Sampling scheme

The slaughterhouses that participated in the study were selected using a non-probability sampling plan, based on the following criteria: 1) companies located in one of the three ecological-livestock areas of Mexico (arid and semiarid, wet tropics, and dry tropics); 2) carcass halves are cut between the twelfth and thirteenth ribs; 3) the company is a Federal Inspection Type (TIF) slaughterhouse, and 4) availability of working areas with technical lighting and space conditions, access to primary cuts, and cutting area with a refrigeration temperature of 6-8 ºC.

According to these criteria, four TIF slaughterhouses were selected, located in Baja California, Sinaloa, Querétaro, and Tabasco states, in which a total of 1,165 samples were analyzed (Figure 1). Before color evaluation, meat was exposed to oxygen for 30 min; this created a much slower working rate compared to the process flow. Therefore, it was not possible to analyze all the carcasses processed in one day. The goal was to take at least 170 carcasses per slaughterhouse, which was the minimum sample size determined, considering a confidence level of 95 %, an accuracy of 0.5 units in the instrumental color variables, and a variance of 10.69, estimated from preliminary tests. Thus, using the equation n=Zα2*S2d2, it was obtained a sample size of 164, which was rounded to 170. However, in order to obtain the highest possible variation, in each establishment, all the loins that could be collected in an 8-hour shift for 3 or 4 days were analyzed. Therefore, the actual sample size per slaughterhouse fluctuated between 172 and 405, according to the slaughter volume in each establishment.

Figure 1 Number of loin samples analyzed in TIF slaughterhouses of four Mexican states between November 2012 and July 2013 

The population of animals from which the data were obtained was made up, in about 80%, of non-castrated entire males, the remaining percentage were females. As the research was carried out on slaughterhouses, it was not always possible to know the age at slaughter. However, in a sample of around 300 animals in which this data was available, 86 % were 24-mo-old or less. Overall, this was the age expected for the studied sample, considering previous reports that document the preponderance of young bulls in the beef cattle slaughter population in Mexico²³.

Color measurement

Color measurements, both visual and instrumental, were carried out following the American Meat Science Association guidelines²⁴. Readings were made at 24 h postmortem in the cutting and deboning area of each TIF slaughterhouse, with a controlled room temperature of 6 to 8 ºC. The complete loins (Longissimus dorsi muscle), recently separated from the carcass and with a temperature not higher than 3 ºC, were collected. The selected loins were left to rest for 30 min on a stainless-steel table, with the loin eye area between the twelfth and thirteenth ribs exposed to air, to allow optimal oxygenation of myoglobin before the readings.

The visual color evaluation was performed under standardized conditions. As a light source, it was used a photographic quality Osram incandescent lamp with an intensity of 150 foot candles (1,614 luxes) and a color temperature of 3,200 ºK, placed at a 45° angle concerning the surface of the chop. A professional photographer took high-resolution photos of each loin chop evaluated; for this purpose, was used a 12-megapixel Nikon D300S camera with a Sigma 24-70 f. 2.8 zoom lens. Photographs were taken against a black background to eliminate color differences associated with the meat surface. The photographs function as illustrative images of the different hues of the meat, which could be included in the visual color scale that was being developed. Subsequently, using the United States eight-level color pattern for beef (Figure 2)⁹, an experienced researcher performed the visual color evaluation in order to have a preliminary reference about the number of possible visual categories in the studied sample.

Figure 2 Visual scale for color evaluation in beef carcasses developed in the United States of America⁹  

For instrumental measurements, was used a Hunter MiniScan EZ 4500L spectrophotometer with a 45/0 geometry and a port size opening of 25 mm (Hunter Associates Laboratory, Inc, Reston, Virginia, United States). The instrument was configured as follows: A/10° illuminant/observer combination and specular component excluded. Furthermore, the instrument was remotely operated using the program OnColor QC Lite, version 6 (CyberChrome, Inc., New Paltz, New York, United States) for computer data capture. Calibrations were performed before starting measurements and after every 100 readings or after 1 hour (whichever occurred first), using the black trap and white tile supplied by the manufacturer. From each loin chop, four readings were obtained, and their average was used to calculate the lightness (L*), redness (a*), yellowness (b*), hue (h*), and Chroma (C*) of each loin chop in the CIELAB scale. Along with these data, we also collected the spectral curves, which constitute the color footprint and were necessary for the professional impression of the scale.

Data analysis and visual scale conformation

For the statistical analysis was used the Statgraphics XV Centurion software for Windows (Statpoint, Inc., The Plains, Virginia, United States). The means of the color instrumental values were compared between the quality categories visually assigned. For this, a one-way analysis of variance (ANOVA) was performed using the Generalized Linear Model (GLM) procedure. As a result of the different number of observations per level, when significant differences were found, the means were discriminated using the Bonferroni multiple comparison procedure.

Different prediction models were tested to construct the scale, by means of the GLM procedure, using the visual category as the dependent variable and different combinations of the instrumental variables and their interactions as explicative variables (See Table S1 in Supplementary Information). Among the generated models, the one that explained the highest percentage of the observed variation between visual categories was chosen, which turned out to be the one that uses L* as the only explicative variable. Therefore, 95 % confidence intervals of L* were constructed within each visual category. To illustrate the scale, we selected the photographs that corresponded to L* loin chop values within the confidence interval of the mean in each category.

Lastly, it was necessary to determine if the categories could be differentiated visually. For this, it was calculated the total color difference (∆E*ab), which is the sum of the modular differences of L*, a*, and b* between two samples: (∆E*ab= ΔL*2+Δa*2+Δb*2). The greater the ∆E*ab, the easier it is to visually distinguish the color difference between two samples. However, usually, samples with ∆E*ab values >2 units are considered easily distinguishable²⁵.

Preliminary validation experiment

In order to validate the visual scale, we evaluated the newly developed visual standards in an industrial setting, under the conditions described below. The visual scale was tested in a slaughterhouse not included in the initial sampling. For validation, a trained sensory panel was formed, with six people familiar with the proposed color pattern. Panelists were selected using the Farnsworth-Munsell test (http://www.color-blindness.com/farnsworth-munsell-100-hue-color-vision-test/#prettyPhoto), which allows dismissing people with color appreciation deficiencies. Selected panelists were required to have a color vision deficiency score of “none” or “mild” (0 to 70 points). The initial group was formed by 20 candidates selected according to the generally accepted guidelines for the planning and selection of trained judges²⁶.

As a result of the selection process, the sensory panel ended up with six members. Each panelist participated in six evaluation sessions, in which they evaluated six meat samples (2 of normal-colored meat (N), 2 with moderate DC, and 2 with extreme DC), for a total of 36 measurements per judge. For the evaluations, the judges received instructions on how to use a structured 10-cm-long scale, ranging from very light red (L* = 50) to very dark red (L* = 30):

According to the proposed visual scale, extreme DC meat has L* values between 30 and 34; moderate DC between 35-37; and N meat, 38 or more. Judges were asked to mark the scale with a cross to indicate the color that corresponded to the evaluated sample. The position of the cross was measured with a ruler to obtain the estimated L* value. The data were analyzed by a three-factor analysis of variance (judge, sample, session, and their interactions); this corroborated that the only source of significant variation in the evaluations was the sample. Finally, a correlation analysis was performed between the L* values estimated from the color evaluation made by the judges and the real L* values determined with the spectrophotometer. In order to validate the proposed visual scale for beef color evaluation, it was necessary to obtain a significant discriminatory power of the judges, a significant high-magnitude correlation between the real L* values of the samples and the L* values estimated by the judges, as well as a positive hit rate > 80% when assigning a color category to each sample. In previous studies with orange juice and wine, the visual differences of the samples were validated with a much lower hit percentage of the sensory panel (e.g., >50 %)^27,28. However, the present study looked for a higher hit rate in order to determine if the visual scale worked for most people, which is why color evaluations were also conducted with an untrained panel (n = 6). Consumer panelists were asked to indicate if they observed any difference between N and DC meat samples. The latter meant to corroborate the reliability of the scale when used by people without color evaluation training. Samples were selected and displayed under the same conditions previously described for the trained panel. However, the untrained panel was only presented with pairs of samples, and panelists were asked to choose the sample they preferred based on color.

Definition of visual categories and their relation with instrumental measurements

Visual evaluation allowed to identify, in the studied sample, seven of the eight categories represented in the North American visual standards because none of the samples had an appearance similar to category 5 (slightly dark red). Figure 3 shows the distribution of samples in each of the seven identified classes.

Figure 3 Distribution of samples (n = 1,165) in each of the visual categories identified using the North American pattern for beef color evaluation (the description of each category is the same as in Figure 2). 

The meat proportion associated with quality defects (categories 7 and 8 represent different degrees of DC appearance) was relatively low (7 %). In contrast, more than 70 % of the samples corresponded to the categories 2-4, an appearance generally associated with normal quality meat.

Moreover, the ANOVA was significant (P<0.0001) for all the instrumental color variables (Table 1). However, L* was the only variable with significant differences between the means across all the categories. Furthermore, the ANOVA of L* was the only one with a high coefficient of determination (R²=0.9171), while in the other variables, it ranged from 0.2744 to 0.4284.

Although other prediction models were tested (see Table S1 in supplementary information), none of them were better than the one obtained with L* as the only explicative variable and, hence, were discarded. Therefore, 95 % confidence intervals of the mean value of L* were used as a criterion to select the representative photos from each of the visual categories that form the developed color scale (Figure 4). An important feature of the latter is that meat appearance in some adjacent categories is very similar. For example, categories 1 and 2 represent light meat. Similarly, categories 3 and 4 are representative of the bright cherry-red hue, which is usually the most attractive to consumers at retail. Therefore, at first, we considered merging both pairs of categories into one. However, the ∆E*ab between these pairs was of almost three CIELAB units, which means that they are clearly differentiable from each other with the naked eye. Therefore, instead of merging them, it was used a denomination that would allow the user to identify that both categories are associated with similar hue meat. Hence, categories 1 and 2 were renamed as 1A and 1B, respectively, while 3 and 4 were denominated 2A and 2B. These adjustments forced the name modification of the remaining categories (5, 6, and 7) into 3, 4, and 5; their appearance is clearly differentiable from each other, with a much higher ∆E*ab between them (3.3 to 5.5 CIELAB units).

L* values correspond to the 95% confidence interval of the mean for each category. The numbers at the bottom indicate the total color differences (∆E*ab) between adjacent categories.

Figure 4 Visual scale for evaluating beef meat color in Mexico 

Preliminary validation of the descriptive scale under industrial conditions

The first validation criterion for the color scale was for the sensory panel to detect the differences between the samples. Judges, both individually and as a group, showed sufficient discriminatory power (significant ANOVA, P<0.0001); therefore, results were satisfactory. Furthermore, the sample was the only factor that significantly influenced (P=0.0002) the score of the judges (see Tables S2 to S4 in supplementary information).

The correlation between the real L* values of the samples and the values assigned by the judges (Figure 5), which was the second validation criterion, was significant and of high magnitude (r=0.9338, P<0.0001). These results correspond to the percentage of correct answers by the judges, third and last validation criterion of the scale; judges, on average, correctly assigned the samples to the visual category of the pattern in 92.6 % of the evaluations. The scale showed even better performance in the detection of DC; the judges correctly assigned samples with this condition in 100 % of the evaluations. Furthermore, the scale also allowed to distinguish the different degrees of the DC defect; trained judges confused moderate DC with extreme DC in only 5 % of the evaluations. Finally, the evaluation performed by the untrained panel corroborated the relevance of the categories that describe the DC defect, since in 85.3 % of the evaluations performed by untrained panelists, the color of N meat was preferred over the color of DC meat. Overall, the results of the sensory tests demonstrate the technical viability of the proposed color scale since both trained personnel and consumers were able to use it to dictate the appearance of meat correctly.

The model includes the bilateral prediction limits. ***P<0.0001.

Figure 5 Regression analysis between the L* estimated by the trained sensory panel and the real L* of the samples measured using the spectrophotometer (n=108)