Experimental design
The experiment was conducted using a 2 x 2 x 2 factorial design to evaluate three factors: chilling method (air chilling; AC vs. water immersion chilling; WC), fabrication method (bone-in vs boneless) and dark storage period (7 days vs. 14 days). Eviscerated, hot chicken carcasses (n = 256) were obtained from a commercial processing facility in California and transported to the USDA-inspected Meat Science Laboratory at the University of California, Davis (UCD; Davis, CA) within two hours of harvest. Carcasses were transported in sterile 150-quart coolers (MaxCold Cooler; Igloo Products Corp., Katy, Texas) at a mean temperature of 30.25 °C. Upon arrival at UCD, carcasses were divided into sampling groups following the scheme in Figure1. Sixteen carcasses were identified for a taste panel and placed four each in the treatment groups that were placed under 7-day dark storage (AC-bone-in, AC-boneless, WC-bone-in, WC-boneless). Of the other carcasses, 20 were sampled immediately for hot carcass samples and the remainder were randomly and evenly assigned into either AC or WC (n = 110 birds/chilling method). Following chilling (described below), 10 carcasses from each AC and WC treatment group were sampled, and the remaining were evenly assigned to fabrication pathways (n = 50 birds/fabrication) yielding either bone-in or boneless chicken breasts (n = 20 breasts/fabrication pathway). Immediately following fabrication, 10 chicken breasts from each group were sampled, and the remaining breasts were placed on expanded polystyrene trays and overwrapped with polyvinyl chloride film (40-gauge, Berry AEP1504310). Overwrapped trays were placed in rigid cardboard boxes (n = 8 trays/box) and stored at 4 °C (3-6°C) for either 7 or 14 days. At each storage interval (7 or 14 days), individual packages of chicken breasts (bone-in and boneless) were removed from dark storage. Ten breasts from each group were sampled immediately after removal from storage, and the remaining packaged breasts were placed in a retail display case (Barker, Keosauqua, IA, average light intensity 1061 LUX) maintained at 4°C (3-6°C) for 3 days.
Chicken processing
Procurement of Chicken Carcasses. A commercial chicken processing facility in California was utilized to procure hot chicken carcasses for this research. Live birds were subjected to standard poultry harvest protocols as implemented by the commercial processing facility. Carcasses used for this experiment were obtained following defeathering, evisceration, and application of an initial post-harvest antimicrobial carcass spray. Prior to chilling, the carcasses (n = 256) were removed from the processing line, placed in sterile plastic bags (n = 30-32 carcasses/bag), and bags placed in sealed sterile coolers for transportation to the Meat Science Laboratory at UC-Davis (Davis, CA). Additionally, temperature recorders (LogTag Tred30-16r; LogTag, Auckland, NZ) were placed in the coolers to monitor temperature during transportation.
Processing and Chilling of Chicken Carcasses. It should be noted that this process, while designed to mimic industrial systems, was performed on a much smaller scale. Upon arrival at the UC Davis Meat Science Laboratory, 20 chicken carcasses were randomly selected for initial evaluations (described below), while the remaining carcasses were randomly and evenly assigned to one of two chilling methods (AC or WC; Fig. 5A). Weights (g) of individual carcasses were obtained prior to chilling for comparison with weights obtained following chilling (described below). Sixteen carcasses were reserved for taste evaluation (described below) after seven days of storage and three days of retail display. This subset of carcasses was subjected to both chilling methods and fabrication methods (described below), leaving 240 carcasses for laboratory analyses. Carcasses designated for WC were submerged in one of two simulated water chill tanks (Fig. 5C). Simulated water chill tanks were constructed from commercial water tanks (Structural Form Stock Tanks, 150 gal, Rubbermaid), and a slurry of water and ice was formulated using potable water and commercial ice. Water temperature was monitored throughout chilling and birds were agitated while submerged using a paddle. Carcass temperature was monitored regularly using a thermometer (Multitrip Data Logger, Temprecord, New Zealand) probe inserted into the thickest portion of the chicken breast. When the average internal carcass temperature reached 4 °C, the chicken carcasses were removed from the water chilling system and placed on sterile wire racks for 10 minutes. Post-chilling weight and temperature were obtained after the 10-minute holding period. Additionally, five carcasses were randomly selected for analyses (described below).
To simulate air chilling, an isolated cold room in the UC-Davis Meat Science Laboratory was outfitted with a high-velocity fan (Model # BF60BDORGPRO, Maxx Air; 60’ fan with 19000 CFM, providing an airflow of 1.23 m/s.) as shown in Figure 5C. Chicken carcasses were placed on sterile wire racks located approximately six m from the commercial fan. The wire racks were rotated throughout the chilling process to assure equitable exposure to the chilling conditions. Carcass temperature was monitored throughout by inserting a thermometer probe into the thickest portion of the breast, and once the average internal temperature reached 4 °C, the carcasses were removed from the AC room. A post-chilling weight and internal temperature were obtained from individual carcasses. Additionally, 10 carcasses from each method were randomly selected for analyses (described below).
Fabrication of Chicken Carcasses, Packaging of Chicken Breasts, and Dark Storage.
Immediately following chilling, carcasses within each chilling method (AC and WC) were randomly and evenly assigned to one of two fabrication methods (n = 50/ fabrication method) for the generation of boneless (BL) and bone-in (BI) chicken breasts. Carcasses were fabricated by trained personnel in the UC Davis Meat Science Laboratory using sterile instruments and WC and AC carcasses were fabricated separately. Immediately following fabrication, 10 chicken breasts from each group were sampled, and the remaining breasts were placed on expanded polystyrene trays and overwrapped with polyvinyl chloride film (40-gauge, Berry AEP1504310). Overwrapped trays were placed in rigid cardboard boxes (n = 4 trays/box) and stored at 4 °C (3-6 °C) for either 7 or 14 days.
Retail display
At each storage interval (7 or 14 days), individual packages of chicken breasts (BI and BL) were removed from dark storage. Ten breasts from each group were sampled immediately after removal from storage, and the remaining packaged breasts were placed in a retail display case (Barker, Keosauqua, IA, average light intensity 1061 LUX) maintained at 4 °C (3 °C - 6 °C). Packages remained in the retail display case for three days. Instrumental meat color, measured via evaluating the lean color of the boneless samples and skin color of the bone-in samples, was taken using a portable spectrophotometer (MiniScan EZ; Hunter Association Laboratory Inc., Reston, VA) that was standardized before each use. A total of three readings of the International Commission on Illumination (CIE) L* (lightness), a* (redness) and b* (yellowness) values were taken using an illuminant A/10° observer for each breast. Measurements were taken through the packaging material at three separate locations on the chicken breast and were averaged prior to analyses. Packages were rotated in the display case every 12 h to assure equitable temperature and light exposure.
Microbial sample collection/processing
At each sampling point, the microbial communities of the chicken products were collected using a rinsate method. At pre-fabrication timepoints, the entire chicken carcass was placed in a sterile collection bag (Whirl-Pak; Nasco, Fort Atkinson, WI) with 200 ml of phosphate-buffered saline (PBS; National Diagnostics, Atlanta, GA) and agitated for 60 seconds to dislodge surface bacteria. After this, the carcass was removed from the rinsate and saved for physicochemical analysis. At the post-fabrication timepoints, each chicken breast was divided in half. One half was placed in a sterile collection bag (Whirl-Pak; Nasco, Fort Atkinson, WI) with 50 mL of PBS and agitated for 60 seconds. The second half was reserved for physicochemical analysis. At all time points, the rinsate was collected from the sampling bag into 50 mL falcon tubes (Corning Science, Mexico) for analysis. A 10 mL aliquot of each sample was separated to be used for aerobic bacteria population analysis and the remainder was frozen to -80 °C and transported to Colorado State University (Fort Collins, CO) PI Metcalf Laboratory for microbial ecology analysis.
The rinsate sample collected for microbiome analysis was further divided into 30 mL aliquot before DNA extraction. Cells within the rinsate were concentrated into a pellet by centrifugation at 4,600 g for 15 minutes in a swing bucket rotor (Sorvall Legend X1R; Thermo Scientific, Waltham, MA)). The supernatant was poured off and a portion of the pellet equivalent to approximately 600 uL was used for analysis. DNA was extracted from the pellet following standard protocols utilizing the Qiagen PowerSoil DNA 96 well extraction kit (Qiagen, Hilden, Germany) following manufacturers protocol for low biomass samples, which included the additional step of allowing the EB solution to be heated to 65°C before adding to the DNA plate wells for five minutes before eluting. DNA was eluted in two steps. Initially, 60 µL was eluted and considered our more concentrated DNA extraction. Next, we eluted an additional 80 µL of DNA. Each 96-well plate included 8 mock extractions (no sample added) and 1 positive control (ZymoBIOMICS D6300). The 16S rRNA gene (V4 region) was amplified using primers 515F and 806R universal primers with the forward primer barcoded to allow for multiplexing during sequencing, following the Earth Microbiome Project protocols (http://press.igsb.anl.gov/earthmicrobiome/protocols-and-standards/16s/). The forward primer 515F included the unique sample barcode following Parada et al. [35], and both primers included degeneracies as described in Parada et al. and Apprill et al. [35, 36]. Two PCR reactions using Invitrogen Platinum Hot Start PCR 2x Mastermix (Invitrogen, Carlsbad, CA) with 1 µL of DNA and a final concentration of 0.2 μM primer were run for each sample and combined to a total of 75µL. The PCR product was quantified and then pooled into a single pool in equimolar concentrations with the exception of samples that did not meet a minimum concentration, in which case 25 µL were added (this allowed the inclusion of mock extraction controls in the sequencing run). The resulting pool was cleaned using a Minelute PCR purification kit (Qiagen) and sequenced with a Miseq reagent v2 500 cycle kit at the CSU Next Generation Sequencing Core on the Illumina Miseq platform.
After sequencing, microbiome data were analyzed using QIIME2 [37] and R software version 3.5.1. Sequences were demultiplexed and quality-filtered in QIIME2 using error-correcting Golay barcodes that prevent misassignment. Reads were trimmed to 250 bp, then amplicon sequence variants (ASVs) were inferred using DADA2 [38]. Taxonomy was then assigned using the QIIME2 feature-classifier plugin [39] against the SILVA-132 99% database [40]. Non-microbial sequences that assigned to mitochondria and chloroplasts were filtered from the dataset. Samples were rarefied to 6,152 sequences and diversity metrics were calculated using the QIIME2 core metrics pipeline. Statistical comparisons for alpha diversity were made using the Kruskal-Wallis test with an alpha of 0.05 and statistical comparisons for beta diversity were made using PERMANOVA with multiple testing correction and an alpha level of 0.05. The composition of the microbiomes was compared by testing the differential abundance of taxa using the ANCOM plugin in QIIME2 [41]. The ability of microbial communities to predict quality and spoilage outcomes was assessed using the QIIME2 sample-classifier classify-samples plugin [42, 43]. Models were trained and tested using k-fold cross-validation and the Random Forest classifier with hyperparameter tuning. Visualizations were generated using QIIME2 and R software with ggplot2 [37, 44].
Phylogenetic trees
All Pseudomonas ASV sequences were extracted from the feature table by filtering based on assigned taxonomy. An alignment of all type strain 16S rRNA gene sequences for this genus was downloaded from the Ribosomal Database Project [45], along with an appropriate outgroup. A maximum likelihood backbone tree was generated using RAxML 8.2.12 [46] using the GTRGAMMA substitution matrix and 100 rapid bootstraps on the RDP alignment. An information file was then generated to be used in SATé-Enabled Phylogenetic Placement (SEPP) which was modified to fit the parsing parameters from pplacer v1.1.alpha13-0-g1ec7786 (removed 1 line according to documentation on the SEPP website https://github.com/smirarab/sepp/issues/40). SEPP 4.3.10 was then run with the following parameters (-P=33 -A 10) to optimize the alignment breakdown using the ASVs file, the RAxML tree, the RAxML info file and the reference alignment as input.
Quality measurements
Aerobic Bacteria Populations. As described by Martin et al. (2013), aerobic bacterial populations are strong indicators of the end of shelf-life. Thus, quantifying the aerobic populations present--in addition to the characterization of the microbiome--will provide insight into the shelf-life impacts of the microbial population. At each sampling interval, the carcass of one sample from each chicken was rinsed using 200 ml PBS for the carcass and 50 ml for the breast. One milliliter of the rinsate was serially diluted in 0.1% buffered peptone water (BPW; Becton, Dickinson and Company, Sparks, MD) and plated in duplicate onto Petrifilm aerobic count plates (3M Microbiology, St. Paul, MN). Plates were then incubated at 7 °C for 10 days and 35 °C for 48 h.
Physicochemical Analysis. Numerous biochemical and physicochemical changes that affect shelf-life occur in post-harvest meat products [47]. Thus, an assessment of these changes during processing was conducted to obtain information regarding product quality. At each sampling point, after the rinsate was collected, the sample was fabricated to a boneless breast if it was not already, though the skin was left on for carcass and bone-in samples. Then, the breast was flash-frozen in liquid nitrogen and homogenized using a blender (Magic Bullet; Capital Brands, Los Angeles, CA). To evaluate the carcass samples and bone-in breasts, the breast meat was removed from the bone at the time of sampling. The frozen homogenate was then transported to the Colorado State University Center for Meat Safety and Quality (Fort Collins, CO) for physicochemical and compositional analyses.
Fatty Acid Composition. Fatty acid composition was obtained using gas chromatography following methods described by Engle et al. and Kang and Wang [48, 49]. First, fat was extracted using the Folch method (Folch, 1957). One gram of the frozen homogenate was combined with 20 mL of 2:1 chloroform:methanol mixture and homogenized, then filtered using Whatman No. 1 filter paper (Fisher Scientific; Waltham, MA). Then, 4 mL of 0.9% NaCl solution was added per 20 mL chloroform:methanol and the solution was incubated at 4 °C overnight. During this time the solvent separated into two phases; the lower phase contained the lipid extract, which was separated and dried in a dry matter oven at 100 °C for 16 h. After this extraction, the lipid extract was methylated by adding 1 mL of 0.5 N KOH in MeOH and heated in a water bath. Samples were then prepared for gas chromatography by mixing with 2 mL HPLC-grade hexane and 2mL saturated NaCl, which was then back-extracted and reconstituted to concentrate the fatty acids. The reconstituted lipid was measured by gas chromatography (Agilent 6890 plus; Agilent, Wilmington, DE) with standard fatty acid methyl ester mixtures and SUPELCO FAME standard (Millipore Sigma, Darmstadt, Germany) to calibrate. Fatty acids were identified by matching relative peak retention times to those of the standards, calculated as normalized area percentages of fatty acids.
Lipid Oxidation. Lipid oxidation was measured using the thiobarbituric acid assay (TBARs), as described in Yin et al. [50]. Briefly, 5 g of the frozen homogenate was mixed with trichloroacetic acid, homogenized using a standing homogenizer, and filtered using Whatman No.1 filter paper (Fisher Scientific; Waltham, MA). A 1mL aliquot of the filtrate was mixed with 1 mL of 10 mM thiobarbituric acid and incubated at 25 °C for 20 h, after which the absorbance at 532 nm was measured using a spectrophotometer (UV-2401, Shimadzu Inc., Columbia, MD).
Proximate Analysis. Nutrient composition analysis (proximate analysis) was conducted to determine the dry matter, moisture, ash, crude fat, and crude protein composition within each sample. Dry matter and moisture were measured using the AOAC oven drying method, 950.46 and 934.01[51]. Two grams of frozen homogenate were weighed and placed in a standard laboratory convection oven for 24 h at 60 °C, then re-weighed. Percent moisture was calculated using the formula: % moisture content = [[(wet weight - dry weight) / wet weight] * 100]. Percent dry matter was calculated as 100 - moisture content. Ash content was determined using the ash oven method as described in the AOAC 923.03 and 920.153 [52]. Approximately 1 g of the frozen homogenate was placed into a dry crucible, then inserted into a Thermolyne box furnace at 600 °C for 18 h. Percent ash was calculated using the formula: % ash = (ash weight / wet weight) * 100. Crude fat was measured using the Folch method, as described above. Finally, crude protein was measured following AOAC method 992.15 [53], which used a TrueSpec CN nitrogen determinator (LECO, St. Joseph, MI). Percent protein was calculated using the formula: % protein = total % N * 6.25. Results were represented on a dry matter basis. Statistical analyses on all physiochemical tests were conducted using ANOVA and the emmeans package [54] with a 2 x 2 x 2 factorial design with an alpha level was 0.05.
Sensory Analysis.
Eight untrained participants were asked to evaluate the acceptability of three sensory attributes (color, odor, texture) during retail display using a three-point sensory scale described by Lytou et al. [55] . In addition, subjective color (desirable, acceptable, unacceptable) and willingness to purchase (would purchase, would not purchase, would purchase at a discounted price) was evaluated by these untrained panelists every 12 h during retail display. At the end of each three-day retail display period (10 day and 17 day) the same participants were then asked to evaluate subjective odor, texture and purchase selection using the same approach. Evaluation scores were analyzed as continuous data using mixed procedures of SAS (version 9.4; SAS Inst. Inc., Cary, NC). Participants were treated as random variables and the alpha level was defined as 0.05.
In addition to evaluation of chicken breast color, odor, and texture, trained taste panelists were asked to evaluate various palatability attributes (chicken flavor intensity, off-flavor intensity, springiness, cohesiveness of mass, and moistness) of bone-in and boneless chicken breasts representing both chilling methods following 7 days of dark storage. Panelists consisted of graduate students from the CSU Center for Meat Safety and Quality and were trained to recognize the aforementioned attributes using methods and references described by Solo [56]. Samples for evaluation were randomized and panels were conducted over two days to avoid sensory fatigue. Chicken breasts were cooked to an internal temperature of 76 °C and cut into 2.54 x 2.54-cm cubes before being served to panelists under red lights. Panelists then ranked each breast portion for each of the above attributes on a 100-point scale. Data were analyzed using an ANOVA and the emmeans package in R [54] with an alpha level of 0.05.
Chilling system techno-economic analysis
An economic evaluation of each chilling system, AC and WC, was also performed. The work included the development of baseline models of each system that allowed for a direct comparison on the metric of economics. The baseline models used the same system boundaries that limited this techno-economic assessment to the chilling process and used harmonized model inputs when possible for consistency. Some facilities include maturation as an extension of the chilling process; however, these baseline models don’t include anything outside the chilling process. The models used standard nth plant economic assumptions from literature and assume 10% internal rate of return (IRR), 20-year facility life, 8% loan interest rate on a 10-year loan with 40% equity, and the 2019 U.S. corporate tax rate of 21% [57–59]. The above economic assumptions were combined with capital costs, operational costs, linear depreciation, and poultry processing rate to perform a 20-year discounted cash flow rate of return (DCFROR) for each poultry chilling system. These models use the IRR as the discount rate to determine the minimum processing cost ($/metric ton) associated with poultry chilling while providing a net present value (NPV) of zero. This minimum processing cost represents a levelized cost of chilling poultry carcasses that supports a 10% IRR over the 20-year life of the system.
All baseline values were taken from literature or acquired through communication with poultry chilling equipment manufacturers and poultry processing facilities. In particular, the system layout and energy consumptions reported in the literature were found to be antiquated and thus most of these data were acquired through communications with industry. The mutual baseline inputs were plant operation (250 d per year, 16 h per day), poultry processing rate (16.5 tonne/hour), water treatment ($1.5/m3; [9, 34, 60, 61]) and electricity ($0.10/kWh) prices, and fixed annual maintenance cost (5% of total capital cost). While AC and WC fixed and variable labor requirements might vary slightly, the labor requirements were assumed to be equivalent for the baseline cases for both systems. Based on input from a chilling equipment suppliers, the AC and WC models reflect the major differences between the two systems (WC, AC); floor space requirements (100 m2, 500 m2), water use (3200 L/tonne; 300 L/tonne), and chilling energy costs (20.9 kWh/tonne, 31.4 kWh/tonne). Due to the complexity and high variability in WC and AC system designs and operational characteristics, sensitivity analyses were used to evaluate how system parameters can impact the levelized cost associated with chilling poultry carcasses. The sensitivity analysis varied the following model values by ±50%: capital cost ($/tonne), chilling floorspace (m2), processing rate (tonne/hour), chilling energy (kWh/tonne), worker capacity (tonne/hour), water use (L/tonne), water price plus water treatment cost ($/L), and electricity cost ($/kWh). The end result is a direct comparison of the two technologies in terms of costs with sensitivity used to highlight high impact variables.