 135( 40), 53 ( 11) , 55( 7) ,137( 7) , 81( 6) 90 2-nonanone L MS, RT 58(100) , 43( 65) r 57( 24), 59( 21), 71( 21) , 41( 10] , 55( 8) ,142( 5) 91 undecane G,P,L MS, RT 57(100) , 43( 53) , 71( 33), 85( 20), 41( 14) , 56( 13] , 69( 11) , 70 ( 11) 92 nonanal G,P,L MS, RT 57(100) , 43( 57) , 56( 54), 41 ( 52), 55 ( 45) , 44( 45] , 70( 37) , 98( 26) 93 2, 2, 5-trimethyl-3,4- L MS 43(100) , 57( 38) , 71( 22) 58( 21), hexadione 69( 15] , 98( 12] , 41( 10] , 85( 7) 94 6-nonynoic acid G,P MS 79(100) , 94( 49) , 67( 44) 81( 38), 55( 22] , 41( 17 , 93( 19] , 77 ( 16) 95 1,4-undecadiene P MS 81(100) , 67( 89) , 68 ( 69) 55( 48), 79( 42] , 43( 32 , 95( 28] , 41( 19) 96 unknown L MS 95(100) , 55( 64) , 69( 61) r 58( 58), 41( 55] , 83( 47 , 57( 39 , 70 ( 36) 97 2-pentyl-thiophene L MS 97(100) , 83( 54) , 55 ( 34) , 56( 33), 98 ( 26] , 84( 20 , 57 ( 19 , 41( 17) 98 6-nonynoic acid G,P MS 79(100) , 67( 30) , 94( 24) , 81( 21), 93 ( 20 , 41( 18 , 55 ( 17 ,108( 15) 99 tetramethyl-benzene P MS 119(100) ,134( 33) , 79( 13) , 91( 11), 67( 10] ,120( 8 , 77 ( 7 ,133( 7) 100 decahydro-2-methyl- P MS 152(100) , 95( 83) , 82( 82) , 81( 66), naphthalene 67 ( 62 , 96( 60 ),137( 55 , 55( 55) 155 Table 30. V o l a t i l e compounds t e n t a t i v e l y i d e n t i f i e d i n canned pink 3almon of grades A and r e j e c t , and canned l a t e run chum salmon by TTS/GC/MS (cont.d). Peak Compound name Sample 3 ID Mass S p e c t r a l Data: no. mass to charge r a t i o (abundance) 101 methyl-cyclodecane P MS 69(100) 68 ( 88), 55( 75), 41 ( 58), 83( 55) , 70( 48) , 97( 39) , 56( 35) 102 diethylmethyl-benzene P MS 133(100), H9( 96), 148( 32), 43 ( 28), 91( 24) ,120( 18) , 45( 16) , 57( 15) 103 naphtalene P MS, RT 128(100) 137 ( 82), 79( 64), 91 ( 28), 57( 17) ,H9( 11) ,129( 11) , 63( 11) 104 2-decanone G,L MS, RT 58(100) 43( 64), 59( 30), 71( 28), 57{ 15) , 41( 9) ,156( 9) , 55( 8) 105 dodecane G,P, L MS, RT 57(100) 43( 62), 71( 40), 85( 25), 41( 24) , 56( 21) , 55( 17) , 70( 14) 106 decanal G,P, L MS, RT 57(100) , 43( 92), 41( 79), 55 ( 62), 44 ( 56) , 70( 51) , 82( 50) , 71( 48) 107 unknown P MS 81(100) , 53( 8), 160( 3), 82 ( 2), 137( 2) , 57( 2) ,1H( 1) ,119{ 1) 108 2-hexyl-thiophene L MS 97(100) , 41( 36), 98( 34), 43 ( 33), 70( 31) , 57( 31) , 69( 29) , 83( 29) 109 unknown G,P MS 43(100) , 57( 90), 81( 90), 79( 89), 108( 80) , 91( 60) , 77( 51) ,120( 41) 110 2-undecanone P MS, RT 43(100) , 58( 91), '71( 41), 59 ( 36), 81( 15) , 55( 10) , 57( 8) , 85( 8) 111 unknown P MS 57(100) , 95( 76), 43 ( 7(4), 71( 40), 79( 38) ,108( 35) ,109( J23) ,120( 21) 112 tridecane G,P MS, RT 57(100) , 71( 62), 43( 56), 85( 41), 41( 27) , 56( 19) , 55( 17) , 70 ( 15) 113 1-tridecene L MS 57(100) , 55( 86), 43( 81), 69 ( 55), 70( 46) , 56( 40) , 41( 40) , 83( 39) 114 undecanal L MS, RT 43(100) , 57( 94), 41( 72), 55 ( 69), 82{ 54] , 56( 44) , 44( 44! , 29( 43) 115 unknown (C12H130) L MS 55(100) , 44( 95), 56( 82), 82 ( 77), 68( 55] , 41( 43) , 81 ( 43) , 69( 41) 116 6-tridecene L MS 97(100) , 43( 87), 44( 71), 41 ( 68), 71( 53] , 70( 48) , 98( 47; , 69( 40) 117 unknown (C12H18) P MS 91(100) , 80 ( 99) , 67( 81) 41( 79), 69( 76] ,105( 68) , 92( 63) , 93( 43) 118 3-dodecanone L MS 57(100) , 72( 68), 43( 23), 155 ( 23), 85( 22] , 55( 22) , 71( 13] , 41( 12) 119 tetradecane G,P, L MS, RT 57(100) , 43( 73) 71( 44) 85( 21), 41( 11] , 55 ( 9) , 56( 8] , 79( 6) 120 unknown P MS 79(100) , 55( 47) 93( 39) 80 ( 36), 67-( 36] , 91( 33) , 41( 31] , 77 ( 27) 156 Table 30. Volatile compounds tentatively identified in canned pink 3almon of grades A and reject, and canned late run chum salmon by TTS/GC/MS (cont.d). Peak Compound name Sample3 IDb Mass Spectral Data: no. raaaa to charge ratio (abundance) 121 unknown P,L MS 105(100), 79( 63), 93( 81), 91( 64), 77( 63), 69( 63),107( 58), 55( 58) 122 pentadecane G,P,L MS, RT 57(100), 4K 25), 72( 70), 43( 62), 85( 42), 55( 17), 56( 14), 70( 12) 123 unknown P MS 43(100), 56( 15), 57( 62), 71( 36), 85( 17), 55( 11), 99( 9), 69( 9) 124 hexadecane G,P,L MS, RT 57(100), 41( 15), 43( 82), 71( 46), 85( 35), 70( 11), 56( 10), 55{ 10) 125 unknown L MS 71(100), 56( 6), 43( 31), 41( 7),159( 7), 55( 5), 57( 4), 69( 4) 126 unknown L MS 57(100), 43( 41), 45( 58), 69( 44), 83( 42), 56( 38), 97( 31), 55( 30) 127 2,2,4,4,6,8,8-heptaroethyl-nonane L MS 57(100), 41( 10), 43( 28), 44( 22), 85( 13), 56( 8), 55( 6),113( 6) 128 heptadecane P,L MS, RT 57(100), 41( 28), 71( 77), 43< 61), 85( 56), 70( 23), 69( 15), 99( 8) 129 2,6,10,14-tetramethyl-pentadecane P,L MS 57(100), 56( 20), 71( 92), 43( 46), 85( 31), 55( 18),113( 17), 41( 16) 130 nonadecane P,L MS, RT 57(100), 56( 19), 71( 81), 43( 47), 85( 32), 55( 18), 41( 15),113( 14) G, canned pink salmon of grade A ; P , canned pink salmon of grade reject; L, canned chum salmon of advanced sexual maturity. MS, tentatively identified by mass spectrometry; RT, retention time consistent with that of authentic compounds. 157 hydrocarbons (50), ketones (22), sulfur-containing compounds (6), nitrogen-containing compounds (6), aldehydes (6), alcohols (5), acids (2), and miscellaneous compounds (6). Figures 38, 39, and 40 are FID chromatograms of CCS v o l a t i l e s from samples of canned pink salmon of grade A, canned pink salmon of grade reject, and canned l a t e run chum salmon, respectively. Based on the retention time of authentic standards and the pattern of e l u t i o n , several peaks in these figures have been associated with compounds l i s t e d i n Table 30. Also associated with these figures are Tables 31, 32, and 33 which l i s t the odour a t t r i b u t e , the retention time and the possible i d e n t i t y of the v o l a t i l e that caused the sensation. The CCS samples of canned pink salmon of poor q u a l i t y included substantial concentrations of ethanol. To avoid overloading the column, an i n i t i a l purging was performed for 5 min before cryofocussing. Although CCS allowed a non-discriminatory c o l l e c t i o n of v o l a t i l e s which were subsequently evaluated f o r t h e i r sensory c h a r a c t e r i s t i c s , the retention times of the e a r l y e l u t i n g v o l a t i l e s tended to vary due to the presence of water c o l l e c t e d during purging. As opposed to FID, mass spectrometers are s e n s i t i v e to large quantities of eluted water. Therefore t h i s CCS means could not be used i n conjunction with mass spectrometry. The FID chromatograms obtained using CCS measured only h a l f of the v o l a t i l e s concentrated since a 1:1 s p l i t t e r was used to gather the aromagrams. In spite of t h i s , comparison with both sets of chromatograms (Figures 35, 36, 37 vs 38, 39, 40) shows that, in general, higher OD 40 75 11.0 145 18.0 215 25.0 285 32.0 365 39.0 425 46.0 405 63.0 565 60.0 Retention time (min) Figure 38. Chromatogram obtained by gas chromatography/f lame i o n i z a t i o n detection (GC/FID) of headspace v o l a t i l e components from canned pink salmon of good q u a l i t y (grade A) concentrated using cryofocussing. Compounds are i d e n t i f i e d by peak numbers shown i n Table 30 and l e t t e r s refer to Table 31. 0.0 6.0 10.0 15.0 20.0 26.0 30.0 36.0 40.0 45.0 50.0 65.0 60.0 Retention time (min) Figure 39. Chromatogram obtained by gas chromatography/f lame i o n i z a t i o n detection (GC/FID) of headspace v o l a t i l e components from canned pink salmon of advanced decomposition (grade reject) concentrated using cryofocussing. Compounds are i d e n t i f i e d by peak numbers shown in Table 30 and l e t t e r s r efer to Table 32. I I I I I I I I I II I I I II I I I I A B C O E F G HI JK L MN OPQ R S T I ^ • • ^ I I I I I I I 1 1 1 1 0 0 3J5 6 J 0 10JO 16J0 200 25.0 30.0 36J0 40J0 46.0 600 56.0 600 Retention time (min) Figure 40. Chromatogram obtained by gas chromatography/f lame i o n i z a t i o n detection (GC-FID) of headspace v o l a t i l e components from canned chum salmon of advanced sexual maturity (spawning dark) concentrated using cryof ocussing. Compounds are i d e n t i f i e d by peak numbers shown i n Table 30 and l e t t e r s r e f e r to Table 33. 161 Table 31. Cryofocussing concentration sampling/gas chromatography/odour evaluation (CCS/GC/OE) of volatile components from canned pink salmon of good quality (grade A). ID RT Possible compound Odour attribute (min) A 5 .4 2-butanone green, raw, nutty B 5 .8 unknown burnt C 6 .4 unknown burnt D 6 .9 unknown chlorine-like E 9 .1 l-penten-3-ol p l u 3 unknown rancid, malodorant F 9 .9 dimethyl disulfide sulfurous G 11 .6 2-methyl thiophene green, gasoline-like H 12 .6 unknown f i 3 h y I 13 .5 3-methyl-l,4-heptadiene burnt, medicinal, stinky J 15 .3 1,3,6-octatriene chlorine-like K 15 i .7 2,3,3-trimethyl-l,4-pentadiene and dimethyl benzene pungent, st inky L 17 .4 4-methyl hexanal fish skin M 22 .3 trimethyl benzene concrete, petroleum-like N 29 .6 nonanal aldehydic, spicy, green 0 33 .3 unknown cucumber P 37 .2 decanal aldehydic, fishy Q 40 .5 undecanal green, waxy 162 Table 32. Cryofocu3sing concentration sampling/gas chromatography/odour evaluation (CCS/GC/OE) of volatile components from canned pink salmon of advanced decomposition (grade reject). ID RT Possible compound Odour attribute (min) A 1 .7 ethanol alcoholic B 2 .2 unknown burnt C 5 .8 2-butanone green, nutty D 7 .9 3-pentanone chlorine-like E 8 .1 3-hydroxy-2-butanone buttery, toast-like F 8 .9 3-methyl-l-butanol alcoholic, solvent-like G 12 .6 3-methy1-1,4-heptadiene fishy H 12 .9 3-methyl-l,4-heptadiene f i3hy I 13 .4 methyl pyrazine roasted, baking J 16 .6 4-methyl hexanal sour, sea-like K 17 .4 2,6-dimethyl pyrazine roasted, nutty L 19 .0 dimethyl trisulfide burnt hair M 20 .7 unknown wet dog N 32 .9 unknown spring smell, meaty 0 40 .1 undecanal green, waxy, flowery P 40 .5 unknown cooked rice 163 Table 33. Cryofocussing concentration sampling/gas chromatography/odour evaluation (CCS/GC/OE) of volatile components from canned chum salmon of advanced sexual maturity (spawning dark). ID RT Possible compound Odour attribute (min) A 1 .7 ethanol alcoholic A 4 .3 2-butanone green, nutty, raw B 4 .5 unknown strong odour of hay, straw, cooked malt, cereal, fishy C 4 .8 unknown hay, straw-like D 5 .5 2-methyl-butanal slight odour of hay, straw-like E 8 .5 unknown chlorine, swimming pool F 9 .8 4-methyl-2-pentanone chlorine-like, rubber, quinine G 12 .1 2-methyl thiophene slightly fishy, green, sweet H 13 .3 2-hexanone nutty, green I 13 .8 tetrahydro-2,5-dimethyl furan burnt, ethereal J 17 .2 2-heptanone sweat-like J K 17 .6 4-methyl-hexanal fishy, cardboard, straw-like L 20 .1 4,5-dimethyl thiazole slightly meaty M 21 .3 2-octanone earthy, soil, mushroom, herbal N 21 .9 octanal pungent, green 0 26 .5 tetramethyl pyrazine nutty, grassy, roasted P 26 .6 nonanone green, flower, cream Q 27 .3 nonanal 3picy, green R 29 .0 2,2,5-trimethyl-3,4-hexadione very faint butter-like S 30 .7 unknown spring smell, sweet T 34 .1 undecanal waxy, flowery concentrations of v o l a t i l e s are achieved with the CCS method than with the TTS method. This could be observed p a r t i c u l a r l y i n the f i r s t 10 min of v o l a t i l e e l u t i o n . Tenax possesses a low capacity for the pre-concentration of highly v o l a t i l e organic molecules. In addition, dry purging for 1 h before releasing the v o l a t i l e s was necessary to remove most of the moisture accumulated in the Tenax trap. Compounds of molecular weight lower than about 5 carbons, including a l i p h a t i c alcohols and a l i p h a t i c carboxylic acids, are not q u a n t i t a t i v e l y retained at room temperature (Nunez and Gonzalez, 1984). The hydrocarbons i d e n t i f i e d included a homologous series of n-hydrocarbons ranging from C6 to C19 (Table 30). Watanabe and Sato (1971) suggested that saturated alkanes could r e s u l t from decarboxylation and s p l i t t i n g of carbon-carbon chains of higher f a t t y acids. When heating t r i s t e a r i n i n a i r , the presence of n-hydrocarbons among the v o l a t i l e s was interpreted by Selke et a l . (1975) to a r i s e from the reaction of a l k y l -free r a d i c a l s , which were the product of thermally decomposed hydroperoxides with free hydrogen r a d i c a l s . No defined aroma pertaining to alkanes were d i r e c t l y i d e n t i f i e d i n the present study. Alkanes detected i n beef (Larick and Turner, 1990), c r a y f i s h t a i l meat (Vejaphan et a l . , 1988), and crabmeat (Matiella and Hsieh, 1990) d i d not contribute s i g n i f i c a n t l y to aroma as they possess weak odour. In the past, alkanes from C l l to C15 have been reported to possess sensory c h a r a c t e r i s t i c s which resembled a l i p h a t i c alcohols (Ohloff et a l . , 1985). Seven branched alkanes, two c y c l i c alkanes, s i x straight chain alkenes, s i x branched alkenes, and one c y c l i c alkene were i d e n t i f i e d . 2,6,10,14-Tetramethylpentadecane, i d e n t i f i e d i n t h i s study, was also found in cooked Antarctic k r i l l (Kubota et a l . , 1982) and c r a y f i s h waste (Tanchotikul and Hsieh, 1989) . A c h l o r i n e - l i k e smell was associated with 1,3,6-octatriene. A pungent, unpleasant aromatic sensation was perceived at retention times corresponding to the co-elutions of 2,3,3-trimethyl-1,4-pentadiene and dimethyl benzene. 3-Methyl-l,4-heptadiene was characterized as having a burnt, medicinal, disagreeable odour i n canned pink salmon of grade A while i t was associated with a f i s h y smell i n canned pink salmon of grade r e j e c t . Although unsaturated and aromatic hydrocarbons are known to contribute to marine flavours of s h e l l f i s h and seaweed (Ohloff et a l . , 1985), i t i s possible that an u n i d e n t i f i e d compound responsible for the odour i n grade A canned pink salmon might have co-eluted with 3-methyl-l,4-heptadiene. Although the o r i g i n of most of these compounds remains uncertain, i t i s possible that some of the branched and c y c l i c hydrocarbons are secondary reaction products from the thermal oxidation of carotenoids and other unsaturated l i p i d s (Ohloff et a l . , 1985) . Sixteen aromatic hydrocarbons, including 14 alkylbenzenes and 2 naphthalenes, were i d e n t i f i e d i n canned salmon headspaces. Although they are often found as artefacts (from solvent residues or from Tenax degradation), the experimental method d i d not d i r e c t l y contribute to these benzene d e r i v a t i v e peaks since the i n i t i a l backgrounds were examined under comparable experimental conditions before sample analysis. Several of 166 these alkylbenzenes are known oxidation products of l i p i d s and have been found in tea (Habu et a l . , 1985), corn (Buttery et a l . , 1978), nuts (Crain and Tang, 1975; Walradt et a l . , 1971), meat (Shahidi et a l . , 1986), and many heated foods (Forss, 1972). Even i f none of these compounds have a meat or f i s h odour, they may play a role i n the o v e r a l l flavour (Min et a l . , 1979). Watanabe and Sato (1971) reported the formation of various alkylbenzenes from beef fats during heating. Carotenoids have been proposed to be precursors of toluene, xylene, and benzene d e r i v a t i v e s found i n chicken, beef, butter, and f i s h (Borenstein and Bunnell, 1966; Pippen et a l . , 1969) . Possible routes to the aromatic compounds include the oxidation of unsaturated hydrocarbons (Min et a l . , 1977) or other products of fatty a c i d autoxidation (Nonaka et a l . , 1967). However, some of these hydrocarbons may have come from the aquatic environment. If present i n polluted waters, they can be bioaccumulated by marine f i s h and s h e l l f i s h through the skin, g i l l s , and ingestion of contaminated food (Connell and M i l l e r , 1981). In a study done by Neff et a l . (1976), aromatic hydrocarbons were accumulated to a greater extent and were retained longer than alkanes by f i s h . Straight chain and branched aldehydes were a l l found to be saturated (Table 30). Nonanal was the only aldehyde observed in a l l three types of canned salmon in r e l a t i v e l y high concentration (Figures 35, 36, 37) . Besides nonanal, canned l a t e run chum salmon had high l e v e l s of 2-methyl-butanal, 4-methyl-hexanal, and undecanal. Four v o l a t i l e s t r a i g h t chain aldehydes containing 8 to 11 carbons were i d e n t i f i e d . They contributed spicy, green, waxy notes. A sour, fi s h y , sea-like, cardboard sensation 167 was associated with 4-hexanal (Fig. 38, 39, and 40) . A hay, straw-like aroma reminiscent of l a t e run chum odour was noticed at a retention time corresponding to 2-methyl-butanal. This compound was, however, not the only one involved i n the c h a r a c t e r i s t i c l a t e run odour; two other occurrences s i m i l a r l y characterized as cooked malt, hay, straw, and f i s h -l i k e were found to elute before 2-methyl-butanal (Table 33) . Although detected by the CCS method, these two compounds were part of the breakthrough volume of the tenax material, since a l a t e run odour came out of the trap during purging, and were therefore not i d e n t i f i e d by mass spectrometry. Straight chain aldehydes as well as 2-methyl-butanal have been reported in beef (MacLeod and Ames, 1986), cod (McGill et a l . , 1977) and c r a y f i s h (Vejaphan et a l . , 1988). The formation of alkanals can be attributed to thermal decomposition of hydroperoxides and peroxy r a d i c a l s proposed to be i n i t i a l products of thermally-oxidized fats (Sink, 1973). In some cases, they can originate from the Strecker degradation of amino acids; for example, 2-methyl-butanal may be derived from isoleucine (Dwivedi, 1975). Due to t h e i r low threshold values, aldehydes and ketones are important aroma compounds i n foodstuffs. On the one hand, they contribute to desirable aroma, but on the other hand, they are responsible for rancid odour and flavour during spoilage of fats and f a t t y foods (Forss, 1972). Wilson and Katz (1972) concluded that saturated and unsaturated aldehydes and ketones were important components of the des i r a b l e aroma o f cooked chicken. However, a number of aldehydes and ketones i n sp o i l e d f i s h were i d e n t i f i e d by Wong et a l . (1967) and M i l l e r et a l . (1972b). McGill et a l . 168 (1974; 1977) also showed that aldehydes including 4-cis-heptenal were responsible for the t y p i c a l off-odours of frozen cold stored cod. Ketones were the second largest c l a s s of v o l a t i l e components found i n canned salmon. A larger number of ketones compared to aldehydes have also been reported as flavour v o l a t i l e s i n shrimp (Kubota et a l . , 1986). Ketones were the major v o l a t i l e components i n c r a y f i s h waste and t a i l meat (Vejaphan et a l . , 1988). One aromatic, one unsaturated, and seventeen saturated ketones as well as two alkadiones and one hydroxy ketone were i d e n t i f i e d i n the present i n v e s t i g a t i o n . The most abundant components i n the three types of canned salmon were 2-butanone and 3-pentanone. A green, nutty odour originated from 2-butanone and a c h l o r i n e - l i k e , rubber odour was associated with 3-pentanone and 4-methyl-2-pentanone. 4-Methyl-2-pentanone found i n canned chum salmon of advanced sexual maturity was shown to be present i n beef (MacLeod and Ames, 1986) , crabmeatj (Matiella and Hsieh, 1990), and c r a y f i s h waste (Tanchotikul and Hsieh, 1989). 2, 3-Butanedione (diacetyl) and 3-hydroxy-2-butanone (acetoin) are constituents of many food aromas and provide a buttery flavour (Arctander, 1969). A t o a s t - l i k e , buttery odour associated with acetoin was p a r t i c u l a r l y noted in canned pink salmon of advanced decomposition (Table 32) . Both d i a c e t y l and acetoin have been i d e n t i f i e d i n chicken (Minor et a l . , 1965) as well as in cooked beef and pork (Mottram et a l . , 1982) . In addition, a series of methyl and et h y l ketones (C4 to C12) and a number of branched methyl ketones were detected in the canned salmon samples. A number of them were prevalent i n canned late run chum salmon 169 (Table 30) and were perceived to have various aromas. 2-Butanone and 2-hexanone were l a b e l l e d with a green, nutty sensation compared to 2-heptanone and 2-octanone which r e s p e c t i v e l y evoked sweat-like and earthy, s o i l odours. Autoxidation of f a t t y acids, p a r t i c u l a r l y unsaturates (via hydroperoxides) , has been proposed as a mechanism for the formation of methyl ketones (Thomas et a l . , 1971). Selke et a l . (1975) reported the formation of a homologous serie s of methyl ketones from heated t r i s t e a r i n and concluded that they could be the res u l t of ft-oxidation (from the carbonyl end) of the carbon chain followed by decarboxylation. S i m i l a r l y , various other ketones could p o s s i b l y be derived from d i s t i n c t secondary degradation reactions i n v o l v i n g diverse substances from the l i p i d f r a c t i o n during heating. Among the 5 alcohols detected by the TTS method, two were found i n high concentration: l-penten-3-ol i n canned pink salmon of grade A and 3-methyl-l-butanol i n canned pink salmon of grade r e j e c t . Because of t h e i r r e l a t i v e l y high threshold concentrations, alcohols are generally minor odour contributors unless present at high concentrations or unless they contain unsaturated bonds. In canned grade A pink salmon, a rancid, malodorant aroma was perceived when l-penten-3-ol co-eluted with an unknown compound (Table 31) . l-Penten-3-ol has been recognized i n A t l a n t i c and P a c i f i c oysters (Josephson et a l . , 1985), roasted and b o i l e d shrimp (Kubota et a l . , 1986), p i c k l e d smelt (Josephson et a l . , 1987), fresh and oxidized frozen whitefish (Josephson et a l . , 1983; Josephson et a l . , 1984b), and c r a y f i s h waste (Tanchotikul and Hsieh, 1989). Alcohols may be formed by decomposition of secondary hydroperoxides of f a t t y acids. In a similar pathway to that involved in the generation of 1,5-octadien-3-ol (Wurzenburger and Grosch, 1986), a rearrangement and cleavage of hydroperoxides from l i n o l e i c or arachidonic acids could y i e l d 1-penten-3-ol. Cyclopentanol, which has also been i d e n t i f i e d i n A t l a n t i c and P a c i f i c oysters, can be formed from the c y c l i z a t i o n of l-penten-3-ol by a mechanism analogous to that proposed for the production of l-octen-3-o l to cyclooctanol (MacLeod and Panchasara, 1983). Two al c o h o l i c odours corresponding to ethanol and 3-methyl-l-butanol were smelled during e l u t i o n of v o l a t i l e s from canned pink salmon of advanced decomposition. Ethanol was not detected by the TTS method but i t s presence was deduced based on previous analyses with the s t a t i c headspace method developed i n Chapter I I . Along with ethanol and other v o l a t i l e s , 3-methyl-l-butanol was found i n Chapter III to be an important discriminator for q u a l i t y c l a s s i f i c a t i o n of canned pink salmon. A l i p h a t i c alcohols occur l a r g e l y i n f r u i t s , vegetables, and fermented foods. During the period of r e f r i g e r a t e d storage, known psychrotrophic spoilage microorganisms of f i s h , such as Gram-negative ba c t e r i a of the genera Pseudomonas and Achromobacter, were assumed to have p r o l i f e r a t e d and produced alcohols and other possible compounds. Ethanol can be derived from a variety of fermentable substrates, most sugars, many amino acids, certain organic acids, purines, pyrimidines and other miscellaneous substances whose catabolism leads to the pyruvate-acetaldehyde-ethanol transformation (Brock, 1979) . - M i l l e r et a l . (1973b, c) showed that S.  putrefaciens, P. fluorescens, P. perolens, and Achromobacter produced 3-methyl-l-butanol during growth on f i s h muscle. The presence of t h i s compound has also been observed in yeast fermentations and metabolic by-products of f r u i t c e l l s . Four and five-carbon methyl-branched alcohols were found to be generated by Saccharomyces cerevisae from amino a c i d precursors (Sentheshanmuganathan, 1960) . L a b e l l i n g experiments using postclimacteric banana tissue s l i c e s and l a b e l l e d amino acids demonstrated 14 the transformation of (U- C)-leucine to 3-methyl-l-butanol (Myers et a l . , 1969; 1970) possibly through transamination and decarboxylation. The only phenolic substance detected i n t h i s i n v e s t i g a t i o n was 4-ethylphenol. This compound has previously been found i n cod (McGill et a l . , 1977) and c r a y f i s h waste (Tanchotikul and Hsieh, 1989). It was, however, detected i n trace amounts and d i d not seem to play an important role i n advanced spoilage. During purging of the canned pink salmon of grade A and reject, the effluent going through the cryogenic i r a p and e x i t i n g through the s n i f f i n g port had a sulfurous odour. This odour was not perceived in canned la t e run chum salmon. The compound responsible f o r the aroma was believed to be hydrogen s u l f i d e based on i t s low b o i l i n g point and i t s p r i o r i d e n t i f i c a t i o n among the v o l a t i l e s found i n Chapter I I . Two straight chain sulfur-containing compounds, dimethyl s u l f i d e and dimethyl t r i s u l f i d e , were i d e n t i f i e d by GC-MS of the v o l a t i l e s concentrated by the TTS method (Table 30) . A sulfurous odour corresponding to dimethyl d i s u l f i d e was perceived from canned pink salmon of good quality while dimethyl t r i s u l f i d e gave a burnt hair smell to the e f f l u e n t of canned pink salmon of grade r e j e c t (Tables 31 and 32). Due to t h e i r very low sensory 172 threshold values, v o l a t i l e organic s u l f u r compounds are an important f r a c t i o n of aroma i n numerous foods. For example, when s u l f u r compounds were removed from the d i s t i l l a t e during the cooking of chicken, the t y p i c a l meat aroma also disappeared (Minor et a l , 1965). These compounds have previously been reported i n raw and thermally processed f i s h and crustaceans (Sipos and Ackman, 1964; Josephson et a l . , 1985; Hughes, 1964; Ronald and Thompson, 1964; W h i t f i e l d et a l . , 1981a,b; and Vejaphan et a l . , 1988) . The numbers and concentrations of low molecular weight s u l f u r compounds were c h a r a c t e r i s t i c of the off-odours a r i s i n g during co l d storage of s p o i l i n g cod (Herbert et Shewan, 1976; McGill et a l . , 1977). They can be formed p r i n c i p a l l y during heat treatments from the free, peptidic, and p r o t e i n i c sulfur-amino acids as well as the glutathione pool i n f i s h t i s s u e . In r e f r i g e r a t e d s p o i l i n g f i s h , they may also be of microbial o r i g i n derived from b a c t e r i a such as Pseudomonas and Achromobacter (Kadota and Ishida, 1972; M i l l e r et a l . , 1973b, c) . Using r a d i o a c t i v e l y - l a b e l l e d precursors, Herbert and Shewan (1975, 1976) conducted studies i n which hydrogen s u l f i d e was found to come from cysteine while methionine served as the precursor of methanethiol. Polysulfides may subsequently form as oxidation products of methanethiol (Maruyama, 1970; Christensen et a l . , 1981). Four 2-alkylthiophenes and one a l k y l t h i a z o l e 'were i d e n t i f i e d (Table 30) . A green, sweet, g a s o l i n e - l i k e odour was associated with 2-methyl thiophene and noticed i n canned pink salmon of good q u a l i t y and i n canned late run chum salmon (Fors, 1983) . Thiophenes are important v o l a t i l e s i n the flavour of cooked meat as they can also contribute a mild sulfurous odour. 2-Methylthiophene has been i d e n t i f i e d i n b o i l e d and pasteurized crabmeat (Matiella and Hsieh, 1990), b o i l e d c r a y f i s h t a i l meat (Vejaphan et a l . , 1988), and roasted shrimp (Kubota et a l . , 1986). A meaty aroma compound, 4,5-dimethyl t h i a z o l e , was recognized i n canned l a t e run chum salmon. Fors (1983) also reported t h i s compound to possess a braised, roasted, and meaty flavour. A large number of substituted h e t e r o c y c l i c thiophenes and thiazoles have been reported in thermally processed meat such as beef, pork and chicken (Shahidi et a l . , 1986). The s u l f u r in thiophenes and thiazoles may be derived from amino acids (cysteine, cystine, methionine) or from vitamin B1. It has been suggested that thiophenes are formed by the action of hydrogen s u l f i d e on sugar degradation products such as dehydroreductones and furans (Vernin and Parkanyi, 1982), f u r f u r a l (Shibamoto, 1977), and furanones (van der Ouwelend and Peer, 1975) during the course of the M a i l l a r d reaction. Phospholipid oxidation products such as 1,4-ketoaldehydes or unsaturated aldehydes have also been suggested to i n t e r a c t with ammonia and hydrogen s u l f i d e , derived from the Strecker degradation of cysteine, to give 2-alkylthiophenes (Mottram and Whitfield, 1987) . Similar pools of M a i l l a r d reactants allow for the i n c l u s i o n of nitrogen i n the heterocycles to form derivatives such as t h i a z o l e s . Traces of 3 furan derivatives were i d e n t i f i e d as v o l a t i l e components of canned pink salmon of grade reject while one furanoid compound, tetrahydro-2,5-dimethyl furan, was found i n canned chum salmon of advanced sexual maturity (Table 30). A burnt, ethereal odour was associated with t h i s l a s t compound. The v o l a t i l e 2-pentyl furan, separated and i d e n t i f i e d in t h i s study, has been reported to impart reversion, beany, grassy, and l i c o r i c e - l i k e flavours i n soybean o i l , and can be produced by oxidation of f a t t y acids (Taylor and Mottram, 1990; Smouse and Chang, 1967). Furans can be found i n dehydrated and thermally degraded condensates of carbohydrate, or formed by Amadori rearrangement pathways (Whistler and Daniel, 1985). A number of furanoid v o l a t i l e s have previously been found in brewed coffee (Shimoda and Shibamoto, 1990), f r i e d bacon (Ho et a l . , 1983), roast beef (Min et a l . , 1979), and chicken (Nonaka et a l . , 1967). Besides the a l k y l t h i a z o l e , a t o t a l of 5 nitrogen-containing compounds including 4 alkylpyrazines and 1 pyridinamine were detected i n the canned late-run chum salmon while 2 alkylpyrazines were found i n canned pink salmon of poor q u a l i t y (Table 30) . The alkylpyrazines a l l contributed nutty, roasted odorous notes (Tables 32 and 33) . Pyrazines are known to r e s u l t from the c l a s s i c M a i l l a r d browning reaction involving a combination of NH3 or amino-containing compounds with sugars or other carbonyl compounds (Fors, 1983). Several alkylpyrazines also derive from pyro l y s i s of hydroxyamino compounds such as threonine, serine, ethanolamine, and glucosamine (Kato et a l . , 1970). They have been i d e n t i f i e d i n roasted lamb fat (Buttery et a l . , 1977), cocoa butter (Rizzi, 1967), coffee (Shimoda and Shibamoto, 1990), cooked pork l i v e r (Mussinan, and Walradt, 1974), f r i e d chicken (Tang et a l . , 1983), and pressure cooked beef (Mussinan et a l . , 1973) . Pyrazines are recognized as important contributors to the flavours of a l l roasted, toasted, or s i m i l a r l y heated foods (Maga, 1982) . Although a variety of compounds appear to have contributed towards the t o t a l flavour, no single odour impact compound was found responsible for the c h a r a c t e r i s t i c canned salmon aroma. As has been the case with meat i n the past, t h i s t y p i c a l aroma more l i k e l y r e s u l t s from the sum of the sensory e f f e c t s of a complex array of v o l a t i l e s generated during thermal processing. In addition, various components such as trimethylamine and esters, that have previously been detected i n decomposed f i s h , and were implicated in undesirable aromas, were not i d e n t i f i e d . More work i s needed to complement the exploratory data accumulated thus far and to investigate the i n t e r a c t i o n s of the important aroma components in order to elucidate canned salmon flavour. 176 V. CONCLUSIONS A l l s i x factors studied using the s t a t i c headspace gas chromatographic technique a f f e c t e d the s e n s i t i v i t y and/or r e s o l u t i o n of the chromatograms, and needed to be optimized. Increases i n weight of f i s h flakes, incubation temperature and time brought about s i g n i f i c a n t chromatographic improvement measured by the increase i n t o t a l area count. RCO allowed the completion of several treatment runs at a time i n contrast to previous single-step sequence optimization programs. This optimization method responded to the need of performing several runs consecutively, and was e f f e c t i v e at reaching the optima of i n i t i a l oven temperature, column headpressure, and t o t a l flowrate. At the optimized conditions obtained, the capacity of detection was raised to 80 peaks, of which 34 were i d e n t i f i e d by retention time matching of reference standards and/or by GC-MS. Multivariate interdependence methods such as PC and CFA were useful to decipher the underlying quantitative data structures of the 44 v o l a t i l e s accumulated from the developed SHGC method and allowed a s i m p l i f i c a t i o n of the number of variables for further s t a t i s t i c a l use. Subsequent discriminant analyses applied on the PC scores from canned salmon v o l a t i l e s segregated the species of P a c i f i c salmon, the stages of sexual maturity of chum salmon, and the q u a l i t y grades of pink salmon with high rates of correct c l a s s i f i c a t i o n . Similar effectiveness was obtained when the same SHGC and s t a t i s t i c a l methods were performed on fresh salmon of d i f f e r e n t q u a l i t y grades. Use of the appropriate error estimation method coupled with the t e s t i n g of s t a t i s t i c a l assumptions were required to obtain r e l i a b l e and stable functions. The remedial use of non-parametric discriminant functions was found necessary in a l l treatments investigated and provided equivalent or superior r e s u l t s . Judicious interpretation of the PC's scores was also b e n e f i c i a l at extracting meaningful functions. The instrumental method was advantageous for helping sensory evaluation at c l a s s i f y i n g grade B canned pink salmon samples since the e a r l y stages of decomposition were very d i f f i c u l t to determine by flavour evaluation alone. The r e s u l t s of the r e f r i g e r a t e d storage highlighted the important role of ethanol and 3-methyl-l-butanol among other possible compounds i n the development of decomposition indices for fresh and canned salmon. These deserve more attention in future research. On a p r a c t i c a l basis, the f i s h processing industry may use the SHGC method for q u a l i t y assurance purposes. In combination with sensory evaluation, i t can serve as a screening t o o l by regulatory agencies to assess and confirm the q u a l i t y of sampled l o t s of canned P a c i f i c salmon, e s p e c i a l l y those of marginal grades. The method can also a s s i s t i n the standardization of the sensory grading process by r e l a t i n g grades to v o l a t i l e concentrations and consequently helping the t r a i n i n g of inspectors. In addition, the growing aquaculture industry could possibly u t i l i z e t h i s technique to assess the impact of various combinations of diets on the flavour of farmed raised P a c i f i c salmon. 178 Use of dynamic headspace analysis showed c e r t a i n groups of compounds to be important contributors to canned salmon flavours. Hydrocarbons, carbonyls, s u l f i d e s , and heterocyclic compounds were detected and associated with various flavour compounds. No single v o l a t i l e constituents had a t y p i c a l canned salmon aroma, in d i c a t i n g that a combination of compounds may be involved i n t h i s c h a r a c t e r i s t i c sensory perception. In addition, a few odour active compounds were noted using the odour evaluation technique but were not detected by the GC-MS method used. Refrigerated spoilage of salmon p r i o r to canning caused the occurrence of a d d i t i o n a l compounds, p a r t i c u l a r l y hydrocarbons, alcohols, ketones, furans, and pyrazines. A number of these compounds were correlated with aroma a t t r i b u t e s . Furthermore, several alkylpyrazines and alkylthiophenes were present i n the canned chum salmon samples. Large peak areas of several ketones and aldehydes were analyzed i n canned l a t e run chum salmon and were important contributors to the o v e r a l l odour. 2-Methyl-butanal and two other unknown low-boiling point v o l a t i l e s had hay, cooked malt, and straw-like c h a r a c t e r i s t i c s t y p i c a l of canned la t e run chum salmon. Based on the aroma c h a r a c t e r i s t i c s and the retention times, i t i s be postulated that these two unknown compounds may also be aldehydes of molecular weight equal to or lower than 2-methyl-butanal. 179 REFERENCES Ackman, R.G. and McLeod, C. 1988. To t a l l i p i d s and n u t r i t i o n a l l y important fat t y acids of some Nova Scotia f i s h and s h e l l f i s h food products. Can. Inst. Food S c i . Technol. J . 21:390. Ahamed, A. and Matches, J.R. 1983. Alcohol production by f i s h spoilage bacteria. J. Food Prot. 46:1055. Aishima, T. 1979a. C l a s s i f i c a t i o n of soy sauce on p r i n c i p a l components i n GC p r o f i l e s . Agric. B i o l . Chem. 43:1905. Aishima, T. 1979b. Objective evaluation of soy sauce by s t a t i s t i c a l analysis of GC p r o f i l e s . A g r i c . B i o l . Chem. 43:1935. Aishima, T. 1983. Relationships between gas chromatographic p r o f i l e s of soy sauce v o l a t i l e s and organoleptic c h a r a c t e r i s t i c s based on multivariate analysis. In "Instrumental analysis of foods," ed. G. Charalambous, Vol. 1, p. 37. Academic Press, London. Aishima, T. and Nakai, S. 1987. Pattern recognition of GC p r o f i l e s f o r c l a s s i f i c a t i o n of cheese v a r i e t y . J . Food S c i . 52:939. Aishima, T., Yada, R.Y., Mohler-Smith, A., Nakai, S. 1986. M u l t i v a r i a t e analysis of structure-related data to explain milk c l o t t i n g a c t i v i t y of proteolytic enzymes. J . Ag r i c . Food Chem. 34:675. Aksnes, A., Gjerde, B., and Roald, S.O. 1986. B i o l o g i c a l , chemical,j and organoleptic changes during maturation of farmed A t l a n t i c salmon, Salmo salar. Aquaculture 53:7. Alford, J.A., Smith, J.L., and L i l l y , H.D. 1971. Relationship of m i c r o b i a l a c t i v i t y to changes i n l i p i d s of food. J. Appl. B a c t e r i o l . 34:133. Amantea, G.F., Skura, B.J., and Nakai, S. 1986. Culture e f f e c t on ripening c h a r a c t e r i s t i c s and r h e o l o g i c a l behavior of Cheddar cheese. J.- Food Sc i . 51:912. Ando, S., Hatano, M., and Zama, K. 1985a. Deterioration of chum salmon (Oncorhynchus keta) muscle during spawning migration - I. Changes in proximate composition of chum salmon muscle during spawning migration. Comp. Biochem. P h y s i o l . 80B:303. Ando, S., Hatano, M., and Zama, K. 1985b. A consumption of muscle l i p i d during spawning migration of chum salmon Oncorhynchus keta. B u l l . Jpn. Soc. S c i . Fish. 51:1817. Apgar, M.E. and Hultin, H.O. 1982. L i p i d peroxidation i n f i s h muscle microsomes in the frozen s t a t e . C r y o b i o l . 19:154. 180 Aquaculture and Commercial F i s h e r i e s Branch, 1989. F i s h e r i e s production s t a t i s t i c s of B r i t i s h Columbia. Commercial F i s h e r i e s Section, Ministry of Agri c u l t u r e and Fi s h e r i e s , BC. Aries, R.E., Gutteridge, C.S., and Evans, R. 1986. Rapid characterization of orange juice by p y r o l y s i s mass spectrometry. J. Food S c i . 51:1183. Arctander, S. 1969. "Perfume and Flavor Chemicals (Aroma chemicals)," Vol. 1-2. Steffen Arctander, Publisher. NJ. Autran, J.C., Abecassis, J., and F e i l l e t , P. 1986. S t a t i s t i c a l evaluation of d i f f e r e n t technological and biochemical t e s t s for q u a l i t y assessment i n durum wheats. Cereal Chem. 63:390. Barber, K., Weintraub, S.E., and Vanderstoep, J. 1987. N u t r i t i o n a l analysis of B r i t i s h Columbia canned salmon. Unpublished report. Fisheries Council of B r i t i s h Columbia, Vancouver, BC. Barber, K., Weintraub, S.E., and Vanderstoep, J . 1988. N u t r i t i o n a l analysis of B r i t i s h Columbia fresh/frozen and cooked salmon. Unpublished report. F i s h e r i e s Council of B r i t i s h Columbia, Vancouver, BC. Borenstein, B. and Bunnell, R.H. 1966. Carotenoids: properties, occurence and u t i l i z a t i o n i n foods. Food Res. 15:195. Botta, J.R., Lauder, J.T., and Jewer, M.A. 1984. E f f e c t of methodology on t o t a l v o l a t i l e basic nitrgen (TVB-N) determination as an index of qual i t y of fresh A t l a n t i c cod. J. Food S c i . 49:734. Braddock, R.J. and Dugan L.R.Jr. 1972. Phospholipid changes i n muscle from frozen stored Lake Michigan Coho salmon. J . Food S c i . 37:426. Brock, T.D. (Ed.) 1979. "Biology of Microorganisms," 3rd ed. Prentice-Hall Inc., Englewood C l i f f s , NJ. Brown, D.G.W. and Clapperton, J.F. 1978. Discriminant analysis of sensory and instrumental data on beer. J. Inst. Brewing 84:318. Buttery, R.G., Ling, L.C., Teranishi, R., and Mon, T.R. 1977. Roasted lamb fa t : basic v o l a t i l e compoents. J. Agric. Food Chem. 25:1227. Buttery, R.G., Ling, L.C., and Chang, B.G. 1978. V o l a t i l e s of corn kernels and husks: possible corn ear worm attra c t a n t s . J. Agric. Food Chem. 26:866. Buttery, R.G., Bomben, J.L., Guadagni, D.G., and Ling, L.C. 1971. Some considerations of the v o l a t i l i t i e s of organic f l a v o r compounds in foods. J. Agric. Food Chem. 19:1045. I 181 Chang, S.S., and Peterson, R.J. 1977. Symposium: the basis of q u a l i t y i n muscle foods. Recent developments i n the f l a v o r of meat. J. Food S c i . 42:298. Chen, H.M., Meyers, S.P., Hardy, R.W., and Biede, S.L. 1984. Color s t a b i l i t y of astaxanthin pigmented rainbow trout under various packaging conditions. J. Food S c i . 49:1337. Childerhose, R.J. and Trim, M. 1981. P a c i f i c Salmon. Douglas and Mclntyre Ltd., Vancouver, BC. Christensen, B.W., Kjaer, A., and Madsen, J.O. 1981. V o l a t i l e s u l f u r compounds and other headspace constituents of north sea f i s h o i l s . J . Am. O i l . Chem. Soc. 58:1053. Chung, J.R. 1968. Postmortem degradation of f i s h muscle proteins: the role of p r o t e o l y t i c Pseudomonas spp. and t h e i r mechanism of action. Ph.D. Dissertation, U n i v e r s i t y of Washington, Seattle, WA. Clapperton, J.F. and Piggott, J.R. 1979. D i f f e r e n t i a t i o n of ale and lager flavors by p r i n c i p a l components analysis of f l a v o r characterization data. J. Inst. Brewing 85:271. C o l l i n s , L.S. 1989. Quali t y enhancement of canned late-run chum salmon (Oncorhynchus keta). M.Sc. Thesis, Univ. of B r i t i s h Columbia, Vancouver, BC. Connell, J.J. (Ed.) 1980. Control of Fish Quality. 2nd ed. Fishing News Ltd., Surrey, Endland. I I Connell, D.W. and M i l l e r , G.J. 1981. Petroleum hydrocarbons i n aquatic ecosystems - Behavior and ef f e c t s of sublethal concentrations: Part 2. XI Tainting. CRC C r i t . Rev. Environ. Control, 11:104. Cowey, C.B., Daisley, K.W., and Parry, G. 1962. Amino acids free or as components of protein, and some B vitamins i n the tissues of the A t l a n t i c salmon during spawning and migration. Comp. Biochem. Physiol. 7:29-38. Crain, W.O. and Tang, C S . 1975. V o l a t i l e components of roasted macadamia nuts. J. Food S c i . 40:207. Crosgrove, D.M. 1978. A rapid method for estimating ethanol i n canned salmon. J. Food S c i . 43:641. Davies, W.L. and G i l l , E. 1936. Investigations on f i s h y f l a v o r s . Chem. Ind. (London) 55:1415. Davis, P.L. and Chace, W.G.Jr. 1969. Determination of alcohol i n c i t r u s juice by gas chromatographic analysis of headspace. HortSci. 4:117. 182 Department of Fis h e r i e s and Oceans, 1988. Annual report 1987-1988. Communications Directorate, Ottawa, Ontario. Department of Fisheries and Oceans, 1989. Quality Grade Standard f o r Frozen Gutted P a c i f i c Salmon. Inspection D i v i s i o n , F i e l d Services Branch, Department of Fi s h e r i e s and Oceans, Vancouver, BC. D i l l o n , W.R. and Goldstein, M. (Ed.) 1984. Mu l t i v a r i a t e Analysis. John Wiley and Sons, New York, NY. Dravnieks, A., R e i l i c h , H.G., Whi t f i e l d , J., and Watson, C A . 1973. C l a s s i f i c a t i o n of corn odor by s t a t i s t i c a l analysis of gas chromatographic patterns of headspace v o l a t i l e s . J. Food S c i . 38:34. Dravnieks, A. and Watson, CA. 1973. Corn odor c l a s s i f i c a t i o n from low-resolution gas chromatographic p r o f i l e s of head space v o l a t i l e s . J. Food S c i . 38:1024. Dunteman, G.H. (Ed.). 1984. "Introduction to Mu l t i v a r i a t e Analysis," Sage Publications Ltd, London, England. Dwivedi, B.K. 1975. Meat f l a v o r . CRC C r i t . Rev. Food Technol. 5:487. Easley, D.M., Kleopfer, R.D., and Carasea, A.M. 1981. Gas chromatographic-mass spectrometric determination of v o l a t i l e organic compounds i n f i s h . J. Assoc. Off. Anal. Chem. 64:653. Ehira, S., and Uchiyama, H. 1987. Determination of f i s h freshness using the K-value and comments on some other biochemical changes i n re l a t i o n to freshness. In ^'Seafood Qua l i t y Determination," D.E. Kramer and J. Liston (Ed.), p.185. E l s e v i e r Science Publishing Co., New York, NY. Eiten m i l l e r , R.R., Orr, J.H., and Wallis, W.W. 1982. Histamine formation i n f i s h : microbiological and biochemical conditions. In "Chemistry and Biochemistry of Marine Food Products." R.E. Martin, G.G. F l i c k , C E . Hebard, and D.R. Ward (Ed.), p. 39. AVI Publishing Co., Westport, CN. Entz, R.C. and H o l l i f i e l d , H.C 1982. Headspace gas chromatographic analysis of foods for v o l a t i l e halocarbons. J. Agric. Food Chem. 30:84. Fors, S. 1983. Sensory properties of v o l a t i l e M a i l l a r d reaction products and related compounds. In "The M a i l l a r d Reaction i n Foods and Nu t r i t i o n , " G.R. Waller and M.S. Feather (Ed.), p. 185. ACS Symp. Series No. 215, American Chemical Society, Washington, DC. Forss, D.A. 1972. Odor and f l a v o r compounds from l i p i d s . In "Progress i n the Chemistry of Fats and other L i p i d s , " R.T. Holman (Ed.), Pergamon Press, New York, NY. 183 Frazier, W.C., and Westhoff, D.C. (Eds.). 1978. Food microbiology. 3rd ed. McGraw-Hill, New York, NY. Gait, A.M. and MacLeod, G. 1983. The a p p l i c a t i o n of factor a n a l y s i s to cooked beef aroma d e s c r i p t o r s . J . Food S c i . 48:1354. Glick, N. 1978. Additive estimators f o r p r o b a b i l i t i e s of correct c l a s s i f i c a t i o n . Pattern Recogn., 10:211. Greene, C.W. 1926. The physiology of spawning migration. P h y s i o l . Rev. 6:201. Grosch, W. 1984. Neuere Vorstellungen uber die l i p o x i d a t i o n . Lebensm. Ge r i c h t l . Chem. 38:81. Guadagni, D.G., Buttery, R.G., and Okano, S. 1963. Odour threshold of some organic compounds associated with food flavours. J. S c i . Food Agric. 14:761. Gunnar, F. 1982. Enzymatic ammonia production i n penaied shrimp held on ice. In "Chemistry and Biochemistry of Marine Food Products," R.E. Martin, G.G. F l i c k , C E . Hebard, and D.R. Ward (Ed.), p. 323. AVI Publishing Co., Westport, CN. Habu, T., Flath, R.A., Mon, T.R., and Morton, J.F. 1985. V o l a t i l e components of rooibos tea (Aspalathus l i n e a r i a ) . J. A g r i c . Food Chem. 33:249. Hand, D.J. 1982. Kernel Discriminant Analysis. Research Studies Press, New York, NY. j Hardy, R., McGill, A.S., and Gunstone, F.D. 1979. L i p i d and autoxidative changes in cold stored cod (Gadus morhua). J. S c i . Food Agric. 30:999. Hatae, K., Yoshimatsu, F., and Matsumoto, J.J. 1984. Discriminative characterization of d i f f e r e n t texture p r o f i l e s of various cooked f i s h muscles. J . Food S c i . 49:721. Headley, L.M. and Hardy, J.K. 1989. C l a s s i f i c a t i o n of whiskies by p r i n c i p a l component a n a l y s i s . J . Food S c i . 54:1351. Hearn, T.L., Sgoutas, S.A, Hearn, J.A., and Sgoutas, D.S. 1987. Polyunsaturated f a t t y acids and fat i n f i s h f l e s h for selected species for health b e n e f i t s . J . Food S c i . 52:1209. Heath H.B. and Reineccius, G. (Ed.). 1986. Flavor and i t s study. Ch. 1. In "Flavor Chemistry and Technology," p. 3. AVI Publishing Co., Westport, CN. 184 Hebard, C.E., F l i c k , G.J., and Martin, R.E. 1982. Occurence and s i g n i f i c a n c e of trimethylamine oxide and i t s derivatives in f i s h and s h e l l f i s h . In "Chemistry and Biochemistry of Marine Food Products," R.E. Martin, G.G. F l i c k , C E . Hebard, and D.R. Ward (Ed.), p.149. AVI Publishing Co., Westport, CN. Helbert, J.R. and Hoff, J.T. 1974. Discriminant analysis - a powerful t o o l for brewery research. Proc. Amer. Soc. Brew. Chem. 32:43. Herbert, R.A. and Shewan, J.M. 1975. Precursors of the v o l a t i l e sulphides i n s p o i l i n g North Sea cod (Gadus morhua). J. S c i . Food Agric. 26:1195. Herbert, R.A. and Shewan, J.M. 1976. Roles played by b a c t e r i a l and a u t o l y t i c enzymes i n the production of v o l a t i l e sulphides in s p o i l i n g North Sea cod (Gadus morhua). J. S c i . Food Agric. 27:89. Herranz, A., de l a Serna, P., and Barro, C. 1990. M u l t i v a r i a t e s t a t i s t i c a l methods applied to the d i f f e r e n t i a t i o n of rum brands. J . S c i . Food Agric. 51:555. Heyman, H. and Noble, A.C 1989. Comparison of canonical variate and p r i n c i p a l component analyses of wine d e s c r i p t i v e analysis data. J. Food S c i . 54:1355. H i l l i g , F. 1958. Determination of alcohol i n f i s h and egg products. J. Assoc. Off. Anal. Chem. 41:776. H i l l s , M. 1966. A l l o c a t i o n rules and t h e i r error rates. J. Royal Stat. Soc., B28:l. Ho, C , Lee, K., and J i n , Q.Z. 1983. I s o l a t i o n and i d e n t i f i c a t i o n of v o l a t i l e f l a v o r compounds i n f r i e d bacon. J . Agric. Food Chem. 31:336. Hoff, J.T., Helbert, J.R., and Chicoye, E. 1975. C l a s s i f i c a t i o n of lager beers by computer analysis of v o l a t i l e p r o f i l e s . Tech. Q., Master Brew. Assoc. Amer. 12:209. Hoff, J.T. and Herwig, W.C 1976. C o r r e l a t i o n of t r i a n g u l a r taste t e s t panel r e s u l t s with s t a t i s t i c a l differences between headspace v o l a t i l e p r o f i l e s . J. Amer. Soc. Brew. Chem. 34:1. Holaday, D.A. 1939. The alcohols as a measure of spoilage in canned f i s h . J . Assoc. Off. Anal. Chem. 22:418. Hollingworth, T.A.Jr. and Throm, H.R. 1983. A headspace gas chromatographic method for the rapid analysis of ethanol in canned salmon. J. Food S c i . 48:290. 185 Hollingworth, T.A.Jr., Throm, H.R., Wekell, M., Trager, W.F., and O'Donnell, M.W.Jr., 1986. Headspace gas chromatographic method f o r determination of ethanol i n canned salmon: collaborative study. J. Assoc. Off. Anal. Chem. 69:524. Hosfield, G.L., Ghaderp, A., and Uebersax, M.A. 1984. A factor analysis of y i e l d and sensory and physico-chemical data from tests used to measure cu l i n a r y q u a l i t y i n dry edible beans. Can. J. Plant S c i . 64:285. Hughes, R.B. 1964. Chemical studies on the herring (Clupea harenqus) . IX.-Preliminary gas-chromatographic study of v o l a t i l e sulphur compounds produced during the cooking of herring. J . S c i . Food Agric. 15:290. Human, J . and Khayat, A. 1981. Quality evaluation of raw tuna by gas chromatography and sensory evaluation. J . Food S c i . 46:868. Iida, H., Nakamura, K., Tokunaga, T., Sato, K., Hoshino, C , Terano, S., and Furukawa, T. 1982. Study on the several chemical tests for estimating the q u a l i t y of canned product. I I . On the canned sardine and masu-salmon. B u l l . Tokai Reg. F i s h . Res. Lab. 104:11. Iida, H., Tokunta, T., and Nakamura, K. 1981a. Usefulness of ethanol as a qu a l i t y index of f i s h and f i s h products - I. B u l l . Tokai Reg. Fi s h . Res. Lab. 104:77. Iida, H., Tokunta, T., Nakamura, K., and Oota, Y. 1981b. Usefulness of ethanol as a q u a l i t y index of f i s h and f i s h products - II. B u l l . Tokai Reg. F i s h . Res. Lab. 104:83. Jahns, F.D., Howe, J.L., Coduri, R.J., and Rand, A.G. 1976. A rapid enzyme te s t to assess f i s h freshness. Food Technol. 30(7):27. Jones, N.R. 1969. F i s h as raw material for freezing: factors influencing the q u a l i t y of products frozen at sea. In "Freezing and I r r a d i a t i o n of F i s h , " R. Kreuzer (Ed.), p. 31. FAO, Fishi n g News Ltd, London, UK. Josephson, D.B. 1987. Mechanisms for the formation of v o l a t i l e s i n fresh seafood s. Ph.D. Dissertation, University of Wisconsin-Madison, Madison, WI. Josephson, D.B., Lindsay, R.C, and Struiber, D.A. 1983. I d e n t i f i c a t i o n of compounds char a c t e r i z i n g the aroma of fresh whitefish (Coreqonus  clupeaformis). J. Agric. Food Chem. 31:326. Josephson, D.B., Lindsay, R.C, and Stuiber, D.A. 1984a. Variations i n the occurences of enzymatically-derived v o l a t i l e aroma compounds i n s a l t - and freshwater f i s h . J. Agric. Food Chem. 32:1344. 186 Josephson, D.B., Lindsay, R.C, and Struiber, D.A. 1984b. I d e n t i f i c a t i o n of v o l a t i l e aroma compounds from oxidized frozen whitefish (Coreqonus clupeaformis). Can. Inst. Food S c i . Technol. J. 17:178. Josephson, D.B., Lindsay, R.C, and Struiber, D.A. 1985. V o l a t i l e compounds characterizing the aroma of fresh A t l a n t i c and P a c i f i c oysters. J. Food S c i . 50:5. Josephson, D.B., Lindsay, R.C, and Struiber, D.A. 1987 . Influence of processing on the v o l a t i l e compounds characterizing the f l a v o r of pickled f i s h . J. Food S c i . 52:10. Kadota, H. and Ishida, Y. 1972. Production of v o l a t i l e s u l f u r compounds by microorganisms. Ann. Rev. M i c r o b i o l . 26:127. Kaiser, H.F. 1960a. The a p p l i c a t i o n of e l e c t r o n i c computers to factor analysis. Educ. Psychol. Measurement 20:141. Kaiser, R.E. 1960b. "Gas Chromatographic," Geest and Portig, Leipzig. Kamiya, A. and Ose, Y. 1984. Study on the odorous compounds produced by putrefaction of foods, V: Fatty acids, s u l f u r compounds, and amines. J. Chromatogr. 292:383. Kaneko, T., Takeuchi, M., I s h i i , S., Higashi, H. and Kikuchi, T. 1966. Effect of dietary l i p i d s on f i s h under c u l t i v a t i o n . IV. Changes i n f a t t y acid composition i n f l e s h l i p i d s i n rainbow trout on non-feeding. B u l l . Jpn. Soc. S c i . F i s h . 33:56. Kato, S., Kurata, T., Ishitsuka, R., and Fujimaki, M. 1970. P y r o l y s i s of p-hydroxy amino acids, e s p e c i a l l y L-serine. Agric. B i o l . Chem. 36:951. Ke, P.J., Linke, B.A., and Ackman, R.G. 1978. Acceleration of l i p i d oxidation in frozen mackerel f i l l e t by pretreatment with microwave heating. J. Food S c i . 43:38. Kelleher, S. and Z a l l , R.R. 1983. Ethanol accumulations i n muscle tis s u e as a chemical i n d i c a t o r of f i s h spoilage. J. Food Biochem. 7:78. Khayat, A. 1979. Correlation of off-odor scores of canned tuna with gas chromatographic data. J . Food S c i . 44:37. Kitahara, T. 1983. Behavior of carotenoids i n the chum salmon (Oncorhynchus keta) during anadromous migration. Comp. Biochem. Physiol. 76B:97. Krzynowek, J. and Murphy, J. 1987. Proximate composition, energy, f a t t y acid, sodium, and c h o l e s t e r o l content of f i n f i s h , s h e l l f i s h , and t h e i r products. U.S. Department of Commerce, NOAA, Seattle. 187 Kubota, K., Kobayashi, A., and Yamanishi, T. 1982. Basic and neutral compounds in the cooked odor from a n t a r c t i c k r i l l . Agric. B i o l . Chem. 46:2835. Kubota, K., Shijimaya, H., and Kobayashi, A. 1986. V o l a t i l e components of roasted shrimp. Agric. B i o l . Chem. 50:2867. Kwak, H.S., Jeon, I.J., and Perng, S.K. 1989. S t a t i s t i c a l patterns of lipase a c t i v i t i e s on the release of short-chain f a t t y acids i n Cheddar cheese s l u r r i e s . J. Food S c i . 54:1559. Lachenbruch, P.A. 1967. An almost unbiased method of obtaining confidence intervals for the p r o b a b i l i t y of m i s c l a s s i f i c a t i o n i n discriminant analysis. Biometrics 23:639. Lachenbruch, P.A., Sneeringer, C , and Revo, L.T. 1973. Robustness of the linear and quadratic discriminant function to c e r t a i n types of non-normality. Commun. Stat. 1:39. Larick, D.K. and Turner, B.E. 1990. Headspace v o l a t i l e s and sensory c h a r a c t e r i s t i c s of ground beef from forage- and grain-fed h e i f e r s . J. Food S c i . 54:649. Leblanc, R.J., and G i l l , T.A. 1984. Ammonia as an objective q u a l i t y index in squid. Can. Inst. Food S c i . Technol. J . 17:734. Leblanc, E.L., Leblanc, R.J., and Blum, I.E. 1988. Predi c t i o n of q u a l i t y in frozen cod (Gadus morhua) f i l l e t s . J. Food S c i . 53:328. Lerke, P.A. and Huck, R.W. 1977. Objective determination of canned tuna quality: i d e n t i f i c a t i o n of ethanol as a p o t e n t i a l l y useful index. J. Food S c i . 42:755. Leland, J.V., Reineccius, G.A., and Lahif f , M. 1987. Eavaluation of copper-induced oxidized f l a v o r i n milk by discriminant analysis of c a p i l l a r y gas chromatographic p r o f i l e s . J. Dairy S c i . 70:524. Li-Chan, E., Nakai, S., and Wood, D.F. 1987. Muscle.protein structure-function r e l a t i o n s h i p s and discrimination of f u n c t i o n a l i t y by multivariate a n a l y s i s . J. Food S c i . 52:31. Lindsay, R.C. 1977. Objective measurement of the flavour q u a l i t y of beer. Brewers' Dig. 52:44. Lindsay, R.C, Josephson, D.B., and O l a f s d o t t i r , G. 1986. Chemical and biochemical indices for assessing the qu a l i t y of f i s h packaged i n controlled atmospheres. In "Seafood Quality Determination," D.E. Kramer and J. Liston (Eds.), p.221. E l s e v i e r Science Publishing Co., New York, NY. 188 Liston, J. 1982. Recent advances i n the chemistry of iced f i s h spoilage. In "Chemistry and Biochemistry of Marine Food Products," R.E. Martin, G.G. F l i c k , C E . Hebard, and D.R. Ward (Ed.), p. 27. AVI Publishing Co., Westport, CN. Love, R.M. 1970. Maturity and Spawning. In "The Chemical Biology of Fishes I." Academic Press, London, UK. Love, R.M. 1980. Maturation. In "The Chemical Biology of Fishes I I . " Academic Press, London, UK. Lovern, J.A. 1934. Fat metabolism i n f i s h e s . M o b i l i z a t i o n of depot f a t i n the salmon. Biochem. J. 28:1955. Lyon, B.G. 1988. Descriptive p r o f i l e analysis of cooked, stored, and reheated chicken p a t t i e s . J. Food S c i . 53:1086. MacLeod, G. and Ames, J.M. 1986. C a p i l l a r y gas chromatography-mass spectometric analysis of cooked ground beef aroma. J. Food S c i . 51:1427. MacLeod, A.J. and Panchasara, S.D. 1983. V o l a t i l e aroma components, p a r t i c u l a r l y glucosinolate products, of cooked edible mushrooms (Aqaricus bisporus) and cooked d r i e d mushrooms. Phytochem. 22:705. Maga, J.A. 1982. Pyrazines in flavour. In "Food Flavours," Part A, I.D. Morton and A.J. MacLeod (Eds.), p. 283. E l s e v i e r Science Publishing Co., New York, NY. Marais, J., van Rooyen, P.C, and du P l e s s i , C S . 1 9 8 2 C l a s s i f i c a t i o n of white c u l t i v a r wines by o r i g i n using v o l a t i l e aroma components. 1973-1981. Agric. Res., p.46. Marks, S. and Dunn, O.J. 1974. Discriminant functions when covariance matrices are unequal. J. Amer. Stat. Assoc. 69:555. Maruyama, F.T. 1970. I d e n t i f i c a t i o n of dimethyl t r i s u l f i d e as a major aroma component of cooked Brassicaceous vegetables. J. Food S c i . 35:540. Matiella, J.E. and Hsieh, T.C-Y. 1990. Analysis of crabmeat v o l a t i l e compounds. J. Food S c i . 55:962. McGill, A.S., Hardy, R., Burt, J.R., and Gunstone F.D. 1974. Hept-cis-4-enal and i t s contribution to the o f f - f l a v o r i n col d stored cod. J . S c i . Food Agric. 25:1477. McGill, A.S., Hardy, R., and Gunstone, F.D. 1977. Further analysis of the v o l a t i l e components of frozen c o l d stored cod and the influence of these on flavour. J. S c i . Food Agric. 28:200. 189 Mietz, J.L., and Karmas, E. 1978. Polyamine and histamine content of rockfish, salmon, lobster, and shrimp as an indica t o r of decomposition. J. Assoc. Off. Anal. Chem. 61:139. M i l l e r III, A., Scanlan, R.A., Lee, J.S., and Libbey, L.M. 1972a. Quantitative and s e l e c t i v e gas chromatographic analysis of dimethyl-and trimethylamine in f i s h . J. Agric. Food Chem. 20:709. M i l l e r III, A., Scanlan, R.A., Lee, J.S., and Libbey, L.M. 1972b. V o l a t i l e compounds produced in ground muscle tissue of canary r o c k f i s h (Sebastes pinniqer) stored on i c e . J . Fish. Res. Bd. Can. 29:1125. M i l l e r III, A., Scanlan, R.A., Lee, J.S., and Libbey, L.M. 1973a. Id e n t i f i c a t i o n of v o l a t i l e compounds produced in s t e r i l e f i s h muscle (Sebastes melanops) by Pseudomonas f r a q i . Appl. M i c r o b i o l . 25:952. M i l l e r III, A., Scanlan, R.A., Lee, J.S., and Libbey, L.M. 1973b. V o l a t i l e compounds produced in s t e r i l e f i s h muscle (Sebastes melanops) by Pseudomonas perolens, Pseudomonas fluorescens and an Achromobacter species. Appl. Microbiol. 26:18. M i l l e r III, A., Scanlan, R.A., Lee, J.S., Libbey, L.M., and Morgan, M.E. 1973c. V o l a t i l e compounds produced i n s t e r i l e f i s h muscle (Sebastes  melanops) by Pseudomonas perolens. Appl. M i c r o b i o l . 25:257. Min, D.B.S., Ina, K., Peterson, R.J., and Chang, S.S. 1977. The alkylbenzenes i n roast beef. J. Food Sci.42:503. -Min, D.B., Ina, K., Peterson, R.J., and Chang, S.S. 1979. Preliminary i d e n t i f i c a t i o n of v o l a t i l e f l a v o r compounds i n the neutral f r a c t i o n of roast beef. J. Food S c i . 44:639. Minor, L.J., Pearson, A.M., Dawson, L.E., and Shweigert, B.S. 1965. Chicken flavor: the i d e n t i f i c a t i o n of some chemical components and the importance of s u l f u r compounds i n the cooked v o l a t i l e f r a c t i o n . J. Food S c i . 30:686. Mohler-Smith, A. and Nakai, S. 1990. C l a s s i f i c a t i o n of cheese v a r i e t i e s by multivariate analysis of HPLC p r o f i l e s . Can. Inst. Food S c i . Technol. J. 23:53. Montedero G. and B e r t u c c i o l i , M. 1983. Application of GC and HPLC i n the predictive quality evaluation of wines. In "Instrumental Analysis Of Foods," ed. G. Chalarambous, Vol. 2, p. 409. Academic Press, London, UK. Moret, I., Capodaglio, G., and Scarponi, G. 1986. S t a t i s t i c a l evaluation of the group structures of f i v e Venetian wines from chemical measurements. Anal. Chim. Acta 191:331. Morrison, D.F. 1976. Multivariate S t a t i s t i c a l Methods. 2nd ed. McGraw-H i l l Book Co., New York, NY. 190 Mottram, D., Edwards, R., and MacFie, J. 1982. A comparison of the f l a v o r v o l a t i l e s from cooked beef and pork meat systems. J. S c i . Food Agric. 33:934. Mottram, D.S. and Whitfield, F.B. 1987. The e f f e c t of l i p i d on the formation of v o l a t i l e h e t e r o c y c l i c compounds i n the M a i l l a r d reaction. In "Flavour Science and Technology," M. Martens, G.A. Dalen, and H. Russwurm j r . (Ed.), p.29. John Wiley and Sons Ltd., New York, NY. Mulders, E.J. 1973. The odour of white bread. IV. Quantitative determination of constituents i n the vapour and t h e i r odour values. Z. Lebensm.-Unters. Forsch. 151:310. Mussinan, C.J. and Walradt, J.P. 1974. Constituents of pressure cooked pork l i v e r . J. Agric. Food Chem. 22:827. Mussinan, C.J., Wilson, R.A., and Katz, I. 1973. I s o l a t i o n and i d e n t i f i c a t i o n of pyrazines present i n pressure cooked beef. J . Agric. Food Chem. 21:871. Myers, M.J., Issenberg, P., and Wick, E.L. 1969. Vapor analysis of the production by banana f r u i t of cer t a i n v o l a t i l e constituents. J . Food S c i . 34:504. Myers, M.J., Issenberg, P., and Wick, E.L. 1970. L-Leucine as a precursor of isoamyl alcohol and isoamylacetate v o l a t i l e aroma constituents of banana f r u i t d i s c s . Phytochem. 9:1693. I Nakai, S. 1989. A new random-centroid optimization for food research and processing. Paper No. 302, presented at 50th Annual Meeting of Inst, of Food Technologists, Chicago, IL, June 25-29. Nakai, S., Koide, K., and Eugster, K. 1984. A new mapping super-simplex optimization for food product and process development. J . Food S c i . 49:1143. Nawar, W.W. 1985. L i p i d s . Ch. 4. In "Food Chemistry," 0. Fennema (Ed.), p. 139. Marcel Dekker Inc., New York, NY. Neff, J.M., Cox, B.A., D i x i t , D., and Anderson, J.W. 1976. Accumulation and release of petroleum-derived aromatic hydrocarbons by four species of marine animals. Marine B i o l . 38:279. Noble, A.C., Flath, R.A., and Ralph, R.R. 1980. Wine headspace analysis, rep r o d u c i b i l i t y and a p p l i c a t i o n to v a r i e t a l c l a s s i f i c a t i o n . J. Agric. Food Chem. 28:346. Nonaka, M., Black, D.R., and Pippen, E.L. 1967. Gas chromatographic and mass spectral analysis of cooked chicken meat v o l a t i l e s . J. Agric. Food Chem. 15:713. 191 Nunez, A.J. and Gonzalez L.F. 1984. Pre-concentration of headspace v o l a t i l e s for trace organic analysis by gas chromatography. J . Chromtogr. 300:127. Ohloff, G., Flament, I., and Pickenhagen, W. 1985. Flavor Chemistry. Food Rev. Internat. 1:99. Parliment, T.H. 1986. Sample preparation techniques for g a s - l i q u i d chromatographic analysis of b i o l o g i c a l l y derived aromas. Ch. 4. In "Biogeneration of Aromas," T.H. Parliment and R. Croteau (Ed.), p. 34. American Chemical Society, Symposium Series 317, Washington, DC. Pesek, C.A., Wilson, L.A., and Hammond E.G. 1985. Spice q u a l i t y : e f f e c t of cryogenic and ambient grinding on v o l a t i l e s . J. Food S c i . 50:599. Pham, A.M. and Nakai, S. 1984. Application of stepwise discriminant analysis to high pressure l i q u i d chromatography p r o f i l e s of water extract for judging ripening of Cheddar cheese. J. Dairy S c i . 67:1390. Pino, J.A., T o r r i c e l l a , R.G., Orsi, F., and Figueras, L. 1986. Application of multivariate s t a t i s t i c s for the q u a l i t y c l a s s i f i c a t i o n of s i n g l e -strength g r a p e f r u i t j u i c e . J. Food Qual. 9:205. Pippen, E.L., Mecchi, E.P., and Nonaka, M. 1969. O r i g i n and nature of aroma i n fat of cooked poultry. J. Food S c i . 34:436. Pokorny, J. 1980. E f f e c t of browning on the formation of f l a v o r substances. Nahrung 24:115. j Powers, J.J., Godwin, D.R., and Bargmann, R.E. 1977. Relations between sensory and objective measurements for q u a l i t y evaluation of green beans. In "Flavor Quality: Objective Measurement," R.A. Scanlan (Ed.) Symp. Series 51. Am. Chem. S o c , Washington, DC. Powers, J.J. and Keith, E.S. 1968. Stepwise discriminant analysis of gas chromatographic data as an ai d i n c l a s s i f y i n g the f l a v o r q u a l i t y of foods. J. Food S c i . 33:207. Powers, J.J., Smit, C.J.B., and Godwin, D.R. 1978. Relations among sensory and objective a t t r i b u t e s of canned rabbiteye (Vaccinium ashei Reade) blueberries. I I . Cluster and discriminant analysis examination. Lebensmit.-Wissens. Technol. 11:279. Powrie, W.D. 1984. Chemical e f f e c t s during storage of frozen foods. J . Chem. Education. 61:340. Quarmby, A.R. and Ratkowshy, D.A. 1988. Free-choice flavour and odour p r o f i l i n g of f i s h spoilage: does i t achieve i t s objective? J. S c i . Food Agric. 44:89. 192 Rao, C.R. 1973. Linear S t a t i s t i c a l Inference and i t s Application. 2nd ed. John Wiley and Sons Inc., New York, NY. Reinert, K.H., Hunter, J.V., and Sabatino, T. 1983. Dynamic heated headspace analyses of v o l a t i l e organic compounds present i n f i s h t issue samples. J. Agric. Food Chem. 31:1057. Resurreccion, A.V.A. and Shewfelt, R.L. 1985. Relationships between sensory attr i b u t e s and objective measurements of postharvest q u a l i t y of tomatoes. J. Food S c i . 50:1242. R i z z i , G.P. 1967. The occurence of simple a l k y l pyrazines in cocoa butter. J. Agric. Food Chem. 15:549. Ronald, A.P. and Thomson, W.A.B. 1964. The v o l a t i l e sulphur compounds of oysters. J. Fish Res. Bd. Canada. 21:1481. Rubico, S.M., Resurreccion, A.V.A., and Beuchat, L.R. 1988. Comparison of sensory properties and headspace v o l a t i l e s of a peanut beverage processed at d i f f e r e n t temperature and time conditions. J.Food S c i . 53:176. Sanford, K.A., G u l l e t t , E.A., and Roth, V.J. 1988. Optimization of the sensory properties of frozen peas by p r i n c i p a l components an a l y s i s and multiple regression. Can. Inst. Food S c i . Technol. J . 21:174. Sapers, G.M., Sullivan, J.F., and T a l l e y , F.B. 1970. Flavor q u a l i t y i n explosion puffed dehydrated potato. 1. A gas chromatographic method for the determination of aldehydes associated with flavor q u a l i t y . J. Food S c i . 35:728. j SAS I n s t i t u t e Inc. 1989. SAS User's Guide: S t a t i s t i c s , version 5 e d i t i o n , release 6.03. Cary, NC. Sawyer, F.M., Cardello, A.V., and P r e l l , P.A. 1988. Consumer evaluation of the sensory properties of f i s h . J . Food S c i . 53:12. Schreier, P. and Reiner, L. 1979. Characterization and d i f f e r e n t i a t i o n of grape brandies by multiple discriminant analysis. J. S c i . Food Agric. 30:319. Seki, N. 1971. Nucleotides i n aquatic animals and seaweeds. B u l l . Jpn. Soc. S c i . F i s h . 37:777. Selke, E., Rohwedder, W.K., and Dutton, H.J. 1975. V o l a t i l e components from t r i s t e a r i n heated i n a i r . J . Amer. O i l Chem. Soc. 52:232. Sentheshanmuganathan, S. 1960. The mechanism of the formation of higher alcohols from amino acids by Saccharomyces cerevisiae. Biochem. J. 74:568. 193 Shahidi, F. Rubin, L.J., and D'Souza, L.A. 1986. Meat f l a v o r v o l a t i l e s : a review of the composition, techniques of analysis, and sensory evaluation. CRC C r i t . Rev. Food S c i . Nutr. 24:141. Shibamoto, T. 1977. Formation of s u l f u r - and nitrogen-containing compounds from the reaction of f u r f u r a l with hydrogen s u l f i d e and ammonia. J. Agric. Food Chem. 25:206. Shimoda, M. and Shibamoto, T. 1990. I s o l a t i o n and i d e n t i f i c a t i o n of headspace v o l a t i l e s from brewed coffee with an on-column GC/MS method. J. Agric. Food Chem. 38:802. Silverman, B.W. 1986. Density Estimation for S t a t i s t i c s and Data Analysis. Chapman and H a l l , New York, NY. Simpson, K.L. 1982. Carotenoid pigments i n seafood. In "Chemistry and Biochemistry of Marine Food Products," R.E. Martin, G.G. F l i c k , C E . Hebard, and D.R. Ward (Ed.), p. 115. AVI Publishing Co., Westport, CN. Sink, J.D. 1973. L i p i d soluble components of meat , odor and t h e i r biochemical o r i g i n . J . Amer. O i l Chem. Soc. 50:470. Sipos, J.C and Ackman, R.G. 1964. Association of dimethylsulphide with the "blackberry" problem i n cod from the Labrador area. J . F i s h . Res. Bd. Canada 21:423. Skrede, G., and Storebakken, T. 1986. C h a r a c t e r i s t i c s of color i n raw, baked and smoked wild and pen-reared a t l a n t i c salmon. J . Food S c i . 51:804. | i Smeyers-Verbeke, J., Massart, D.L., and Coomans, D. 1977. Application of line a r discriminant analysis to the d i f f e r e n t i a t i o n of pure milk from d i f f e r e n t species and mixtures. J. Assoc. Off. Anal. Chem. 60:1382. Smouse, T.H. and Chang, S.S. 1967. A systematic c h a r a c t e r i z a t i o n of the reversion f l a v o r of soybean o i l . J. Amer. O i l Chem. Soc. 44:509. Stansby, M.E. 1962. Speculations on f i s h y odors and f l a v o r s . Food Technol. 16(4):28. Stryjewska, E, Rubel, S., Henrion, A, and Henrion, G. 1987. S t a t i s t i c a l estimation of the influence of mineralization methods on the res u l t s of heavy metals determination i n cereals by the DPASV method. Fresenius Z. Anal. Chem. 327:679. Syarief, H., Hamann, D.D., Giesbrecht, F.G., Young, C.T., and Monroe, R.J. 1985. Interdependency and underlying dimensions of sensory f l a v o r c h a r a c t e r i s t i c s of selected foods. J . Food S c i . 50:631. 194 Tang, J., J i n , Q.Z., Shen, G.H., Ho, C.T., and Chang, S.S. 1983. I s o l a t i o n and i d e n t i f i c a t i o n of v o l a t i l e compounds from f r i e d chicken. J . Agri c . Food Chem. 31:1287. Tanchotikul, U. and Hsieh, T.C.-Y. 1989. V o l a t i l e f l a v o r components i n c r a y f i s h waste. J. Food S c i . 54:1515. Taylor, A.J. and Mottram, D.S. 1990. Composition and odour of v o l a t i l e s from autoxidized methyl arachidonate. J. S c i . Food Agric. 50 :407. Thomas, CP., Dimick, D.S., and McNeil, J.H. 1971. Sources of f l a v o r i n poultry skin. Food Technol. 25 (4) :109. Tokunaga, T., Iida, H., Nakamura, K., Terrano, S-, Furukawa, K., and Hoshino, C. 1982. Study on the several chemical t e s t s f o r estimating the q u a l i t y of canned product - I. On the canned mackerel. B u l l . Tokai Reg. Res. Lab. 107:1. van der Ouwelend, G.A.M. and Peer, H.G. 1975. Components co n t r i b u t i n g to beef f l a v o r . V o l a t i l e compounds produced by the reaction of 4-hydroxy-5-methyl-3(2H)-furanone and i t s thio analog with hydrogen s u l f i d e . J. Agric. Food Chem. 23:501. Vangtal, A. and Hammond, E.G. 1986. Correlation of the f l a v o r c h a r a c t e r i s t i c s of Swiss-type cheeses with chemical parameters. J. Dairy S c i . 69:2892. Van Ness, J.W. and Simpson, C. 1976. On the ef f e c t s of dimension i n discriminant analysis. Technometrics 18:175. Vejaphan, W., Hsieh, T.C.-Y., and Williams, S.S. 1988. V o l a t i l e f l a v o r components from b o i l e d c r a y f i s h (Procambarus c l a r k i i ) t a i l meat. J. Food S c i . 53:1666. Vernin, G. and Parkanyi, C. 1982. "Chemistry of Heterocyclic Compounds i n Flavours and Aromas," G. Vernin and E. Horwood (Ed.) . Halsted press, Chichester, West Sussex. Walradt, J.P., P i t t e t , A.O., K i n l i n , T.E., Muralidhara, R., and Sanderson, A. 1971. V o l a t i l e components of roasted peanuts. J . Agric. Food Chem. 19:972. Watanabe, K. and Sato, Y. 1971. Gas chromatographic and mass s p e c t r a l analyses of heated f l a v o r compounds of beef f a t s . Agric. B i o l . Chem. 35:756. Wheaton, F.W., and Lawson, T.B. (Eds.) 1985. In "Processing Aquatic Food Products." John Wiley and Sons, N.Y. Whistler, R.L. and Daniel, J.R. 1985. Carbohydrates. Ch. 3. In "Food Chemistry," 0. Fennema (Ed.), p. 69. Marcel Dekker Inc., New York, NY. 195 Whitfield, F.B., Freeman, D.J., and Bannister, P.A. 1981a. Dimethyl t r i s u l p h i d e : an important o f f - f l a v o u r component of the royal red prawn (Hymenopenaeus siboqae). Chem. Ind. (London) October, p. 692. Whitfield, F.B., Freeman, D.J., Last, J.H., and Bannister, P.A. 1981b. Bis-(methylthio)-methane: an important o f f - f l a v o u r component of prawns and sand lobs t e r s . Chem. Ind. (London) March, p. 158. Whitfield, F.B., Freeman, D.J., Last, J.H., Bannister, P.A., and Kennett, B.H. 1982. Oct-l-en-3-ol and (5Z)-octa-1,5-dien-3-ol, compounds important i n the flavour of prawns and sand-lobsters. Aust. J . Chem. 35:373. Wilkinson, L. 1990. SYSTAT: The System for S t a t i s t i c s , version 4.2. Evanston, IL. Wilson, R.A. and Katz, J. 1972. Review of l i t e r a t u r e on chicken f l a v o r and report of i s o l a t i o n of several new chicken f l a v o r components from aqueous cooked chicken. J . Agric. Food Chem. 30:741. Wong, N.P., Damico, J.N.P., and Slawin, H. 1967. Investigation of v o l a t i l e compounds i n cod f i s h by gas chromatography and mass spectrometry. J. Assoc. Off. Anal. Chem. 50:8. Woo, A.H. and Lindsay, R.C. 1983. S t a t i s t i c a l c o r r e l a t i o n of qu a n t i t a t i v e f l a v o r i n t e n s i t y assessment and i n d i v i d u a l free f a t t y a c i d measurements for routine detection and prediction of h y d r o l y t i c r a n c i d i t y o f f - f l a v o r s i n butter. J. Food S c i . 48:1761. Wurzenburger, M. and Grosch, W. 1986. Enzymic oxidation of l i n o l e n i c a c i d to 1,Z5-octadien-3-ol, Z2, Z5-octadien-l-ol and 10-oxo-E-8-decenoic acid by a protein f r a c t i o n from mushrooms ( P s a l l i o t a bispora) . Lipids 21:261. Yasuhara, A. 1987. Comparison of v o l a t i l e components between fresh and rotten mussels by gas chromatography-mass spectrometry. J. Chromatogr. 409:251. Young, C.T. and Hovis, A.R. 1990. A method for the rapid analysis of headspace v o l a t i l e s of raw and roasted peanuts. J. Food S c i . 55:279. Young, L.L., Bargmann, R.E., and Powers, J . J . 1970. Co r r e l a t i o n between gas chromatographic patterns and f l a v o r evaluation of chemical mixtures and of cola beverages. J . Food S c i . 35:219. Zar, J.H. (Ed.) 1984. " B i o s t a t i s t i c a l - Analysis," 2nd ed. P r e n t i c e - H a l l Inc., Englewood C l i f f s , NJ 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

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"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0100443/manifest

Comment

Related Items