- Email: [email protected]
MS and the electronic nose
Meat Science 65 (2003) 1125–1138 www.elsevier.com/locate/meatsci
A comparison of warmed-over flavour in pork by sensory analysis, GC/MS and the electronic nose M.G. O’Sullivana,*, D.V. Byrnea, M.T. Jensena, H.J. Andersenb, J. Vestergaardc a
Department of Dairy and Food Science, The Royal Veterinary and Agricultural University, DK-1958, Frederiksberg C, Copenhagen, Denmark b Department of Animal Product Quality, Danish Institute of Agricultural Sciences, PO Box 50, DK-8830, Tjele, Denmark c YTI Research Centre, PO Box 181, FIN-50101, Mikkeli, Finland Received 24 June 2002; received in revised form 3 October 2002; accepted 4 December 2002
Abstract Pork muscle samples (M. longissimus dorsi and M. psoas major) were obtained from pigs given one of 4 dietary treatments, (i) control diet, (ii) supplemental iron (300 mg iron (II) sulphate/kg feed), (iii) supplemental vitamin E (200 mg dl-a-tocopheryl acetate/kg of feed) and (iv) supplemental vitamin E+supplemental iron. Warmed-over flavour (WOF) was evaluated by a trained sensory panel (n=8) for the four treatments cooked and refrigerated at 4 C for up to 5 days. Gas chromatography mass spectrometry (GC/MS) and Electronic nose analysis was performed on a subset of the full design which included samples of M. longissimus dorsi, treatments (ii) and (iii) and M. psoas major with treatment (i) for 0 days of WOF development. Day 5 of WOF development was included in the subset and represented by samples of M. longissimus dorsi, treatment (iv) and M. psoas major, treatments (ii) and (iii). Bi-linear modeling was used to determine the correlation of GC/MS and electronic nose data to sensory data. Also, the reproducibility and reliability of electronic nose data was evaluated by repeating the analysis of samples in a different laboratory and with a time difference of approximately 11 months. Mean-centring was used to normalise the data from these two different electronic noise data sets. GC/MS data correlated to sensory data with specific compounds (e.g., pentanal, 2-pentylfuran, octanal, nonanal, 1-octen-3-ol and hexanal), proving to be good indices of oxidation in cooked samples of M. longissimus dorsi and M. psoas major. Electronic nose data correlated to sensory data and separated the sensory variation. The reproducibility of this data was high with the second set of samples being predictive of the first set. # 2003 Elsevier Ltd. All rights reserved. Keywords: Pork; Sensory; Warmed-over flavour; Iron; Vitamin E; GC/MS; Electronic nose
1. Introduction Up to the present, the analysis of characteristic food odours has been commonly carried out by human assessment and headspace/direct gas chromatography mass spectrometry (GC/MS) (Grigioni, Margaria, Pensel, Sa´nchez, & Vaudagna, 2000). One compound, hexanal, has been identified and used to evaluate the oxidative state and correlated to sensory scores of meat from different animals (e.g. Shahidi, Yun, Rubin, & Wood, 1987, pork; Drumm & Spanier, 1991, beef, Byrne, Bredie, Mottram, & Martens, 2002, chicken). Hexanal is a secondary breakdown product formed * Corresponding author. Tel.: +45-352-83174; fax: +45-35283210. E-mail address: [email protected] (M.G. O’Sullivan). 0309-1740/03/$ - see front matter # 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0309-1740(02)00342-X
during the oxidation of linoleic acid (C18:2) (Frankel, Neff, & Selke, 1981). Other compounds have also been associated with oxidation, but specifically with warmedover flavour (WOF) development. St.Angelo et al. (1987) proposed that pentanal and 2,4-decadienal could be considered for use as marker compounds to follow the development of WOF and its associated rancid flavours in cooked meats. The usefulness of GC/MS is clear for the detection of WOF in cooked chill stored meat products, but as a technique it has certain drawbacks. Instrumental techniques, such as GC/MS, have high operating costs and are time consuming (Pryzylski & Eskin, 1995). However, the electronic nose may provide a practical advantage over other methods and may have an application in an On-Line/At-Line capacity for the quality determination of meat products with respect to WOF development.
1126
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
The electronic nose is usually an array of chemical gas sensors with a broad and partly overlapping selectivity for measurement of volatile compounds within the headspace over a sample combined with computerised multivariate statistical data processing tools (Gardner & Bartlett, 1994). In contrast to the well known analytical (GC/MS) and sensory techniques that have been used for the analysis of flavour compounds, the electronic nose does not give any information about the compounds causing the investigated aroma; neither about the identity of the compounds nor their sensorial properties. Using the electronic nose the aroma is judged by the so-called ‘‘aroma pattern’’, which should be characteristic to the investigated substrate (Siegmund & Pfannhauser, 1999). The sensor array of an electronic nose has a very large information potential and will give a unique overall pattern of the volatiles. In principle, both the electronic and the human nose operate by sensing simultaneously a high number of components giving rise to a specific response pattern (Haugen & Kvaal, 1998). The principal aims of this experiment were: (1) To correlate sensory and GC/MS analysis and to identify compounds that could be used as indices of lipid oxidation. (2) To assess the suitability of the electronic nose for the measurement of WOF development in different types of cooked pork meats. (3) To determine the reproducibility of electronic nose measurements of similar samples over time and space, i.e., two sets of similar samples measured using the same instrument, but in two separate laboratories and with a time separation of 11 months.
2. Methods and materials 2.1. Animals and diets Slaughter pigs were reared at The Danish Institute of Agricultural Sciences (DIAS), Foulum, Denmark. Pork muscle samples (M. longissimus dorsi and M. psoas major) were obtained from seven female DLY (Duroc/ Landrace/Yorkshire) cross bred pig from each of four dietary treatments. A control diet, a supplemental iron diet [300 mg iron (II) sulphate/kg feed], a supplemental vitamin E diet (200 mg dl-a-tocopheryl acetate/kg of feed) and a supplemental vitamin E (200 mg dl-a-tocopheryl acetate/ kg of feed)+supplemental iron diet [300 mg iron (II) sulphate/kg feed]. A subset of the full design listed earlier (Table 1) was used in all subsequent analysis and included samples of M. longissimus dorsi and the vitamin E and iron treatments, M. psoas major and the control group with day 0 of WOF development and samples of M. longissimus dorsi and the iron/vitamin E treatment, M. psoas major and vitamin E and iron treatments with day 5 of WOF development. This subset of samples was
selected in a preliminary evaluation using experts with product experience, and reflected the variation in muscle type, treatment and degree of WOF development. A previous sensory experiment (O’Sullivan, Byrne, Nielsen, Andersen, & Martens, 2002) using the same samples as the current study was conducted on the full set of samples. The pigs were penned individually and fed three times daily for the first month and twice thereafter. Water was provided ad libitum throughout the study. Iron supplementation was undertaken 1 month prior to slaughter and vitamin E supplementation 2 months prior to slaughter. All animals received vitamin C (9 g vitamin C/kg feed) in the ration 1 month prior to slaughter to facilitate iron uptake and were fed to all treatment groups to set an equal base level for all groups. 2.2. Sampling procedure and storage Slaughter was performed as described by Rosenvold et al. (2001). Following slaughter, carcasses were split centrally and chilled to 4 C for 24 h prior to muscle excision. The M. longissimus dorsi and M. psoas major muscles were removed from the right side of each animal. Muscles were then vacuum-packed and frozen at 20 C and stored in the dark for approximately 2 weeks. 2.3. Pattie preparation To facilitate ease of cutting and grinding the frozen muscles were placed in a refrigerator at 5 C for 12 h. Connective tissue and visible fat, both subcutaneous and inter muscular, were removed prior to cutting the muscles into cubes (approx. 2 cm3). The seven muscles from each individual treatment were then thoroughly mixed before grinding in a rotary screw mincer (model X 70, Scharfen GmbH & Co. Maschinenfabrik KG, Witten, Germany) through a 4.5-mm hole plate. The minced samples were again mixed before shaping in to 70 g patties (approx. 1 cm thick) using a commercial burger maker (i.d. 9 cm). Patties were then vacuumpacked in oxygen impermeable plastic laminate bags (polyamide/polyethylene 20/70) before being frozen at 20 C for up to 4 weeks (Byrne, Bak, Bredie, Bertelsen, & Martens, 1999; O’Sullivan, Byrne, Martens, & Martens, 2002). 2.4. Heat treatment and storage Patties were thawed in warm water to a temperature of approximately 20 C and removed from their vacuum bags. All samples were then in turn batch cooked in a preheated convection oven at 140 C for 30 min and turned every 5 min to facilitate even cooking. Internal cooking temperatures reached were found to be
1127
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138 Table 1 Sample codes, muscles, supplemental diets and days of warmed-over flavour (WOF) Sample code
Muscle
LD-E-0-1, -2 LD-I-0-1, -2 LD-IE-5-1, -2 PS-E-5-1, -2 PS-I-5-1, -2 PS-C-0-1, -2
M. longissimus dorsi M. longissimus dorsi M. longissimus dorsi M. psoas major M. psoas major M. psoas major
Vitamin Ea
Ironb
Vitamin Cc
@ x @ @ x x
x @ @ x @ x
@ @ @ @ @ @
Days WOF 0 0 5 5 5 0
-1=Sample set 1, analysed at The Royal Veterinary and Agricultural University, Copenhagen, Denmark. -2=Sample set 2, analysed at The YTI Research Centre, PO Box 181, FIN-50101, Mikkeli, Finland after 11 months frozen storage at 20 C. @=Present, x=Absent. a 200 mg dl-a-tocopheryl acetate/kg of feed. b 3000 mg iron (II) sulphate/kg feed. c 9 g/kg feed.
approximately 80 C. All samples were cooked in the same oven to reduce cooking variation. After cooking, samples were allowed to cool to 5 C in a refrigerator before being placed in oxygen permeable bags (low density polyethylene 1502500.025 mm) and stored at 4 C in the dark for up to 5 days to allow the development of WOF (Byrne et al., 1999; O’Sullivan, Byrne, Martens et al., 2002). 2.5. Re-heating treatment Patties were divided into eight equal triangular pieces and individually vacuum-packed in oxygen permeable plastic laminate bags (polyamide/polyethylene 20/70) before being placed in a water filled steel tray (at ambient temperature) and re-heated in a convection oven at 140 C for 19 min. Mean serving temperature of samples was found to be 65 C (Byrne et al., 1999; Byrne, O’Sullivan, Dijksterhuis, Bredie, & Martens, 2001). 2.6. Training and profiling An eight member sensory panel (four males/four females, aged from 24 to 62 years) was recruited from the public and students of The Royal Veterinary and Agricultural University, Frederiksberg, Denmark. Selection criteria for panellists were availability and motivation to participate on all days of the experiment. Sensory training was carried out prior to sensory profiling (Byrne et al., 1999; O’Sullivan, Byrne, Martens et al., 2002). Sensory analysis was carried out in the panel booths at the university sensory laboratory that conforms to ISO (1988) international standard. During both training and sensory profiling, analysis was performed with assessors not having any knowledge of sample history so as not to introduce bias into their sensory assessment. Unstructured 15-cm line scales anchored on the left by the term ‘none’ and on the right by the term ‘extreme’ were used for all the sensory descriptors (Meilgaard, Civille, & Carr, 1999). The responses of the panellists were recorded by measuring
the distance in mm (1–150) from the left side of the scale for the odour, flavour, taste and after taste sensory terms as described in Table 2. 2.6.1. GC/MS sample pre-treatment and extraction The meat patties were received vacuum packed in oxygen-impermeable plastic bags frozen at 20 C. To ensure a constant temperature of the patties before cooking, the plastic bags were placed in a water bath and held at 20 C until they were judged to be isothermal to the surrounding water. The thawed patties were cooked in a preheated convection oven identically to the sensory samples. Of the cooked patties 3 g was transferred to flame-heated (and subsequently cooled) 10-ml vials and equilibrated in 5 min at 65 C (the serving temperature in the sensory analysis) in the heating box of the autosampler (CTC analytics, Zwingen, Switzerland). To induce warmed over flavour, the cooked patties were packed in oxygen permeable polyethylene bags and stored at 5 C for 5 days. An equilibration time of 5 min at 65 C was selected from a pilot experiment, which showed that the aldehyde production was complete after 10 min, an equilibration time longer than 5 min would therefore mask the fine structure of the volatiles produced. The volatiles were sampled with an SPME-fibre (75 mm carboxen/polydimethylsiloxane) in 1 min. The fibre was thereafter transferred to the injection port of the gas chromatograph and desorbed at 250 C. To ensure the samples were not oxidised during storage before a run on GC/MS, only one sample was analysed at a time. 2.6.2. GC/MS analysis The gas chromatograph was a Varian Star 3400 CX coupled with a Varian Saturn 2000 mass spectrometer. The column was a DB-1701, length of 30 m, 0.25 mm, (i.d. 1 mm) operated with a gas flow of 1.6 ml min 1 at 35 C. The temperature gradient started at 35 C and was held at that temperature for 10 min until a gradient with an end temperature of 150 C with an increase of 3 C/min was initiated. The temperature of 150 C was
1128
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
Table 2 Sensory descriptive terms and references developed for sensory profiling Sensory term
Definition (with appropriate reference)
Odour 1. Cardboard-O 2. Linseed oil-O 3. Rubber/Sulphur-O 4. Nut-O 5. Green-O 6. Fatty-O
Odour reference Wet cardboard Warmed linseed oil/linseed oil based paint Warmed rubber/the white of a boiled egg Crushed fresh hazel nuts Fresh green French beans Pig back fat (fresh, non-oxidised)
Taste 7. Sweet-T 8. Salt-T 9. Sour-T 10. Bitter-T 11. MSG/Umami-T
Taste reference Sucrose 1 g/l aqueous soln. Sodium chloride 0.5 g/l aqueous soln. Citric acid monohydrate 0.3 g/l aqueous soln. Quinine chloride 0.05 g/l aqueous soln. Monosodium glutamate 0.5 g/l aqueous soln.
Flavour 12. Metallic-F/ Bloody-F 13. Fresh cooked pork-F 14. Rancid-F 15. Lactic acid/fresh sour-F 16. Vegetable oil-F 17. Piggy/Animal-F 18. Fish-F 19. Tinny-F 20. Livery-F
Flavour reference Ferrous sulphate 0.1 g/l aqueous soln. Oven cooked pork without browning Oxidised vegetable oil Natural yoghurt Fresh vegetable oil Skatole 0.06 mg/ml refined vegetable oil Fish stock in boiling water Stainless steel strip Cooked beef liver
Aftertaste 21. Astringent-AT
Aftertaste reference Aluminium sulphate 0.02 g/l aqueous soln.
Suffix to sensory terms indicates method of assessment by panelists; -O=Odour, -F=Flavour, -T=Taste, -AT=Aftertaste.
held for 5 min, then a column cleaning gradient was initiated with an increase of 30 C/min to 250 C and held at 250 C for 5 min. The mass spectrometer had a mass range of 35–400 m/z with a scan time of 1 s in electron impact mode with an electron energy of 70 eV and an electron multiplier voltage of 1650 V. The chromatograms were integrated and the retention times and peak areas, transferred to an Excel spreadsheet. Peaks with retention time of more than 40 min were deleted as they were high boiling compounds with no aroma. The data were then aligned in an aligning program (Borggaard, 2000) with sample no. P-I-5 as reference chromatogram and the lowest area limit was set to 5000. 2.7. Electronic nose Electronic nose analysis was conducted initially using the samples outlined in Table 1 (sample set 1) and in a laboratory at The Royal Veterinary and Agricultural University, Copenhagen, Denmark. Eleven months later the experiment was repeated using similar samples to those presented in Table 1 (sample set 2) at The Department of Sensory Science, Institute of Environmental Technology, Mikkeli, Finland. Samples were held at 20 C during the interim period. The MGD-1
(Environics Ltd, Finland) is based on the principle of ion mobility. The ion mobility cell IMCELL1 contains six channels (IMC-1 to IMC-6). IMC-1, IMC-2 and IMC-3 register signals from positive ions. IMC-4, IMC-5 and IMC-6 register signals from negative ions. The MGD-1 is also equipped with a semiconductor cell (SCCELL) as a ‘‘seventh’’ sensor. One gram of sample was extracted in 5 ml isooctane. The sample was diluted in isooctane (1:8) and 5 ml injected into a desorber, which is a special device containing a ceramic plate developed for the MGD-1 and used for laboratory experiments. The ceramic plate was heated to 300 C and MGD-1 signals were obtained for 90 s (IMCELL temperature 75 C). Measurements were carried out with two replicates. Electronic nose profile data were collected by summing signals for each of the channels (0–60 s). 2.8. Data analysis ANOVA–Partial Least Squares Regression (APLSR) was used to investigate product sensory variation in the profiling data and was performed using full-cross validation. The X-matrix was designated as 0/1 design variables for assessor, replicate and product. The Ymatrix was designated as sensory and instrumental
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
variables. In this model assessor and replicate level effects were removed (Martens & Martens, 2000). In order to compare the electronic nose data sets (1 and 2) a mean-centring (Martens & Martens, 2001; chapter 5) of these instrumental measurements was undertaken, which determined the patterns of variation between samples and in effect normalised these data. All analyses were performed using the Unscrambler Software, version 7.6 (CAMO ASA, Trondheim, Norway).
3. Results and discussion 3.1. Sensory sample variation ANOVA–APLSR was used to process the mean data accumulated from eight test subjects during sensory profiling. The X-matrix was designated as 0/1 design variables for meat samples and days of WOF development. The Y-matrix was designated as sensory variables (Table 2). The optimal number of components in the APLSR model presented was determined to be 3 Principal Components. Thus PC 1 versus 2 (Fig. 1) and PC 2 versus 3 were explored. PC 2 versus 3 is not presented here, as additional information was not gained through their examination. Further PCs did not provide any predictive improvement in the Y-matrix obtained. In this model assessor and session level effects were removed. The calibrated explained variance for this model was PC1=22% and PC2=13%. Fig. 1 is presented as a correlation loadings plot. In this fig. sample LD-E-0, a cooked sample that has been stored in a
1129
refrigerator at 4 C for 0 days, covaried with Fresh Cooked Pork-F and Lactic acid/fresh sour-F. Sample LD-I-0 also covaried with both these descriptors, but to a lesser extent. These samples reflect 0 days of WOF development. Sample PS-C-0 covaried with Nut-O, Fatty-O and to a lesser degree to the oxidative terms Rubber/Sulphur-O and Cardboard-O. The M. psoas major is more prone to oxidative attack compared with M. longissimus dorsi. This is in confirmation with the results of O’Sullivan, Byrne, and Nielsen et al. (2002) who performed a WOF experiment with a full design (2 muscles, M. longissimus dorsi and M. psoas major 4 dietary treatments, (i), (ii), (iii) and (iv) 4 days of WOF development, 0, 1, 3 and 5) of which the results presented here are a subset. These authors observed a greater degree of WOF development for M. psoas major compared with M. longissimus dorsi and postulated that this may reflect the greater iron levels in M. psoas major, thus predisposing it to a greater oxidative effect expressed as degree of WOF development. Samples LD-IE-5 and PS-E-5, cooked samples that have been stored in a refrigerator at 4 C for 5 days, covaried with Bitter-T and to a lesser extent to Sour-T. However, PS-I-5 covaried with the descriptors Astringent-AT, MSG-T and to the oxidative descriptors Rancid-F, Fish-F and Linseed oil-O. These results are again in confirmation with the more complete experiment conducted by O’Sullivan, Byrne, and Nielsen et al. (2002) who showed that iron supplementation increased the susceptibility to warmedover flavour development for both M. longissimus dorsi and M. psoas major compared with the other dietary treatments. The ability of heme pigments and non-heme
Fig. 1. An overview of the variation found in the mean data from the ANOVA–Partial Least Squares Regression (APLSR) correlation loadings plot for sample set 1. Shown are the loadings of the X- and Y-variables for the first 2 PCs. =sensory descriptor and ~=sample and days of WOF. The concentric circles represent 100 and 50% explained variance, respectively.
1130
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
iron to accelerate the propagation step of the free-radical chain mechanism can explain the rapid rate of oxidation in cooked meats (Pearson, Love, & Shortland, 1977). The samples selected for this experiment thus reflect varying degrees of WOF development and Fig. 1 displays this dimension along PC 1. Samples stored for 5 days under refrigerated conditions displayed the greatest levels of WOF. The M. psoas major muscle produced greater WOF compared to M. longissimus dorsi and iron supplementation of animal diets produced greater WOF than any of the other dietary treatments. This variation in WOF between the different muscle samples thus provided an interesting model for subsequent GC/MS and electronic nose analysis. 3.2. Sensory and GC/MS variation ANOVA–APLSR was also used to process the mean data accumulated from sensory evaluation by the test subjects and GC/MS analysis. Table 3 corresponds to compounds identified by GC/MS analysis of meat samples. The X-matrix was designated as 0/1 design variables for meat samples and sensory variables (Table 2). The Y-matrix was designated as GC/MS measurements. Fig. 2 corresponds to M. longissimus dorsi and Fig. 3 to M. psoas major. The optimal number of components in these APLSR models was determined to be 3 Principal Components. Thus PC 1 versus 2 (Figs. 2 and 3) and PC
2 versus 3 were explored. PC 2 versus 3 is not presented here, as additional information was not gained through their examination. The validated explained variance for Fig. 2 is PC 1=33% and PC 2=29% and for Fig. 3, validated explained variance PC 1=60% and PC 2=17%. M. longissimus dorsi (Fig. 2) appears to be separated along PC 1 with samples containing vitamin E on the left side of the plot and samples high in iron content along the right side of the correlation loadings plot. Sample LD-E-0-1 correlated to Fresh Cooked Pork-F and Sweet-T, sensory descriptors associated with 0 days of WOF development. Sample LD-IE-5-1 covaried in the upper portion of the correlation loadings plot and LD-I-0-1 in the lower right hand quadrant. Both these samples appear more oxidised due to their greater association with descriptor Rancid-F. Sample LD-I-0-1 also covaried with Green-O and the GC/MS identified compounds pentanal, 2-pentylfuran, octanal, nonanal, 1-octen-3-ol and hexanal. Therefore, this sample which was frozen after 0 days of WOF development underwent oxidation during frozen storage. Similarly in Fig. 3 sample PS-I-5-1 covaried with the sensory descriptor Green-O and the GC/MS identified compounds pentanal, 2-pentylfuran, octanal, nonanal, 1-octen-3-ol, and hexanal. These compounds (Figs. 2 and 3) have been found to be indicative of lipid oxidation (Byrne et al., 2002; Siegmund & Pfannhauser, 1999). Sample PS-C-0-1 covaried with the oxidative
Fig. 2. An overview of the variation found in the mean data from the ANOVA–Partial Least Squares Regression (APLSR) correlation loadings plot for sample set 1, M. longissimus dorsi samples. Shown are the loadings of the X- and Y-variables for PC 1 versus 2. ~=sensory descriptor and sample and =GC/MS isolated compounds. The concentric circles represent 100 and 50% explained variance, respectively.
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
1131
Fig. 3. An overview of the variation found in the mean data from the ANOVA–Partial Least Squares Regression (APLSR) correlation loadings plot for sample set 1, M. psoas major samples. Shown are the loadings of the X- and Y-variables for PC 1 versus 2. ~=sensory descriptor and sample and =GC/MS isolated compounds. The concentric circles represent 100 and 50% explained variance, respectively.
descriptor Cardboard-O indicating that this sample also underwent a degree of oxidation during frozen storage. The oxidative descriptors Rancid-F and Linseed oil-O were positioned midway between samples PS-I-5-1 and PS-E-5-1 indicating that they were common to both samples and thus had undergone WOF development. Another APLSR plot (Fig. 4) was constructed and included both M. longissimus dorsi and M. psoas major samples in order to determine the relative relationship of these white (M. longissimus dorsi) and red muscle samples (M. psoas major) with respect to WOF development and some of the chemical compounds especially associated with oxidation. Again, Table 3 corresponds to compounds identified by GC/MS analysis of meat samples. The X-matrix was designated as 0/1 design variables for meat samples and days of WOF development and sensory variables (Table 2). The Y-matrix was designated as GC/MS measurements The optimal number of components in the APLSR model presented was determined to be 3 Principal Components. Thus PC 1 versus 2 and PC 2 versus 3 (Fig. 4) were explored. PC 1 versus 2 is not presented here many of the compounds that relate to oxidation were scattered around the origin and thus showed that they were related to many of the samples analysed. However, PC 2 versus 3 (Fig. 4) separated these compounds and allowed the comparison of samples, sensory terms and GC/MS isolated compounds. The validated explained variance for this model
was PC 2=17% and PC 3=9%. In Fig. 4 M. longissimus dorsi appears to be separated along PC 3 and M. psoas major along PC 2. Samples containing vitamin E appear on the left side of the plot and samples high in iron content along the right side of the correlation loadings plot. Sample PS-C-0 had almost twice the level of iron compared with the highest levels found in M. longissimus dorsi (O’Sullivan, Byrne, & Nielsen et al., 2002). Again the sample PS-I-5 appears in the upper right hand quadrant of the PLSR plot and is correlated to Rancid-F and Fish-F. In Fig. 4 the compound hexanal covaried with the oxidised sample PS-I-5 (also Fig. 3) and LD-I-0-1 (also Fig. 2). Hexanal is a secondary breakdown product formed during the oxidation of linoleic acid (C18:2) (Frankel et al., 1981). Hexanal, has been identified and used to evaluate the oxidative state and correlated to sensory scores of red meat from different animals and the results presented agree with Shahidi et al. (1987), who worked on cooked pork and Drumm and Spanier, (1991), who worked with beef. Also, hexanal has a distinctive odour described as ‘green’ or ‘grassy’ (Gasser & Grosch, 1988). The sensory descriptor Green-O covaried with hexanal for both M. longissimus dorsi and M. psoas major (Figs. 2–4). The reference material for this descriptor during sensory training was fresh French beans. During the preliminary selection of reference materials it was decided by an expert panel that French beans produced an odour
1132
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
Table 3 Gas chromatography/mass spectrometry (GC/MS) isolated compounds Compound
Kovats value
Identification/ionsa
Unknown Pentane Pentanol Dimethyl sulfide Trimethylamine Methylbranched alkane 2-Methylpropanal 2-Methylfuran Butanal 2-Butanone 3-Methylhexane Benzene 3-Methylbutanal 2-Ethylfurane Pentanal Aliphatic alcohol Aliphatic alcohol Unknown An octadiene, e.g. 1,3-octadiene Hexanal Nonane Ethylbenzene A dimethylbenzene 2-Butylfuran Butanoic acid 2-Heptanone Heptanal 4-Ethyl-1-octyn-3-ol 2-Pentylfuran A methylbranched alkane 1-Octen-3-ol Benzaldehyde Octanal trans-2-Octenal Nonanal
NA 506 519 573 586 627 703 706 715 721 725 731 747 750 782 813 824 852 855 890 905 915 925 938 974 986 991 1032 1036 1054 1084 1089 1094 1179 1198
43, 45, 47 MS of AS MS of AS MS of AS MS of AS 41, 56, 85 MS MS MS MS of AS MS MS MS of AS MS MS of AS 41, 55, 70 41, 55, 70 42, 43, 45 67, 81, 110 MS of AS MS of AS MS MS MS MS of AS MS MS of AS MS MS of AS 41, 57, 71 MS of AS MS of AS MS of AS MS of AS MS of AS
a MS: Good fit with mass spectra in NIST library. MS of AS: fit with mass spectra of authentic standard. Ions: where the identification was ambiguous, the three most prominent ions in the mass spectra are given. NA: not applicable.
described as being green or grassy. Thus, Green-O was used in an attempt to characterise hexanal, isolated through subsequent GC/MS analysis and was found to covary with this compound. Other compounds have also been identified with oxidation, but specifically with warmed-over flavour development. In Fig. 4 the compounds 1-octen-3-ol, 2-pentylfuran covaried with the sensory oxidative descriptors Rancid-F, Fish-F and Rubber/Sulphur-Like-O. These results agree with those of Siegmund and Pfannhauser (1999) who found that the relative concentrations of the lipid oxidation products hexanal, 1-octen-3-ol and 2-pentylfuran increased in cooked chill stored chicken meat as storage time increased. Pentanal was also found to covary to a greater extent with the sensory sample PS-I-5, the sample with the greatest degree of WOF development (Figs. 3 and 4). Sample LD-I-0-1 (Fig. 2) also covaried
with pentanal again indicating that the pro-oxidative nature of iron during frozen storage. The result of Fig. 2, 3 and 4 are in agreement with St.Angelo et al. (1987) who proposed that pentanal and 2,4-decadienal could be considered for use as marker compounds to follow the development of WOF and its associated rancid flavours in cooked meats. In summary, many of the compounds associated with oxidation of lipids were found to correlate with the oxidative sensory descriptors and the samples with the greater levels of WOF development, i.e. pentanal, 2-pentylfuran, octanal, nonanal, 1octen-3-ol and hexanal. 3.3. Electronic nose analysis, mean-centring and sample reproducibility ANOVA–APLSR was used to compare the electronic nose sensor readings from sample sets 1 and 2 combined. The X-matrix was designated as 0/1 design variables for muscle, treatment and days of WOF development (main effects). The Y-matrix was designated as electronic nose measurements. The optimal number of components in the APLSR model presented was determined to be 3 PCs. Fig. 5 shows a correlation loadings plot (validated explained variances PC1=12% and PC2=9%) of the main design effects of sample sets 1 and 2. M. longissimus dorsi samples were inversely correlated with those of M. psoas major. Each dietary treatment is separated and occupies a quadrant of the correlation loadings plot with the iron/vitamin E and vitamin E treatments separated along PC 1 and iron and iron/vitamin E treatments separated along PC 2. The Control group occupies the bottom right quadrant and day 0 and day 5, days of WOF development, are inversely correlated. In effect the electronic nose device used in this experiment could qualitatively separate samples on the basis of muscle type, treatment and degree of WOF development. ANOVA–APLSR was again used to compare the electronic nose sensor readings from sample sets 1 and 2, analysed with the same instrument, but at different locations and with a time separation of 11 months. The X-matrix was designated as 0/1 design variables for meat sample type and days of WOF development. The Y-matrix was designated as electronic nose measurements. The optimal number of components in the APLSR model presented was determined to be 3 Principal Components. Thus PC 1 versus 2 (Fig. 6a), PC 2 versus 3 (Fig. 6b) and PC 1 versus 3 (Fig. 6c) were explored and all are presented. Further PCs did not provide any predictive improvement in the Y-matrix obtained. In this model session level effects were removed and the data were normalised through meancentring. The calibrated explained variances for these models were PC1=35%, PC2=9% and PC3=3%, respectively.
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
1133
Fig. 4. An overview of the variation found in the mean data from the ANOVA–Partial Least Squares Regression (APLSR) correlation loadings plot for sample set 1. Shown are the loadings of the X- and Y-variables for PC 2 versus 3. =sensory descriptor and GC/MS isolated compounds and ~=sample and days of WOF. The concentric circles represent 100 and 50% explained variance, respectively.
Fig. 5. An overview of the variation found in the mean data from the ANOVA–Partial Least Squares Regression (APLSR) correlation loadings plot for sample set 1 and sample set 2 combined. Shown are the loadings of the X- and Y-variables for the first 2 PCs. =electronic nose sensor and ~=design variables (main effects). The concentric circles represent 100 and 50% explained variance, respectively.
1134
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
Fig. 6. An overview of the variation found in the mean data from the ANOVA–Partial Least Squares Regression (APLSR) correlation loadings plot for sample set 1 and 2. Shown are the loadings of the X- and Y-variables for (a) PC 1 versus 2, (b) PC 2 versus 3, (c) PC 1 versus 3. =Electronic nose sensor and ~=sample and days of WOF. The concentric circles represent 100 and 50% explained variance, respectively.
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
1135
Fig. 6. (continued)
Fig. 6a reveals samples LD-I-0-1, 2 and PS-C-0-1, 2 covaried with day 0, which corresponds to 0 days of warmed-over flavour development and samples LD-IE5-1, 2 covaried with day 5 which corresponds to 5 days of warmed-over flavour development. PS-C-0-1, 2 also covaried with day 0, but to a lesser extent. The remaining samples appear clustered around the origin, but PSI-5-1 covaried to a greater degree to day 5 of WOF development. In Fig. 6b the various samples are separated by the order of oxidation LD-E-0 < LD-I-0 < PSC-0 < LD-IE-5< PS-E-5 < PS-I-5. Similar results are also found in Fig. 6c, with samples being separated according to their respective degrees of WOF development. In summary the electron nose data from sample sets 1 and 2 agree with the sensory data in Fig. 1 and the device was effective in determination of the oxidative state of the samples analysed. In Fig. 6a samples (Table 1) LD-E-0-1 and LD-E-0-2 covaried and similarly, LD-I-0-1 and LD-I-0-2, PS-C-01 and PS-C-0-2, PS-I-5-1 and PS-I-5-2 covaried. However, samples LD-IE-5-1 and LD-IE-5-2, PS-E-5-1 and PS-E-5-2 displayed greater levels of variation. Similar results are seen in Figs. 6b and c except for samples PSE-5-1 and PS-E-5-2 in Fig. 6b which varied to a greater extent. One of the principle drawbacks to date of electronic noses is the large variation in data acquisition. In order to compare results over time (weeks, months or
years) it is required that these instruments give the same signal when identical samples are being measured over time. However, due to dynamic processes taking place in the sensors over time, the signal from a sensor array may vary significantly (Haugen & Kvaal, 1998). Gardner and Bartlett (1996) proposed odour standards and analytical expressions to define the working range and resolving power of electronic noses. A first set of standards, defined as ‘‘Type A’’, used organoleptic chemicals to determines the range of an electronic nose and a further set of standards (Type B), of similar samples were used to determine the resolving power of an electronic nose. In the present study electronic nose analysis could be reproduced using similar samples measured using the same instrument, but in two separate laboratories and with a time separation of 11 months. The greatest variation amongst samples in Fig. 6a–c appears to occur with the samples PS-E-5 and LD-IE-5, both of which display high levels of WOF development. This variation may be due to oxidative changes which occurred during the interim 11 months of frozen storage at 20 C prior to the second electronic nose analysis. Both samples were frozen when WOF development was quite advanced and also both samples contain endogenous vitamin E. Wen, Morrissey, Buckley, and Sheehy (1996) found that vitamin E level in cooked frozen ( 20 C) turkey burgers decreased from 5.6 to 2.88 and from 3.29 to 1.85
1136
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
mg/g in burgers made from turkeys supplemented with 600 and 300 mg a-tocopheryl acetate/kg feed respectively. Thus, a reduction in vitamin E levels may have occurred during frozen storage resulting in changes in the oxidation potential of these samples and expressed in Figs. 6a–c by the reduced degree of covariance of PSE-5 and LD-IE-5 as measured by the electronic nose. Moreover, lipids of pork meat can suffer alterations during frozen storage, owing to lipolysis and oxidation (Hernandez, Navarro, & Toldra, 1999). AbdelKader (1996) found that malondialdehyde, a chemical marker for lipid oxidation, increased in cooked chicken breast frozen stored for 3 and 6 months. Sample PS-I-5-1 and PS-I-5-2 covaried highly in Fig. 6a–c and displayed a high level of reproducibility as measured by the electronic nose. The apparent lack of variation in these samples as compared to the other meat samples with 5 days of WOF may be explained by the fact that these samples displayed the greatest degree of WOF (Fig. 1) and also iron supplementation led to a much more accelerated degree of WOF development. Thus, the oxidation reaction of these samples may have been more advanced and samples were more stable during subsequent frozen storage. The free-radical chain mechanism involved in the oxidation of unsaturated fatty acids is thought to occur in three stages: (1) initiation, the formation of free-radicals; (2) propagation, the free-radical chain reactions; (3) termination, the formation of non-radical products (Tappel, 1962). This lipid oxidation mechanism proposed by Tappel (1962) depends on the presence of pre-formed fatty acid hydoperoxides, which react with heme compounds and undergo homolytic decomposition. Thus, the high covariance of the electronic nose data for samples LD-E-0-1 and LD-E-0-2, LD-I-01 and LD-I-0-2 and PS-C-0-1 and PS-C-0-2, was due to their stability as oxidation had not fully commenced. The high covariance of samples PS-I-5-1 and PS-I-5-2 covaried because oxidation was most likely in the termination stage. However, samples LD-IE-5-1 and LD-IE5-2, PS-E-5-1 and PS-E-5-2 produced more variable results because these samples were most probably in the propagation stage of oxidation which continued, albeit at a lower rate, during frozen storage. In the present experimental samples were stored at 20 C. However, if samples were stored at 80 C, the variation between LD-IE-5-1, LD-IE-5-2 and PS-E-5-1, PS-E-5-2 would most probably be reduced because of the lower reaction rate at this temperature. The electronic nose has both large differences in sensitivity and selectivity from the human nose (Haugen & Kvaal, 1998). To date electronic noses have been employed in the analysis of a large variety of meat products (e.g. Eklo¨v, Johansson, Winquist, & Lundstro¨m, 1998, fermented sausage; O´lafsdo´ttir, Martindo´ttir, & Jo´nsson, 1997, fish) and in the warmed-over flavour analysis of various meat products (e.g. Siegmund &
Pfannhauser, 1999, chicken; Grigioni et al., 2000, beef). These instruments all differed in the individual sensor array set-up, numbers of sensors and the data analysis used to process the characteristic signature data of the various meat types. These data analytical techniques employed included, Neural Networks and Principal Component Regression (PCR). Balaban, Korel, Odabasi, and Folkes (2000) investigated the conversion of electronic nose datasets for the purpose of accumulating seamless data sets. They found that matrix conversion was most satisfactory and neural network transformations using multi-layer perceptrons and trained with back propagation did not perform as well. None of the afore-mentioned authors attempted to reproduce their results in time and space (i.e. at a different time and location). In the present study, differences in the electronic nose measurements between sample sets 1 and 2 were mean-centred using Unscrambler Software, version 7.6 (CAMO ASA, Trondheim, Norway). In effect this normalised the data for subsequent analysis using PLS regression which allowed the direct comparison of data from sample sets 1 and 2. Another potential drawback of the electronic nose is that sensors have a limited life, they must be replaced after some time and new sensors from another batch will differ in performance (Haugen & Kvaal, 1998). One potential method of solving these fundamental problems is to use a reliable data analytical tool to correct for variations over time, as discussed earlier and secondly to use an electronic nose in which the sensors do not require replacement. The sensor technology employed in the electronic nose used in this experiment is based on the principle of ion mobility and ionisation of gas molecules (MGD-1, Environics Ltd, Finland). The clusters formed through ion-molecule reactions are brought into different electrical fields perpendicular to the sample flow and detected. In effect the sensors do not wear out as volatiles do not come in contact directly with the sensors and this is advantageous over other existing electronic nose technology where sensors degrade due to direct contact with the substances being measured.
4. Conclusions The samples selected for this experiment reflect varying degrees of WOF development. Cooked samples stored for 5 days under refrigerated conditions displayed the greatest levels of WOF. The M. psoas major muscle produced greater WOF compared with M. longissimus dorsi and iron supplementation of animal diets resulted in greater WOF than any of the other dietary treatments. This variation in WOF between samples thus provided a good model for subsequent GC/MS and electronic nose analysis.
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
Many of the compounds associated with oxidation of lipids were found to correlate with the oxidative sensory descriptors and the samples with the greater levels of WOF development, i.e., pentanal, 2-pentylfuran, octanal, nonanal, 1-octen-3-ol and hexanal. The electronic nose device coupled with the multivariate methodology used in this experiment could clearly separate samples on the basis of muscle type, treatment and degree of WOF development. The electron nose data from sample sets 1 and 2 agreed with and correlated to sensory analysis. The electronic nose was effective in the determination of the oxidative state of the experimental samples. Electronic nose analysis could be reproduced using similar samples measured using the same instrument, but in two separate laboratories and with a time separation of 11 months. Oxidative changes during frozen storage may have resulted in the reduced reproducibility of electronic nose data for samples PS-E-5 and LD-IE-5. Mean-centring was effective in normalising the data from sample sets 1 and 2 prior to subsequent PLSR. The sensor technology employed in the electronic nose used in this experiment, had an advantage over existing electronic nose devices in that the sensors do not wear because the molecules measured do not come in direct contact with the sensors.
Acknowledgements This project was part of a project titled ‘‘Improved Iron Status through Utilization of Meat Proteins for an Improved Food Iron Availability and through Safe Iron Fortification’’. The project was funded by FØTEK, Norma og Frode S. Jacobsen Fund and Danske Slagterier, and is led by Professor Leif Skibsted. The Frame Program AQM (Advanced Quality Monitoring) in the Food Production Chain is also acknowledged for financial and academic contributions. Also acknowledged is Dorte Blaaberg Jensen for GC/ MS sample preparation.
References AbdelKader, Z. M. (1996). Lipid oxidation in chicken as affected by cooking and frozen storage. Nahrung Food, 40, 21–24. Balaban, M. O., Korel, F., Odabasi, A. Z., & Folkes, G. (2000). Transportability of data between electronic noses: mathematical methods. Sensors and Actuators B, 60–77. Borggaard, C. (2000). Danish Meat Research Institute, Maglega˚rdsvej 2, DK-4000 Roskilde, Denmark. Byrne, D. V., & Bak, L. S., Bredie, W. L. P., Bertelsen, G., & Martens, M. (1999). Development of a sensory vocabulary for warmed-over flavour 1: in porcine meat. Journal of Sensory Studies, 14, 47–65. Byrne, D. V., O’Sullivan, M. G., Dijksterhuis, G. B., Bredie, W. L.P, & Martens, M. (2001). Sensory panel consistency during development
1137
of a vocabulary for warmed-over flavour. Food Quality and Preference, 12, 171–187. Byrne, D. V., Bredie, W. L. P., Mottram, D. S., & Martens, M. (2002). Sensory and chemical investigations on the effect of oven cooking on warmed-over flavour in chicken meat. Meat Science, 61, 127– 139. Drumm, T. D., & Spanier, A. M. (1991). Changes in the lipid content of autoxidation and sulphur-containing compounds in cooked beef during storage. Journal of Agriculture and Food Chemistry, 49, 336– 343. Eklo¨v, T., Johansson, G., Winquist, F., & Lundstro¨m, I. (1998). Monitoring sausage fermentation using an electronic nose. Journal of Science and Food Agriculture, 76, 525–532. Frankel, N. E., Neff, W. E., & Selke, E. (1981). Analysis of autoxidised fats by gas chromatography-mass spectrometry: VII. Volatile thermal decomposition products of pure hydroperoxides from autoxidised and photosensitised oxidised methyl oleate, linoleate and linolenate. Lipids, 16, 279–285. Gardner, J. W., & Bartlett, P. N. (1994). A brief history of electronic noses. Sensors and Actuators B, 46-, 47, 211–220. Gardner, J. W., & Bartlett, P. N. (1996). Performance definition and standardization of electronic noses. Sensors and Actuators B, 60–77. Gasser, U., & Grosch, W. (1988). Flavour deterioration of soya bean oil: identification of intense odour compounds during flavour reversion. Fett Wissenschart Technologie, 90, 332–336. Grigioni, G. M., Margaria, C. A., Pensel, N. A., Sa´nchez, G., & Vaudagna, S. R. (2000). Warmed-over flavour in low temperaturelong time processed meat by an ‘‘electronic nose’’. Meat Science, 56, 221–228. Haugen, J. E., & Kvaal, K. (1998). Electronic nose and artificial neural network. Meat Science, 49, S273–S286.. Hernandez, P., Navarro, J., & Toldra, F. (1999). Effect of frozen storage on lipids and lipolytic activities in the longissimus dorsi muscle of the pig. Zeitschrift fu¨r Lebensmittel-Untersuchung und Foschung A, 208, 110–115. ISO 8589. (1988). Sensory analysis. General guidance for the design of test rooms. Paris: International Organisation for Standardisation. Martens, H., & Martens, M. (2000). Modified Jack-knife estimation of parameter uncertainty in bilinear modelling by partial least squares regression (PLSR). Food Quality and Preference, 11, 5–16. Martens, H., & Martens, M. (2001). In Multivariate analysis of quality. An introduction (pp. 99–101). Chichester: John Wiley and Sons. Meilgaard, M. C., Civille, G. V., & Carr, B. T. (1999). Sensory evaluation techniques (3rd ed.). Boca Raton, FL: Academic Press. O´lafsdo´ttir, G., Martindo´ttir, E., & Jo´nsson, E. H. (1997). Rapid gas sensor measurements to determine spoilage of capelin (Mallotus villosus). Journal of Agriculture and Food Chemistry, 45, 2654– 2659. O’Sullivan, M. G., Byrne, D. V., Martens, H., & Martens, M. (2002). Sensory and data analytical methodologies in the development of a vocabulary for evaluation of meat quality. Journal of Sensory Studies, 64, 175–189. O’Sullivan, M. G., Byrne, D. V., Nielsen, J. H., Andersen, H. J., & Martens, M. (2002). Sensory and chemical assessment of pork supplemented with iron and vitamin E. Meat Science, 64, 175–189. Pearson, A. M., Love, J. D., & Shortland, F. B. (1977). Warmed-over flavour in meats, poultry and fish. Advances in Food Research, 23, 1– 74. Pryzylski, R., & Eskin, N. A. M. (1995). Methods to measure volatile compounds and the flavour significance of volatile compounds. In K. Warner, & N. A. M. Eskin (Eds.), Methods to assess quality and stability of oils and fat-containing foods (pp. 107–133). IL: AOCS Press. Rosenvold, K., Petersen, J. S., Lærke, H. N., Jensen, S. K., Therkildsen, M., Karlsson, A. H., Møller, H. S., & Andersen, H. J. (2001). Muscle glycogen and meat quality as affected by strategic finishing of slaughter pigs. Journal of Animal Science, 79, 382–391.
1138
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
Shahidi, F., Yun, J., Rubin, L. J., & Wood, D. F. (1987). The hexanal content as an indicator of oxidative stability and flavour acceptability in cooked ground pork. Canadian Institute of Food Science and Technology Journal, 20, 104–106. Siegmund, B., & Pfannhauser, W. (1999). Changes of the volatile fraction of cooked chicken meat during chill storing: results obtained by the electronic nose in comparison to GCMS and GC olfactometry. Zeitschrift fu¨r Lebensmittel-Untersuchung und Foschung A, 208, 336–341. St.Angelo, A. J., Vercellotti, J. R., Legendre, M. G., Vennett, C. H.,
Kuan, J. W., James Jr., C., & Dupuy, H. P. (1987). Chemical and instrumental analysis of warmed-over flavour in beef. Journal of Food Science, 52, 1163–1168. Tappel, A. L. (1962). Heme compounds and lipoxidase as biocatalysts. In Symposium on foods: lipids and their oxidation (p. 122). Westport, CT: AVI. Wen, J., Morrissey, P. A., Buckley, D. J., & Sheehy, P. J. A. (1996). Oxidative stability and alpha-tocopherol retention in turkey burgers during refrigerated and frozen storage as influenced by dietary alpha-tocopheryl acetate. British Poultry Science, 37, 787–795.
A comparison of warmed-over flavour in pork by sensory analysis, GC/MS and the electronic nose M.G. O’Sullivana,*, D.V. Byrnea, M.T. Jensena, H.J. Andersenb, J. Vestergaardc a
Department of Dairy and Food Science, The Royal Veterinary and Agricultural University, DK-1958, Frederiksberg C, Copenhagen, Denmark b Department of Animal Product Quality, Danish Institute of Agricultural Sciences, PO Box 50, DK-8830, Tjele, Denmark c YTI Research Centre, PO Box 181, FIN-50101, Mikkeli, Finland Received 24 June 2002; received in revised form 3 October 2002; accepted 4 December 2002
Abstract Pork muscle samples (M. longissimus dorsi and M. psoas major) were obtained from pigs given one of 4 dietary treatments, (i) control diet, (ii) supplemental iron (300 mg iron (II) sulphate/kg feed), (iii) supplemental vitamin E (200 mg dl-a-tocopheryl acetate/kg of feed) and (iv) supplemental vitamin E+supplemental iron. Warmed-over flavour (WOF) was evaluated by a trained sensory panel (n=8) for the four treatments cooked and refrigerated at 4 C for up to 5 days. Gas chromatography mass spectrometry (GC/MS) and Electronic nose analysis was performed on a subset of the full design which included samples of M. longissimus dorsi, treatments (ii) and (iii) and M. psoas major with treatment (i) for 0 days of WOF development. Day 5 of WOF development was included in the subset and represented by samples of M. longissimus dorsi, treatment (iv) and M. psoas major, treatments (ii) and (iii). Bi-linear modeling was used to determine the correlation of GC/MS and electronic nose data to sensory data. Also, the reproducibility and reliability of electronic nose data was evaluated by repeating the analysis of samples in a different laboratory and with a time difference of approximately 11 months. Mean-centring was used to normalise the data from these two different electronic noise data sets. GC/MS data correlated to sensory data with specific compounds (e.g., pentanal, 2-pentylfuran, octanal, nonanal, 1-octen-3-ol and hexanal), proving to be good indices of oxidation in cooked samples of M. longissimus dorsi and M. psoas major. Electronic nose data correlated to sensory data and separated the sensory variation. The reproducibility of this data was high with the second set of samples being predictive of the first set. # 2003 Elsevier Ltd. All rights reserved. Keywords: Pork; Sensory; Warmed-over flavour; Iron; Vitamin E; GC/MS; Electronic nose
1. Introduction Up to the present, the analysis of characteristic food odours has been commonly carried out by human assessment and headspace/direct gas chromatography mass spectrometry (GC/MS) (Grigioni, Margaria, Pensel, Sa´nchez, & Vaudagna, 2000). One compound, hexanal, has been identified and used to evaluate the oxidative state and correlated to sensory scores of meat from different animals (e.g. Shahidi, Yun, Rubin, & Wood, 1987, pork; Drumm & Spanier, 1991, beef, Byrne, Bredie, Mottram, & Martens, 2002, chicken). Hexanal is a secondary breakdown product formed * Corresponding author. Tel.: +45-352-83174; fax: +45-35283210. E-mail address: [email protected] (M.G. O’Sullivan). 0309-1740/03/$ - see front matter # 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0309-1740(02)00342-X
during the oxidation of linoleic acid (C18:2) (Frankel, Neff, & Selke, 1981). Other compounds have also been associated with oxidation, but specifically with warmedover flavour (WOF) development. St.Angelo et al. (1987) proposed that pentanal and 2,4-decadienal could be considered for use as marker compounds to follow the development of WOF and its associated rancid flavours in cooked meats. The usefulness of GC/MS is clear for the detection of WOF in cooked chill stored meat products, but as a technique it has certain drawbacks. Instrumental techniques, such as GC/MS, have high operating costs and are time consuming (Pryzylski & Eskin, 1995). However, the electronic nose may provide a practical advantage over other methods and may have an application in an On-Line/At-Line capacity for the quality determination of meat products with respect to WOF development.
1126
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
The electronic nose is usually an array of chemical gas sensors with a broad and partly overlapping selectivity for measurement of volatile compounds within the headspace over a sample combined with computerised multivariate statistical data processing tools (Gardner & Bartlett, 1994). In contrast to the well known analytical (GC/MS) and sensory techniques that have been used for the analysis of flavour compounds, the electronic nose does not give any information about the compounds causing the investigated aroma; neither about the identity of the compounds nor their sensorial properties. Using the electronic nose the aroma is judged by the so-called ‘‘aroma pattern’’, which should be characteristic to the investigated substrate (Siegmund & Pfannhauser, 1999). The sensor array of an electronic nose has a very large information potential and will give a unique overall pattern of the volatiles. In principle, both the electronic and the human nose operate by sensing simultaneously a high number of components giving rise to a specific response pattern (Haugen & Kvaal, 1998). The principal aims of this experiment were: (1) To correlate sensory and GC/MS analysis and to identify compounds that could be used as indices of lipid oxidation. (2) To assess the suitability of the electronic nose for the measurement of WOF development in different types of cooked pork meats. (3) To determine the reproducibility of electronic nose measurements of similar samples over time and space, i.e., two sets of similar samples measured using the same instrument, but in two separate laboratories and with a time separation of 11 months.
2. Methods and materials 2.1. Animals and diets Slaughter pigs were reared at The Danish Institute of Agricultural Sciences (DIAS), Foulum, Denmark. Pork muscle samples (M. longissimus dorsi and M. psoas major) were obtained from seven female DLY (Duroc/ Landrace/Yorkshire) cross bred pig from each of four dietary treatments. A control diet, a supplemental iron diet [300 mg iron (II) sulphate/kg feed], a supplemental vitamin E diet (200 mg dl-a-tocopheryl acetate/kg of feed) and a supplemental vitamin E (200 mg dl-a-tocopheryl acetate/ kg of feed)+supplemental iron diet [300 mg iron (II) sulphate/kg feed]. A subset of the full design listed earlier (Table 1) was used in all subsequent analysis and included samples of M. longissimus dorsi and the vitamin E and iron treatments, M. psoas major and the control group with day 0 of WOF development and samples of M. longissimus dorsi and the iron/vitamin E treatment, M. psoas major and vitamin E and iron treatments with day 5 of WOF development. This subset of samples was
selected in a preliminary evaluation using experts with product experience, and reflected the variation in muscle type, treatment and degree of WOF development. A previous sensory experiment (O’Sullivan, Byrne, Nielsen, Andersen, & Martens, 2002) using the same samples as the current study was conducted on the full set of samples. The pigs were penned individually and fed three times daily for the first month and twice thereafter. Water was provided ad libitum throughout the study. Iron supplementation was undertaken 1 month prior to slaughter and vitamin E supplementation 2 months prior to slaughter. All animals received vitamin C (9 g vitamin C/kg feed) in the ration 1 month prior to slaughter to facilitate iron uptake and were fed to all treatment groups to set an equal base level for all groups. 2.2. Sampling procedure and storage Slaughter was performed as described by Rosenvold et al. (2001). Following slaughter, carcasses were split centrally and chilled to 4 C for 24 h prior to muscle excision. The M. longissimus dorsi and M. psoas major muscles were removed from the right side of each animal. Muscles were then vacuum-packed and frozen at 20 C and stored in the dark for approximately 2 weeks. 2.3. Pattie preparation To facilitate ease of cutting and grinding the frozen muscles were placed in a refrigerator at 5 C for 12 h. Connective tissue and visible fat, both subcutaneous and inter muscular, were removed prior to cutting the muscles into cubes (approx. 2 cm3). The seven muscles from each individual treatment were then thoroughly mixed before grinding in a rotary screw mincer (model X 70, Scharfen GmbH & Co. Maschinenfabrik KG, Witten, Germany) through a 4.5-mm hole plate. The minced samples were again mixed before shaping in to 70 g patties (approx. 1 cm thick) using a commercial burger maker (i.d. 9 cm). Patties were then vacuumpacked in oxygen impermeable plastic laminate bags (polyamide/polyethylene 20/70) before being frozen at 20 C for up to 4 weeks (Byrne, Bak, Bredie, Bertelsen, & Martens, 1999; O’Sullivan, Byrne, Martens, & Martens, 2002). 2.4. Heat treatment and storage Patties were thawed in warm water to a temperature of approximately 20 C and removed from their vacuum bags. All samples were then in turn batch cooked in a preheated convection oven at 140 C for 30 min and turned every 5 min to facilitate even cooking. Internal cooking temperatures reached were found to be
1127
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138 Table 1 Sample codes, muscles, supplemental diets and days of warmed-over flavour (WOF) Sample code
Muscle
LD-E-0-1, -2 LD-I-0-1, -2 LD-IE-5-1, -2 PS-E-5-1, -2 PS-I-5-1, -2 PS-C-0-1, -2
M. longissimus dorsi M. longissimus dorsi M. longissimus dorsi M. psoas major M. psoas major M. psoas major
Vitamin Ea
Ironb
Vitamin Cc
@ x @ @ x x
x @ @ x @ x
@ @ @ @ @ @
Days WOF 0 0 5 5 5 0
-1=Sample set 1, analysed at The Royal Veterinary and Agricultural University, Copenhagen, Denmark. -2=Sample set 2, analysed at The YTI Research Centre, PO Box 181, FIN-50101, Mikkeli, Finland after 11 months frozen storage at 20 C. @=Present, x=Absent. a 200 mg dl-a-tocopheryl acetate/kg of feed. b 3000 mg iron (II) sulphate/kg feed. c 9 g/kg feed.
approximately 80 C. All samples were cooked in the same oven to reduce cooking variation. After cooking, samples were allowed to cool to 5 C in a refrigerator before being placed in oxygen permeable bags (low density polyethylene 1502500.025 mm) and stored at 4 C in the dark for up to 5 days to allow the development of WOF (Byrne et al., 1999; O’Sullivan, Byrne, Martens et al., 2002). 2.5. Re-heating treatment Patties were divided into eight equal triangular pieces and individually vacuum-packed in oxygen permeable plastic laminate bags (polyamide/polyethylene 20/70) before being placed in a water filled steel tray (at ambient temperature) and re-heated in a convection oven at 140 C for 19 min. Mean serving temperature of samples was found to be 65 C (Byrne et al., 1999; Byrne, O’Sullivan, Dijksterhuis, Bredie, & Martens, 2001). 2.6. Training and profiling An eight member sensory panel (four males/four females, aged from 24 to 62 years) was recruited from the public and students of The Royal Veterinary and Agricultural University, Frederiksberg, Denmark. Selection criteria for panellists were availability and motivation to participate on all days of the experiment. Sensory training was carried out prior to sensory profiling (Byrne et al., 1999; O’Sullivan, Byrne, Martens et al., 2002). Sensory analysis was carried out in the panel booths at the university sensory laboratory that conforms to ISO (1988) international standard. During both training and sensory profiling, analysis was performed with assessors not having any knowledge of sample history so as not to introduce bias into their sensory assessment. Unstructured 15-cm line scales anchored on the left by the term ‘none’ and on the right by the term ‘extreme’ were used for all the sensory descriptors (Meilgaard, Civille, & Carr, 1999). The responses of the panellists were recorded by measuring
the distance in mm (1–150) from the left side of the scale for the odour, flavour, taste and after taste sensory terms as described in Table 2. 2.6.1. GC/MS sample pre-treatment and extraction The meat patties were received vacuum packed in oxygen-impermeable plastic bags frozen at 20 C. To ensure a constant temperature of the patties before cooking, the plastic bags were placed in a water bath and held at 20 C until they were judged to be isothermal to the surrounding water. The thawed patties were cooked in a preheated convection oven identically to the sensory samples. Of the cooked patties 3 g was transferred to flame-heated (and subsequently cooled) 10-ml vials and equilibrated in 5 min at 65 C (the serving temperature in the sensory analysis) in the heating box of the autosampler (CTC analytics, Zwingen, Switzerland). To induce warmed over flavour, the cooked patties were packed in oxygen permeable polyethylene bags and stored at 5 C for 5 days. An equilibration time of 5 min at 65 C was selected from a pilot experiment, which showed that the aldehyde production was complete after 10 min, an equilibration time longer than 5 min would therefore mask the fine structure of the volatiles produced. The volatiles were sampled with an SPME-fibre (75 mm carboxen/polydimethylsiloxane) in 1 min. The fibre was thereafter transferred to the injection port of the gas chromatograph and desorbed at 250 C. To ensure the samples were not oxidised during storage before a run on GC/MS, only one sample was analysed at a time. 2.6.2. GC/MS analysis The gas chromatograph was a Varian Star 3400 CX coupled with a Varian Saturn 2000 mass spectrometer. The column was a DB-1701, length of 30 m, 0.25 mm, (i.d. 1 mm) operated with a gas flow of 1.6 ml min 1 at 35 C. The temperature gradient started at 35 C and was held at that temperature for 10 min until a gradient with an end temperature of 150 C with an increase of 3 C/min was initiated. The temperature of 150 C was
1128
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
Table 2 Sensory descriptive terms and references developed for sensory profiling Sensory term
Definition (with appropriate reference)
Odour 1. Cardboard-O 2. Linseed oil-O 3. Rubber/Sulphur-O 4. Nut-O 5. Green-O 6. Fatty-O
Odour reference Wet cardboard Warmed linseed oil/linseed oil based paint Warmed rubber/the white of a boiled egg Crushed fresh hazel nuts Fresh green French beans Pig back fat (fresh, non-oxidised)
Taste 7. Sweet-T 8. Salt-T 9. Sour-T 10. Bitter-T 11. MSG/Umami-T
Taste reference Sucrose 1 g/l aqueous soln. Sodium chloride 0.5 g/l aqueous soln. Citric acid monohydrate 0.3 g/l aqueous soln. Quinine chloride 0.05 g/l aqueous soln. Monosodium glutamate 0.5 g/l aqueous soln.
Flavour 12. Metallic-F/ Bloody-F 13. Fresh cooked pork-F 14. Rancid-F 15. Lactic acid/fresh sour-F 16. Vegetable oil-F 17. Piggy/Animal-F 18. Fish-F 19. Tinny-F 20. Livery-F
Flavour reference Ferrous sulphate 0.1 g/l aqueous soln. Oven cooked pork without browning Oxidised vegetable oil Natural yoghurt Fresh vegetable oil Skatole 0.06 mg/ml refined vegetable oil Fish stock in boiling water Stainless steel strip Cooked beef liver
Aftertaste 21. Astringent-AT
Aftertaste reference Aluminium sulphate 0.02 g/l aqueous soln.
Suffix to sensory terms indicates method of assessment by panelists; -O=Odour, -F=Flavour, -T=Taste, -AT=Aftertaste.
held for 5 min, then a column cleaning gradient was initiated with an increase of 30 C/min to 250 C and held at 250 C for 5 min. The mass spectrometer had a mass range of 35–400 m/z with a scan time of 1 s in electron impact mode with an electron energy of 70 eV and an electron multiplier voltage of 1650 V. The chromatograms were integrated and the retention times and peak areas, transferred to an Excel spreadsheet. Peaks with retention time of more than 40 min were deleted as they were high boiling compounds with no aroma. The data were then aligned in an aligning program (Borggaard, 2000) with sample no. P-I-5 as reference chromatogram and the lowest area limit was set to 5000. 2.7. Electronic nose Electronic nose analysis was conducted initially using the samples outlined in Table 1 (sample set 1) and in a laboratory at The Royal Veterinary and Agricultural University, Copenhagen, Denmark. Eleven months later the experiment was repeated using similar samples to those presented in Table 1 (sample set 2) at The Department of Sensory Science, Institute of Environmental Technology, Mikkeli, Finland. Samples were held at 20 C during the interim period. The MGD-1
(Environics Ltd, Finland) is based on the principle of ion mobility. The ion mobility cell IMCELL1 contains six channels (IMC-1 to IMC-6). IMC-1, IMC-2 and IMC-3 register signals from positive ions. IMC-4, IMC-5 and IMC-6 register signals from negative ions. The MGD-1 is also equipped with a semiconductor cell (SCCELL) as a ‘‘seventh’’ sensor. One gram of sample was extracted in 5 ml isooctane. The sample was diluted in isooctane (1:8) and 5 ml injected into a desorber, which is a special device containing a ceramic plate developed for the MGD-1 and used for laboratory experiments. The ceramic plate was heated to 300 C and MGD-1 signals were obtained for 90 s (IMCELL temperature 75 C). Measurements were carried out with two replicates. Electronic nose profile data were collected by summing signals for each of the channels (0–60 s). 2.8. Data analysis ANOVA–Partial Least Squares Regression (APLSR) was used to investigate product sensory variation in the profiling data and was performed using full-cross validation. The X-matrix was designated as 0/1 design variables for assessor, replicate and product. The Ymatrix was designated as sensory and instrumental
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
variables. In this model assessor and replicate level effects were removed (Martens & Martens, 2000). In order to compare the electronic nose data sets (1 and 2) a mean-centring (Martens & Martens, 2001; chapter 5) of these instrumental measurements was undertaken, which determined the patterns of variation between samples and in effect normalised these data. All analyses were performed using the Unscrambler Software, version 7.6 (CAMO ASA, Trondheim, Norway).
3. Results and discussion 3.1. Sensory sample variation ANOVA–APLSR was used to process the mean data accumulated from eight test subjects during sensory profiling. The X-matrix was designated as 0/1 design variables for meat samples and days of WOF development. The Y-matrix was designated as sensory variables (Table 2). The optimal number of components in the APLSR model presented was determined to be 3 Principal Components. Thus PC 1 versus 2 (Fig. 1) and PC 2 versus 3 were explored. PC 2 versus 3 is not presented here, as additional information was not gained through their examination. Further PCs did not provide any predictive improvement in the Y-matrix obtained. In this model assessor and session level effects were removed. The calibrated explained variance for this model was PC1=22% and PC2=13%. Fig. 1 is presented as a correlation loadings plot. In this fig. sample LD-E-0, a cooked sample that has been stored in a
1129
refrigerator at 4 C for 0 days, covaried with Fresh Cooked Pork-F and Lactic acid/fresh sour-F. Sample LD-I-0 also covaried with both these descriptors, but to a lesser extent. These samples reflect 0 days of WOF development. Sample PS-C-0 covaried with Nut-O, Fatty-O and to a lesser degree to the oxidative terms Rubber/Sulphur-O and Cardboard-O. The M. psoas major is more prone to oxidative attack compared with M. longissimus dorsi. This is in confirmation with the results of O’Sullivan, Byrne, and Nielsen et al. (2002) who performed a WOF experiment with a full design (2 muscles, M. longissimus dorsi and M. psoas major 4 dietary treatments, (i), (ii), (iii) and (iv) 4 days of WOF development, 0, 1, 3 and 5) of which the results presented here are a subset. These authors observed a greater degree of WOF development for M. psoas major compared with M. longissimus dorsi and postulated that this may reflect the greater iron levels in M. psoas major, thus predisposing it to a greater oxidative effect expressed as degree of WOF development. Samples LD-IE-5 and PS-E-5, cooked samples that have been stored in a refrigerator at 4 C for 5 days, covaried with Bitter-T and to a lesser extent to Sour-T. However, PS-I-5 covaried with the descriptors Astringent-AT, MSG-T and to the oxidative descriptors Rancid-F, Fish-F and Linseed oil-O. These results are again in confirmation with the more complete experiment conducted by O’Sullivan, Byrne, and Nielsen et al. (2002) who showed that iron supplementation increased the susceptibility to warmedover flavour development for both M. longissimus dorsi and M. psoas major compared with the other dietary treatments. The ability of heme pigments and non-heme
Fig. 1. An overview of the variation found in the mean data from the ANOVA–Partial Least Squares Regression (APLSR) correlation loadings plot for sample set 1. Shown are the loadings of the X- and Y-variables for the first 2 PCs. =sensory descriptor and ~=sample and days of WOF. The concentric circles represent 100 and 50% explained variance, respectively.
1130
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
iron to accelerate the propagation step of the free-radical chain mechanism can explain the rapid rate of oxidation in cooked meats (Pearson, Love, & Shortland, 1977). The samples selected for this experiment thus reflect varying degrees of WOF development and Fig. 1 displays this dimension along PC 1. Samples stored for 5 days under refrigerated conditions displayed the greatest levels of WOF. The M. psoas major muscle produced greater WOF compared to M. longissimus dorsi and iron supplementation of animal diets produced greater WOF than any of the other dietary treatments. This variation in WOF between the different muscle samples thus provided an interesting model for subsequent GC/MS and electronic nose analysis. 3.2. Sensory and GC/MS variation ANOVA–APLSR was also used to process the mean data accumulated from sensory evaluation by the test subjects and GC/MS analysis. Table 3 corresponds to compounds identified by GC/MS analysis of meat samples. The X-matrix was designated as 0/1 design variables for meat samples and sensory variables (Table 2). The Y-matrix was designated as GC/MS measurements. Fig. 2 corresponds to M. longissimus dorsi and Fig. 3 to M. psoas major. The optimal number of components in these APLSR models was determined to be 3 Principal Components. Thus PC 1 versus 2 (Figs. 2 and 3) and PC
2 versus 3 were explored. PC 2 versus 3 is not presented here, as additional information was not gained through their examination. The validated explained variance for Fig. 2 is PC 1=33% and PC 2=29% and for Fig. 3, validated explained variance PC 1=60% and PC 2=17%. M. longissimus dorsi (Fig. 2) appears to be separated along PC 1 with samples containing vitamin E on the left side of the plot and samples high in iron content along the right side of the correlation loadings plot. Sample LD-E-0-1 correlated to Fresh Cooked Pork-F and Sweet-T, sensory descriptors associated with 0 days of WOF development. Sample LD-IE-5-1 covaried in the upper portion of the correlation loadings plot and LD-I-0-1 in the lower right hand quadrant. Both these samples appear more oxidised due to their greater association with descriptor Rancid-F. Sample LD-I-0-1 also covaried with Green-O and the GC/MS identified compounds pentanal, 2-pentylfuran, octanal, nonanal, 1-octen-3-ol and hexanal. Therefore, this sample which was frozen after 0 days of WOF development underwent oxidation during frozen storage. Similarly in Fig. 3 sample PS-I-5-1 covaried with the sensory descriptor Green-O and the GC/MS identified compounds pentanal, 2-pentylfuran, octanal, nonanal, 1-octen-3-ol, and hexanal. These compounds (Figs. 2 and 3) have been found to be indicative of lipid oxidation (Byrne et al., 2002; Siegmund & Pfannhauser, 1999). Sample PS-C-0-1 covaried with the oxidative
Fig. 2. An overview of the variation found in the mean data from the ANOVA–Partial Least Squares Regression (APLSR) correlation loadings plot for sample set 1, M. longissimus dorsi samples. Shown are the loadings of the X- and Y-variables for PC 1 versus 2. ~=sensory descriptor and sample and =GC/MS isolated compounds. The concentric circles represent 100 and 50% explained variance, respectively.
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
1131
Fig. 3. An overview of the variation found in the mean data from the ANOVA–Partial Least Squares Regression (APLSR) correlation loadings plot for sample set 1, M. psoas major samples. Shown are the loadings of the X- and Y-variables for PC 1 versus 2. ~=sensory descriptor and sample and =GC/MS isolated compounds. The concentric circles represent 100 and 50% explained variance, respectively.
descriptor Cardboard-O indicating that this sample also underwent a degree of oxidation during frozen storage. The oxidative descriptors Rancid-F and Linseed oil-O were positioned midway between samples PS-I-5-1 and PS-E-5-1 indicating that they were common to both samples and thus had undergone WOF development. Another APLSR plot (Fig. 4) was constructed and included both M. longissimus dorsi and M. psoas major samples in order to determine the relative relationship of these white (M. longissimus dorsi) and red muscle samples (M. psoas major) with respect to WOF development and some of the chemical compounds especially associated with oxidation. Again, Table 3 corresponds to compounds identified by GC/MS analysis of meat samples. The X-matrix was designated as 0/1 design variables for meat samples and days of WOF development and sensory variables (Table 2). The Y-matrix was designated as GC/MS measurements The optimal number of components in the APLSR model presented was determined to be 3 Principal Components. Thus PC 1 versus 2 and PC 2 versus 3 (Fig. 4) were explored. PC 1 versus 2 is not presented here many of the compounds that relate to oxidation were scattered around the origin and thus showed that they were related to many of the samples analysed. However, PC 2 versus 3 (Fig. 4) separated these compounds and allowed the comparison of samples, sensory terms and GC/MS isolated compounds. The validated explained variance for this model
was PC 2=17% and PC 3=9%. In Fig. 4 M. longissimus dorsi appears to be separated along PC 3 and M. psoas major along PC 2. Samples containing vitamin E appear on the left side of the plot and samples high in iron content along the right side of the correlation loadings plot. Sample PS-C-0 had almost twice the level of iron compared with the highest levels found in M. longissimus dorsi (O’Sullivan, Byrne, & Nielsen et al., 2002). Again the sample PS-I-5 appears in the upper right hand quadrant of the PLSR plot and is correlated to Rancid-F and Fish-F. In Fig. 4 the compound hexanal covaried with the oxidised sample PS-I-5 (also Fig. 3) and LD-I-0-1 (also Fig. 2). Hexanal is a secondary breakdown product formed during the oxidation of linoleic acid (C18:2) (Frankel et al., 1981). Hexanal, has been identified and used to evaluate the oxidative state and correlated to sensory scores of red meat from different animals and the results presented agree with Shahidi et al. (1987), who worked on cooked pork and Drumm and Spanier, (1991), who worked with beef. Also, hexanal has a distinctive odour described as ‘green’ or ‘grassy’ (Gasser & Grosch, 1988). The sensory descriptor Green-O covaried with hexanal for both M. longissimus dorsi and M. psoas major (Figs. 2–4). The reference material for this descriptor during sensory training was fresh French beans. During the preliminary selection of reference materials it was decided by an expert panel that French beans produced an odour
1132
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
Table 3 Gas chromatography/mass spectrometry (GC/MS) isolated compounds Compound
Kovats value
Identification/ionsa
Unknown Pentane Pentanol Dimethyl sulfide Trimethylamine Methylbranched alkane 2-Methylpropanal 2-Methylfuran Butanal 2-Butanone 3-Methylhexane Benzene 3-Methylbutanal 2-Ethylfurane Pentanal Aliphatic alcohol Aliphatic alcohol Unknown An octadiene, e.g. 1,3-octadiene Hexanal Nonane Ethylbenzene A dimethylbenzene 2-Butylfuran Butanoic acid 2-Heptanone Heptanal 4-Ethyl-1-octyn-3-ol 2-Pentylfuran A methylbranched alkane 1-Octen-3-ol Benzaldehyde Octanal trans-2-Octenal Nonanal
NA 506 519 573 586 627 703 706 715 721 725 731 747 750 782 813 824 852 855 890 905 915 925 938 974 986 991 1032 1036 1054 1084 1089 1094 1179 1198
43, 45, 47 MS of AS MS of AS MS of AS MS of AS 41, 56, 85 MS MS MS MS of AS MS MS MS of AS MS MS of AS 41, 55, 70 41, 55, 70 42, 43, 45 67, 81, 110 MS of AS MS of AS MS MS MS MS of AS MS MS of AS MS MS of AS 41, 57, 71 MS of AS MS of AS MS of AS MS of AS MS of AS
a MS: Good fit with mass spectra in NIST library. MS of AS: fit with mass spectra of authentic standard. Ions: where the identification was ambiguous, the three most prominent ions in the mass spectra are given. NA: not applicable.
described as being green or grassy. Thus, Green-O was used in an attempt to characterise hexanal, isolated through subsequent GC/MS analysis and was found to covary with this compound. Other compounds have also been identified with oxidation, but specifically with warmed-over flavour development. In Fig. 4 the compounds 1-octen-3-ol, 2-pentylfuran covaried with the sensory oxidative descriptors Rancid-F, Fish-F and Rubber/Sulphur-Like-O. These results agree with those of Siegmund and Pfannhauser (1999) who found that the relative concentrations of the lipid oxidation products hexanal, 1-octen-3-ol and 2-pentylfuran increased in cooked chill stored chicken meat as storage time increased. Pentanal was also found to covary to a greater extent with the sensory sample PS-I-5, the sample with the greatest degree of WOF development (Figs. 3 and 4). Sample LD-I-0-1 (Fig. 2) also covaried
with pentanal again indicating that the pro-oxidative nature of iron during frozen storage. The result of Fig. 2, 3 and 4 are in agreement with St.Angelo et al. (1987) who proposed that pentanal and 2,4-decadienal could be considered for use as marker compounds to follow the development of WOF and its associated rancid flavours in cooked meats. In summary, many of the compounds associated with oxidation of lipids were found to correlate with the oxidative sensory descriptors and the samples with the greater levels of WOF development, i.e. pentanal, 2-pentylfuran, octanal, nonanal, 1octen-3-ol and hexanal. 3.3. Electronic nose analysis, mean-centring and sample reproducibility ANOVA–APLSR was used to compare the electronic nose sensor readings from sample sets 1 and 2 combined. The X-matrix was designated as 0/1 design variables for muscle, treatment and days of WOF development (main effects). The Y-matrix was designated as electronic nose measurements. The optimal number of components in the APLSR model presented was determined to be 3 PCs. Fig. 5 shows a correlation loadings plot (validated explained variances PC1=12% and PC2=9%) of the main design effects of sample sets 1 and 2. M. longissimus dorsi samples were inversely correlated with those of M. psoas major. Each dietary treatment is separated and occupies a quadrant of the correlation loadings plot with the iron/vitamin E and vitamin E treatments separated along PC 1 and iron and iron/vitamin E treatments separated along PC 2. The Control group occupies the bottom right quadrant and day 0 and day 5, days of WOF development, are inversely correlated. In effect the electronic nose device used in this experiment could qualitatively separate samples on the basis of muscle type, treatment and degree of WOF development. ANOVA–APLSR was again used to compare the electronic nose sensor readings from sample sets 1 and 2, analysed with the same instrument, but at different locations and with a time separation of 11 months. The X-matrix was designated as 0/1 design variables for meat sample type and days of WOF development. The Y-matrix was designated as electronic nose measurements. The optimal number of components in the APLSR model presented was determined to be 3 Principal Components. Thus PC 1 versus 2 (Fig. 6a), PC 2 versus 3 (Fig. 6b) and PC 1 versus 3 (Fig. 6c) were explored and all are presented. Further PCs did not provide any predictive improvement in the Y-matrix obtained. In this model session level effects were removed and the data were normalised through meancentring. The calibrated explained variances for these models were PC1=35%, PC2=9% and PC3=3%, respectively.
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
1133
Fig. 4. An overview of the variation found in the mean data from the ANOVA–Partial Least Squares Regression (APLSR) correlation loadings plot for sample set 1. Shown are the loadings of the X- and Y-variables for PC 2 versus 3. =sensory descriptor and GC/MS isolated compounds and ~=sample and days of WOF. The concentric circles represent 100 and 50% explained variance, respectively.
Fig. 5. An overview of the variation found in the mean data from the ANOVA–Partial Least Squares Regression (APLSR) correlation loadings plot for sample set 1 and sample set 2 combined. Shown are the loadings of the X- and Y-variables for the first 2 PCs. =electronic nose sensor and ~=design variables (main effects). The concentric circles represent 100 and 50% explained variance, respectively.
1134
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
Fig. 6. An overview of the variation found in the mean data from the ANOVA–Partial Least Squares Regression (APLSR) correlation loadings plot for sample set 1 and 2. Shown are the loadings of the X- and Y-variables for (a) PC 1 versus 2, (b) PC 2 versus 3, (c) PC 1 versus 3. =Electronic nose sensor and ~=sample and days of WOF. The concentric circles represent 100 and 50% explained variance, respectively.
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
1135
Fig. 6. (continued)
Fig. 6a reveals samples LD-I-0-1, 2 and PS-C-0-1, 2 covaried with day 0, which corresponds to 0 days of warmed-over flavour development and samples LD-IE5-1, 2 covaried with day 5 which corresponds to 5 days of warmed-over flavour development. PS-C-0-1, 2 also covaried with day 0, but to a lesser extent. The remaining samples appear clustered around the origin, but PSI-5-1 covaried to a greater degree to day 5 of WOF development. In Fig. 6b the various samples are separated by the order of oxidation LD-E-0 < LD-I-0 < PSC-0 < LD-IE-5< PS-E-5 < PS-I-5. Similar results are also found in Fig. 6c, with samples being separated according to their respective degrees of WOF development. In summary the electron nose data from sample sets 1 and 2 agree with the sensory data in Fig. 1 and the device was effective in determination of the oxidative state of the samples analysed. In Fig. 6a samples (Table 1) LD-E-0-1 and LD-E-0-2 covaried and similarly, LD-I-0-1 and LD-I-0-2, PS-C-01 and PS-C-0-2, PS-I-5-1 and PS-I-5-2 covaried. However, samples LD-IE-5-1 and LD-IE-5-2, PS-E-5-1 and PS-E-5-2 displayed greater levels of variation. Similar results are seen in Figs. 6b and c except for samples PSE-5-1 and PS-E-5-2 in Fig. 6b which varied to a greater extent. One of the principle drawbacks to date of electronic noses is the large variation in data acquisition. In order to compare results over time (weeks, months or
years) it is required that these instruments give the same signal when identical samples are being measured over time. However, due to dynamic processes taking place in the sensors over time, the signal from a sensor array may vary significantly (Haugen & Kvaal, 1998). Gardner and Bartlett (1996) proposed odour standards and analytical expressions to define the working range and resolving power of electronic noses. A first set of standards, defined as ‘‘Type A’’, used organoleptic chemicals to determines the range of an electronic nose and a further set of standards (Type B), of similar samples were used to determine the resolving power of an electronic nose. In the present study electronic nose analysis could be reproduced using similar samples measured using the same instrument, but in two separate laboratories and with a time separation of 11 months. The greatest variation amongst samples in Fig. 6a–c appears to occur with the samples PS-E-5 and LD-IE-5, both of which display high levels of WOF development. This variation may be due to oxidative changes which occurred during the interim 11 months of frozen storage at 20 C prior to the second electronic nose analysis. Both samples were frozen when WOF development was quite advanced and also both samples contain endogenous vitamin E. Wen, Morrissey, Buckley, and Sheehy (1996) found that vitamin E level in cooked frozen ( 20 C) turkey burgers decreased from 5.6 to 2.88 and from 3.29 to 1.85
1136
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
mg/g in burgers made from turkeys supplemented with 600 and 300 mg a-tocopheryl acetate/kg feed respectively. Thus, a reduction in vitamin E levels may have occurred during frozen storage resulting in changes in the oxidation potential of these samples and expressed in Figs. 6a–c by the reduced degree of covariance of PSE-5 and LD-IE-5 as measured by the electronic nose. Moreover, lipids of pork meat can suffer alterations during frozen storage, owing to lipolysis and oxidation (Hernandez, Navarro, & Toldra, 1999). AbdelKader (1996) found that malondialdehyde, a chemical marker for lipid oxidation, increased in cooked chicken breast frozen stored for 3 and 6 months. Sample PS-I-5-1 and PS-I-5-2 covaried highly in Fig. 6a–c and displayed a high level of reproducibility as measured by the electronic nose. The apparent lack of variation in these samples as compared to the other meat samples with 5 days of WOF may be explained by the fact that these samples displayed the greatest degree of WOF (Fig. 1) and also iron supplementation led to a much more accelerated degree of WOF development. Thus, the oxidation reaction of these samples may have been more advanced and samples were more stable during subsequent frozen storage. The free-radical chain mechanism involved in the oxidation of unsaturated fatty acids is thought to occur in three stages: (1) initiation, the formation of free-radicals; (2) propagation, the free-radical chain reactions; (3) termination, the formation of non-radical products (Tappel, 1962). This lipid oxidation mechanism proposed by Tappel (1962) depends on the presence of pre-formed fatty acid hydoperoxides, which react with heme compounds and undergo homolytic decomposition. Thus, the high covariance of the electronic nose data for samples LD-E-0-1 and LD-E-0-2, LD-I-01 and LD-I-0-2 and PS-C-0-1 and PS-C-0-2, was due to their stability as oxidation had not fully commenced. The high covariance of samples PS-I-5-1 and PS-I-5-2 covaried because oxidation was most likely in the termination stage. However, samples LD-IE-5-1 and LD-IE5-2, PS-E-5-1 and PS-E-5-2 produced more variable results because these samples were most probably in the propagation stage of oxidation which continued, albeit at a lower rate, during frozen storage. In the present experimental samples were stored at 20 C. However, if samples were stored at 80 C, the variation between LD-IE-5-1, LD-IE-5-2 and PS-E-5-1, PS-E-5-2 would most probably be reduced because of the lower reaction rate at this temperature. The electronic nose has both large differences in sensitivity and selectivity from the human nose (Haugen & Kvaal, 1998). To date electronic noses have been employed in the analysis of a large variety of meat products (e.g. Eklo¨v, Johansson, Winquist, & Lundstro¨m, 1998, fermented sausage; O´lafsdo´ttir, Martindo´ttir, & Jo´nsson, 1997, fish) and in the warmed-over flavour analysis of various meat products (e.g. Siegmund &
Pfannhauser, 1999, chicken; Grigioni et al., 2000, beef). These instruments all differed in the individual sensor array set-up, numbers of sensors and the data analysis used to process the characteristic signature data of the various meat types. These data analytical techniques employed included, Neural Networks and Principal Component Regression (PCR). Balaban, Korel, Odabasi, and Folkes (2000) investigated the conversion of electronic nose datasets for the purpose of accumulating seamless data sets. They found that matrix conversion was most satisfactory and neural network transformations using multi-layer perceptrons and trained with back propagation did not perform as well. None of the afore-mentioned authors attempted to reproduce their results in time and space (i.e. at a different time and location). In the present study, differences in the electronic nose measurements between sample sets 1 and 2 were mean-centred using Unscrambler Software, version 7.6 (CAMO ASA, Trondheim, Norway). In effect this normalised the data for subsequent analysis using PLS regression which allowed the direct comparison of data from sample sets 1 and 2. Another potential drawback of the electronic nose is that sensors have a limited life, they must be replaced after some time and new sensors from another batch will differ in performance (Haugen & Kvaal, 1998). One potential method of solving these fundamental problems is to use a reliable data analytical tool to correct for variations over time, as discussed earlier and secondly to use an electronic nose in which the sensors do not require replacement. The sensor technology employed in the electronic nose used in this experiment is based on the principle of ion mobility and ionisation of gas molecules (MGD-1, Environics Ltd, Finland). The clusters formed through ion-molecule reactions are brought into different electrical fields perpendicular to the sample flow and detected. In effect the sensors do not wear out as volatiles do not come in contact directly with the sensors and this is advantageous over other existing electronic nose technology where sensors degrade due to direct contact with the substances being measured.
4. Conclusions The samples selected for this experiment reflect varying degrees of WOF development. Cooked samples stored for 5 days under refrigerated conditions displayed the greatest levels of WOF. The M. psoas major muscle produced greater WOF compared with M. longissimus dorsi and iron supplementation of animal diets resulted in greater WOF than any of the other dietary treatments. This variation in WOF between samples thus provided a good model for subsequent GC/MS and electronic nose analysis.
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
Many of the compounds associated with oxidation of lipids were found to correlate with the oxidative sensory descriptors and the samples with the greater levels of WOF development, i.e., pentanal, 2-pentylfuran, octanal, nonanal, 1-octen-3-ol and hexanal. The electronic nose device coupled with the multivariate methodology used in this experiment could clearly separate samples on the basis of muscle type, treatment and degree of WOF development. The electron nose data from sample sets 1 and 2 agreed with and correlated to sensory analysis. The electronic nose was effective in the determination of the oxidative state of the experimental samples. Electronic nose analysis could be reproduced using similar samples measured using the same instrument, but in two separate laboratories and with a time separation of 11 months. Oxidative changes during frozen storage may have resulted in the reduced reproducibility of electronic nose data for samples PS-E-5 and LD-IE-5. Mean-centring was effective in normalising the data from sample sets 1 and 2 prior to subsequent PLSR. The sensor technology employed in the electronic nose used in this experiment, had an advantage over existing electronic nose devices in that the sensors do not wear because the molecules measured do not come in direct contact with the sensors.
Acknowledgements This project was part of a project titled ‘‘Improved Iron Status through Utilization of Meat Proteins for an Improved Food Iron Availability and through Safe Iron Fortification’’. The project was funded by FØTEK, Norma og Frode S. Jacobsen Fund and Danske Slagterier, and is led by Professor Leif Skibsted. The Frame Program AQM (Advanced Quality Monitoring) in the Food Production Chain is also acknowledged for financial and academic contributions. Also acknowledged is Dorte Blaaberg Jensen for GC/ MS sample preparation.
References AbdelKader, Z. M. (1996). Lipid oxidation in chicken as affected by cooking and frozen storage. Nahrung Food, 40, 21–24. Balaban, M. O., Korel, F., Odabasi, A. Z., & Folkes, G. (2000). Transportability of data between electronic noses: mathematical methods. Sensors and Actuators B, 60–77. Borggaard, C. (2000). Danish Meat Research Institute, Maglega˚rdsvej 2, DK-4000 Roskilde, Denmark. Byrne, D. V., & Bak, L. S., Bredie, W. L. P., Bertelsen, G., & Martens, M. (1999). Development of a sensory vocabulary for warmed-over flavour 1: in porcine meat. Journal of Sensory Studies, 14, 47–65. Byrne, D. V., O’Sullivan, M. G., Dijksterhuis, G. B., Bredie, W. L.P, & Martens, M. (2001). Sensory panel consistency during development
1137
of a vocabulary for warmed-over flavour. Food Quality and Preference, 12, 171–187. Byrne, D. V., Bredie, W. L. P., Mottram, D. S., & Martens, M. (2002). Sensory and chemical investigations on the effect of oven cooking on warmed-over flavour in chicken meat. Meat Science, 61, 127– 139. Drumm, T. D., & Spanier, A. M. (1991). Changes in the lipid content of autoxidation and sulphur-containing compounds in cooked beef during storage. Journal of Agriculture and Food Chemistry, 49, 336– 343. Eklo¨v, T., Johansson, G., Winquist, F., & Lundstro¨m, I. (1998). Monitoring sausage fermentation using an electronic nose. Journal of Science and Food Agriculture, 76, 525–532. Frankel, N. E., Neff, W. E., & Selke, E. (1981). Analysis of autoxidised fats by gas chromatography-mass spectrometry: VII. Volatile thermal decomposition products of pure hydroperoxides from autoxidised and photosensitised oxidised methyl oleate, linoleate and linolenate. Lipids, 16, 279–285. Gardner, J. W., & Bartlett, P. N. (1994). A brief history of electronic noses. Sensors and Actuators B, 46-, 47, 211–220. Gardner, J. W., & Bartlett, P. N. (1996). Performance definition and standardization of electronic noses. Sensors and Actuators B, 60–77. Gasser, U., & Grosch, W. (1988). Flavour deterioration of soya bean oil: identification of intense odour compounds during flavour reversion. Fett Wissenschart Technologie, 90, 332–336. Grigioni, G. M., Margaria, C. A., Pensel, N. A., Sa´nchez, G., & Vaudagna, S. R. (2000). Warmed-over flavour in low temperaturelong time processed meat by an ‘‘electronic nose’’. Meat Science, 56, 221–228. Haugen, J. E., & Kvaal, K. (1998). Electronic nose and artificial neural network. Meat Science, 49, S273–S286.. Hernandez, P., Navarro, J., & Toldra, F. (1999). Effect of frozen storage on lipids and lipolytic activities in the longissimus dorsi muscle of the pig. Zeitschrift fu¨r Lebensmittel-Untersuchung und Foschung A, 208, 110–115. ISO 8589. (1988). Sensory analysis. General guidance for the design of test rooms. Paris: International Organisation for Standardisation. Martens, H., & Martens, M. (2000). Modified Jack-knife estimation of parameter uncertainty in bilinear modelling by partial least squares regression (PLSR). Food Quality and Preference, 11, 5–16. Martens, H., & Martens, M. (2001). In Multivariate analysis of quality. An introduction (pp. 99–101). Chichester: John Wiley and Sons. Meilgaard, M. C., Civille, G. V., & Carr, B. T. (1999). Sensory evaluation techniques (3rd ed.). Boca Raton, FL: Academic Press. O´lafsdo´ttir, G., Martindo´ttir, E., & Jo´nsson, E. H. (1997). Rapid gas sensor measurements to determine spoilage of capelin (Mallotus villosus). Journal of Agriculture and Food Chemistry, 45, 2654– 2659. O’Sullivan, M. G., Byrne, D. V., Martens, H., & Martens, M. (2002). Sensory and data analytical methodologies in the development of a vocabulary for evaluation of meat quality. Journal of Sensory Studies, 64, 175–189. O’Sullivan, M. G., Byrne, D. V., Nielsen, J. H., Andersen, H. J., & Martens, M. (2002). Sensory and chemical assessment of pork supplemented with iron and vitamin E. Meat Science, 64, 175–189. Pearson, A. M., Love, J. D., & Shortland, F. B. (1977). Warmed-over flavour in meats, poultry and fish. Advances in Food Research, 23, 1– 74. Pryzylski, R., & Eskin, N. A. M. (1995). Methods to measure volatile compounds and the flavour significance of volatile compounds. In K. Warner, & N. A. M. Eskin (Eds.), Methods to assess quality and stability of oils and fat-containing foods (pp. 107–133). IL: AOCS Press. Rosenvold, K., Petersen, J. S., Lærke, H. N., Jensen, S. K., Therkildsen, M., Karlsson, A. H., Møller, H. S., & Andersen, H. J. (2001). Muscle glycogen and meat quality as affected by strategic finishing of slaughter pigs. Journal of Animal Science, 79, 382–391.
1138
M.G. O’Sullivan et al. / Meat Science 65 (2003) 1125–1138
Shahidi, F., Yun, J., Rubin, L. J., & Wood, D. F. (1987). The hexanal content as an indicator of oxidative stability and flavour acceptability in cooked ground pork. Canadian Institute of Food Science and Technology Journal, 20, 104–106. Siegmund, B., & Pfannhauser, W. (1999). Changes of the volatile fraction of cooked chicken meat during chill storing: results obtained by the electronic nose in comparison to GCMS and GC olfactometry. Zeitschrift fu¨r Lebensmittel-Untersuchung und Foschung A, 208, 336–341. St.Angelo, A. J., Vercellotti, J. R., Legendre, M. G., Vennett, C. H.,
Kuan, J. W., James Jr., C., & Dupuy, H. P. (1987). Chemical and instrumental analysis of warmed-over flavour in beef. Journal of Food Science, 52, 1163–1168. Tappel, A. L. (1962). Heme compounds and lipoxidase as biocatalysts. In Symposium on foods: lipids and their oxidation (p. 122). Westport, CT: AVI. Wen, J., Morrissey, P. A., Buckley, D. J., & Sheehy, P. J. A. (1996). Oxidative stability and alpha-tocopherol retention in turkey burgers during refrigerated and frozen storage as influenced by dietary alpha-tocopheryl acetate. British Poultry Science, 37, 787–795.