Influence of storage temperature on microbial spoilage characteristics of haddock fillets (Melanogrammus aeglefinus) evaluated by multivariate quality prediction

International Journal of Food Microbiology 111 (2006) 112 – 125 www.elsevier.com/locate/ijfoodmicro

Influence of storage temperature on microbial spoilage characteristics of haddock fillets (Melanogrammus aeglefinus) evaluated by multivariate quality prediction G. Olafsdottir a,⁎, H.L. Lauzon a , E. Martinsdottir a , K. Kristbergsson b a

b

Icelandic Fisheries Laboratories, Skulagata 4, 101 Reykjavík, Iceland University of Iceland, Department of Food Science, Hjardarhagi 6-7, 107 Reykjavík, Iceland Received 13 July 2005; received in revised form 23 March 2006; accepted 25 April 2006

Abstract The proliferation of specific spoilage organisms (SSO) and quality changes were evaluated in haddock fillets stored in styrofoam boxes at 0, 7 and 15 °C and under temperature fluctuations. A rapid electronic nose technique was used to monitor different classes of compounds, representing microbial metabolites that were characteristic for the onset of spoilage odors. Photobacterium phosphoreum predominated among the spoilage bacteria and high levels of TVB-N were observed at sensory rejection. Pseudomonas spp. appeared to be responsible for the development of sweet, fruity spoilage odors in haddock fillets coinciding with increasing response of the electronic nose CO sensor. H2S-producing bacteria, most likely Shewanella putrefaciens, were associated with the H2S sensor's response at abusive temperature conditions. Partial Least Squares Regression (PLSR) was used as an explorative tool to provide a better understanding of the spoilage potential of SSOs, by evaluating models based on electronic nose responses and counts of specific spoilage organisms to predict sensory quality (Torry scores). The best prediction of the sensory quality was obtained by PLSR models based on five variables: the electronic nose sensors (CO, NH3 and H2S), pseudomonads counts and a time–temperature variable. Good agreement between the predicted and experimental data indicates that these variables characterize the sensory quality of haddock fillets stored under different temperatures. © 2006 Elsevier B.V. All rights reserved. Keywords: Specific spoilage organisms; Electronic nose; Fish quality; Sensory analysis; TVB-N; PLSR; DMFit; SSSP

1. Introduction Export of fresh fish as fillets from Iceland to the markets in Europe and in the USA has become increasingly important in recent years, while the export of whole fish in ice has declined. It is well documented that packed fillets spoil more rapidly and have different spoilage pattern than whole fish (Lindsay et al., 1986; Jørgensen et al., 1988; Huss, 1995; Lauzon et al., 2002). Consequently, the vested interest of the industry partners is to maintain the freshness of the fish fillets, by optimal handling and transport conditions to ensure the high quality of the products on the market. Temperature control during all stages of the production and in the distribution chain is emphasized in guide⁎ Corresponding author. Tel.: +354 5308600; fax: +354 5308601. E-mail address: [email protected] (G. Olafsdottir). 0168-1605/$ - see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.ijfoodmicro.2006.04.045

lines on good manufacturing practices. However, temperature fluctuations can occur in the production or in the distribution chain because of unforeseeable events such as improper icing of the raw material or delays in transport. In this case the rate of the deteriorative changes occurring in fish caused by microbial growth and oxidative changes will be influenced and therefore, the information on storage days or days from catch will not be reliable to determine the actual quality (Botta et al., 1987; Huss, 1995). Recent developments in sensor technologies and data analysis procedures have stimulated interests to develop rapid techniques to detect postmortem quality changes in muscle food (Olafsdottir et al., 1997a, 2004; Ellis and Goodacre, 2001). For accurate evaluation of quality no single index can encompass all the complex changes occurring during spoilage of fish (Martin et al., 1978) and therefore a multisensor concept has

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125

been introduced based on simultaneous instrumental analysis of various sensory related attributes (Olafsdottir et al., 2004). The detection of specific spoilage organisms (SSO) like Shewanella putrefaciens, Pseudomonas ssp. and Photobacterium phosphoreum is considered more reliable than total viable counts (TVC) to accurately evaluate the freshness or spoilage level of fish products (Dalgaard, 2000; Gram et al., 2002). The potential of H2Sproducing organisms, like S. putrefaciens, to develop sulfur compounds contributing to off-flavor described as onion, cabbage and putrid spoilage odors at advanced stages of storage in fish is well known (Miller et al., 1973a; Herbert et al., 1975). S. putrefaciens has been identified as the main SSO in whole cod (Herbert and Shewan, 1976; Jørgensen et al., 1988) as well as in aerobically stored haddock fillets (Levin, 1968; Chai et al., 1971). Development of ammonia-like and ‘fishy’ off-flavors has been related to fish spoilage bacteria like S. putrefaciens and P. phosphoreum that can reduce TMAO to TMA (Jørgensen and Huss, 1989; Dalgaard, 1995). P. phosphoreum was identified as an important TMA producer in iced cod and in cod fillets (Van Spreekens, 1974; Van Spreekens and Toepoel, 1981) and has been suggested as the main spoilage organism in packed cod fillets, although this bacterium does not produce intensive off odors (Dalgaard et al., 1993; Dalgaard, 1995). Other well known spoilage bacteria like Pseudomonas spp. do not produce TMA from TMAO (Castell et al., 1959). Earlier studies on the formation of odorous degradation compounds in fish (cod, haddock and rockfish) from the North Atlantic area showed that pseudomonads, in particular P. fragi, were responsible for quality changes and development of sweet, fruity off odors in chilled fish (Castell and Greenough, 1957; Castell et al., 1959; Miller et al., 1973b). More recently the importance of pseudomonads in spoilage in fish species from the Mediterranean Sea (Koutsoumanis and Nychas, 1999) and American plaice (Lauzon, 2000) has been reported. Growth of pseudomonads species was modeled as a function of storage time and correlated well with organoleptical shelf-life in Mediterranean species (Koutsoumanis and Nychas, 2000). Predictive microbiology as a tool to study the microbial ecology in food has gained considerable interest (Ross and McMeekin, 1994). Kinetic models based on the Arrhenius equation have been developed to describe the relationship between temperature and relative spoilage rate for muscle foods. Recently models based on enumeration of specific spoilage organisms (SSO) to determine the remaining shelf-life of fish products have been developed (Dalgaard et al., 1997a; Koutsoumanis and Nychas, 2000; Koutsoumanis, 2001; Dalgaard, 2002). Comparison of the chemical profiles of spoiled seafoods and of the metabolites produced by potential spoilage organisms has only been used to a limited extent for characterization of SSOs (Gram and Dalgaard, 2002). More knowledge on the chemical characterization of spoilage processes of muscle food and their correlations with sensory and microbiological changes is needed (Dainty, 1996). Multi-compound indices based on gas chromatography analysis of the main compounds produced by bacteria during spoilage have been suggested to evaluate the complex changes occurring in fish products stored under different conditions (Lindsay et al., 1986; Olafsdottir and Fleurence, 1998; Jørgensen et al., 2000, 2001; Olafsdottir et al., 2005).

113

Detection of microbial metabolites would be useful as an alternative or supporting information to microbial enumeration to use in models for a more rapid evaluation of the quality and shelf-life of fish products. Electronic noses based on selective detection of the main classes of volatile compounds produced contributing to the spoilage odor, like amines, sulfur compounds, alcohols, aldehydes and esters, can be used to measure quality changes in fish in a rapid way (Olafsdottir et al., 1997b, 2000, 2002, 2003; Di Natale et al., 2001). The data presented herein is from two storage experiments of haddock fillets performed in the years 2001 and 2003. The different temperature conditions during storage were selected to study the influence of abusive conditions on the proliferation of the microflora. The objective was to characterize spoilage in haddock fillets and determine the spoilage domain of the specific spoilage bacteria by studying their growth and production of spoilage metabolites under a range of temperature conditions. The maximum potential growth rate of the SSOs was evaluated by using the curve fitting model DMFit (Baranyi and Roberts, 1994). Partial least squares regression (PLSR) models were used to explore the spoilage potential of the microflora by studying the correlation of SSOs, volatile compounds measured by an electronic nose, TVB-N and sensory analysis. Determination of the end of shelf-life based on sensory analysis was compared to the estimation of shelf-life based on the currently available SSSP model. 2. Materials and methods 2.1. Storage experiments of haddock fillets at constant environmental temperatures — 2001 The fish was caught by longline in November 2001 in fishing grounds Southwest of Iceland. The average seawater temperature was around 8 °C. The fish was iced and stored ungutted on board and received at the processing factory within 12 h from catch. After gutting, filleting and skinning, the fillets were packaged in styrofoam (EPS, expanded polystyrene) boxes lined with a plastic bag and an adsorbing pad at the bottom. Each box (110 × 400 × 263 mm) contained 11 fillets and a cooling mat was placed on top. The boxes were transported to the laboratory and stored at different temperatures (0, 7 and 15 °C) until sensory rejection. Analyses of duplicate samples were performed on days 1, 3, 5, 7, 10 and 14 after catch for samples stored at 0 °C. Because of more rapid spoilage at higher temperatures additional samples were taken on day 2 for groups stored at 7 and 15 °C and the last sampling days were on days 7 and 4, respectively. 2.2. Storage experiments of haddock fillets at fluctuating environmental temperatures — 2003 Additional storage experiments were conducted in November 2003 on haddock fillets obtained from the same factory as in 2001 to collect further data to study the influence of fluctuating temperatures during storage. The fish was caught by longline close to Sandgerði, Southwest of Iceland. The average seawater temperature was around 8 °C in November. The fish was transported the same day to the factory where it was stored ungutted

114

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125

in ice slurry overnight and processed traditionally at the factory the following day. The fish was then hand filleted and cooled in icewater before being skinned in a machine and packed in plastic bags in the styrofoam boxes as before and transported to the laboratory the same day. Two groups of samples were prepared (“0 °C a” and “0 °C + abuse”). Both sample groups were initially stored at 0 to 1 °C, but on the fourth day post-catch, group “0 °C + abuse” was transferred from the cooler and stored overnight (16 h) at room temperature (RT) and then moved back to the cooler. The first samples were taken on the day following processing (2 day postcatch). Samples were analyzed regularly until sensory rejection: group “0 °C a” on days 2, 4, 7, 9 and 11 post-catch; group “0 °C + abuse” on days 2, 4, 5, 7 and 9 post-catch. At each day of sampling duplicate analyses were done using one box containing 11 fillets for the various analyses. Four fillets were used for sensory analysis. Another four fillets were pooled into 2 samples and used for microbial analysis (TVC and SSO counts), chemical analysis of total volatile bases (TVB-N) and pH. Three fillets were then used for the analysis of volatile compounds with an electronic nose with electrochemical sensors.

225 mL of cooled Maximum Recovery Diluent (MRD, Oxoid) in a stomacher for 1 min. Successive 10-fold dilutions were made as required. Total viable psychrotrophic counts (TVC, 15 °C) were evaluated by spread-plating aliquots onto modified Long and Hammer's medium; while counts of H2S-producing bacteria and presumptive pseudomonads were evaluated on spread-plated Iron Agar (15 °C) and modified CFC medium (22 °C), respectively (Lauzon et al., 2002). Counts of P. phosphoreum were estimated by using the PPDM-Malthus conductance method (Dalgaard et al., 1996), as described by Lauzon (2003). 2.6. TVB-N and pH measurements Total volatile basic nitrogen content (TVB-N) was measured by a steam distillation method described by Malle and Poumeyrol (1989), using the rest of the flesh mince prepared for microbiological analysis within 30 min of preparation. The pH was measured in 5 g of mince moistened with 5 mL of deionised water. The pH meter was calibrated using the buffer solutions of pH 7.00 ± 0.01 and 4.01 ± 0.01 (25 °C) (Radiometer Analytical A/S, Bagsvaerd, Denmark).

2.3. Temperature recording 2.7. Electronic nose measurements Temperature data loggers were used in the experiments (Stow Away®, Onset Computer Corporation, Bourne, Mass., USA) to monitor the temperature of the fillets and the ambient storage environment. The loggers were located underneath and above the fillets inside the styrofoam boxes, i.e. one box for each temperature treatment, and on top of the box to follow the environmental variations of each storage condition. Temperature recordings were at 5 min intervals. Data is only shown from loggers located underneath the fillets. 2.4. Sensory analysis Sensory analysis to evaluate the freshness of the fillets was performed by 8–12 previously trained panelists from the IFL sensory panel (age range 25 to 65 years), both female and male. They were all trained according to international standards (ISO, 1993) for evaluation of different seafood, including detection and recognition of tastes and odors by use of scales and descriptors. The Torry scheme (Shewan et al., 1953) was used to assess the freshness (odor and flavor) of cooked fish. The scheme ranges from 10 = very fresh to 3 = very spoiled, with a rejection level at 5.5. Fish from each sampling group was portioned into an aluminium box and cooked in a steam oven (98 to 100 °C for 5 min). Samples were anonymously coded and assessed in duplicate. A computerized system (FIZZ, Version 2.0, 1994–2000, Biosystèmes, Couternon, France) was used for data recording and for further data processing. Average scores of the judges were calculated for each sample assessed and the reported value was the average of the duplicate samples. 2.5. Microbiological evaluation Fillets were aseptically minced, pooling 2 fillets for each sample. Twenty-five grams of minced flesh were mixed with

Electronic nose measurements were performed using a gas sensor instrument called “FreshSense”, developed by the Icelandic Fisheries Laboratories and Maritech (Iceland). The instrument is based on electrochemical gas sensors (CO, H2S, and SO2: Dräger, Lübeck, Germany; NH3: City Technology, Portsmouth, United Kingdom). The measurement technique and the accuracy of the measurements were described earlier by Olafsdottir et al. (1998, 2002). The instrument consisted of a 2.3 L closed glass sampling container and a dynamic sampling system. Duplicate samples of approximately 500 g of fish fillets were analyzed. The samples were maintained at room temperature before the measurements started, for about 15 to 30 min or until they reached 8 to 12 °C. Measurements were taken every 10 s for 5 min. The reported value (current) is the average of the last three measurements of the 5 min measurement cycle minus the average of three values before the measurement starts. 2.8. Data analysis Microsoft Excel 97 was used for data processing, to calculate means and standard deviations for all multiple measurements and to generate graphs. Analysis of variance (ANOVA) was applied to the sensory, chemical and electronic nose data using the Number Cruncher Statistical Software (NCSS 2000, NCSS, Kaysville, Utah, USA). Significant differences were determined by One way ANOVA and Duncan's Multiple-Comparison Test was used to determine the statistical difference between samples. An effect was considered significant at the 5% level ( p < 0.05). Multivariate analysis was performed by the Unscrambler (version 9.1.2, CAMO A/S, Trondheim, Norway). The relationships between sensory, microbial, chemical and electronic nose measurements were explored by Partial Least Squares Regression

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125

Fig. 1. Temperature profiles of haddock fillets stored in styrofoam boxes at 15 °C, 7 °C, 0 °C and 0 °C + temperature abused on day 4 for 16 h. Temperature loggers located underneath the fillets at the bottom of the boxes.

(PLSR). The effect of temperature of fillets and time was also included as an independent variable by calculating, for all samples at each sampling day, the accumulative influence of temperature (T) and time (t): Taccumulative = ∑(T − Tmin) × dt; Tmin was defined as the minimum temperature of fillets in storage. PLSR was performed on the data using average values of replicates standardized to equal variance (weighting with 1/ standard deviation). PLSR models were calculated with microbial, chemical and electronic nose variables as X predictors and sensory data as Y response factors (Martens and Martens, 2001). The Jack-knife method (Martens and Martens, 1999) was used to determine significant variables in X with a significance level of 5% ( p < 0.05). Outlier detection is built into the Unscrambler software and detects samples that are badly described by the models. All models were calculated with full-leave one-out cross validation. The bacterial growth data collected were fitted to determine maximum specific growth rates (h− 1) by using DMFit (http:// www.ifr.ac.uk/safety/DMFit/), an in-house program of IFR (UK) which is based on a reparameterized version of the model of Baranyi and Roberts (1994). DMFit is an Excel add-in program to fit curves where a linear phase is preceded and followed by a stationary phase. The Seafood Spoilage and Safety Predictor (SSSP v. 2.0) software (http://www.dfu.min.dk/micro/sssp) was used to predict the effect of storage temperature on shelf-life of haddock using the recorded temperature profiles. The SSSP is based on kinetic models for growth of specific spoilage organisms and empirical relative rates of spoilage secondary models (Dalgaard et al., 2002). The S. putrefaciens-like H2S-producing bacteria model for aerobically stored fresh fish was used to predict the shelf-life of haddock fillets. 3. Results Temperature of the fillets was monitored during storage under the different environmental conditions (Fig. 1). The rapid increase

115

in temperature at the higher temperature storage is obvious and fillets had reached 14 °C after 1.5 days of storage at 15.1 ± 1.1 °C. The temperature of the fillets stored at 7.2 ± 0.2 °C had increased to 7 °C within 3 days, while the fillets stored at 0.9 ± 0.3 °C had cooled down in one day (day 2 from catch). The initial temperature of the fillets was higher in 2001 (∼4 °C) because no cooling of the fillets was employed after filleting and deskinning, while the initial temperature of the fillets in 2003 was lower (0.9 to 2.3 °C) when the fillets were cooled during processing (Table 2). The temperature of the fillets of sample group “0 °C + abuse” increased gradually to 10 °C in 16 h when stored at room temperature. When the boxes were transferred back to chilled conditions it took almost three days to cool the fillets to 0 °C. The thermal insulating properties of the styrofoam boxes that are commonly used for export of fillets are obviously important when temperature fluctuates, influencing both the warming and cooling time of the fish. 3.1. Sensory analysis and shelf-life determination of haddock fillets The end of shelf-life based on Torry sensory score of 5.5 was estimated after 13.5 days from catch for samples stored at 0 °C; after 6.5 days when stored at 7 °C and after approximately 4.5 to 5 days when stored at 15 °C (Fig. 2). The temperature abused fillets from the experiments in 2003 had a shorter shelf-life than the fillets stored at 0 °C (9 days and 11.5 days, respectively). The different shelf-life of the fillets stored at 0 °C in the experiments in 2001 and 2003 (13.5 and 11.5 days, respectively) may be caused by the difference in the initial handling and/or biological variation of the raw material resulting in higher initial bacterial counts in 2003 (Table 2) and slower bacterial growth in 2001 (Fig. 3). Although the seawater temperature was around 8 °C for both the years 2001 and 2003 in November, it is of interest that the average seawater temperature in Icelandic waters was generally higher in 2003 than in the year 2001 (http://www.

▪

Fig. 2. Assessment of shelf-life by sensory analysis (Torry) of haddock fillets stored at 0 °C (- -), 7 °C (-♦-) and 15 °C (-x-) from the experiment in 2001, and at 0 °C a (-▴-) (traditional process) and 0 °C + abuse (-Δ-) (traditional process and temperature abuse) in 2003.

116

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125

▪

Fig. 3. Counts of total viable psychrotrophic bacteria (TVC) (a), P. phosphoreum (Pp) (b), presumptive Pseudomonas spp. (c) and H2S-producing bacteria (d) in haddock fillets stored in styrofoam boxes at 0 °C (- -), 7 °C (-♦-) and 15 °C (-x-) from experiment in 2001 and at 0 °C a (-▴-) (traditional process) and 0 °C + abuse (-Δ-) (traditional process and temperature abuse) in 2003.

hafro.is/Sjora/). However, it is not clear if this may have influenced the higher initial counts of bacteria in 2003. In 2003 the raw material was stored ungutted overnight in ice slurry, but was stored in ice before processing in 2001. This may have influenced the more rapid spoilage in 2003 as storage progressed, hence a shorter shelf-life, despite higher sensory scores initially. Earlier studies on packed haddock fillets have reported shelf-life of about 9 to 11 days post-catch for the control sample stored in ice in a study on MA-packed fillets (Dhananjaya and Stroud, 1994). Similar shelf-life of 9 to 10 days from catch was reported for haddock fillets stored in styrofoam boxes (Olafsdottir et al., 2003). Origin, catching techniques, initial handling and temperature during storage will inevitably influence the spoilage rate and shelf-life, as well as seasonal differences as shown by Olafsdottir et al. (2003). A more rapid spoilage rate of whole haddock caught in spring was observed compared to fish caught in the fall. Therefore, it is important to study the relationship between the sensory shelf-life and the spoilage domain of the specific spoilage bacteria taking into account the various extrinsic and intrinsic factors to achieve

better understanding of the complex spoilage processes occurring in fish fillets. 3.2. Microbiological evaluation The initial microbial counts (psychrotrophic TVC) of the haddock fillets, processed one day post-catch were log 3.7 CFU/g in the experiment in 2001. In the 2003 experiment a ten-fold higher average TVC (log 4.7 CFU/g) was measured after one day of cooled storage in styrofoam boxes (day 2 post-catch). Initial load of spoilage bacteria differed between genera, with Pseudomonas spp. being found in highest numbers (log 3.2–3.5 CFU/g) (Fig. 3c) and H2S-producing bacteria at lowest levels, i.e. at or below the detection level (log 1.3 CFU/g). P. phosphoreum was below the detection limit (log 1.3 CFU/g) on the first day of storage in 2003, while it was detected (log 2.6 CFU/g) on the first sampling day in November 2001. Slow proliferation of the microflora occurred during the first 4 to 5 days of storage at 0 °C, while it developed at a faster rate with increasing temperature (Fig. 3a). A similar behavior was

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125

117

Table 1 Maximum specific growth rate (h− 1) of different bacterial groups (Pseudomonas spp., P. phosphoreum, S. putrefaciens-like H2S producing bacteria) assessed in haddock fillets stored in styrofoam boxes under different temperature conditions from experiments in 2001 and 2003 as estimated by DMFit and compared to SSSP values for growth rate estimations of S. putrefaciens at constant temperature based on average temperature of fillets

2001 — 0 °C 2003 — 0 °C a 2003 — 0 °C + abuse 2001 — 7 °C 2001 — 15 °C

Temperature a

TVC

Pseudomonas spp.

P. phosphoreum

S. putrefaciens

°C

DMFit

DMFit

DMFit

DMFit

SSSP

0.012 0.022 c 0.047 d 0.033 0.048

b

0.033 0.053 0.081 0.078 0.145

0.045 0.047 0.069 0.175 0.325e

NA 0.2 ± 0.2 1.7 ± 2.5 6.9 ± 0.8 12.0 ± 3.7

0.016 0.013 0.024 0.037 0.068

0.191 0.079 0.116 0.105 0.203

Average temperature of fillets during storage (temperature recording for 2001 0 °C was not available and a value of 0°C was estimated) initial delay of 105.6 hb, 69.6 hc, 60.5 hd, followed by rapid growth. e Out of range of the model (>10 °C). a

observed for P. phosphoreum (Pp) (Fig. 3b) reaching TVC levels and dominating the spoilage microflora on day 7 at 0 °C and on days 3 at 7 and 15 °C. Growth of Pseudomonas spp. (Fig. 3c) and H2S-producing bacteria (Fig. 3d) was slow but steady throughout the storage period at lower temperature, but their growth accelerated with increasing temperature as expected. Similar maximum specific growth rate was observed for Pseudomonas spp. and H2S-producing bacteria in the temperature range of 0 to 7 °C, as evaluated by DMFit (Table 1). However, because of the higher initial Pseudomonas counts (about 100 fold), this bacterial group had reached higher levels at sensory rejection, ranking it in second place of importance among the SSOs evaluated (Table 2). At 15 °C, larger difference in maximum specific growth rate was observed among these genera, the fastest rate observed for P. phosphoreum (0.203),

second for H2S-producing bacteria (0.145) and slowest for Pseudomonas spp. (0.048). Temperature abuse of raw material and processed fillets clearly triggered the development of P. phosphoreum, as well as the other bacterial groups assessed but to a lesser extent (Table 1). The fact that P. phosphoreum rapidly reached TVC levels in all samples and dominated the spoilage microflora suggests that this bacterium may contribute considerably to the overall spoilage of aerobically stored haddock fillets. Other workers have reported the presence of this bacterium in aerobically spoiled fish (Van Spreekens, 1974; Dalgaard et al., 1997b; Esaiassen et al., 2004). This bacterium has already been reported as the main spoilage organism in fish packed under modified atmosphere (Dalgaard, 1995; Dalgaard et al., 1997b; Emborg et al., 2002). The maximum specific growth rate of pseudomonads and H2S-producing

Table 2 Overview of measured initial values for the microbial, TVB-N, pH, electronic nose and sensory data and the estimated values for the experimental data at sensory rejection based on Torry score = 5.5, including estimation of shelf-life by TVB-N (35 mg N/100 g) and the SSSP S. putrefaciens model for haddock fillets stored in styrofoam boxes from storage experiments in 2001 and 2003 at different temperatures and temperature fluctuations Initial values 2001

Estimated shelf-life: Sensory analysis (Torry score = 5.5) a TVB-N (35 mg N/100 g) a Sensory analysis (Torry score = 5.5) b SSSP model: H2S-producer counts b Microbial counts: TVC (log10 CFU/g) H2S-producer counts % H2S-producers/TVC Pseudomonas counts % Pseudomonas spp./TVC P. phosphoreum counts % Pp/TVC Sensory: Torry score TVB-N (mg N/100 g) pH Electronic nose: CO sensor Electronic nose: NH3 sensor Electronic nose: H2S sensor a b

Estimated values at sensory rejection 2003

3.7 ± 0.3 1.4 ± 0.1

4.7 ± 0.0 1.3 ± 0.9

3.2 ± 0.3

3.5 ± 0.5

2.6 ± 0.2

1.3 ± 0.0

8.7 ± 0.3 14.0 ± 0.1 6.55 ± 0.01 64 ± 19 <10 38 ± 6

9.0 ± 0.1 11.8 ± 0.4 6.68 ± 0.05 31 ± 5 <10 31 ± 27

2001

2003

0 °C

7 °C

15 °C

0 °C a

0 °C + abuse

13.5 days 12–13 days 12.5 days 10 days

6.5 days 4.5 days 5.5 days 3.5 days

4.5 days 3 days 3.5 days 2 days

11.5 days 12–13 days 10 days 10–10.5 days

9 days 8 days 8 days 7 days

8.1 5.9 0.6% 6.9 6.3% 8.0 79.4% 5.5 45 6.7 280 <10 52

8.4 6 0.4% 7.1 5.0% 8.1 50.2% 5.5 77 6.7 470 40 130

8.2 6.1 0.8% 6.4 1.6% 8.2 100% 5.5 60 6.8 510 30 92

7.5 5.8 2.4% 6.7 15.1% 7.5 100% 5.5 27 6.7 730 10 20

8.2 6.1 0.8% 7.3 15.2% 8.0 61.2% 5.5 45 6.9 570 28 50

Estimated total shelf-life from catch (including time delay between catch and processing). Estimated shelf-life after processing.

118

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125

bacteria was higher in 2003 than in 2001. Based on their ability to produce off odors as reported earlier (Castell et al., 1959; Levin, 1968; Chai et al., 1971; Miller et al., 1973a,b; Jørgensen et al., 1988), it is suggested that these bacteria, especially Pseudomonas spp., may have contributed to the characteristic spoilage odors of fillets and shorter shelf-life in 2003. 3.3. TVB-N and pH measurements The formation of volatile bases (TVB-N) in the haddock fillets increased at the end of storage in all sample groups (Fig. 4a). Although large standard deviations for TVB-N were noticed for the last days of storage, especially for the groups stored at elevated temperatures (0 °C + abuse, 7 °C and 15 °C), the increase was significant (p < 0.05). The influence of the abusive temperature treatment on the sample group from 2003 (0 °C + abuse) was evidenced by a more rapid TVB-N production compared to the sample group stored continuously at 0 °C. Comparison of the

▪

TVB-N values obtained for groups stored at 0 °C showed that a lower level (p < 0.05) was reached in 2003 (27 mg N/100 g) than in 2001 (45 mg N/100 g) at sensory rejection, on days 11 and 13.5, respectively (Table 2). The lower TVB-N value observed in 2003 concurred with the lower P. phosphoreum counts reached at sensory rejection (log 7.5 CFU/g) than in 2001 (log 8 CFU/g) (Fig. 3b). Based on the finding of Dalgaard (1995) that P. phosphoreum was a 30 times more active TMA producer than S. putrefaciens it is likely that P. phosphoreum was mainly responsible for the high TVB-N content. Slight changes were observed in pH of samples during storage (Table 2). The range of pH values measured in the samples was 6.5 to 6.9. Lower initial pH values were measured in fillets from 2001 (6.5 to 6.6) which was maintained during most of the storage period, reaching 6.73 on the last sampling day (0 °C). The lower pH value of the fillets in 2001 may have influenced the extended shelf-life. An initial pH value of 6.68 was observed in 2003 reaching 6.71 at the end of chilled storage. Higher pH values were

Fig. 4. TVB-N values (a) and electronic nose responses measured by CO sensor (b), H2S sensor (c) and NH3 sensor (d) in haddock fillets stored in styrofoam boxes at 0 °C (- -), 7 °C (-♦-) and 15 °C (-x-) from storage experiment in 2001 and at 0 °C a (-▴-) (traditional process) and 0 °C + abuse (-Δ-) (traditional process and temperature abuse) in 2003. Only the temperature abused samples are shown for the H2S and NH3 sensors.

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125

reached (6.8 to 6.9) at elevated temperatures. Increase in pH value, as storage time progressed, coincided with the peak production of TVB-N. The dominating growth of P. phosphoreum may possibly be explained by the pH and conditions that probably favored its growth. It is likely that microaerophilic conditions developed in plastic wrapped fillets in the styrofoam boxes (containing 11 fillets, approximately 5 kg) with the production of bacterial carbon dioxide influencing the pH. However, this can not be confirmed since the CO2 level in the headspace of the boxes was not measured. 3.4. Electronic nose measurements — evaluation of microbial metabolites Response of the electronic nose sensors as a function of storage time was characterized by an initial lag phase and a steep slope at the end as seen for the CO sensor response (Fig. 4b). This can partly be explained by the low sensitivity of the sensors (Olafsdottir et al., 2002) and the fact that high microbial cell counts are needed to produce metabolites in high enough concentrations to be detected by the sensors. The large standard deviation observed for the electronic nose responses (Fig. 4b, c, d) is related to their volatile nature and hence their sensitivity to slight changes in temperature during sampling. (Olafsdottir, 2003). The electrochemical gas sensors in the FreshSense electronic nose have different selectivity and sensitivity towards selected standards representative for the main classes of compounds produced in fish during spoilage (Olafsdottir et al., 1998, 2002). The CO sensor detects volatile alcohols, aldehydes and esters, the NH3 sensor is selective for ammonia and amines and the H2S and SO2 sensors detect sulfur compounds. However, some cross sensitivity has been observed for the CO and the H2S sensors (Olafsdottir et al., 2002). 3.4.1. CO sensor The CO sensor was the most sensitive of the sensors and increasing response was noticed earlier than for the other sensors, suggesting the presence of alcohols or aldehydes early in the spoilage process. The response of the CO sensor (Fig. 4b) showed the same trend for the spoilage rate of the sample groups of haddock fillets stored at different temperatures as shown by the sensory analysis. An increase (p < 0.05) was first detected between days 2 and 3 for samples stored at 15 °C, between days 2 and 7 for samples stored at 7 °C, between days 4 and 9 for the 0 °C a and between days 5 and 7 for the 0 °C + abuse. The difference in spoilage rate of the sample groups stored at 0 °C in 2001 and 2003 observed by sensory analysis was supported by the more rapid increase in the CO sensor response in 2003, indicating more production of metabolites like alcohols, aldehydes and esters resulting in the shorter sensory shelf-life. Earlier gas chromatography studies on volatile compounds in haddock fillets showed that volatile esters and alcohols in haddock stored in ice coincided with the response of the CO sensor and development of fruity, sweet odors detected in the spoiled fillets (Olafsdottir, 2003). Pseudomonas spp., including P. fragi, has been associated with the sweet, fruity off odors and onionlike odors often encountered on commercially prepared fillets

119

rather than round or eviscerated fish (Castell and Greenough, 1957; Castell et al., 1959). Miller et al. (1973b) identified esters, aldehydes and sulfides as the main contributors of these odors caused by growth of P. fragi. 3.4.2. NH3 and H2S sensors Increasing response at the end of the storage time was noticed for the H2S and NH3 sensors for samples stored under abusive temperature treatments and data is only shown for those conditions (Fig. 4c and d). The insignificant responses of the SO2 sensor (data not shown) indicate that sulfides were not present or in very low levels in all samples. The response of the NH3 sensors was low towards all sample groups (Fig. 4d). Increase in the NH3 response for samples stored at 7, 15 °C and 0 °C + abused coincided with increase in the TVB-N values in these groups. However, a significant increase (p <0.05) for the NH3 sensor was only detected for the sample stored at 7 °C for 7 days. Increase for the H2S sensor (p <0.05) was observed for samples stored under abusive conditions at 7 and 15 °C at the end of shelf-life (Fig. 4c). This may indicate the influence of H2S-producing bacteria under abusive temperature conditions as suggested by Lauzon (2000). Based on the high responses of the CO sensor and low responses of H2S and NH3 sensors, it is likely that alcohols, aldehydes and esters, coinciding with pseudomonads growth, were contributing to the spoilage odors in haddock fillets at low temperature in this study. Similar results for the characteristic responses of the electronic nose to spoilage development of haddock fillets stored in ice were reported earlier (Olafsdottir et al., 2003). It should be emphasized that the low sensitivity of the sensors (Olafsdottir et al., 2002) and the semi-quantitative approach in the electronic nose measurements limit the accurate evaluation of the actual concentration of the highly volatile compounds like hydrogen sulfide and ammonia in the headspace in this study. However, comparison of the sensor responses between samples as done herein gives indication of quantitative changes with time. 4. Discussion 4.1. Evaluation of the end of shelf-life of haddock fillets by sensory, chemical and microbial criteria Different storage temperatures influenced the shelf-life of the haddock fillets as estimated by sensory and microbial methods in agreement with earlier studies on aerobically stored cod fillets (Einarsson, 1992). Estimation of the shelf-life was dependent on the evaluation method used as seen in Table 2. Values of the different measurement variables were different at sensory rejection indicating the very complex nature of the spoilage processes in fish. Microbial metabolites measured by the electronic nose were produced in greater amount with increasing storage temperatures as shown by the higher value of the sensors' responses at sensory rejection (Fig. 4b, c, d). This occurred despite the fact that similar bacterial levels were found at sensory rejection. The spoilage microflora appeared to have greater ability to produce metabolites at higher temperatures, as was reported by Davis (1990).

120

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125

4.1.1. TVB-N Based on the fixed TVB-N limit (35 mg N/100 g) as quoted in the EU regulations for gadoids (European Union, 1995) a slightly shorter shelf-life was estimated for all sample groups (see Fig. 4a), than when using the sensory Torry score criteria of 5.5 (Table 2). TVB-N is only useful to detect advanced spoilage because values only begin to increase at later stages of storage (Oehlenschläger, 1998; Baixas-Nogueras et al., 2003). It has been pointed out that TVB-N and TMA often give ambiguous information about the quality of the products as their levels are influenced by the storage method like in modified atmosphere packaging (Davis, 1990; Dalgaard et al., 1993; Debevere and Boskou, 1996; Lauzon et al., 2002) and in pre-frozen chilled fillets (Magnússon and Martinsdóttir, 1995; Guldager et al., 1998). 4.1.2. TVC and SSO Total viable psychrotrophic counts (TVC) can give controversial estimation of the end of shelf-life and different limits have been set based on product types. Einarsson (1992) reported that shelf-life of cod fillets estimated by microbial growth was in general shorter than when evaluated by sensory analysis. In this study, the spoilage bacteria became dominated by P. phosphoreum (50 to 100%) at all storage temperatures tested reaching levels of log 7.5 to 8.2/g at sensory rejection (Table 2). Pseudomonas spp. represented about 5 to 15% of the spoilage bacteria reaching loads of log 7.3/g under abused conditions, while H2S-producing bacteria were found in minority (0.4 to 2.4%) reaching levels of about log 6/g. At 15 °C Pseudomonas spp. and H2S-producing bacteria were found at very low levels, 1.6 and 0.8%, respectively. Contradictory to previous findings on the role of S. putrefaciens in TMA production and spoilage of fresh fish/fillets (Jørgensen and Huss, 1989), their low levels (about log 6/g) observed at sensory rejection did not support the high TVB-N levels measured. High TVB-N levels at sensory rejection were suggested to be related to high P. phosphoreum counts that reached TVC levels (>log 8/g) for most sample groups. Interestingly, lower TVB-N levels were found in samples from 2003 when TVC and Pp counts were
with Pseudomonas spp. at high temperature (10–15 °C) in a fish model system. In a model system with single bacterial cultures, detection of metabolites coincides with high counts, however, evidence suggests that in co-culture systems the metabolite production may be high even though the counts of the individual bacteria are lower (Lauzon, unpublished data). The interaction of the spoilage microflora has been studied by many researchers to understand better the spoilage development in fish (Gram et al., 2002). Pseudomonas spp. have been suggested to inhibit the growth of S. putrefaciens (Gram and Melchiorsen, 1996; Lauzon, 2000). Interestingly, the H2S sensor response appeared to decline in the group 0 °C a, at the end of storage, at the same time as a rapid increase was observed in the CO response suggesting active production of metabolites by pseudomonads. This may indicate the ability of Pseudomonas spp. to delay the growth of S. putrefaciens, and as a result limits the production of the H2S metabolite, but further studies are needed to verify this. 4.1.3. Seafood Spoilage and Safety Predictor (SSSP) The estimated shelf-life (SL) obtained by the S. putrefacienslike H2S-producing bacteria (Sp) model of the SSSP underestimated the observed SL after processing for the abusive temperature conditions (Table 2). The estimated growth rates of Sp at different temperatures, on which the models are based, were higher than those estimated by DMFit for the products tested in 2001 (Table 1) and therefore the predicted SL was shorter than observed (Table 2). However, the shelf-life predicted by the SSSP model for the 0 °C sample in 2003 corresponded to the observed value of about 10 days (Table 2) and similar growth rates were estimated by DMFit and SSSP (Table 1). The Sp-model is based on the growth of S. putrefaciens in a liquid model system, but it has been validated in some fish species, haddock among others, and should only be used within a temperature range of 0–10 °C. More rapid growth rates of SSO in such model systems can be expected (Koutsoumanis, 2001; Dalgaard et al., 2002), as experimental conditions may not resemble the product targeted

Fig. 5. PLSR correlation loadings based on all the measured variables: microbial (TVC, pseudomonads, H2S counts), Photobacterium phosphoreum (PP), chemical (TVB-N, pH) and electronic nose (CO, NH3, SO2 and H2S sensors) and Tacc as predictors for the Torry scores as a response variable for the haddock samples stored at different temperatures. The outer and the inner ellipses indicate 100% and 50% explained variance, respectively. Significant variables are symbolized with small circles.

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125

121

Table 3 PLSR model to predict Torry sensory scores (Y) based on different combination of the measured variables (X): TVB-N, pH, microbial counts (TVC, Pseud, Pp, H2S counts), electronic nose sensors (CO, NH3, H2S, SO2) and temperature treatment (Tacc) for haddock fillets (N = 21)

Microbial counts SSO

Electronic nose

Combination e-nose + SSO + temp

Classical

X

No. of variables

All measured All significant TVC, Pseud, Pp, H2S counts, Tacc Pseud, Pp, H2S counts, Tacc Pseud, Pp, Tacc Pseud, H2S counts, Tacc Pp, H2S counts, Tacc CO, NH3, H2S, Tacc CO, NH3, Tacc CO, H2S, Tacc CO, Tacc CO, NH3, H2S + H2S counts + Tacc CO, NH3, H2S + Pseud + Tacc CO, NH3, H2S + Pp + Tacc CO, NH3, H2S + TVC + Tacc TVB-N, TVC, Tacc

11 7 5 4 3 3 3 4 3 3 2 5 5 5 5 3

Expl X

Expl Y

PC1

PC2

PC1

PC2

62 79 80 76 71 74 76 70 74 68 73 67 64 66 67 75

15 10 15 19 23 25 21 18 19 24 27 20 22 19 19 18

86 92 89 89 86 89 82 54 59 54 60 74 75 69 71 74

7 1 3 3 5 3 7 17 13 14 7 16 14 15 16 15

Correlation r2

RMSEP

0.94 0.95 0.92 0.92 0.92 0.92 0.89 0.77 0.77 0.78 0.70 0.92 0.93 0.87 0.89 0.91

0.44 0.40 0.50 0.50 0.51 0.51 0.59 0.89 0.89 0.85 0.93 0.51 0.49 0.64 0.60 0.54

All models calculated using 2 principal components and full cross validation.

because of differences in growth substrates, structure and microbial diversity, hence resulting in shorter estimated SL. SSSP models have been found to predict SL of naturally contaminated seafood with an accuracy of 25% (Dalgaard et al., 2002), but a higher prediction error (0 to 46%) was found in our case. This may be explained by the fact that S. putrefaciens was not the dominating SSO in the haddock fillets. The SSSP model is based on the initial concentration of the SSO in the product and as clearly explained by Dalgaard (2002) several assumptions are made regarding the growth of the SSO. For example it is assumed that other bacteria present in the microflora will not influence the growth of the SSO and that the SSO will grow without a lag phase and produce metabolites responsible for spoilage. The latter criterion was fulfilled since no apparent lag phase was observed for Sp in the haddock fillets (Table 1). Moreover, the response of the H2S electronic nose sensor correlated best with the H2S counts, suggesting the production of metabolites, but the assumption that the SSO has to reach a minimal spoilage level (107 CFU/ g) when the product is rejected by sensory analysis was not justified. In the haddock fillets the counts of H2S-producing bacteria were much lower (about log 6 CFU/g). For S. putrefaciens, cell counts of about 107 CFU/g were needed to achieve detectable TMA production in inoculated fish juice and 108 CFU/ g for off odor detection (Jørgensen and Huss, 1989). Higher counts (108 CFU/g) were observed when TMA was detected in packed cod (Dalgaard, 1995) and when TVB-N levels reached 26 mg N/100 g in sterile fish block inoculated with S. putrefaciens (Koutsoumanis and Nychas, 2000). The SSSP model also assumes that micro-organisms other than the SSO should be of no importance for spoilage, but herein the predominance of the P. phosphoreum was evident and the pseudomonads appeared to have spoilage potential as well. Therefore, this emphasizes the importance of identifying the SSOs and their spoilage domain in the respective products before applying a model based on only one SSO for shelf-life prediction.

4.2. Evaluation of multi-indicator criteria for the quality of haddock fillets — PLSR correlation and prediction modeling Evaluation of the spoilage characteristics of haddock by PLSR was done to determine which variables best described the quality degradation of haddock fillets stored at different temperatures by evaluating their correlations and potential to predict sensory quality. 4.2.1. PLSR — correlation of variables The correlation of the variables was explored by PLSR modeling using average values of all the samples (N = 22) from different storage conditions. The correlation loading plot (Fig. 5) shows that all the microbial methods are highly correlated along the first principal component (PC1) explaining 62% of the variation in the data set and describing the spoilage level of the samples. The negative correlation coefficients for the microbial variables and the Torry sensory score based on their actual values were high (r = − 0.92; − 0.96; − 0.93 and − 0.88 for TVC, H2S-, Pseud- and Pp-counts, respectively). The responses of the CO sensor correlated better with the sensory score (r = − 0.82) than the NH3 sensor (r = −0.63). The NH3 sensor responses on the other hand had higher correlation to TVB-N (r = 0.80) than the CO sensor (r = 0.64) as can be expected because of its sensitivity to amines. The variables TVB-N, NH3, pH, H2S and Tacc (Tmin = −0.1) on the upper half of the plot (Fig. 5) describe the variation in the data explained by the second principal component (PC2). Their loadings were characteristic for the spoiled samples that were influenced by high temperatures (7 and 15 °C). The variables H2S and NH3 were not significant according to the Jack-knife method to predict the Torry score. However, it is worthwhile to consider their contribution, since the PC2 is explaining 15% of the variation in the data. The H2S and NH3 sensors were best correlated to the Tacc variable (r = 0.74 and 0.75, respectively) and describing samples that were influenced by high temperature.

122

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125

The SO2 sensor had insignificant responses to the haddock fillets and as a result its location is in the middle of the plot, indicating no contribution to the model and no correlation to the Torry sensory score. 4.2.2. PLSR — evaluation of models to predict sensory quality It is of interest to study further how well the individual variables can predict the Torry scores. Table 3 summarizes the explained variance in X and Y, the correlation and the error of prediction (root mean square error of prediction, RMSEP) for different PLSR models. Sample 0 °C day 5 was badly described by the models and was determined an outlier. The sample had unexpectedly low microbial counts and high sensory score, which can only be explained by individual variation of the fillets. This sample was left out in calculations of all models. Samples 7 °C day 7 and 15 °C day 4 were also detected as outliers. The reason for this was the increased values for the variables TVB-N, NH3, pH, H2S and Tacc suggesting their high spoilage level and also the high standard deviation for the chemical variables measured. Deviation from predicted values was sometimes considerably lower when the outliers were left out of the models. However, it is important to select representative samples for the spoilage characteristics of haddock fillets, and in fact the extreme samples were characteristic for abusive conditions and were, therefore, not excluded from the models. Different combinations of variables were explored to find suitable models to predict the sensory Torry score (Table 3). The best model was obtained when using all the significant variables, indicating the importance of all the variables to predict the sensory score. However, it was of interest to explore the possibility to use a rapid technique like the electronic nose to predict the sensory quality. The use of the Tacc variable is important since time and temperature profiles are easily obtained and therefore ideal to include in models to predict the quality of fish in combination with other rapid techniques. The Tacc variable was used for calculating all models and contributed to higher correlation and lower prediction error in all cases. Models based on the electronic nose sensors had the lowest correlation (r2 ∼ 0.77) and highest error (RMSEP ∼ 0.9) (Table 3). This can be expected partly, because of the characteristic responses of the chemical indicators in fish as explained before, but also the low sensitivity of the sensors and large standard deviations experienced for volatile compounds. Models based on the microbial variables had in general higher correlations and a lower prediction error (RMSEP ∼ 0.5) in Torry score values. Taking into consideration the different information obtained in microbial and the electronic nose data, it is of interest to combine the e-nose variables and selected microbial variables. A combination of the CO, NH3 and H2S sensors with the temperature variable (Tacc) and selected SSOs showed that the best models to predict Torry score were obtained with pseudomonads or H2S counts (r2 = 0.93 and 0.92, respectively; RMSEP = 0.49 and 0.51, respectively). Therefore, it appears that the spoilage characteristics of haddock fillets stored under different temperature conditions are best described by the pseudomonads and H2S counts when evaluated in combination with the electronic nose sensors. When Pp counts or TVC were used for the prediction of Torry score, the models had lower

correlations (r 2 = 0.87 and r 2 = 0.89, respectively) and higher prediction errors. This may be related to the fact that the low NH3 sensor response inadequately reflected the high TVB-N production observed and the corresponding increasing Pp counts. For comparison with traditional methods a model based on TVB-N, TVC and the time–temperature variable gave a slightly inferior model (r2 = 0.91 and RMSEP = 0.54) (Table 3). However, when using TVB-N instead of the electronic nose sensors with the same combination of the SSO, models with higher correlations were obtained (r2 = 0.94 and RMSEP = 0.43 to 0.44). This indicates that TVB-N is highly relevant to predict the sensory quality in combination with the SSO, although, when used as a single quality criterion shorter shelf-life was predicted compared to sensory analysis. 4.2.3. Validation of the PLSR prediction of sensory quality Prediction of sensory quality (Torry score) was done using PLSR models with different subsets of the data and a test set with one sample group (0 °C + abuse) to verify the ability of the models to predict quality of “unknown” samples. After exploring different models five variables that gave the best correlation for the whole data set (three electronic nose sensors (CO; NH3 and H2S), pseudomonads counts and Tacc) were selected for a PLSR model based on a subset of the data (N = 18) with samples from 0 °C, 7 °C and 15 °C storage (2001 experiments) and samples stored at 0 °C in 2003. This model was similar (r2 = 0.94; RMSEP = 0.49) as the one calculated with the whole data set (r2 = 0.93; RMSEP = 0.49, see Table 3). Models with the other SSOs instead of the pseudomonads had lower correlation and higher errors (H2Sproducing bacteria counts r2 = 0.90 and RMSEP = 0.54; Pp counts r2 = 0.88 and RMSEP = 0.59) suggesting the importance of pseudomonads in the spoilage of haddock fillets. A biplot of the scores and loadings for the model illustrates well that the PC2 is explaining the influence of the temperature as described before and the temperature abused samples are on the upper half while the samples stored at 0°C are located on the lower half, indicating the different spoilage pattern (Fig. 6). The prediction of the Torry score for the test samples (N = 5) from (0 °C + abuse) group

Fig. 6. Biplot for the principal components of a PLSR model (r2 = 0.94; RMSEP = 0.49) based on a subset of the data (N = 18) of haddock fillets stored at different temperatures and 5 variables: H2S, CO and NH3 electronic nose sensors, pseudomonads counts and Tacc as predictors (X) for the Torry score as a response variable (Y). Samples are labeled with storage temperature and storage days.

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125

showed a good agreement between the predicted and experimental data with a minimum deviation of 0.28 in Torry score units and a maximum deviation of 0.63. The average percent difference between predicted and experimental data was 4.8%. 5. Conclusions The results showed that P. phosphoreum appeared to be very important based on their counts and growth rate in the spoilage of fresh haddock fillets packed in bulk (approximately 5 kg) under aerobic conditions in styrofoam boxes. High counts of P. phosphoreum coincided with high TVB-N values. P. phosphoreum and Pseudomonas spp. were dominating under temperature abusive conditions. Similar counts were observed for each bacterial group at sensory rejection in samples stored at 0 °C, 7 °C and 15 °C. However, microbial metabolites were produced in higher levels with increasing storage temperature as shown by the increasing responses of the electronic nose sensors and increased TVB-N values at sensory rejection. Pseudomonas spp. were suggested to contribute to the characteristic sweet, fruity spoilage odors of haddock fillets associated with high responses of the CO sensor. The spoilage characteristics of haddock fillets evaluated as Torry sensory scores were best described by pseudomonads counts in combination with the electronic nose responses when evaluated by PLSR models. The CO sensor was useful to detect incipient spoilage and characteristic spoilage of fillets stored at low temperatures while the H2S and NH3 sensors detected advanced spoilage and the influence of temperature abuse. Further development of the electronic nose instrument including a sampling system with temperature control and more sensitive sensors could improve its performance to detect the different chemical spoilage indicators. This would permit the inclusion of reliable data on microbial metabolites into shelf-life models for fish fillets along with data on different SSOs and time temperature history of the products. For accurate prediction of sensory quality and shelf-life of fish fillets combined criteria of the SSOs and their metabolites are needed. Acknowledgements The authors thank The Icelandic Centre for Research for partly financing the project, the staff at IFL for their valued contribution in chemical, microbial and sensory analyses of samples as well as the staff at Tros, Sandgerði, for their participation in the project. References Baranyi, J., Roberts, T.A., 1994. A dynamic approach to predicting bacterial growth in food. International Journal of Food Microbiology 23, 277–294. Baixas-Nogueras, S., Bover-Cid, S., Veciana-Nogues, M.T., Vidal-Carou, M.C., 2003. Suitability of volatile amines as freshness indexes for iced Mediterranean Hake. Journal of Food Science 68, 1607–1610. Botta, J.R., Bonnell, G., Squires, B.E., 1987. Effect of method of catching and time of season on sensory quality of fresh raw Atlantic cod (Gadus morhua). Journal of Food Science 52 (4), 928–931. Castell, C.H., Greenough, M.F., 1957. The action of Pseudomonas on fish muscle: 1. Organisms responsible for odour produced during incipient spoilage of chilled fish muscle. Journal of the Fisheries Research Board of Canada 12 (4), 617–625.

123

Castell, C.H., Greenough, M.F., Dale, J., 1959. The action of Pseudomonas on fish muscle: 3. Identification of organisms producing fruity and oniony odours. Journal of the Fisheries Research Board of Canada 16 (1), 13–19. Chai, T., Chien, C., Rosen, A., Levin, R.E., 1971. Detection and incidence of specific species of spoilage bacteria on fish. Applied Microbiology 16 (11), 1738–1741. Dainty, R.H., 1996. Chemical/biochemical detection of spoilage. International Journal of Food Microbiology 33 (1), 19–33. Dalgaard, P., 1995. Qualitative and quantitative characterization of spoilage bacteria from packed fish. International Journal of Food Microbiology 26, 319–333. Dalgaard, P., 2000. Fresh and lightly preserved seafood, In: Man, C.M.D., Jones, A.A. (Eds.), Shelf Life Evaluation of Foods, 2nd edition. Aspen Publishing Inc., Maryland, USA, pp. 110–139. Dalgaard, P., 2002. Modelling and predicting the shelf-life of seafood. Chapter 12. In: Bremner, H.A. (Ed.), Safety and Quality Issues in Fish Processing. Woodhead Publishing Ltd., Cambridge, England, pp. 191–219. Dalgaard, P., Gram, L., Huss, H.H., 1993. Spoilage and shelflife of cod fillets packed in vacuum or modified atmospheres. International Journal of Food Microbiology 19, 283–294. Dalgaard, P., Mejlholm, O., Huss, H.H., 1996. Conductance method for quantitative determination of Photobacterium phosphoreum in fish products. Journal of Applied Bacteriology 81, 57–64. Dalgaard, P., Mejlholm, O., Huss, H.H., 1997a. Application of an iterative approach for development of a microbial model predicting the shelf-life of packed fish. International Journal of Food Microbiology 38, 169–179. Dalgaard, P., Mejlholm, O., Christiansen, T.J., Huss, H.H., 1997b. Importance of Photobacterium phosphoreum in relation to spoilage of modified atmosphere-packed fish products. Letters in Applied Microbiology 24, 373–378. Dalgaard, P., Buch, P., Silberg, S., 2002. Seafood spoilage predictor— development and distribution of a product specific application software. International Journal of Food Microbiology 73, 343–349. Davis, H.K., 1990. Some effects of modified atmosphere packaging gases on fish and chemical tests for spoilage. Chilling and Freezing of New Fish Products, International Institute of Refrigeration, Commission C2, Aberdeen 1990-3, pp. 201–207. Debevere, J., Boskou, G., 1996. Effect of modified atmosphere packaging on the TVB/TMA-producing microflora of cod fillets. International Journal of Food Microbiology 31, 221–229. Dhananjaya, S., Stroud, G.D., 1994. Chemical and sensory changes in haddock and herring stored under modified atmosphere. International Journal of Food Science and Technology 29, 575–583. Di Natale, C., Olafsdottir, G., Einarsson, S., Mantini, A., Martinelli, E., Paolesse, R., Falconi, C., D'Amico, A., 2001. Comparison and integration of different electronic noses for the evaluation of freshness of cod fish fillets. Sensors and Actuators B 77, 572–578. Einarsson, H., 1992. Predicting the shelf life of cod (Gadus morhua) fillets stored in air and modified temperatures between −4 °C and +16 °C. In: Huss, H.H., Jacobsen, M., Liston, J. (Eds.), Quality Assurance in the Fish Industry. Elsevier Science Publishers B.V., Amsterdam, The Netherlands, pp. 479–488. Ellis, D.I., Goodacre, R., 2001. Rapid and quantitative detection of the microbial spoilage of muscle foods: current status and future trends. Trends in Food Science and Technology 12, 414–424. Emborg, J., Laursen, B.G., Rathjen, T., Dalgaard, P., 2002. Microbial spoilage and formation of biogenic amines in fresh and thawed modified atmosphere packed salmon (Salmo salar) at 2 °C. Journal of Applied Microbiology 92, 790–799. Esaiassen, M., Nilsen, H., Joensen, S., Skjerdal, T., Carlehog, M., Eilertsen, G., Gundersen, B., Elvevoll, E., 2004. Effects of catching methods on quality changes during storage of cod (Gadus morhua). Lebensmittel-Wissenschaft und Technologie 37, 643–648. European Union, 1995. 95/149/EC: commission decision of 08 March 1995 fixing the total volatile basic nitrogen (TVB-N) limit values for certain categories of fishery products and specifying the analysis method to be used. Official Journal L 097, 0084–0087, 29/04/1995. Gram, L., Dalgaard, P., 2002. Fish spoilage bacteria — problems and solutions. Current Opinion in Biotechnology 13, 262–266.

124

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125

Gram, L., Melchiorsen, J., 1996. Interaction between fish spoilage bacteria Pseudomonas sp. and S. putrefaciens in fish extracts and on fish tissue. Journal of Applied Bacteriology 80, 589–595. Gram, L., Ravn, L., Rasch, M., Bruhn, J.B., Christensen, A.B., Givskov, M., 2002. Food spoilage — interactions between food spoilage bacteria. International Journal of Food Microbiology 78, 79–97. Guldager, H.S., Bøknæs, N., Østerberg, C., Nielsen, J., Dalgaard, P., 1998. Thawed cod fillets spoil less rapidly than unfrozen fillets when stored under modified atmosphere at 2 °C. Journal of Food Protection 61, 1129–1136. Herbert, R.A., Shewan, J.M., 1976. Roles played by bacterial and autolytic enzymes in the production of volatile sulphides in spoiling North Sea cod (Gadus morhua). Journal of the Science of Food and Agriculture 27, 89–94. Herbert, R.A., Ellis, J.R., Shewan, J.M., 1975. Isolation and identification of the volatile sulphides produced during chill-storage of North Sea cod (Gadus morhua). Journal of the Science of Food and Agriculture 26, 1195–1202. Huss, H.H., 1995. Quality and quality changes in fresh fish. FAO Fisheries Technical Paper, vol. 348. Food and Agriculture Organization of the United Nations, Roma, pp. 31–50. ISO, 1993. Sensory Analysis — General Guidance for the Selection, Training and Monitoring of Assessors. Part 1: Selected Assessors, 8586-1. The International Organization for Standardization, Geneva, Switzerland, pp. 1–10. Jørgensen, B.R., Huss, H.H., 1989. Growth and activity of Shewanella putrefaciens isolated from spoiling fish. International Journal of Food Microbiology 9, 51–62. Jørgensen, B.R., Gibson, D.M., Huss, H.H., 1988. Microbiological quality and shelf life prediction of chilled fish. International Journal of Food Microbiology 6, 295–307. Jørgensen, L.V., Dalgaard, P., Huss, H.H., 2000. Multiple compound quality index for cold-smoked salmon (Salmo salar) developed by multivariate regression of biogenic amines and pH. Journal of Agricultural and Food Chemistry 48, 2448–2453. Jørgensen, L.V., Huss, H.H., Dalgaard, P., 2001. Significance of volatile compounds produced by spoilage bacteria in vacuum-packed cold-smoked salmon (Salmo salar) analyzed by GC–MS and multivariate regression. Journal of Agricultural and Food Chemistry 49, 2376–2781. Koutsoumanis, K., 2001. Predictive modeling of the shelf life of fish under nonisothermal conditions. Applied and Environmental Microbiology 67 (4), 1821–1829. Koutsoumanis, K., Nychas, G.J.E., 1999. Chemical and sensory changes associated with microbial flora of Mediterranean Boque (Boops boops) stored aerobically at 0, 3, 7, and 10 °C. Applied and Environmental Microbiology 65 (2), 698–706. Koutsoumanis, K., Nychas, G.J.E., 2000. Application of a systematic experimental procedure to develop a microbial model for rapid fish shelf life predictions. International Journal of Food Microbiology 60, 171–184. Lauzon, H.L., 2003. Notkun Malthus leiðnitækni til hraðvirkra örverumælinga. IFL Project Report, vol. 30-03. 30 pp. (in Icelandic). Lauzon, H.L., 2000. Shelf-life and bacteriological spoilage of American plaice (Hippoglossoides platessoides). In: Shahidi, F. (Ed.), Seafood in Health and Nutrition, Transformation in Fisheries and Aquaculture: Global Perspectives, a Science Tech. Book. St. John's, NF, Canada, pp. 327–354. Lauzon, H.L., Stefánsson, G., Jónsson, G.P., Sveinsdóttir, K., 2002. Sensory quality and shelf life of MAP redfish (Sebastes marinus) fillets as affected by previous bulk storage of whole fish. Journal of Aquatic Food Product Technology 11 (3/4), 251–273. Levin, R.E., 1968. Detection and incidence of specific species of spoilage bacteria on fish. Applied Microbiology 16 (11), 1734–1737. Lindsay, R.C., Josephson, D.B., Olafsdóttir, G., 1986. Chemical and biochemical indices for assessing the quality of fish packaged in controlled atmospheres. In: Kramer, D.E., Liston, J. (Eds.), Proceedings of an International Symposium, University of Alaska Sea Grant Program, Anchorage, Alaska. Elsevier Science Publishers B.V., Amsterdam, The Netherlands, pp. 221–234. Magnússon, H., Martinsdóttir, E., 1995. Storage quality of fresh and frozen– thawed fish in ice. Journal of Food Science 60, 273–278. Malle, P., Poumeyrol, M., 1989. A new chemical criterion for the quality control of fish: trimethylamine/total volatile basic nitrogen (%). Journal of Food Protection 52 (6), 419–423.

Martens, H., Martens, M., 1999. Modified Jack-knife estimation of parameter uncertainty in bilinear modelling by partial least squares regression (PLSR). Food Quality and Preference 11, 5–16. Martens, M., Martens, H., 2001. Multivariate Analysis of Quality. An Introduction. John Wiley and Sons, England. 445 pp. Martin, R.E., Gray, R.J.H., Pierson, M.D., 1978. Quality assessment of fresh fish and the role of the naturally occurring microflora. Food Technology 5, 188–192. Miller III, A., Scanlan, R.A., Lee, J.S., Libbey, L.M., 1973a. Volatile compounds produced in sterile fish muscle (Sebastes melanops) by Pseudomonas putrifaciens, Pseudomonas fluorescens, and an Achromobacter species. Applied Microbiology 26 (1), 18–21. Miller III, A., Scanlan, R.A., Lee, J.S., Libbey, L.M., 1973b. Identification of the volatile compounds produced in sterile fish muscle (Sebastes melanops) by Pseudomonas fragi. Applied Microbiology 25 (6), 952–955. Oehlenschläger, J., 1998. Suitability of ammonia-N, dimethylamine-N, trimethylamine-N, trimethylamine oxide-N and total volatile basic nitrogen as freshness indicators in seafoods. In: Olafsdottir, G., Luten, J., Dalgaard, P., Careche, M., Verrez-Bagnis, V., Martinsdottir, E., Heia, K. (Eds.), Methods to Determine the Freshness of Fish in Research and Industry. International Institute of Refrigeration, Paris, pp. 92–99. Olafsdottir, G., 2003. Developing rapid olfaction arrays for determining fish quality. In: Tothill, I.E. (Ed.), Rapid and On-Line Instrumentation for Food Quality Assurance. Woodhead Publishing Ltd, Cambridge, England, pp. 339–360. Olafsdottir, G., Jonsdottir, R., Lauzon, H.L., Luten, J., Kristbergsson, K., 2005. Characterization of volatile compounds in chilled cod (Gadus morhua) fillets by gas chromatography and detection of quality indicators by an electronic nose. Journal of Agricultural and Food Chemistry 53 (26), 10140–10147. Olafsdottir, G., Fleurence, J., 1998. Evaluation of fish freshness using volatile compounds—classification of volatile compounds in fish. In: Olafsdottir, G., Luten, J., Dalgaard, P., Careche, M., Verrez-Bagnis, V., Martinsdottir, E., Heia, K. (Eds.), Methods to Determine the Freshness of Fish in Research and Industry. International Institute of Refrigeration, Paris, pp. 55–69. Olafsdottir, G., Martinsdóttir, E., Oehlenschläger, J., Dalgaard, P., Jensen, B., Undeland, I., Mackie, I.M., Henehan, G., Nielsen, J., Nilsen, H., 1997a. Methods to evaluate fish freshness in research and industry. Trends in Food Science and Technology 8, 258–265. Olafsdottir, G., Martinsdóttir, E., Jónsson, E.H., 1997b. Rapid gas sensor measurements to predict the freshness of capelin (Mallotus villosus). Journal of Agricultural and Food Chemistry 45 (7), 2654–2659. Olafsdottir, G., Högnadóttir, Á., Martinsdóttir, E., 1998. Application of gas sensors to evaluate freshness and spoilage of various seafoods. In: Olafsdottir, G., Luten, J., Dalgaard, P., Careche, M., Verrez-Bagnis, V., Martinsdottir, E., Heia, K. (Eds.), Methods to Determine the Freshness of Fish in Research and Industry. International Institute of Refrigeration, Paris, pp. 100–109. Olafsdottir, G., Högnadóttir, Á., Martinsdóttir, E., Jónsdóttir, H., 2000. Application of an electronic nose to predict total volatile bases in capelin (Mallotus villosus) for fishmeal production. Journal of Agricultural and Food Chemistry 48 (6), 2353–2359. Olafsdottir, G., Li, X., Lauzon, H.L., Jonsdottir, R., 2002. Precision and application of electronic nose measurements for freshness monitoring of redfish (Sebastes marinus) stored in ice and modified atmosphere bulk storage. Journal of Aquatic Food Product Technology 11 (3/4), 229–249. Olafsdottir, G., Jonsdottir, R., Tryggvadóttir, S.V., Einarsson, S., Lauzon, H.L., 2003. Prediction of freshness quality of haddock fillets using electronic nose, texture and TVB-N measurements. First Joint Trans-Atlantic Fisheries Technology Conference (TAFT) 33rd WEFTA Meeting and 48th Atlantic Fisheries Technology Conference, 11–14 June 2003, Reykjavik–Iceland, pp. 148–150. http://vefur.rf.is/TAFT2003. Olafsdottir, G., Nesvadba, P., Di Natale, C., Careche, M., Oehlenschläger, J., Tryggvadóttir, S.V., Schubring, R., Kroeger, M., Heia, K., Esaiassen, M., Macagnano, A., Jørgensen, B.M., 2004. Multisensor for fish quality determination. Trends in Food Science and Technology 15, 86–93. Ross, T., McMeekin, T.A., 1994. Predictive microbiology. International Journal of Food Microbiology 23, 241–264.

G. Olafsdottir et al. / International Journal of Food Microbiology 111 (2006) 112–125 Shewan, J.M., Macintosh, R.G., Tucker, C.G., Ehrenberg, A.S.C., 1953. The development of a numeric scoring system for the sensory assessment of the spoilage of wet white fish stored in ice. Journal of the Science of Food and Agriculture 4, 283–298. Van Spreekens, K.J.A., 1974. The suitability of Long and Hammer's medium for the enumeration of more fastidious bacteria from fresh fishery products. Archiv fur Lebensmittelshygiene 25 (10), 213–219.

125

Van Spreekens, K.J.A., Toepoel, L., 1981. Quality of fishery products in connection with the psychrophilic and psychrotrophic bacterial flora. In: Roberts, T.A., Hobbs, G., Christian, J.H.B., Skovgaard, N. (Eds.), Psychrotrophic Microorganisms in Spoilage and Pathogenicity. Academic Press, London, UK, pp. 283–294.