Multivariate Relationships Between Carcass, Muscle, and Beef Eating Quality Traits in Limousine Cull Cows

Animal and carcass grading

The study involved 102 carcasses of Limousine cull cows, with records of their slaughter age and carcass weight. The EUROP grid of conformation and fat scores was converted into a continuous 15-point scale as described by Hickey et al. (2007). Two expert graders accredited by the International Meat Research 3G Foundation determined the ossification score, hump height, eye muscle area, fat thickness, as well as the fat color, meat color, and marbling according to the guidelines of the MSA grading scheme (Santinello et al., 2024a). Because carcass cutting differs between the MSA system (10^th–12^th rib) and European commercial practice (5^th rib; Santinello et al., 2024a), fat color, meat color, and marbling were graded at both the 5^th rib and 10^th rib sites to ensure comparability between grading systems. At 24 h postmortem, beef ultimate pH and temperature were measured using a portable pH meter (HI 99163, Hanna Instruments, Germany). After a blooming period of 20 min, the CIE L*, a*, b* were determined using a spectrophotometer (Konica Minolta CM-600d, Osaka, Japan) equipped with an 8 mm aperture, using illuminant D65 and a 10° standard observer.

The representative samples of 2 muscles, striploin (m. longissimus thoracis, STR) and rump (m. gluteus medius, RMP) were then collected from the left or the right of each carcass. All the samples were cut into steaks and aged in vacuum bags at 4°C for 10 d, and then stored at −20°C. This ageing protocol was standardized across all carcasses to ensure comparable postmortem proteolysis.

Muscle physicochemical traits measurements

Rheological values of shear and compression forces were measured. The WBSF of raw meat was measured using the MTS Synergie 200 (Ellies-Oury et al., 2009). A 1 cm² (square in cross-section) sample with muscle fibers parallel to the longitudinal axis of the sample was placed on a tabletop under the V-blade and sliced through as the blade moved downward at a constant speed (10 mm min⁻¹). Compression force at 80% (compression) was analyzed using a Shimadzu EZ-SX tetrameter, coupled with Trapezium X software, operating in the compression mode and using a 25 kg load cell (Soulat et al., 2021). Results are expressed in Newton (N). Six replicates per sample were recorded for these determinations. Replicates were averaged before analysis to minimize within-sample variability.

The intramuscular fat (IMF) content was determined after hydrolysis with hydrochloric acid and extraction with petroleum ether (boiling point: 35 C; Soxhlet standard method; NF V 04-402, 1968). The FA composition was determined by chromatographic analysis after transmethylation. FAs underwent methylation by mixing boron trifluoride (BF3) and methanol 14%. Methyl esters of FAs were then extracted with hexane and recovered after centrifugation at 1000 × g for 10 min at room temperature. The FA composition was determined by gas chromatography flame ionization (CPG/FID, Shimadzu, Kyoto, Japan) using an Omegawax 250 capillary column (30 m, 0.25 mm ID). Injector and detector temperatures were 230°C and 250°C, respectively. The temperature was increased by 5°C/min. The FAs were identified and quantified, and the results are expressed as content (mg/100 g muscle; Hamdi et al., 2018).

Beef eating quality traits evaluated by untrained consumers

Consumer assessment of eating quality was done according to the MSA protocol described by Watson et al. (2008). Each muscle was sliced into 75 × 60 × 25 mm steaks. Steaks were grilled at a maintained temperature of 200°C for 2.5 min sitting on a Silex clamshell grill (Silex Grills Australia Pty Ltd., Marrickville, NSW, Australia). Each consumer first evaluated a standard reference steak used for sensory calibration, followed by 6 steak samples for testing, with no salt or seasoning added. The served order of testing samples was designed on a 6 × 6 Latin square to balance order effects across consumers. A total of 360 untrained French consumers were recruited and assigned to 6 experimental plans, each including 60 consumers. Sensory evaluations were conducted over 3 testing sessions, as previously described by Legrand et al. (2013).

According to the MSA protocol, the consumers assessed each steak based on tenderness, juiciness, flavor liking, and overall liking on hedonic scales from 0 to 100 mm, and assigned each sample to one of 4 MSA quality grades (2 stars [unsatisfactory], 3 stars [good everyday], 4 stars [better than everyday], and 5 stars [premium]) that generally described their overall preference of the sample. A general meat quality score (MQ4) was calculated for each steak using the following formula (Thompson et al., 2010): MQ4 = 0.3×tenderness + 0.1 × juiciness + 0.3 × flavor liking + 0.3 × overall liking.

In addition, a whole carcass weighted MSA index was predicted using the SP2009 version of the MSA prediction model, which integrates animal sex, carcass weight, hanging method, hump height, ossification score, MSA marbling score, rib fat depth, ultimate pH, and days of ageing (Bonny et al., 2018).

Statistical analysis

All statistical analyses were conducted in R software (version 4.3.2). Pearson correlation coefficients of muscle characteristics with carcass traits were first calculated separately for each muscle and all animals. The correlation matrix was found to be very similar in the 2 muscles, and most of the coefficients were not significantly different between them. The pooled relationship between muscle characteristics was therefore calculated on all the observations (n = 24) after removing the muscle effect; that is, the deviation of each observation from the corresponding muscle mean. This centering approach allows the clear analysis of within-muscle variability. Besides, since there were no significant differences in beef eating quality, muscle texture, and pH between the left and right sides of the carcass (Santinello et al., 2024a; Pipek et al., 2003), this study did not consider the effects of carcass grading side. Associations among carcass traits and centered meat quality traits were then assessed using Pearson’s correlation coefficients, computed on pairwise complete observations. Principal component analysis (PCA) was then performed and visualized using the “ggplot2” package.

Linear mixed-effects models were fitted with the ‘lme4’ package to evaluate the contribution of each variable to eating quality traits. Four dependent variables (tenderness, juiciness, flavor liking, and overall liking) were analyzed with marbling score at 10^th rib (Marbling10) specified as a fixed effect and Animal ID as a random effect (core model). Additional variables were sequentially included one by one when 2 conditions were met: (i) it had a significant marginal effect, and (ii) its inclusion reduced the Akaike information criterion and Bayesian information criterion values. Statistical significance of fixed effects was declared at P ≤ 0.05. Model adequacy was then evaluated based on marginal and conditional R², where marginal R² represents the variance explained by fixed effects alone, and conditional R² represents the variance explained by both fixed and random effects.

Data description of carcass, muscle, and beef eating quality traits

The data structure for 102 carcasses is presented in Table 1, and descriptive data for STR and RMP muscles are shown in Tables 2 and 3, respectively. The Limousine cows were primarily old, with an average age of 125.8 months (nearly 10.5 years). According to the EUROP grid, conformation scores ranged from 7 to 12 (R- to U+), and all except 2 carcasses received a fat score of 3. Then, the critical control points (CCPs) under the MSA principles were measured. Carcass pH averaged 5.68, with approximately 42% of carcasses above 5.70. Despite a coefficient of variation of 33.7% for slaughter age, 88% of ossification scores were concentrated between 550 and 590, including 75% at 590. Fat color, meat color, and marbling scores were recorded at both the 5^th and 10^th ribs, representing French and Australian processing practices, respectively. These CCPs were used to predict the MSA index, which ranged from 45.47 to 58.52 in this dataset. Laboratory and sensory indicators were assessed independently for each muscle due to their differing biochemical properties. Consumer-evaluated sensory scores ranged from 59.3 to 63.6 for STR, and from 57.3 to 61.6 for RMP.

Traits such as fat score, ossification, and ω3 content displayed highly skewed or extremely narrow distributions and were therefore excluded from multivariate analyses due to their limited discriminatory power.

Correlation across carcass, muscle, and beef eating quality traits

Pooled Pearson’s correlations among carcass and meat quality traits are shown in Figure 1, with nonsignificant coefficients (P > 0.05) masked. All the eating quality traits of the studied cuts evaluated by untrained consumers were strongly positively correlated (0.67 ≤ r ≤ 0.96) but showed weaker correlations with the MSA index (0.20 ≤ r ≤ 0.28). Carcass yield correlated traits (carcass weight, conformation, and eye muscle area) were strongly intercorrelated (r > 0.70). These traits also correlated positively with all eating quality scores (0.18 ≤ r < 0.54), particularly with the MSA index (0.46 ≤ r ≤ 0.54) and juiciness score (0.40 ≤ r ≤ 0.43). Although their correlations with MSA meat color and fat color were limited (|r| < 0.32), positive associations were observed with CIE a* and b* (r > 0.30). Notably, slaughter age was negatively correlated with these traits (r < −0.2), and as expected, with eating quality (r ≤ −0.18). Among traits associated with fat deposition, indicators of intramuscular fat deposition (Marbling5, Marbling10, and IMF) were positively intercorrelated (r ≥ 0.37), but less with fat thickness (0.16 ≤ r ≤ 0.26). In addition, IMF and marbling were significantly correlated with fatty acid content (P < 0.05), showing negative correlations with SFA, PUFA, and ω6 contents (r < −0.15), and positive correlation with monounsaturated fatty acid (MUFA) content (r > 0.16). However, these fat deposition traits had no significant correlation with eating quality traits (P > 0.05), but a positive correlation with the MSA index (0.27 ≤ r ≤ 0.47). The saturated fatty acids (SFA) had a significant negative correlation with juiciness score (r = −0.17) and the MSA index (r = −0.35), while the MUFA and ω6 were positively correlated with the MSA index (r ≥ 0.14). Notably, the strong negative correlation observed between SFA and MUFA (r = −0.80) was largely driven by the data being expressed as a percentage of IMF, rather than a true biological association. When the fatty acid profile was expressed on a muscle basis (mg/100 g muscle), the IMF was positively correlated with both SFA and MUFA (r ≥ 0.19), and the correlation between SFA and MUFA became negligible (r = −0.08). Muscle physical properties (WBSF and compression) were positively intercorrelated (r = 0.55). However, only compression showed significant negative correlations with eating quality traits (−0.24 ≤ r ≤ −0.20). The WBSF was negatively correlated with carcass yield-related traits (r ≤ −0.21), but not with eating quality traits (P > 0.05).

Figure 1.

Correlation matrix across carcass and pooled-muscle characteristics. Suffixes “5” and “10” of “Fat color,” “Meat color,” and “Marbling” denote measurements taken at the 5^th and 10^th rib positions, respectively; MQ4 = 0.3 × tenderness + 0.1 × juiciness + 0.3 × flavor liking + 0.3 × overall liking; WBSF, Warner-Bratzler shear force; IMF, intramuscular fat; SFA, saturated fatty acids; MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids; TransFA, trans fatty acids. All the correlation values are significant (P ≤ 0.05), and nonsignificant coefficients (P > 0.05) are masked with blank.

PCA clustering of significant traits associated with eating quality

Variables showing significant correlations with eating quality traits were retained for PCA clustering (Figure 2). The first 2 principal components explained 47.3% of the total variance, with PC1 accounting for the largest share. In the PCA plot, consumer-evaluated eating quality traits were inversely related to compression force, defining the main axis of discrimination. On the other hand, the MSA index was closely associated with several carcass and muscle traits, including carcass weight, conformation, eye muscle area, muscle color parameters (a* and b*), IMF, marbling, and fat thickness. In contrast, age, fat color, and SFA showed negative associations with MSA index and carcass and muscle traits. Eating quality traits presented limited direct correlations with both groups of variables.

Figure 2.

Distribution of eating quality-associated traits in PCA. EMA, eye muscle area; CW, carcass weight; IMF, intramuscular fat; MUFA, monounsaturated fatty acids; Marbling10, MSA marbling score at 10^th rib; FT, fat thickness; Compression, compression at 80%; Fat color5, MSA fat color score at 5^th rib; SFA, saturated fatty acids; MQ4 = 0.3 × tenderness + 0.1 × juiciness + 0.3 × flavor liking + 0.3 × overall liking; MSAs, MSA grade evaluated by untrained consumers; Low means a low-MSA grade (namely 2), and high means the MSA grade (3, 4, or 5) for the 2 studied cuts.

Based on the MSA grades rated by consumers, samples from 2 cuts were divided into high (n = 101) and low (n = 93) groups using a threshold of 3. High-MSA grade (i.e., 3, 4, or 5) samples were predominantly associated with favorable carcass and muscle traits linked to eating quality. Conversely, low-MSA grade samples tended to be associated with older animals, higher compression force, higher pH, and higher SFA levels. These patterns suggest that high-quality beef integrates multiple favorable biological features, while low-quality beef is characterized by homogeneous and less favorable structural and physiological traits.

Model performance of the stepwise linear mixed-effects model of eating quality traits

Variables selected from the PCA were used to run linear mixed-effects models (Table 4). In all models of consumer-evaluated eating quality, Animal ID was included as a random effect, with Marbling10 and other carcass traits entered sequentially as fixed effects. Model performance was evaluated using both marginal R² (variance explained by fixed effects) and conditional R² (variance explained by both fixed and random effects). The model with all data and with only Animal ID (random effect) and Marbling10 (fixed effect) had a small impact (marginal R² ≤ 0.01). In contrast, the conditional R² reached 0.33 − 0.43, indicating that Animal ID accounted for nearly all of the explained variance. Adding age, compression, and conformation step by step led to gradual increases in marginal R² up to 0.21, whereas the conditional R² remained largely stable (around 0.33 for tenderness up to around 0.45 for overall liking). As the MSA index is an indicator at the carcass level, Animal ID was not employed as a random effect in this model, and the R² value contributed by other indicators reached 0.62.

When considering subsets of samples based on MSA grade, models for samples with MSA grade < 3 showed generally lower marginal R² (≤ 0.03) and R² of the MSA index models (0.57). In contrast, for samples with MSA grade ≥ 3, the addition of age, compression, and conformation significantly improved marginal effects up to 0.16 for tenderness and 0.33 for juiciness (P < 0.05), and presented a better model performance than the model with all samples. Notably, in the MSA index models, Marbling10 (core predictor) contributed a much higher R² (0.42) in the samples with low-MSA grades compared with the other groups.

Correlations between carcass, muscle, and eating quality traits in Limousine cull cows

As expected, the eating quality traits of beef from old cows were strongly intercorrelated (r = 0.67–0.93), consistent with previous findings reported in calves, young cattle, and diverse European beef cattle populations (Thompson, 2004; Polkinghorne et al., 2011; Bonny et al., 2017; Liu et al., 2020b; Pogorzelski et al., 2020). However, in the PCA plot, the MSA index clustered with carcass traits and was nearly orthogonal (i.e., almost independent) of sensory traits, indicating its independence from MQ4. This relatively weak relationship suggests that the MSA index, developed from multiple muscles and production contexts, may not fully capture eating quality variation in specific muscles of European cull cows. However, it is worth noting that the MSA index was generated by combining MQ4 from 39 different muscles of the whole carcass (McGilchrist et al., 2019), whereas in this study, we considered MQ4 from only 2 muscles, which could explain the moderate correlation (r = 0.23) between MQ4 and the MSA index.

The indicators clustered with the MSA index could be divided into 2 main groups. The first group is carcass yield-related traits, including carcass weight, conformation, and eye muscle area, which were, as expected, strongly correlated with each other (r > 0.70), in line, for example, with the results of Santinello et al. (2025). In the present study involving older animals, these traits were positively correlated with beef eating quality traits (0.18–0.43) and the MSA index (0.54). In contrast, in younger animals, these traits mainly reflect muscular development rather than eating quality (Santinello et al., 2025). The second group comprised fat deposition traits, consisting of IMF, Marbling10, and fat thickness. These variables were positively correlated with the MSA index (0.27–0.47), consistent with findings in younger female cattle (Santinello et al., 2025), but not with eating quality traits evaluated by consumers. This indicates that, in older Limousine cows, marbling carries less importance for French consumer sensory perception of the 2 studied cuts (STR and RMP) than what would be expected based on the MSA model. When included in the core prediction model, marbling also contributed to a negligible marginal impact. This limited influence might be associated with the low-fat content of these cull cow beef, as 61% had an IMF level below 3%, which is considered the minimum threshold for acceptable palatability (O’Quinn et al., 2012). Especially in older cows, a greater amount of IMF is required to disrupt the strongly cross-linked connective tissue compared with younger animals. In such lean carcasses, marbling was the least powerful palatability indicator, while the muscle texture became a stronger determinant (Jackman et al., 2010). In addition, in this study, the compression factor was a key determinant of eating quality, while WBSF showed no significant association. Although both WBSF and compression were strongly correlated with sensory tenderness attributes (r > 0.7; Peachey et al., 2002), compression could better capture the contribution of connective tissue toughness (Lepetit and Ouali, 1986). Furthermore, tenderness in cull cow beef was mainly determined by the characteristics of connective tissues and other myofibrillar properties (Nondorf et al., 2022). These results reinforce the idea that compression measurements are thus particularly relevant in mature animals and should be considered as important indicators for European cull cow evaluation. Overall, these results indicate a shift in the biological determinants of eating quality with animal maturity, with structural muscle traits becoming more influential than fat-related traits. Notably, in models fitted across all samples, the marginal R² values consistently accounted for less than 60% of the conditional R², indicating that a substantial proportion of the explained variance was attributable to random effects (Animal ID). This highlights that a substantial proportion of variability in eating quality remains unexplained by the measured predictors. This may reflect the influence of additional biological factors such as muscle microstructure, metabolic properties, or pre-slaughter conditions, which were not captured in the present study.

Why Marbling may be less important as a consumer eating quality indicator in older cows

Marbling is an important criterion for determining the commercial value of the premium French beef brand “Or Rouge” of the Beauvallet Company, which is exclusively based on the Limousine breed (Santinello et al., 2024a). It is generally positively associated with eating quality attributes such as tenderness, juiciness, and flavor (Choi et al, 2019; Shahrai et al., 2020). However, in this study, marbling did not show any significant association with consumer eating quality traits in cull cows for the 2 studied muscles. This may be attributed to differences in the muscle and fat development stages between cull cows and younger animals. For the young, late-maturing breeds, the marbling score is typically higher at the 5^th rib site than the 10^th rib (Santinello et al., 2024a), supporting the hypothesis that intramuscular fat starts to deposit from the anterior to the posterior part (Acheson et al., 2018). In contrast, Limousine cull cows slaughtered at a fully mature growth phase tend to exhibit homogeneous marbling deposition across grading sites. This pattern was confirmed in our study, where the difference between the Marbling5 and Marbling10 average scores was less than 10 on a scale of 1190. This finding is also consistent with the study of Liu et al. (2021) on Limousine cows of a comparable age (approximately 9 years of age). Despite this stability in marbling, our PCA analysis revealed an opposite cluster direction between age and carcass yield or fat deposition traits. A similar pattern was reported by Latta et al. (2024) in cull cows of various genetic backgrounds. These relationships confirm that older cows are no longer in a growth phase and may begin to deplete both muscle and fat reserves in case of reduced feed intake. Under such conditions, variation in marbling among cows raised under the same feeding system is likely to be driven primarily by genetic factors, as their fat deposition potential has already been largely reached.

These specific physiological relationships related to slaughter age also influence the correlations among carcass traits, particularly muscle development traits and fat deposition traits. In this study, marbling and conformation were positively correlated (r = 0.24). However, several studies have documented a weak or negative association between these traits (Bonny et al. 2016b; Conroy et al., 2010). A broader negative relationship between muscle development traits (eye muscle area, conformation) with fat development traits (fat thickness and marbling) had also been reported in younger animals (Santinello et al., 2025). Similarly, Liu et al. (2020a) observed a negative correlation between marbling and conformation in cows with an average age of around 7 years. Nevertheless, only 56% of the cattle in their study were late-maturing breeds, and they emphasized that such a negative correlation can be observed only when fat synthesis is still maintained. Once the fat deposition phase is complete, as in mature cull cows, both marbling and conformation may increase concurrently, but not fat thickness. Such concurrent variation may also reflect differences in genetic background, lifetime nutritional status, or other management factors influencing muscle and intramuscular fat development.

Consequently, even though marbling was intentionally included in the linear mixed-effects models because of its crucial biological relevance to eating quality, it had almost no contribution to explaining the eating quality of the 2 studied muscles. This limited effect remained consistent across both high and low-MSA grade samples. Thus, it suggests that marbling should not be considered a universal predictor of sensory performance.

Model performance varied by consumer-evaluated MSA grades

Marbling was first included in the core model as a baseline predictor due to its crucial biological relevance to beef eating quality. Age, compression force, and conformation were subsequently incorporated into the linear mixed-effects models of predicting eating quality traits, as the inclusion of these variables significantly increased the marginal R² values. Together with Animal ID, the core model with marbling only explained 33–45% of the variation in consumer eating quality, of which 12–21% explained by the fixed effects alone. Age and conformation are much more important positive determinants of the eating quality of cull cow beef than both marbling and ossification. Incorporating these traits improved the marginal R² of consumer sensory models by up to approximately 0.20, and increased the R² of the MSA index model by 0.40. Indeed, in highly mature cows, because ossification approached and even reached the maximum value, animal age had a greater magnitude of effect on eating quality (Bonny et al., 2016b). In addition, with increasing age, the muscle fiber cross-sectional area gradually increases (Couvreur et al., 2019), which is likely to explain why conformation is likely to be important in this study. Although European classification scores had been proven to be poorly related to the eating quality at the consumer level (Bonny et al., 2016a; Santinello et al., 2024b), those studies were based on cattle younger than 3 years. In mature cull cows, by contrast, higher conformation reflects stronger muscle development, which emerges as a key driver of sensory performance.

However, model performance differed across samples with different MSA grades. Using an MSA grade threshold of 3, which corresponds to an MQ4 score of approximately 61.0 out of 100.0, based on evaluations from French consumers (Legrand et al., 2013; Liu et al., 2023), the models performed substantially better for high grade samples than for low grade ones. In the present dataset, low-MSA samples were generally characterized by older animal age and lower conformation, compared to high-MSA samples. For samples with MSA grade < 3, model performance was lower than for all samples, with conditional R² less than 0.32, and marginal R² values contributed by fixed effects alone less than 0.05. Surprisingly, in the low-MSA grade subset, marbling alone showed an R² of 0.42, accounting for almost 3/4 of the explanatory power of the final model. This suggests that in low-quality beef, the MSA index was more strongly associated with marbling, while the higher quality samples became more dependent on a combination of muscle development traits. Conversely, samples with MSA ≥ 3 showed higher conditional R², marginal R², and R² values. This difference in model performance between high and low-MSA grades may also suggest that a minimum level of marbling is required to achieve acceptable eating quality. Below such a threshold, the contribution of marbling to sensory variation may be limited. However, only 2% of carcasses had Marbling5 or Marbling10 scores above 500 in the present dataset, which indicates a relatively narrow marbling range and makes it difficult to identify such a threshold. Under these conditions, the observed pattern may also reflect a compensatory effect of fat in tougher samples. In addition, French consumers may be relatively accustomed to leaner beef products (Liu et al., 2021), which could further reduce the apparent influence of marbling on eating quality scores.

The PCA distribution was consistent with this observation. High-MSA grade samples were distributed across the entire PC1 axis, extending towards the left quadrant where carcass yield traits, fat deposition traits, and eating quality traits were projected. In contrast, low-MSA grade samples generally exhibit lower fat content and carcass yield traits, as well as higher age, pH, compression force, and SFA content. Because of the homogeneity of these characteristics and the high importance of marbling in the MSA index prediction system (Meat & Livestock Australia, 2023), even small variations in the marbling of low-quality samples were sufficient to drive differences in the MSA index. In addition, when overall eating quality is low, consumer evaluations may rely more strongly on overall liking rather than clear differentiation among individual eating quality attributes, which may weaken the relationship between carcass traits and eating quality scores. Indeed, mature animals typically exhibited lower eating quality scores, likely due to collagen crosslinking formation related to age (Santinello et al., 2025). The narrower range of eating quality variation in low-MSA samples further limited model performance.

By contrast, high-MSA samples exhibit more comprehensive biological qualities and clearer structural mechanisms, in line with the outputs of linear mixed models. Notably, in the high-MSA grades samples, the conditional R² and marginal R² values for juiciness were nearly identical (R²_cond = 0.34; R²_mar = 0.33). This indicates that the fixed effects almost entirely explained the variation in juiciness, while individual animals contributed minimally. Juiciness may thus serve as a sensory attribute that can be reliably predicted through the studied variables.

In summary, the performances of the different models across MSA grades reflect the inherent biological heterogeneity of beef quality. High-quality samples are better suited for sensory prediction using linear structured models. For low-MSA samples, sources of sensory variability may require further investigation, potentially through integration of muscle metabolomics or lipidomics data.