Non-Destructive Measuring Systems for the Evaluation of High Oxygen Stored Poultry: Development of Headspace Gas Composition, Sensory and Microbiological Spoilage

As poultry is known to be a perishable food, the use-by date is set in such a way that food safety is guaranteed even with a higher initial bacterial count. This means, however, that some products are wasted, even if they are still safe to eat. Therefore, non-destructive measurement devices might be a good opportunity for individual shelf-life prediction, e.g., in retail. The aim of this study was therefore to use non-destructive measurement devices based on fluorescence quenching (oxygen detection) and mid-infrared laser spectroscopy (carbon dioxide detection) for the monitoring of high-oxygen-packed poultry in different storage conditions. During 15 days of storage, the gas composition of the headspace was assessed (non-destructively and destructively), while total plate count was monitored and a comprehensive sensory evaluation was performed by a trained panel. We were able to demonstrate that in most cases, non-destructive devices have comparable precision to destructive devices. For both storage conditions, the sensory attribute slime was correlated with reaching the critical microbiological value of 107 CFU/g; the attribute buttery was also useful for the prediction of regularly stored poultry. The change in the gas atmosphere as a sign of premature spoilage, however, was only possible for samples stored in irregular conditions.


Introduction
Food waste is a global issue, estimated by the FAO (Food and Agriculture Organization of the United Nations) to be about 1.3 billion tons per year worldwide [1]. In the wealthier industrialized countries, more than 50% of food waste happens during distribution in the retail sector or at the consumer's home, often due to an expired "best before" or "use by" date [1,2]. The individual quality evaluation of an already packed foodstuff represents a major research topic of our time for different reasons. First of all, the quality standards and consumer expectations of foodstuffs are constantly increasing. To integrate an intelligent sensor, e.g., a colorimetric dye, into a packaging system can serve as both a safeguard for the manufacturer to ensure quality and at the same time to give the consumer the feeling that they are in charge of quality control. The latter, however, brings the disadvantage that consumers are often insecure in estimating the information in the right way, which can cause unnecessary food waste [3]. In addition, the risk remains that the indicator is false negative due to the lack of the relevant metabolites, for example, something that could have very negative consequences if pathogens (e.g., Campylobacter jejuni) are present [3,4]. Contrary to the earlier assumption that chicken meat is best packed in a low-oxygen (O 2) Foods 2022, 11, 592 3 of 18

Panel Training for the Sensory Evaluation
For the sensory evaluation of the samples, a panel (n = 15) consisting of 5 female and 10 male participants with an average age of 29 years was selected.
First, a descriptive analysis was carried out. Therefore, the panelists had to describe two poultry strips in their own words. The first sample was fresh slaughtered poultry and the second a forced-aged sample (4 days stored at 23 • C). A distinction was thereby made between visual and olfactory impression.
Based on the descriptive analysis, fitting references for further training (olfactory references) and for the sensory evaluation (olfactory and visual references) were selected and prepared, respectively. As olfactory references, common odor substances (e.g., diacetyl for buttery) were used. For the visual evaluation, a color chart for fresh and forced-aged poultry was designed. For this purpose, the RGB values of the meat samples were measured and transferred for both fresh and forced-aged poultry to a color chart, which was available to the panelists during the evaluation. In addition, images of fresh and forced-aged poultry were printed as image reference.
Subsequently, the named attributes were collected, and the most frequently described attributes were selected for the second part of the training, consisting of an odor identification test. For this purpose, the participants were given standardized flavors (references) corresponding to the attributes selected above, which they had to match to the correct attributes. In addition, they had to evaluate the odor intensity for each sample with the help of a line scale, ranging from weak (0) to strong (100). This was used to determine how sensitive each panelist was to the flavors.

Determination of the Headspace Gas Atmosphere 2.2.1. Non-Destructive Determination of O 2 Concentration
For the non-destructive determination of O 2 gas concentration, a fluorescence-based measurement system and associated sensor spots containing a fluorescent dye (PreSens Precision Sensing GmbH, Regensburg, Germany) were used. The measurement device works via fiber optics, exciting the embedded fluorophores at a wavelength of 505 nm. The excited fluorophores emit fluorescent light at a wavelength of 650 nm, when returning back to the ground state. Fiber optics has an integrated photodiode (which measures the luminescence lifetime of the emitting fluorophores) that is directly linked with the O 2 concentration, as already described in the Introduction section. To integrate the sensor spots into the lid film, a sensor spot was placed on the inside of the lid film, (PP/PA/PP/PA, 100 µm, allvac Folien GmbH, Waltenhofen, Germany) covered with a PP film (56 µm, Huhtamaki Flexible Packaging Germany GmbH & Co. KG, Ronsberg, Germany) and sealed with a ring-shaped sealing tool at 155 • C. Before sealing, a two-point calibration of the sensor spots in the relevant measuring range (0 and 60% O 2 % (v/v)) was carried out. All O 2 gas concentrations mentioned in the text are to be understood as percent by volume (v/v).

Non-Destructive Determination of CO 2 Concentration
The non-destructive measurement of CO 2 was carried out with a measurement system based on MIR spectroscopy (KNESTEL Technologie und Elektronik GmbH; Hopferbach, Germany). Three different wavelengths in the region of the CO 2 vibrational mode were used: λ1 = 4.26 µm, λ2 = 4.45 µm, and λ3 = 4.27 µm. The beam of a tunable diode laser was directed at 45 • through the corner of the transparent packaging. A two-point calibration with 0% (v/v) and 40% (v/v) CO 2 was carried out on the empty reference trays. All CO 2 gas concentrations mentioned in the text are to be understood as percent by volume (v/v).

Destructive Determination of O 2 and CO 2 Concentration
To estimate the measuring accuracy of the non-destructive devices, a comparison with a destructive measurement system was made. Therefore, a gas analyzer (MAT1500, A.KRÜSS Optronic GmbH, Hamburg, Germany) with a zirconium dioxide sensor for determining O 2 and a non-dispersive infrared sensor for determining CO 2 gas concentration were used. To extract small aliquots (7 mL) of the gas atmosphere, a hollow needle belonging to the measuring device was inserted into the headspace of the packaging via a septum attached to the lid film.

Microbiological Analysis
Samples were prepared for microbiological analysis as already described in Höll et al., 2016 [7], with slight modifications: In a typical analysis, a total of 70 g of chicken strips were weighed in a sample bag (VWR International, Darmstadt, Germany) and homogenized for 120 s with 50 mL Ringer's solution (Merck KGaA, Darmstadt, Germany) in a stomacher (LabBlender400, Gemini BV, Apeldoorn, Netherlands). A tenfold dilution series of chicken homogenate was prepared with Ringer's solution. An amount of 100 µL of each dilution was later spread onto the brain heart infusion agar (Carl Roth GmbH & Co. KG, Karlsruhe, Germany) using sterile glass beads. After incubating the plates aerobically at 30 • C for 3 days, the number of colony-forming units on the plates were counted and the units per gram sample (CFU/g) were calculated.

Influence of the Headspace-to Product Ratio
For the selection of an appropriate headspace:meat ratio, a pre-trial was carried out, with two different headspace:meat ratios-6:1 and 3:1. For this purpose, 200 g fresh chicken strips (Donautal Geflügelspezialitäten, Bogen, Germany) for the 6:1 and 400 g fresh chicken strips for the 3:1 ratio were weighed in transparent polypropylene trays (ES-Plastic GmbH, Hutthurm, Germany) and sealed with a semiautomatic tray sealer (T250, MULTIVAC Sepp Haggenmüller SE & Co. KG, Wolfertschwenden, Germany) under a modified gas atmosphere of 70% O 2 /30% CO 2 . Afterwards, the samples were stored at 10 • C for 11 days, as a greater change in the gas atmosphere was expected at this temperature. The O 2 and CO 2 concentrations were measured with a destructive measuring device (see Section 2.2.3) during storage.

Sample Preparation
For the main trials, a total of 400 g of fresh chicken strips were weighed into transparent polypropylene trays and sealed with the semiautomatic tray sealer at two modified gas atmosphere compositions: 70% O 2 /30% CO 2 or 80% O 2 /20% CO 2 . For each atmosphere, six packages were prepared using lid films with integrated sensor spots as described above. In addition, 44 samples were prepared for each gas atmosphere without integrated sensor materials for the destructive gas concentration measurement as well as for the sensory and microbiological evaluation. Samples were stored at 4 • C and 10 • C. Furthermore, three empty trays were prepared for each temperature and gas combination with sealedin sensor spots to monitor the concentrations of O 2 and CO 2 without product influence during storage. For destructive monitoring, two empty trays were prepared for each day of measurement and each gas concentration or storage temperature (n = 64).

Determination of O 2 and CO 2 Gas Concentration
The gas atmosphere of the prepared trays was monitored for 15 days (except for day 3 for the filled trays and days 3, 5, 6, 12, and 13 for the empty trays) via the non-destructive measurement devices for O 2 and CO 2 detection. In addition, measurements with the destructive measuring device for O 2 and CO 2 detection were carried out on days 0, 1, 4, 6, 8, 11, 13 and 15 to validate the measuring precision of the non-destructive devices for this purpose. The total viable count (TVC) was determined for each temperature and gas composition on days 0, 1, 4,6,8,11,13, and 15 in duplicate.

Sensory Evaluation during Storage
During the storage trial, the samples were investigated by the panelists on days 0, 1, 4, 6, and 8 (4 • C and 10 • C), and on days 11 and 14 for the samples stored at 4 • C. The intensity of the previously specified and trained attributes was evaluated visually and olfactorily on a line scale ranging from 0 to 100 (0 = not perceptible; 100 = strongly perceptible). In addition, it was asked whether the sample was perceived as visually and olfactorily fresh or rotten. This was also classified by a line scale ranging from 0 (fresh) to 100 (rotten). For the evaluation, a sample was defined as no longer acceptable when the average value of the olfactory or visual impression was ≥50.

Statistical Analysis
The drawings and statistical analysis were performed using MS Excel and Origin version 2021 (Origin Lab., Hampton, VA, USA). All data are presented as the arithmetic mean ± standard deviation. For the calculation of the statistical significance, two-sample ttests (two-tailed) were performed (alpha level 0.05). An analysis of variance was performed to test for equal variances prior to t-test analysis.

Panel Training
The first part of the panel training consisted of a descriptive analysis. The summed attributes of the panel are shown in Table 1 for both the visual and olfactory impressions. For the forced-aged sample, many more attributes were found. This is to be expected, as fresh poultry is rather odorless and characterized by a neutral pink coloration [20]. The named attributes for the forced-aged samples were also mostly described in the literature [20][21][22][23]. The two odors, buttery (diacetyl) and pungent (acetic acid), which are also linked with the spoilage of poultry [22,23], were not mentioned by the panel. Nevertheless, they were added to the descriptors and were part of the second part of the training. One possible reason for the missing perception of these attributes during the descriptive analysis might be the forcing at room temperature. In this case, the sulfurous impression became very intense, which probably overlapped with the presence of the diacetyl and pungent odor. For the second part of the training, the references showed in Table 2 were chosen for the odor identification test. All panelists were able to correctly match the flavors to their fitting attributes.  Table 2. Chosen descriptors for the sensory evaluation, based on the descriptive analysis of the panel training. The references and scale for the olfactory impression were also used for the second part of the training-the odor identification test.

Influence of the Headspace to Product Ratio
The measurement of O 2 and CO 2 over a storage period of 11 days at 10 • C with two different headspace:meat volume ratios showed very clearly that the headspace volume is decisive for the effect of O 2 respiration, which was previously described. In Figure 1, it can be seen that the samples with a headspace:meat volume of 3:1 experienced a considerable change in headspace atmosphere, as both O 2 and CO 2 reached a concentration between 45 and 50% on day 11. This means that the O 2 amount decreased and the CO 2 amount increased, as the starting concentration was 70% O 2 and 30% CO 2 . However, the samples with a 6:1 ratio showed hardly any change, with end-concentrations on day 11 of 66.6 ± 0.2% O 2 and 29.8 ± 0.37% CO 2 . For this reason, the following trials were carried out with a headspace:meat volume ratio of 3:1.

Non-Destructive vs. Destructive Determination of O 2 and CO 2 Concentration
To estimate the measuring precision and the suitability of the tested application, a comparison of the non-destructive and destructive measurement systems was carried out and can be seen in Figure 2. Clearly, the measuring points on day 0 and 1 for O 2 were not in agreement with the destructive measurement. That is due to the fact that the sensor spots were integrated into the packaging lid under atmospheric oxygen partial pressure, and therefore the higher O 2 concentration of the MAP had to permeate first into the spot area. For the CO 2 measurement, the values could be recorded directly in real time.
From day 4, the measured values of the non-destructive methods of both O 2 and CO 2 were mostly in agreement with the destructive measurements, with individual exceptions, which are explainable by measurement inaccuracies. However, it is also noticeable that a very high standard of deviations were partially present. Presumably, two effects must be distinguished here. If we look at the measuring point on day 11 of graph d, it becomes clear that this must be an effect in which the gas development in different packages was different, since both the O 2 and the CO 2 content have a similarly large deviation. Here, the individual development of the gas atmosphere by spoilage-which builds the basis of this research-is again clarified.

Development of the Gas Concentration in Empty and Filled Trays
Since the non-destructive measuring systems had similar accuracies compared to the destructive systems, the following results were obtained using only the non-destructive devices. The results are shown in Figure 3.
For the empty trays, almost no changes in the gas content were detected. The amount of O 2 and CO 2 increased and decreased slightly, respectively. In addition, the optical measurement method for O 2 deviated from the real values at the first two to three measurement points. This was because the sensor spots were sealed into the lid film under atmospheric conditions and the higher O 2 concentration in the MAP had to permeate into the spot area first, which was previously discussed in Section 3.3.
Indices indicate a significant difference between the curves with and without poultry: * p < 0.05, ** p < 0.01, and *** p < 0.001. The red circles mark the point when the curve of the respective gas concentration in the filled trays intersects that for the empty trays (cross-over), which indicates a microbiologically induced change in the headspace atmosphere.
By comparing the empty and filled trays, the influence of the product was determined. The samples stored at 4 • C and flushed with 70% O 2 and 30% CO 2 (Figure 3a) showed the least change, only having a noticeable cross-over on day 13 for O 2 . A "cross-over" was defined as the day when the curve of the respective gas concentration in the filled trays intersects that of the empty trays. This is an indication of a microbiologically induced change in the headspace atmosphere. For CO 2 content, however, no change in the headspace atmosphere was observed for the filled and empty trays.
The sample stored at 10 • C (Figure 3b) showed significant greater effects. For these samples, a cross-over for O 2 was detected on day 5, and for CO 2 was detected one day earlier. For both gases, the first significant deviation between empty and filled trays was present two days after having the cross-over (O 2 : day 7; CO 2 : day 6). Afterwards, a rapid decrease in the O 2 and a similar increase in the CO 2 concentration were visible. For both, the O 2 and the CO 2 measurements of the filled trays, high standard deviations were visible from day 11 on. As already described in the prior chapter, this was an effect induced by the individual development of the microbiota.
For the 80% O 2 /20% CO 2 MAP samples stored at 4 • C (Figure 3c), a significant deviation was noted between the filled and empty trays for the CO 2 measurement from days 12 to 15. However, O 2 did not show a statistically significant deviation. The cross-over was on day 12 and then the O 2 content of the filled package decreased steadily until day 15. Samples stored at 10 • C showed the earliest deviation from the empty trays, with a significant change at day 5 for CO 2 and day 6 for O 2 . The cross-over was on day 4 (CO 2 ) and 5 (O 2 ), similar to the samples packed at lower oxygen amounts. The O 2 decrease and CO 2 increase was even faster, with values almost reaching 100% CO 2 , and totally respired O 2 on day 15.
The results clearly indicate that a difference in microbiological spoilage was present. However, the fact that the samples stored at 4 • C showed such a low change in gas concentration stands in contrast to the underlying study of Höll et al., 2016 [7]. Herbert et al. showed more similar results in their study, published in 2015. In that study, the O 2 decrease and CO 2 increase were comparable for the 4 • C samples, packed under 70% O 2 /30% CO 2 MAP. The chicken stored at 10 • C had already reached its cross-over at day 2, and the O 2 decreased to a value of 20%, while the CO 2 increased to 60% on day 5 [8]. These results raise the question of whether the overall degree of microbiological spoilage has a lesser impact, and the composition of the microbiome may play a much greater role. To further answer this, the next chapter shows the determined TVCs.

Microbiological Growth
All samples had a similar starting value of the total viable count of approximately 10 4 CFU/g (see Table 3 and Figure 4). The samples that were stored at 10 • C showed a faster and more immediate growth and reached the defined critical value of 10 7 CFU/g [11] after about 3 (80/20) or 4 days (70/30). The samples that were stored at 4 • C grew significantly slower and reached this value after about 6 (80/20) or 7 (70/30) days. At the end of storage, all the samples were well above the critical limit. The highest value of >10 10 CFU/g was reached by the sample packed with 80% O 2 and 20% CO 2 , which was stored at 10 • C. However, the samples stored at 4 • C also reached final values of >10 9 CFU/g.  [20]. However, other studies showed results more similar to ours, reaching critical values between day 4 and 6 [8,21,22] for the samples stored at 4 • C, and on day 2 for the samples stored at 10 • C [8].
These results show very clearly that the type of microorganisms is decisive for O 2 consumption and CO 2 generation, not necessarily the absolute number of microorganisms. This fact may present difficulties for the prediction of shelf life.

Sensory Evaluation
The sensory panel was able to determine changes for the visual and olfactory attributes that were defined in Table 2. In Figure 5, the sensory evaluation of the samples stored at 4 • C are shown for both gas concentrations. Aside from the visual descriptor, gloss, all attributes increased during the storage time. For the visual evaluation, the overall impression (fresh/rotten), slime (not present/present) and color (pink/grey) increased the most, all signs of spoilage. For the orthonasal evaluation, the overall impression and the intensity of the odor showed the highest change. The descriptor buttery also noticeably increased. Except for the descriptor gloss, all attributes were still below 50 scores, prior to the end of the shelf life. This becomes even clearer when looking at Figure 5c,d.
For both gas concentrations, the overall olfactory impression reached the limit of 50 scores clearly after the critical value of 10 7 CFU/g was achieved. More specifically, the sample packed with a MAP of 70% O 2 /30% CO 2 was classified as rotten only on the last day of evaluation; for the sample packed with 80% O 2 /20% CO 2 , olfactory spoilage was reached on day 11. Poultry stored at 10 • C (see Figure 6) showed a very similar progression to the 4 • C stored sample but at a much faster rate. However, some attributes were even more evident than those in the samples stored at 4 • C. For the visual evaluation, the samples were classified as more greyish and greenish on day 8. For the olfactory evaluation, the descriptors buttery and cheesy/rancid were slightly higher than the 4 • C samples on day 14. However, again, by reaching the critical microbiological limit value, the impressions were below the 50 scores value and therefore, visual and olfactory were rated as still fresh.  Table 4 compares the evaluation of the sensory attributes more precisely. It is also shown which VOC is dominant or typical for the descriptor on high-oxygen-stored poultry and which microorganisms are often responsible for the emitted substances. With respect to this, it should be said that almost all meat spoilage organisms can form the typical spoilage aroma and it is assumed, that the formation of VOCs can be a result of the interactions between different spoilage bacteria, which makes it difficult to assign specific odors to a single species [24]. However, we focused on the involvement of Brochothrix thermosphacta, lactic acid bacteria (LAB) (e.g., Carnobacterium ssp., Leuconostoc ssp., Lactobacillus ssp.) and Pseudomonas ssp. because, based on previous studies, we assumed that these were the main spoilage bacteria in our chicken samples [7,22,25,26].
All samples showed a noticeable change in meat color, changing from pink to grey. For beef, this effect is related to low oxygen levels, as red oxymyoglobin turns into metmyoglobin, which is known to appear as grey-brown [27]. However, this mechanism cannot be the cause for the color change observed in poultry, as several studies have shown that lower levels of oxygen actually result in a more stable color in fresh poultry [28]. In 2017, Franke et al. also showed that storing poultry in two different oxygen-rich atmospheres resulted in similar greying effects, which is consistent with our experiments [22]. From our point of view, the greying must be an oxidation effect initiated by several microorganisms. Gloss seems not to be an attribute that can be linked with the poultry quality, as no change was observed by the panel.
The green discoloration was more present for the samples stored at 10 • C at the end of the storage, compared to the samples stored at 4 • C. However, both samples showed a significant increase during this observation. The greening of food of animal origin is known to be induced by several microorganisms [29,30]; however, several studies were able to connect Pseudomonas with this phenomenon as a result of the formation of sulfur compounds [23,[31][32][33]. Since the green discoloration was clearly increased in the descriptive analysis (see Section 3.1) and a connection with the rotten egg attribute was also seen here, this assumption is supported once again.
In general, slime formation is not a question of which microorganisms dominate, but how large the microbial count is, as the breakdown of the tissue begins at 10 8 CFUg −1 , in other words, after the sample is microbiologically spoiled [34]. This is also consistent with the panel's perception at the time that the critical limit was reached for most of the samples, as the slime formation clearly already increased in this regard. Cheesy/Rancid ++ +++ +++ +++ Isovaleric acid; (2-and 3-methyl) butanoic acid [38,39] LAB [25] B. thermosphacta [32,39] Pungent + + ++ ++ Lactic acid [40], acetic acid [41], propanoic acid [25] LAB [25] B. thermosphacta [32,35,41] The descriptor with the strongest increase-aside from the overall visual and olfactory impression-was buttery, which is linked with the VOC diacetyl (2,3-butanedione). Diacetyl is synthesized by LABs as Carnobacterium divergens or B. thermosphacta [25,42], but it can be also present for Pseudomonas inoculated samples [35]. With the exception of the 70/30 sample stored at 10 • C, the increase in the attribute buttery was rated by the panel as equivalent (++++). Therefore, it is very likely that a mixture of LAB and B. thermosphacta was responsible for this VOC formation.
Typical VOCs, which are perceived as fruity, are ethyl esters-such as ethyl acetate, ethyl butanoate, ethyl 2-hexenoate, etc.-and are very often produced by Pseudomonas [23]. As this aroma was more present for the 10 • C stored samples packed at 80% O 2 /20% CO 2 , we assume that these microorganisms were highly present, especially at the end of storage. This also agrees well with the greening previously described. However, the evaluation of the attributes rotten egg and fishy do not totally fit that presumption, as a highest increase was present here for the 4 • C 80/20 samples.
Volatile fatty acids are responsible for a cheesy or rancid odor. These are often formed by B. thermosphacta, but can also be formed during spoilage where high amounts of LABs are present [25,32,39]. This fits very well with our results, as the samples stored at 10 • C, and the 80/20 sample stored at 4 • C showed an increase in these attributes. These samples also showed a decrease in the concentration of O 2 in the headspace. This is in line with the respiration abilities of these microorganisms, especially with those of B. thermosphacta and Leuconostoc gelidum [10].
The oxygen consumption by B. thermosphacta and LABs for the samples stored at 10 • C can be further underlined by the higher increase in the attribute pungent for these samples, as the formation of acetic and lactic acid is linked with these meat spoilers [32,35,40,41]. Table 5 gives an overview of the tested parameters and indicates some observed correlations. "Correlations" are understood as mutual, temporal relationships. The correlations refer to effects (e.g., the change in gas content and microbiological spoilage) that occurred at the same or a similar point in time. Yellow describes a possible association between the cross-over and microbiological spoilage. Orange indicates a correlation between the cross-over and sensory changes. Green indicates a correlation between the change in gas concentrations in the headspace and sensory changes. Grey stands for a correlation of microbiological spoilage and sensory changes. Table 5. Observed correlations between the tested parameters. p ≤ 0.05 represents the first day where the difference between empty and filled trays was significant with p ≤ 0.05, after the cross-over was reached. "Microbiologically spoiled" indicates a TVC of 10 7 CFUg −1 and "sensory changes" shows the panel classification of the sensory panel with: >50 scores for the overall olfactory impression ("olfactory spoiled") and an increase of >20 scores in the descriptors slime, color, buttery and cheesy. A description of the color marking of the correlations can be found in the previous text.

Correlation of the Results
ples, as the formation of acetic and lactic acid is linked with these meat spoilers [32,35,40,41]. Table 5 gives an overview of the tested parameters and indicates some observed correlations. "Correlations" are understood as mutual, temporal relationships. The correlations refer to effects (e.g., the change in gas content and microbiological spoilage) that occurred at the same or a similar point in time. Yellow describes a possible association between the cross-over and microbiological spoilage. Orange indicates a correlation between the cross-over and sensory changes. Green indicates a correlation between the change in gas concentrations in the headspace and sensory changes. Grey stands for a correlation of microbiological spoilage and sensory changes. Table 5. Observed correlations between the tested parameters. p ≤ 0.05 represents the first day where the difference between empty and filled trays was significant with p ≤ 0.05, after the cross-over was reached. "Microbiologically spoiled" indicates a TVC of 10 7 CFUg −1 and "sensory changes" shows the panel classification of the sensory panel with: >50 scores for the overall olfactory impression ("olfactory spoiled") and an increase of > 20 scores in the descriptors slime, color, buttery and cheesy. A description of the color marking of the correlations can be found in the previous text. As described previously, some studies were carried out that also indicate a correlation between O2 respiration and microbiological spoilage for chicken meat [7,9]. However, this study could not completely confirm these findings. For all samples, no significant change in the gas concentration of the headspace was noted when the limit of 10 7 CFU/g was reached. Microbial spoilage might be visible with the help of the previously described cross-over point for CO2 detection; however, this is only valid for the samples that were stored at 10 °C . This point was reached either on the day (70/30) or 1 day after (80/20) the 3.7.1. Correlation of Change in Gas Concentrations with Microbiological Spoilage As described previously, some studies were carried out that also indicate a correlation between O 2 respiration and microbiological spoilage for chicken meat [7,9]. However, this study could not completely confirm these findings. For all samples, no significant change in the gas concentration of the headspace was noted when the limit of 10 7 CFU/g was reached. Microbial spoilage might be visible with the help of the previously described cross-over point for CO 2 detection; however, this is only valid for the samples that were stored at 10 • C. This point was reached either on the day (70/30) or 1 day after (80/20) the critical value of 10 7 CFU/g was reached. The suggestion that the cross-over could be an indication of premature spoilage, e.g., due to improper storage, is further underlined in the literature, where chicken samples were also stored at 70% O 2 and 30% CO 2 at 10 and 15 • C, and the cross-over of the reference and the packed sample for O 2 and CO 2 measured with a destructive device (=empty tray) were consistent in reaching their critical value [8].

°C
A significant change in the headspace gas atmosphere occurred in the samples stored at 10 • C, and also for CO 2 detection in the 80/20 4 • C samples; however, the TVC was already ≥10 9 CFU/g in these cases. For the samples stored at 10 • C, a very strong change occurred for both gases in the gas atmosphere after 5 or 6 days. At the end of the storage period, the gas atmosphere completely changed in the headspace. Nevertheless, the total microbiological growth was not very different compared with the 4 • C samples after 15 days. This is a strong indication that the type of microorganisms, and not necessarily the quantity, was crucial for O 2 consumption. This was confirmed in a study wherein beef was inoculated with different meat-spoiling bacteria. Samples contaminated with Brochothrix thermospacta showed significant O 2 consumption, while samples contaminated with Carnobacterium divergens and Carnobacterium maltaromaticum showed no consumption at all, even at microbial populations of ≥10 8 /cm 2 [10]. This shows that it is mainly the individual microbiota that determines O 2 consumption and CO 2 production, which makes a shelf-life prediction on this basis difficult. However, the cross-over seems to be a useful tool for identifying inappropriately stored samples, e.g., in case of cold-chain interruption.

Correlation of Change in Gas Concentrations with Sensory Evaluation
There appeared to be a correlation between the gas development and sensory acceptance or descriptors in some cases. For the poultry that was stored at 4 • C at initial gas concentrations of 80% O 2 and 20% CO 2 , a significant change in CO 2 was observed on day 12, and the cross-over happened on day 10. This agreed with the overall olfactory impression on day 11 when the panel classified the sample as not acceptable for the first time. The 4 • C 70/30 sample had its cross-over with O 2 on day 13 and the first classification as olfactory spoiled on day 14. However, the classification on day 11 was slightly below the 50 scores limit, which is why a prediction of shelf life via a determination of the composition of the headspace gases is rather unlikely. The overall olfactory impression was probably mostly induced by the descriptor cheesy/rancid in this case, as both attributes were rated by the panel with an increase of >20 scores compared to day 0 on the same days for each gas composition (80/20: day 11; 70/30: day 14). For the samples stored at 10 • C, however, a very good fit between the gas concentration and the sensory changes was observed. The 80/20 samples experienced their significant change in CO 2 on day 5 and for O 2 on day 6. At this time, the overall olfactory impression was scored as spoiled by the panel and the greyish color, and the descriptors buttery and cheesy/rancid increased by >20 scores. For the 70/30 sample, the gas change occurred one day later; for both gases and the sensory changes, the defined levels were reached on day 8, except for slime.

Correlation of Sensory Evaluation with Microbiological Spoilage
Several sensory attributes show a correlation with microbiological spoilage. For the samples stored at 4 • C, the descriptors slime and buttery increased by >20 scores on day 6 (80/20) or day 7 (70/30), while the critical value of 10 7 CFUg −1 was reached on day 6 for both initial gas compositions. Inappropriately stored samples at 10 • C also showed the slime formation when the critical TVC was reached. This fact demonstrates the need for other detection methods, especially for the samples stored at 4 • C, as the non-destructive gas evaluation seems not to be suitable here. The identification of slime could be possible using other spectroscopic analysis methods such as absorption spectroscopy in the infrared or Raman spectroscopy on the surface of the sample. Another possibility would be the use of hyperspectral imaging. All of these are useable as non-destructive methods for already packed products as well [12].

Influence of the Microbiome
These results allow some conclusions to be drawn about the type of spoilage microorganisms, with some limitations. Franke et al. (2017) showed that chicken breasts packed at high O 2 and stored at 4 • C were mainly populated with B. thermospacta, and developed Carnobacteria sp. and Pseudomonas sp., mainly under MAP with lower CO 2 concentrations (≤15%) [22]. Because Pseudomonas sp. are responsible for the formation of VOCs [20], this agreed well with the earlier sensory spoilage of the 80/20 sample, compared with that of the 70/30 sample. However, O 2 was hardly respired at this point, which could be because of the small population of B. thermospacta; the lack of heme, compared to beef; or because of the low temperature [9,10]. The influence of storage temperature was visible, especially with respect to sensory evaluation and gas development. In addition, this was also observed by Höll et al. (2016). They had a mixed microbiota at the beginning of storage. Later, at 4 • C, a mixture of B. thermospacta, Pseudomonas sp., and Carnobacteria sp. grew, and at 10 • C, the microbiota mainly consisted of Pseudomonas sp. and Serratia sp. at the end of storage period. After 4-8 days, B. thermospacta was present, which likely favored O 2 consumption. Then, VOC-forming Pseudomonas sp. could grow [7]. This effect was likely the same as that observed in this study.
In conclusion, the composition of the microbiome in this work can only be estimated based on the gas change in the headspace and the VOCs produced. So, a reliable statement can be made and the clarification of the species would have to be carried out in the future, e.g., using MALDI-TOF MS.

Conclusions
This study showed that the presented non-destructive measurement systems can monitor the O 2 and CO 2 concentrations in the headspace of a package in real time and showed little to no deviations to the destructive methods in which gas samples were extracted from the headspace. The measurement of the CO 2 with the infrared-based systems showed no problems. The observed variations in headspace gas composition could be attributed to individual compositions of the microbiota on the samples. The systems, however, cannot be used to predict the shelf life of high-O 2 -packed poultry stored under regular conditions and the resulting regular decay mechanisms. For the detection of premature microbial spoilage, however, e.g., due to contamination or an interruption in the cold chain, CO 2 detection might be especially useful because a significant change in the CO 2 fraction of the headspace could be observed before sensory spoilage, and the defined cross-over was correlated with reaching the critical value of 10 7 CFU/g. The predominant sensory changes identified by the successfully trained panel were overall olfactory and visual impression, the formation of slime, color change from pink to grey, and the formation of the odors buttery and cheesy/rancid. However, a correlation with the microbiological shelf life (10 7 CFU/g) was only possible with the formation of slime (4 • C and 10 • C) and buttery (4 • C). As the formation of VOCs also strongly depends on the microbiome, the detection of diacetyl (=buttery) is not a recommendable approach in our view. This is particularly true because the detection of individual substances using simple indicators can be disturbed by the presence of other substances. The optical detection of slime could instead be more promising. In addition, the influence of the heme concentration should be clarified by experiments with beef in the future, as this seems to have a main role in O 2 consumption. Another possible application for the technologies might be the detection of leakages in packages or process control for MAP production lines. For this purpose, further investigations should be carried out on the measurement accuracy of the two non-destructive methods.