A Fresh Look at Ground-Beef Spoilage: Determining Spoilage Thresholds Using Consumer Evaluation and Microbial Analysis

Sample collection

This study used two-hundred and fifty-six 454-g ground-beef (80% lean) packages obtained from a commercial case-ready facility located in the Midwest. Prior to display, all samples were transported to the KSU meat laboratory under refrigerated temperatures (2–4°C) at a single time point. Four individual ground-beef packages were stored in each trigas-flushed mother bag (69.6% nitrogen [N], 30% carbon dioxide, 0.4% CO) and stored at 2°C to 4°C in the absence of light. For each mother bag, paired packages were randomly designated to a display period (0, 2, 4, 6, 8, 10, 12, or 14 d). One sample from each pair was designated for sensory appearance and taste analysis, while the other paired sample was designated for sensory evaluation of odor and touch as well as microbial, lipid oxidation, objective color, and microbiota analysis.

Retail cases were divided into 9 sections (3/case), with samples placed into the cases in a predetermined fashion so that each case section contained 1 pair of samples from each of the 8 display periods on the day of evaluation by consumers. On display day 0 for each display period, samples were placed into the designated section in 3 coffin-style cases (model DMF8; Tyler Refrigeration Corp., Niles, MI) at 2°C to 4°C under continuous fluorescent lights (32 W Del-Warm White 3,000 K; Phillips Lighting Company, Somerset, NJ) averaging 2,143 ± 113 1× emission casewide. The cases were programmed to defrost once per day in the morning to avoid any fogging of packages during evaluation times. Samples were placed in the case at the same time every other day and rotated every 24 h.

Trained sensory panel evaluation

Trained sensory panelists (n ≥ 8) evaluated appearance, odor, touch, and taste of all samples. Sensory panelists were trained according to the American Meat Science Association (AMSA) meat color measurement and sensory evaluation guidelines (AMSA, 2016; King et al., 2023). Sensory panelists were trained on 100-point line scales anchored with 2 end points for all analyses. All visual sensory panelists were subjected to and passed the Farnsworth–Munsell 100 hue color vision test (Munsell Color X-Rite, Grand Rapids, MI) to screen for color blindness. The trained sensory panelists were asked to record their answers on electronic tablets (Fire HD 8; Amazon, Seattle, WA) using a digital survey (Qualtrics Software, Provo, UT).

For visual panels, panelists evaluated the appearance traits of the samples (those sensed visually), which included the discoloration and redness of samples, using methods and anchors outlined by Lybarger et al. (2023). During training sessions, panelists were trained using samples varying in discoloration and redness that represented the entire scale. On evaluation days, panelists were given an hour time period to evaluate samples, approximately 3 h prior to consumer evaluation. A minimum of 8 trained sensory panelists evaluated 32 samples on each day.

Following the conclusion of the visual evaluation, packages designated for odor and touch evaluation were removed from the cases and prepared for evaluation. For touch and odor sensory panels, 2 approximately 2.54-cm by 2.54-cm pieces of the ground-beef loaves were placed in separate 4-oz quilted crystal glass jelly jars (Ball, Westminster, CO) and stored at 2°C to 4°C prior to evaluation. One jar was allocated for touch evaluation, and the other was assigned to odor evaluation. Panelists were trained in multiple sessions using samples that represented a range of display periods and variations in touch characteristics, with attention given to differences related to dryness and sliminess. Touch was evaluated on a 100-point continuous line scale with verbal anchors at both end points, with 0 representing characteristic beef touch and 100 representing noncharacteristic beef touch. Panelists were given a multiple-choice question asking, “If the touch is noncharacteristic, what does the sample feel like?” The answer choices provided were slimy, dry, tacky, or an option to type in their own answer. The samples were evaluated in individual sensory booths under low-intensity (<107.64 lumens) red incandescent lights to mitigate bias related to sample visual appearance.

Once trained sensory panelists touched all samples, they were asked to evaluate the odor of the samples from the paired glass jars. Panelists were trained on odor using a verbally anchored 100-point continuous line scale in conjunction with the touch training sessions and included samples ranging from 0 d to 14 d of retail display and samples that were sour, rancid, and oxidized. The line scale consisted of 2 verbal anchors, where 0 represented no off odor present and 100 represented an extreme off odor present, with a “nonapplicable” option for use with samples with no off odor present. Sensory panelists were asked to describe the odor with a multiple-choice question stating, “If there is off odor present, what does it smell like?” The answers provided were rancid, sour, oxidized, or the option to type in their own answer. For both touch and odor panels, panelists evaluated samples from all days of display in a random order.

Following visual evaluation, loaves designated for taste panels were divided in half, vacuum packaged, and frozen (−20°C) until sensory evaluation. Trained sensory panelists for taste evaluation were familiarized with scales and trained in 6 sessions in the weeks immediately prior to evaluation using the methods and anchors described by Davis et al. (2021). Prior to cooking, samples were thawed 24 h in advance at 2°C to 4°C and split in half, creating 2 approximately 113-g portions and then pressed into 1.2-cm thick patties using a handheld patty maker. Patties were cooked to an end-peak temperature of 82.2°C on a Cuisinart Griddler Deluxe clam-shell style grill (Stamford, CT) and set to a surface temperature of 177°C. To reach the targeted end-point temperature, patties were pulled from the grill at 79.4°C, assessed by a Beckman Industrial Doric 205 thermocouple thermometer (Brea, CA). Patties were cut into 8 equal-sized wedges and served to trained panelists in a random order. At the beginning of each panel session, a representative ground-beef sample was evaluated to allow for panel calibration and prevention of panelist drift. Each sample was evaluated for juiciness, tenderness, texture, beef flavor intensity, and off-flavor intensity. Trained sensory panelists evaluated 1 sample from each of the display periods, in a random order, in individual sensory booths in the same low-intensity red-light conditions as when evaluating touch and odor. Water, apple slices, and unsalted crackers were provided as pallet cleansers along with an expectorant cup and napkin.

Consumer sensory panel evaluation

Consumer sensory panelists (N = 128/panel type) were recruited from Manhattan, Kansas and neighboring communities and were monetarily compensated for their participation. Consumers were asked to visually evaluate samples in the retail case, as well as smell and touch samples in sensory booths under the same red-light conditions as the trained sensory panelists to prevent any bias against the color of the samples for these evaluations. For all sensory evaluations, each consumer evaluated 8 samples, 1 from each day of display (0, 2, 4, 6, 8, 10, 12, and 14 d) in a random order. A continuous 100-point line scale was provided to assess the desirability of the sample as described by Lybarger et al. (2023) and Davis et al. (2021). Consumers were asked a yes/no question if they would purchase the package based on the appearance, touch, odor, or taste, and a yes/no question if they would consider the package spoiled based upon the appearance, touch, odor, or taste.

Consumer samples for odor and touch evaluation were prepared at the same time as the preparation of samples for trained sensory panel evaluation. Similar 2.54-cm by 2.54 -cm pieces were placed in separate glass jars, tightly sealed, and stored at 2°C to 4°C in the absence of light for approximately 2 h prior to evaluation. In sensory booths under red-light conditions, consumers were given the jars without the lids and asked to touch the sample by gently feeling the surface to evaluate the touch of the sample. Then, consumers were handed an additional 8 glass jars with the lids on to assess the odor of the samples. Consumer panelists were asked to remove the lid briefly and smell the sample and then to place the lid back on the sample before passing it to the next consumer to reduce the amount of time each sample remained with the lid off.

Consumer taste evaluation was conducted on the previously frozen samples following approximately 4 mo of frozen storage (−20°C). The same thawing and cooking protocol was followed as outlined for trained sensory panel analysis. Consumers evaluated samples under fluorescent lighting in a large lecture-style classroom. Four sessions of 24 consumers and 2 sessions of 16 consumers were conducted within the same week. Each panel lasted approximately 1 h. In addition to their electronic ballot, each consumer was supplied with water, apple juice, and unsalted crackers to cleanse their pallets as well as an expectorant cup and napkin.

Objective color measurements

Before each trained sensory panel and approximately 30 min after being placed in the retail cases, instrumental color was measured on each sample. Commission Internationale de L’Éclairage L*, a*, and b* values were collected using a Hunter Lab MiniScan spectrophotometer (Illuminant A, 2.54-cm aperture, 10° observer; Hunter Lab Associates Laboratory, Reston, VA), using methods outlined in the AMSA color guidelines (King et al., 2023). Three separate scans were taken on the surface of the ground-beef sample designated for visual evaluation, and the values were averaged. Moreover, spectral data were recorded to calculate hue angle, chroma, percent oxymyoglobin (OMb), and percent metmyoglobin (MMb) according to the AMSA color guidelines (King et al., 2023).

pH measurement

On each evaluation day, the day 0 samples designated for laboratory analysis were measured for pH using an InLab Science Pro-ISM pH probe attached to a Seven Compact pH meter (Meter Toledo, Columbus, OH), calibrated according to manufacturer’s specification previously outlined by Hammond et al. (2022). Five grams of each sample, in duplicate, previously frozen in liquid N and powdered, were weighed into a 100-mL beaker. Fifty mL of Milli-Q water was then added to each beaker, and the sample was mechanically homogenized for 20 s at 10,000 rpm (Homogenizer 850; Fisher Scientific International, Hampton, NH). The pH of the sample was read and recorded.

Microbiological analysis

Microbiological analysis was conducted on all evaluation days to determine aerobic plate counts (APC), Enterobacteriaceae counts (EB), and Escherichia coli/coliform plate counts (ECC). Samples from 0, 2, 4, 6, 8, 10, 12, and 14 d were evaluated. Upon arrival at the laboratory, 25 g of each paired sample were aseptically collected by compositing small aliquots from all 4 corners and the center of the package, ensuring that surface material was included. Each 25-g aliquot was combined with 225 mL of peptone water (PW) and stomached for 60 s (Stomacher 400, Seward, Bohemia, NY). Serial dilutions were prepared by transferring 1 mL of ground-beef homogenate into 9 mL of PW, and creating a 10-fold dilution. Subsequent dilutions were prepared as follows: 10⁻², 10⁻⁴, 10⁻⁶, and 10⁻⁸. Each dilution was plated in duplicate on 3M^™ Petrifilm^™ aerobic count (AC), EB, and ECC count plates by transferring 1 mL per dilution to each plate. Petrifilm AC plates were incubated at 35°C plus or minus 1°C for 48 h (Incubator F Air 6.3CF; VWR, Radnor, PA). EB and ECC plates were incubated at 30°C plus or minus 1°C for 24 h. Following incubation, AC and EB plates were enumerated according to the manufacturer’s instructions. ECC plates were evaluated according to the manufacturer’s protocol and incubated at 35°C plus or minus 1°C for 24 h plus or minus 2 h to enumerate coliforms, followed by an additional 24 h plus or minus 2 h to differentiate E. coli from other coliforms. Colonies were counted from plates with 25 CFU to 250 CFU per standard microbiological practices.

Lipid oxidation

Lipid oxidation was determined using the thiobarbituric acid reactive substances (TBARS) assay following the procedures outlined in Ahn et al. (1998) and similar to previous procedures done at KSU (Dahmer et al., 2022; Beyer et al., 2024). In summary, approximately 0.2 g of a previously powdered sample was weighed and added to 2 mL bead tubes with 1.4 mL of thiobarbituric acid/trichloroacetic acid, and 0.1 mL butylated hydroxytoluene, homogenized (D2400 Homogenizer; Benchmark Scientific, Edison, NJ), and then centrifuged at 3,000 × g for 5 min. Supernatant was removed from the sample tube and placed in a separate tube that was then covered with aluminum foil and incubated for 30 min at 70°C. Samples were chilled in an ice bath for 5 min and placed into a 96-well plate along with 0.2 mL of malondialdehyde (MDA) concentration standards and read using a spectrophotometer and 532 nm. The final MDA concentration was calculated and expressed as μM malonaldehyde using the standard curve developed.

DNA extraction and microbiota sequencing

A microbial DNA sample was extracted from each package of ground beef using Qiagen DNeasy Mericon Food standard protocol (Qiagen, Germantown, MD). Two-gram samples of ground beef were combined with a food lysis buffer and proteinase K solution, vortexed, then incubated (MaxQ Shaker; ThermoFisher Scientific, Waltham, MA) for 1 h with constant shaking at 100 rpm at 60°C. Samples were cooled on ice for 5 min and centrifuged at 2,500 × g for 5 min. Supernatant was removed from the sample tubes, combined with chloroform, vortexed, and centrifuged at 14,000 × g for 15 min. The upper aqueous phase of the sample was removed and added to a phosphate buffer and mixed thoroughly. The sample was placed into a QIAquick spin column and centrifuged at 17,900 × g for 1 min. An AW2 buffer was added to the spin column and centrifuged with the previous settings described. After each centrifuge, the flow through was discarded. The spin column was placed in a new 2-mL tube, and DNA was eluted using an EB buffer in the final centrifugation at 17,900 × g for 1 min. Samples were stored at −80°C until further analysis.

Three negative controls (i.e., no samples added to lysis tube) were included at the DNA extraction step to detect potential contaminants in the kits. Additionally, 3 positive controls (ZymoBIOMICS Microbial Community Standard; Zymo Research, Irvine, CA) were included in the DNA extraction step to verify the extraction efficiency. DNA concentration was quantified with high-sensitivity double-strand DNA kit (Thermo Fisher Scientific) using the Qubit 2.0 Fluorometer (Thermo Fisher Scientific) and sent for amplification of the 16S ribosomal RNA gene V4 region, library preparation, and sequencing to Novogene Co. (Sacramento, CA). Quality control of amplicons was conducted using an Agilent 5400 automatic capillary electrophoresis system (Agilent Technologies, Palo Alto, CA). Polymerase chain reaction (PCR) products were pooled in equimolar concentration, end-repaired, A-tailed, and ligated with Illumina adapters. The quality of the pooled library was then quantified using quantitative PCR. After quality control, libraries were sequenced on an Illumina NovaSeq 6000 using 250-bp paired end reads. The resulting sequences were analyzed using DADA2 (v. 1.30.0) in R (v 4.3.2) with quality filtering (Q2), truncation set to 220 bp, and the default overlap for sequence merging, followed by chimera removal. Sequences were truncated to 220 bp to remove the last 30 bases that were of lower quality.

Compositional data analysis framework was used in R (v 4.3.2) to characterize the bacterial microbiota composition of ground-beef samples from 8 different display periods (Gloor et al., 2016; Gloor et al., 2017). All amplicon sequence variants (ASV) with “0” count values were assigned a small nonzero value with the count zero multiplicative method, using the R package zCompositions (v 1.5.0.4; Palarea-Albaladejo and Martín-Fernández, 2015) prior to applying a center-log ratio (CLR) transformation. To determine whether there was a significant effect of the day of display on the microbiota composition, Aitchison distances were calculated using CLR-transformed data, and pairwise permutational multivariate analysis of variance was used with a linear model that included the effect of the day of display using the R package pairwiseAdonis v0.4.1, with 999 permutations. Specifically, the composition of microbiota samples on display for days 0, 4, 6, and 14 were compared. These days were selected, as consumer purchase intent and spoilage rating predictors suggested samples began to drastically change after day 4 and completely inverted by day 14. P values were corrected for multiple comparisons using the Bonferroni correction. Graphs were plotted in R using ggplot2 (v 3.5.1).

Statistical analysis

The statistical analyses for all sensory data were performed using the procedures of SAS (SAS Inst. Inc., Cary, NC). PROC GLIMMIX was used with day of display comparisons considered significant with an α of 0.05. Paired ground-beef samples served as the experimental unit with data analyzed as a completely random design, with day of display as the fixed effect. Microbial data were analyzed by log transforming all plate-count data and were analyzed using the same model as the sensory and objective measurement data. Using PROC LOGISTIC, logistic regression models were calculated for the probability of a sample being identified as “would purchase” and “spoiled” by consumer sensory panelists. PROC CORR was used to determine Pearson correlation coefficients for sensory and objective measurements. The Kenward–Roger adjustment was used in all analyses.

Demographics

Demographic information for the consumer panelists who participated in appearance, touch, and odor sensory panels, as well as taste panels is located in Table 1. Both groups of consumers were the same size (N = 128). For appearance, touch, and odor panels, consumers were split between 55.2% males and 45.8% females; whereas, the split between males and females was equal (50%) for consumer taste sensory panels. The majority of both groups of consumers were single (69.6% and 54.3%) and reported a household size of 1 or 2 people (70.4% and 62.2%). The majority of both groups were Caucasian (93.2% and 78.0%). The majority of the appearance, touch, and odor consumers reported an annual income less than $50,000 USD (51.2%), while the majority of the taste panel group reported an annual income greater than $50,000 USD (52.7%). Both groups of consumers were educated with some college, a college degree, or a postcollege degree (81.6% and 80.3%), with most consumers under the age of 40 y (69.6% and 58.2%). Moreover, the majority of the consumers reported they consumed beef 1 to 6 times per week (69.6% and 78.8%) and purchased beef in the retail setting 1 to 6 times per month (70.4% and 76.4%).

pH, objective color measurement, and lipid oxidation analysis

The average pH of the ground-beef product used in the study was 5.95 with a SD of 0.16. A summary of the overall mean, minimums, maximums, and variation for all independent variables evaluated is provided in Table 2 and represented a wide range of variation in all variables evaluated to allow for robustness and increased inference space for the calculated models. The least-squares means from objective color measurements are presented in Table 3. L*, a*, and b* values were all the highest for day 0 samples and decreased (P < .05) as the day of display increased. However, a* values increased (P < .05) from day 12 to day 14 of display. Similarly, the calculated percentage of MMb increased (P < .05) from day 0 to day 10, but decreased (P < .05) from day 10 to day 14, where there was also an increase (P < .05) in the calculated percentage of OMb. Chroma values followed a similar trend, with decreasing (P < .05) values from day 0 through day 10, but increasing (P < .05) from day 10 to day 14; whereas, hue angle increased (P < .05) from day 0 through day 10 before decreasing (P < .05) from day 10 through day 14. For lipid oxidation (TBARS), there was a steady increase (P < .05) in oxidation throughout storage, with day 0 samples averaging 0.22 MDA/kg of ground beef and day 14 samples averaging 0.43 MDA/kg of ground beef (Table 3).

Trained sensory analysis

The least-squares means for the traits evaluated by trained descriptive panelists are presented in Table 4. For appearance ratings, redness scores decreased (P < .05) as display time increased from day 0 through day 10 (0 > 2 > 4 > 6 > 8 = 10 d). No difference (P > .05) in redness was found between day 12 and day 14 samples, which were lower (P < .05) than samples displayed for 6 d or less. Discoloration percentage increased (P < .05) from day 0 (1.9%) through day 10 (98.2%), with day 8 to day 12 samples all having greater than 89% discoloration. For odor, from day 2 to day 14 odor intensity increased, with samples from day 10 to day 14 rated greater (P < .05) than samples from day 2 and day 4. Day 14 samples had a greater (P < .05) odor intensity than all other display periods, with the exception of day 0, which had a greater (P < .05) odor intensity than samples from day 2 to day 12, likely due to their previous storage in the gas-flushed mother bags. For touch, few differences were found among display days, with day 12 and day 14 samples having more (P < .05) noncharacteristic beef feel than samples displayed from 4 d to 10 d. For taste characteristics, there were no differences (P > .05) among storage periods for either beef flavor intensity or off-flavor intensity; however, differences among aging times for off-flavor intensity were marginally significant (P = .08), with day 14 samples having the highest numerical mean among all other display days. Differences (P < .05) were observed among samples across days of display for juiciness, tenderness, and texture traits, though most were low in magnitude (<5 U) across the entire display period.

Consumer sensory analysis

Table 4 shows the least-squares means for consumer ratings for the appearance, touch, odor, and taste of retail ground-beef samples. For all sensory traits excluding taste, day of display had a negative impact on consumer ratings. For appearance, consumers rated day 0 and day 2 samples similar (P > .05) and greater (P < .05) than all other days, but ratings decreased (P < .05) with extended display (0 = 2 > 4 > 6 = 8 = 10 = 12 d), with the largest decreases in ratings observed between day 4 and day 8. Similar to the objective measurements, consumer appearance ratings for day 14 samples were greater (P < .05) than day 8 to day 12 samples. For consumer touch and odor ratings, there was no difference (P > .05) among day 1 to day 4 samples, but these days had greater (P < .05) ratings than day 8 to day 12, which were similar (P > .05), with day 14 samples rated the lowest (P < .05) for both traits among all display days. For consumer taste ratings, day of display had no impact (P > .05) from day 0 through day 12, with only day 14 samples rated lower (P < .05) than all other days of display.

Table 5 presents the percentage of samples classified as “would purchase” and “spoiled” by consumer sensory panelists across all days of display. For appearance, day 0 and day 2 samples had the greatest (P < .05) number of samples classified as “would purchase,” with over 96% of samples on these days classified as such. Similar to the consumer appearance ratings, this percentage decreased (P < .05) with increased display time, with the largest decrease occurring from day 4 (88.6%) to day 6 (31.5%) to day 8 (6.85%). From day 8 through day 14, there was no difference (P > .05) in the percentage of samples consumers indicated they would purchase, with each day having less than 7% classified as “would purchase.” The inverse trend was observed for the percentage of samples classified as “spoiled” by consumers. From day 0 through day 4, no difference (P > .05) was reported, with each day having only 3.1% of samples classified as “spoiled.” This increased (P < .05) to 45.6% on day 6, 81.3% on day 8, and 90.8% on day 10, with each of the longest 4 display periods having greater than 78% of samples classified as “spoiled.”

The percentage of samples classified as “would purchase” and “spoiled” for odor and touch followed similar trends, though the magnitude and severity of the decrease was not as drastic as with appearance. For touch, there was no difference (P > .05) in the percentage of samples classified as “would purchase” among day 0 through day 4 samples, with greater than 78% of samples identified as “would purchase.” This percentage steadily decreased with increased days of display, with day 10 to day 14 samples having fewer (P < .05) samples that would be purchased than samples displayed for 6 d or fewer, though even at the longest period of display (day 14) close to half (46.7%) of samples would have still been purchased based on touch. Although differences among display days for the percentage of samples classified as “spoiled” based on touch were found, no more than 33.9% of samples were ever identified as spoiled. For odor, there was no differences (P > .05) in the percentage of samples that would have been purchased among day 0 to day 4 samples, which were greater (P < .05) than the percentage of samples identified as “would purchase” for day 8 to day 12. Samples displayed for day 14 had the lowest (P < .05) percentage (20.4%) of samples identified as “would purchase” among all days. For the percentage of samples classified as “spoiled” based on odor, there were differences (P < .05) among days of display from day 0 to day 12, though with no clear trend, with the percentage of “spoiled” samples ranging from 19.2% to 46.7%. However, day 14 samples had the greatest (P < .05) percentage (65.4%) of samples classified as “spoiled,” close to 20% more than any other day.

Of the traits evaluated, display period had the least impact on consumer perceptions of taste. There was no difference (P > .05) in the percentage of samples identified as “would purchase” across all days of display, with each day having between 55% and 74% of samples identified as such. Differences (P < .05) among display days were found for the percentage of samples identified as “spoiled” based on touch, though fewer than 18% of samples for each day of display were classified as “spoiled.” Across all 4 characteristics—appearance, touch, odor, and taste—taste overwhelmingly had the fewest number of samples identified as spoiled, regardless of display length.

In order to demonstrate how day of display and the associated changes in the percentage of samples that were classified as “would purchase” based on each characteristic, Figure 1 presents these data in relation to a* (A), trained sensory panel redness scores (B), trained sensory panel discoloration scores (C), and calculated percentage of MMb (D). Figure 2 presents the percentage of samples classified as “spoiled” in the same manner. In both figures, appearance had the greatest impact on the percentage of samples classified as both “would purchase” and as “spoiled.” From day 0 to day 4, appearance has fewer samples classified as “spoiled” and the highest percentage of samples classified as “would purchase” of all of the traits evaluated. This changed drastically from day 4 to day 8, in which the percentage of samples identified as “would purchase” dropped by close to 82%, and the percentage of samples identified as “spoiled” increased by close to 80%. This display period also corresponds to a dramatic change in both trained sensory panel redness and discoloration scores, with samples transitioning from a bright, cherry-red color with very little discoloration to a much darker red color with a high percentage of discoloration. Throughout the 14-d display period, changes in no other trait—touch, odor, or taste—resulted such an abrupt shift in consumer spoilage classification and purchase intent, with only day 14 samples for odor ever having more than half of samples classified as “spoiled.”

Figure 1.

Percentage of samples classified by consumers as “would purchase” based upon appearance, touch, odor, and taste across all days of display compared to (A) a*, (B) trained sensory panel redness score, (C) trained sensory panel discoloration score, and (D) percentage MMb. MMb, metmyoglobin.

Figure 2.

Percentage of samples classified by consumers as “spoiled” based upon appearance, touch, odor, and taste across all days of display compared to (A) a*, (B) trained sensory panel redness score, (C) trained sensory panel discoloration score, and (D) percentage MMb. MMb, metmyoglobin.

Microbiological analysis

Results showing the microbiological analyses of APC, EB, and ECC relative to the day of display are presented in Table 3. Moreover, Figures 3 and 4 show the percentage of samples rated as either “would purchase” or as “spoiled” for all evaluated organoleptic traits in relation to APC, EB counts, and ECC. All microbial counts increased (P < .05) throughout the storage period. While APC started at an average of 6.50 Log CFU/g on day 0, consumers were 97.75% likely to purchase the product based on appearance, and only 3.1% of samples were classified as spoiled. EB counts and ECC were lower than APC, both starting at 3.24 Log CFU/g and 1.12 Log CFU/g. At the beginning of the transition from consumer classification of “would purchase” and “not spoiled” to “would not purchase” and “spoiled” (4–6 d) based on appearance, spoilage classification increased by 42.5% and purchase intent dropped by 57.1%; whereas, APC only increased by 0.06 Log CFU/g. From day 6 to day 8, spoilage classification increased an additional 35.7% and purchase intent dropped by 24.7%, while APC increased by 0.52 Log CFU/g. At the conclusion of storage period (day 14), APC, EB counts, and ECC reached 9.48 Log CFU/g, 6.49 Log CFU/g, and 5.25 Log CFU/g, respectively. Consumers were still 55.7% likely to purchase and only 17.9% likely to rate the product as spoiled based on the taste, with only 33.9% and 65.4% of samples classified as spoiled based on touch and odor, respectively.

Figure 3.

Consumer purchase intent based upon appearance, touch, odor, and taste across all days of display compared to (A) APC, (B) EB counts, and (C) ECC. APC, aerobic plate counts; EB, Enterobacteriaceae; ECC, Escherichia coli/coliform counts.

Figure 4.

Consumer spoilage rating based upon appearance, touch, odor, and taste across all days of display compared to (A) APC, (B) EB counts, and (C) ECC. APC, aerobic plate counts; EB, Enterobacteriaceae; ECC, Escherichia coli/coliform counts.

Logistic regression equations

Tables 6 and 7 present the logistic regression equations generated using the objective measurements to predict consumer purchase intent and the likelihood of a consumer spoilage classification based upon the appearance, odor, touch, and taste of the product. Overall, the models showed all objective measurements were predictors (P < .05) of consumer purchase intent based upon appearance, with a* value (Figure 5), calculated percentage MMb (Figure 6), trained sensory panel redness and discoloration scores (Figure 7) all explaining more than 80% of the variation in consumer purchase intent. For spoilage classification based on appearance, all objective measures were also predictors (P < .05) but explained less variation using a* value (70%), MMb (73%), trained sensory panel redness score (77%), and discoloration score (76%).

Figure 5.

Logistic regression for consumer likeliness to purchase retail ground beef or classify a sample as spoiled based on a* value.

Figure 6.

Logistic regression for consumer likeliness to purchase retail ground beef or classify a sample as spoiled based on MMb percentage. MMb, metmyoglobin.

Figure 7.

Logistic regression for consumer likeliness to purchase retail ground beef or classify a sample as spoiled based on trained sensory panel discoloration scores.

When evaluating the models for prediction of both consumer purchase intent and spoilage classification based on appearance using the microbial measures of APC, EB counts, and ECC (Figure 8), the models were predictive (P < .05) but explained much less variation (R² of 0.17–0.59). This was less variation than all other objective measures evaluated, except for TBARS and L*. Similar results were observed in the spoilage classification models (Figure 9), with all 3 microbial models being predictive (P < .05) but explaining less variation (≤46%) than almost all of the other evaluated models.

Figure 8.

Logistic regression for consumer likeliness to purchase retail ground beef based on APC, EB counts, and ECC. APC, aerobic plate counts; EB, Enterobacteriaceae; ECC, Escherichia coli/coliform counts.

Figure 9.

Logistic regression for consumer likeliness to classify retail ground beef as spoiled based on APC, EB counts, and ECC. APC, aerobic plate counts; EB, Enterobacteriaceae; ECC, Escherichia coli/coliform counts.

The logistic regression models for predicting consumer purchase intent and spoilage classification based on both touch and odor using the 3 microbial measures all were predictive (P < .05; Table 7) but explained a very low amount of variation (<11% for purchase intent and <4% for spoilage). None of the 3 microbial measures were able to predict (P > .05) the percentage of samples consumers would purchase based on taste, and while all 3 were predictive (P < .05) of spoilage classification based on taste, each explained less than 4% of the total variation. This provides evidence that APC, EB counts, and ECC were not reasonable predictors for consumer spoilage detection based on touch, odor, or taste and were only moderate predictors of spoilage based on appearance.

Microbiota analysis

Statistical differences in the microbiota composition between day 0, day 4, day 6, and day 14 were assessed at the ASV level for samples collected in the first and second week of the experiment. Microbiota composition on samples analyzed at day 0 did not differ (P > .05) compared to that on samples tested at day 4, day 6, and day 14. However, there was a difference (P < .05) in microbiota composition of week 1 samples on day 4 compared to day 14. This is consistent with the decrease in consumer purchase intent and increase in spoilage rating following day 4 of display. Figure 10 illustrates the relative abundance of ASV by day of display during weeks 1 and 2. Samples from both weeks indicated Lactobacillus and some Pseudomonas species were predominant during the first 4 d of display but then shifted to primarily different species of Pseudomonas. Figure 11 illustrates the relative abundance of bacterial genera according to the day of display during weeks 1 and 2. Between day 0 and day 4, Lactobacillus was the most abundant genus, with a lower relative abundance of Carnobacterium present. Throughout the display period, a more apparent shift is shown in genus abundance between day 4 and day 8, with Pseudomonas becoming the predominant genus in samples up to day 14. Extraction efficiency was higher for Bacillus, Salmonella, Escherichia-Shigella, and Pseudomonas compared to Listeria, Staphylococcus, and Lactobacillus fermentum. This assessment is based on discrepancies between observed relative abundances in sequenced positive control samples and the expected relative abundances specified in the positive control specification.

Figure 10.

Relative abundance of amplicon sequence variants by day of display for (A) week 1 and (B) week 2. ASV present in 1% or higher relative abundance are shown. ASV, amplicon sequence variants.

Figure 11.

Relative abundance of genera by day of display for (A) week 1 and (B) week 2. Genera present in 1% or higher relative abundance are shown.

Consumer perceptions of spoilage

Meat spoilage is a complex interplay among color chemistry, lipid oxidation, and microbiology. Ultimately, the spoilage and shelf life of a product are defined by consumer opinion. Although a product may in many cases be safe for consumption, consumer rejection due to the failure of their organoleptic expectations often results in a discarded product. Consumers use all their senses—sight, smell, taste, touch, and sound—when determining whether a food product would be considered spoiled, thus spoilage may be determined due to the failure of any of these traits (Nollet, 2012; Odeyemi et al., 2020). In the current study, ground-beef spoilage was assessed by consumers and their individual expectations related to each of these traits, allowing for a direct assessment and interpretation of spoilage without having to make inferences using only objective measures. Additionally, all aspects of spoilage in ground beef were evaluated to capture a wholistic representation of the changes that occur through extended display of ground beef that result in spoilage.

Of the traits evaluated, changes in spoilage perception were most closely associated with the appearance of the ground-beef samples. This was clearly demonstrated in the observed changes in both spoilage classification as well as purchase intent that corresponded to large changes in trained sensory panel redness scores, discoloration scores, and reduced a* values. Consumer appearance scores were more closely related (r > [±0.93]) to these variables than any of the other measures evaluated, and the predictive models explained a greater amount of the variation in both spoilage perception (>70%) and purchase intent (>80%) than any of the other measures evaluated for any of the organoleptic measures. Moreover, consumer appearance liking ratings were more closely related to and predictive of spoilage classification based on appearance than any of the liking scores were for the taste, odor, and touch models, with the appearance model explaining twice as much variation in spoilage perception (82 vs. 41%) as the next best model (odor). This indicates consumers more closely tied negative appearance traits with spoilage perception than they did changes in taste, odor, or touch.

Numerous authors have previously reported color as one of the most important factors considered by consumers when purchasing fresh beef (Farmer et al., 2022; Beyer et al., 2024; Decker et al., 2024). However, these reports have often failed to separate the color of redness vs. discoloration within their data. The current work would indicate that both redness and discoloration have a large impact on consumer perceptions of both spoilage and willingness to purchase, with both having a near equal impact. A previous study by Lybarger et al. (2023) modeled consumer willingness to purchase ground beef of varying redness and discoloration scores and found similar results to the current study, with trained sensory panel redness score and discoloration score accounting for more than 77% of the variation in consumer willingness to purchase ground beef if sold at full price. In their study, similar to the current work, both redness and discoloration were some of the best predictors evaluated, with redness score being slightly better (Lybarger et al., 2023). The models presented by Lybarger et al. (2023) produced very similar purchase thresholds for ground-beef discoloration as the current study, but the authors did not assess spoilage within that study, only purchase intent. In the current work, consumers required a greater amount of discoloration for a “spoilage” classification than for a “would purchase” designation. For example, 38.7% discoloration was associated with a 50% chance of a consumer purchase; whereas, a similar amount (38.5%) of discoloration was only associated with a 25% chance of a consumer classifying a sample as spoiled. Thus, consumer purchase intent and spoilage perception are not aligned, with consumers requiring a greater amount of discoloration to classify a sample as “spoiled” than would be required for them to identify the package as “would not purchase.”

Odor is one of the most common organoleptic properties used by consumers to assess spoilage for many products (Dainty, 1996; Kim et al., 2009). With ground beef, this is a trait not easily assessed at the point of sale but is rather assessed upon opening the package prior to cooking. Odor changes in fresh beef are associated with the production of volatile organic compounds, which are recognized as key indicators of spoilage in raw meat, arising from processes such as lipid oxidation and the activity of various microorganisms present in the product (Bleicher et al., 2022). Volatile organic compounds are often lipid-derived compounds produced from the oxidation of fatty acids during autooxidation, which can result in an unsuitable aromatic profile if the product’s shelf life has expired (Domínguez et al., 2019; Dinh et al., 2021). Additionally, certain odors can be an indication of microbial growth. Pseudomonads and other related gram-negative organisms are the primary microorganisms responsible for producing putrid, fruity, cabbagy, and sulfur-associated odors in refrigerated meats (Stutz et al., 1991). Additionally, Dainty et al. (1984) reported Pseudomonads that often dominate the microflora in spoiled products are a common source of “off odors.” In the current work, the percentage of samples classified as “spoiled” by consumers based on odor increased throughout the storage period and odor changes were also indicated by the trained sensory panelists. Aside from appearance, odor production had the greatest impact on spoilage perception by consumers. Although the specific source of the undesirable odors is unclear, the oxidation of lipids and proteins, as well as the microbial growth throughout storage, likely contributed.

Changes in touch and taste and the associated classification of samples as “spoiled” were less pronounced through time than appearance and odor. Of particular note, display period had almost no impact on the taste of the ground beef, with only day 14 samples rated differently than all the previous days of display. This is surprising given the relatively large changes in odor perception and the link between taste and odor perception in food products (Small and Prescott, 2005). Previous authors have reported off-flavor development and decreased consumer liking in beef products aged for extended periods of time (O’Quinn et al., 2016; Foraker et al., 2020), but these reports have primarily been in whole-muscle steaks and included aging periods longer than 35 d rather than the extended retail display of ground-beef products used in the current work.

When assessing spoilage in ground beef, more than 81% of samples were classified as spoiled based on appearance at day 8, at which point far fewer were classified as spoiled based on odor (44.2%), touch (24.2%), and taste (9.4%). The percentage of samples identified as “spoiled” based on appearance remained high throughout the remainder of the storage period, while the percentage of samples classified as “spoiled” for the other 3 traits increased for the remaining storage period. At the conclusion of the storage period (day 14), a much lower percentage of samples were classified as “spoiled” based on touch (33.9%), taste (17.9%), and odor (65.4%; <47% on day 12 and prior) than from day 8 and thereafter for appearance. For the final 6 d of storage, the percentage of samples classified as “spoiled” based on appearance was close to 80% or higher, with the percentage of samples identified as “spoiled,” even with close to an extra week of storage, never reaching this same level based on the other organoleptic traits. Thus, our study indicates that ground-beef spoilage is primarily driven by appearance, with samples failing to meet consumer expectations for appearance long before they fail based on odor, touch, and taste.

Microbial growth

Microbial counts of 7 Log CFU/g are commonly cited as the point at which meat is spoiled (Ayres, 1960; Nollet, 2012; Yang and McMullen, 2024); however, many of these references provide no additional assessment of spoilage to support this claim. This often cited 7-Log threshold originated from the landmark study by Butler et al. in 1953 in which the authors measured spoilage by examining days of storage, color fading, and microbial growth using a Pseudomonas inoculum on beef steaks from the longissimus dorsi muscle. Butler et al. (1953) reported steak color fading corresponded with Pseudomonas levels exceeding 7 Log CFU/g, which correlated with high rates of discoloration. It is noteworthy that the Butler et al. (1953) study used only trained sensory panelists, objective color measures, and microbial growth to determine spoilage, with no consumer-based subjective determination of spoilage included. In the more than half-century since the Butler et al. (1953) study, few studies have attempted to reevaluate the 7-Log CFU/g threshold for spoilage, especially in ground beef.

In the current work, APC started at 6.5 Log CFU/g at day 0 and reached 7 Log CFU/g at day 4. At this same time point, less than 3.5% of samples were classified as “spoiled” by consumer sensory panelists based on appearance. As the majority of changes in appearance, color, and the associated increase in spoilage designation occurred (4–8 d), there was only a 0.6-Log CFU/g increase in APC. Likewise, at day 4, when APC were at 7 Log CFU/g, the percentage of samples classified as “spoiled” for touch, odor, and taste did not differ from samples displayed for a shorter period. The most notable increase in the percentage of samples classified as “spoiled” for odor and touch did not occur until day 14, at which point APC had reached 9.5 Log CFU/g, a much greater level than the cited 7 Log CFU/g. Furthermore, models predicting spoilage based on APC were among the least predictive of the measures evaluated, with touch, odor, and taste models accounting for only a small amount of variation in spoilage (<17%). The APC model for spoilage based on appearance was better, accounting for 46% of the variation yet was still a better predictor of spoilage than only 2 of the nonmicrobial measures evaluated. Taken together, these results indicate that microbial growth was not the largest driver of consumer classification of samples as spoiled in the current study, with myoglobin-related oxidation and the associated production of MMb (Mancini and Hunt, 2005; King et al., 2023) likely having a larger impact on spoilage. Furthermore, the current work does not support a 7-Log CFU/g threshold for spoilage of ground beef.

Although microbial counts showed only a limited ability to indicate spoilage, the changes observed in the composition of the microbiota provided some potential insights. The microbiota in the first 3 display periods were dominated by Lactobacillus, a facultative anaerobic genus, which is likely the direct result of the anaerobic conditions of the mother bags the samples were stored in prior to display. However, between day 4 and day 6 of display, the microbial population shifted to primarily Pseudomonas, an aerobic genus commonly labeled as the “spoilage bacteria” (Sofos et al., 2007), with multiple putative species such as P. fluoroscens, P. putida, P. fragi, P. lundensis, and P. weihenstephanensis previously identified in ground beef (Hilgarth et al., 2019; Kolbeck et al., 2021). Pseudomonas’ contribution to meat spoilage is high due to the microbe’s proteolytic and lipolytic enzymes, which initiate organoleptic changes to the product once released into the cell (Kolbeck et al., 2021). In the current study, this shift in microbial populations occurred simultaneously, with the largest shift in the percentage of samples classified as “spoiled,” potentially contributing to the observed changes. Additionally, Pseudomonas can result in slime formation when the microbial load reaches 10⁷ CFU/g to 10⁸ CFU/g (Wickramasinghe et al., 2019), which would align with the changes in touch characteristics identified by trained sensory panelists and increased percentage of samples identified as “spoiled” based on touch by consumers in the longest display periods. Lastly, Pseudomonas has previously been observed to contribute to a change in color of meat products from a brown to red color through the production of an unidentified red myoglobin derivative (Morita et al., 1996), potentially explaining the observed “re-reddening” of ground beef observed in the current study at day 14. Our results indicate that the community of bacteria present has a larger impact on ground-beef spoilage than microbial counts alone.