2.3. Storage Experiment of Pre-Rigor Processed Cod Loins

Based on results from the initial experiments and existing literature [23,24,27,28,48], it was hypothesized that a combination of CO₂ and O₂ (high levels of both O₂ and CO₂) would show a greater inhibitory effect on bacterial growth than CO₂ and N₂ (with high levels of CO₂) and that vacuum packaging would allow for rapid bacterial growth. Furthermore, it was hypothesized that a short-term freezing process would eliminate a fraction of the initial bacteriota (e.g., Photobacterium) and delay microbial growth further. To test the hypotheses, industrially produced cod loins processed at Plant 3 were packed with different atmospheres as described in Section 2.3.1 (“Vacuum”, “CO₂” and “O₂/CO₂”) and with and without being subjected to a short-term freezing after packaging, followed by a thawing process/refrigerated storage.

Atlantic cod was caught at the coast of northern Norway and stored live in cage (feed-deprived) for about four weeks (Senjahopen, Norway) before slaughter. Filleting was performed within two hours after slaughtering (mean weight was 4.8 kg; headed and gutted).

The fillets were cut into loins (dorsal part of the fillet) using a Valka Cutter water jet machine (Valka, Reykjavik, Iceland). Three different packaging methods were used: (1) MAP with 60% CO₂ and 40% N₂ and with a CO₂-emitter pad (“CO₂”), (2) MAP with 60% CO₂ and 40% O₂ (“O₂/CO₂”), and (3) vacuum packaging (“Vacuum”). In addition, an industrial processed sample of the ordinary production at the processing plant were used (line caught cod at the coast of northern Norway, delivered to the processing plant immediately after catch). The industrial processed samples (designated “Industry”) were filleted, cut into loins, and vacuum packed 2–3 h after processing of the other test samples.

A thermoformer packaging machine (Multivac R145, Multivac, Germany) was used for modified atmosphere packaging (“CO₂”) and vacuum packaging at the processing plant. Drawing depth of vacuum packages was 35 mm (package volume 400 mL) and 40 mm for MAP (627 mL volume, gas/product volume ratio of 1/1). Film used for the bottom web was NICE XX 18 Black (PE/PA/EVOH/PA/PE), thickness of flat material 350 µm (Wipak Oy, Nastola, Finland). For the top web SC XX 3PA (PA/PP/PA/EVOH/PA/PE) with thickness of flat material 80 µm (Wipak Oy) was used. For the “Industry” samples SC 4 PA (PA/PP/PA/PE) was used both for top and bottom web (thickness of flat material 90 µm, Wipak Oy). The industrial processing was part of the processing plants regular production process of frozen single packed fillets.

For safety reasons, the O₂/CO₂ samples were packed at Nofima (packaging machine, Multivac T200, adjusted to high levels of O₂): the fillets were vacuum packaged at the commercial processing plant, transported in EPS Air boxes added ice, and hygienically repackaged into MAP (40% O₂ and 60% CO₂) the day after filleting. Preformed trays (50 mm depth) of HDPE (600 mL volume) and a Biaxer (PET/PE/EVOH/PE) top web were used, with similar gas/product volume ratio as for the “CO₂”. The gas mixture for the “O₂/CO₂” was measured to be 39.2% O₂ and 55.3% CO₂ in an empty package at the day of packaging.

In “CO₂” packages, a CO₂ emitter pad with 35 mL liquid absorption capacity and 230 mL CO₂ emission capacity (XP-CO2-35-230-070175-Y, 70 × 175 mm,; Mc Airlaid’s GmbH, Steinfurt, Germany) was added to achieve an optimal CO₂ availability. The CO₂ emitter develop CO₂ gas inside the package initiated by the liquid loss from the fish sample.

Liquid absorbent pad with 110 mL absorption capacity (MGS-110-070175-70, 70 × 175 mm; Mc Airlaid’s GmbH), was used for the O₂/CO₂ and the Vacuum packages.

The CO₂ and O₂ were analyzed at each sampling time by a CheckMate 9900 O₂/CO₂ analyser (PBI Dansensor, Ringsted, Denmark).

Based on results from the pretest characterizing bacterial strains from fresh and thawed samples, half of the samples (CO₂, O₂/CO₂ and Vacuum, not the Industry samples) were frozen by use of a Torry continuous air blast freezer (IQF; Individual Quick Freezing, Refrigeration Aberdeen LTD, Aberdeen, Scotland, UK) immediately after packaging. Freezing time was 40 min with air temperature monitored to be −30 °C. After freezing, the samples were placed in EPS boxes added wet ice for chill transport, in which the samples were thawed at arrival to the Nofima laboratory (Ås, Norway) the day after.

In the following, thawed samples will be designated as “CO₂-T”, “O₂/CO₂-T” and “Vacuum-T”.

The mean temperature for the thawed samples and the ordinary chilled samples during transport were −2.3 ± 0.9 °C and 0.8 ± 0.6 °C, respectively (temperature loggers used: Kooltrak GmbH, Kiedrich, The Netherlands) and it took about 24 h for the thawed samples to reach the same temperature as the chilled samples (2 °C). Sampling was performed after 1, 5, 8, 13 and 15 days (n = 3 per sampling and treatment) at 2 °C storage (loggers used: Ecolog TN4-L, Elpro-Buchs AG, Buchs, Switzerland) for bacterial analyses. Day 0 was the time of processing and packaging at the industrial processing plant. Volatile components and sensory analysis were performed on day 8 and 13, based on current time of shelf life.

Analyses of total bacterial count were performed as earlier described (Section 2.1.2) of 3 × 3 cm cut from samples (Iron agar).

Forty-five ml of the stomacher solutions were centrifuged at 13,000× g for 5 min. The pellets were frozen at −20 °C until DNA extraction using the Fast DNA-96 Soil Microbe kit (MP Biomedical) and following the manufacture’s MP-96 Inhibitor Removal Plate protocol.

PCR was performed in triplicate and paired end sequencing (2 × 150 bp) was performed using the protocol presented in Caporaso et al. [49] and as described before [50]. The library quantification and sequencing were performed at Nofima. The MiSeq Control Software (MCS) version used was RTA v1.18.54.

The forward and reverse reads were joined in QIIME version 1.9.1, and the barcodes corresponding to the reads that failed to assemble were removed. The sequences were then demultiplexed in QIIME allowing zero barcode errors and a quality score of 30 (Q30) using the QIIME toolkit [51]. The maximum and minimum number of sequences per sample was 171,601 and 64,940, respectively. Reads were assigned to their respective bacterial taxonomy using an openref Operational Taxonomic Unit (OTU) picking workflow. Reads that did not match a reference sequence were discarded resulting in 3779 OTUs with n > 2, each of these represents a phylotype and may be a representative of a bacterial species. The level 6 (genus) table derived from QIIME was used for further statistical analyses.

Dynamic headspace/GC-MS analyses of volatile organic compounds (VOCs) were performed on samples from the packages stored for 8 and 13 days. Samples were cut from the same loin as performed for the bacterial analyses and the sensory analysis. The content of volatiles was analyzed by dynamic headspace/GC-MS as described by [23,52] with small modifications to the method. The peaks were integrated, and compounds tentatively identified with MSD Chemstation software (E.02.02.1431) and NIST/EPA/NIH Mass Spectral Library (version2.0 g, 2012). Concentrations of the individual volatiles were calculated as µg/g sample based on an internal standard.

To describe the objective perception of the various samples, a trained panel performed a Quality Descriptive Analysis (QDA), ISO 13299:2016 (E) of the samples [53]. The panel consisted of 10 subjects employed exclusively to work as sensory assessors at Nofima AS (Ås, Norway). Assessors are selected based on their sensory abilities and trained according to recommendations in ISO 8586-1:2012 (E) [54]. The sensory laboratory is designed according to guidelines in ISO 8589: 2007(E) [55] and electronic registration of data (EyeQuestion^®, Logic8 BV, Utrecht, The Netherlands).

The assessors were trained and calibrated on two samples, fresh and stored cod, for the purpose to agree on the variation in attribute intensity. The samples were evaluated for the intensity of sensory odor attributes sour, seawater, cloying, fermented sour, yeast, chemical, sulfur, ammonia and pungent. Appearance was evaluated for the intensity of whiteness and glossiness (Table A3), evaluated after the odor attributes. In total, 21 loins were cut into ten pieces (one for each assessor) and served during five sessions with at least fifteen minutes break between each session. The coded samples (15 °C at time of evaluation) were served in blind trials randomized according to sample, assessor, and replicate. The samples were served to each assessor as raw samples in plastic cups with a three-digit code, with lid, with an approximate surface size of 3 × 3 cm (samples placed with skin side of the fillet towards the bottom of the cup). The intensity of each odor attribute was perceived by sniffing into the newly opened plastic cup.

The panelists recorded their results at an individual rate on a 15-cm non-structured continuous scale with the left side of the scale corresponding to the lowest intensity, and the right side corresponding to the highest intensity. The software transformed the responses into numbers between 1 (low intensity) and 9 (high intensity).

Bacterial numbers were log10 transformed and mean values and standard error of the mean calculated in Minitab (Minitab 18.1, 2017, www.minitab.com, accessed on 29 May 2019). One-way ANOVA and Fisher least square of differences at 95% confidence were used to calculated differences between means.

Logarithmic change of the selected bacteria strains on cod agar medium trays, was calculated by subtracting the initial number of bacteria from the number after incubation for each strain. Statistical tests were done using variances between strains to calculate error.

Total counts (log10 transformed) were analysed by ANOVA using gas (CO₂, O₂/CO₂, Vacuum), process (Fresh, Thawed), day, and second order interactions as effects. The industry sample was not included as a part of the ANOVA. Terms were considered significant for p-values below 0.05. Post hoc comparisons were done using contrasts for the expected marginal means, Tukey-HSD adjustment of p-values was performed for multiple testing adjustments.

The effects of experimental factors (gas, processing, days of storage) and their interactions on bacteriota and volatile compounds were tested using fifty-fifty MANOVA [56] on each of the datasets separately. Rotation tests [57] were applied for the false discovery rate (FDR) adjustment of p-values (considered significantly different when p-values below 0.05). Log transformation of the data were performed to improve normality conditions. Because there are many zeros in the data for volatile compounds, an offset corresponding to 1% of the minimum value was added to the data prior to log transformation. Bacteriota was analysed at level 6, i.e., the genus level, and only genera with an average abundance above 0.01 or a maximum of 0.1 were included in the analyses, hence the focus was on the most abundant genera, twelve genera passed the filter (Table A4).

The sensory data were analysed using a linear mixed model comprising the factors product (all seven different packaging variants), replicate and assessor and the second order interactions. Assessor and interactions involving assessors were considered random whereas the other factors were fixed.

Principal component analysis (PCA) was performed for each of the data sets from the experiment (bacteriota, sensory profiles and volatile compounds) to visualise and explore effects of the packaging concepts.

Multivariate analyses (fifty-fifty-MANOVA and PCA) were done in Matlab (Mathworks, R2018b), whereas ANOVA on total counts were done using R [58], and the sensory analyses were done with EyeOpenR (EyeQuestion Software 4.4.6, Logic8 BV, The Netherlands).