Integrated volatilomics and lipidomics reveal aroma-related volatile and lipidomic changes associated with Enterococcus faecalis JM263 supplementation in yak meat

Abstract

Lipid composition is closely associated with meat aroma formation, yet the associations between probiotic supplementation and aroma-related profiles in yak meat remain unclear. Here, an integrated analysis combining electronic nose, volatilomics, and lipidomics was performed to investigate the association of Enterococcus faecalis JM263 supplementation with aroma-related changes in yak meat. The electronic nose revealed distinct instrumental aroma profiles between groups. GC–MS and LC–MS/MS screening yielded 24 exploratory candidate differential volatile features and 60 exploratory candidate differential lipids. E. faecalis JM263 supplementation was associated with changes in aldehydes, ketones, esters, hydrocarbons, terpenoids, and lipid metabolic features. Five FDR-supported lipid features, including PE(16,1_17,1), PE(17,0_18,1), PA(18,0_18,1), LNAPE(18,1/N-17,0), and PE(16,0_22,4), were identified. Mantel analysis revealed exploratory associations between electronic nose responses and volatile or lipid profiles. These findings provide exploratory insights into supplementation-associated changes in aroma-related volatile profiles and lipid metabolic features in yak meat.

Graphical abstract

Highlights

• •E. faecalis JM263 modulated yak meat aroma-related volatile profiles.
• •E-nose distinguished control and E. faecalis-supplemented yak meat.
• •Volatile changes were linked to fruity and fatty aroma descriptors.
• •Selected lipids were candidate features of aroma-related metabolic variation.
• •Multi-omics analysis showed exploratory matrix-level associations.

1. Introduction

Yak (Bos grunniens) is the most emblematic livestock species of the Qinghai–Tibet Plateau. Having long lived above 3000 m in cold, hypoxic, and high-UV environments, yaks have developed distinctive physiological and metabolic traits as well as advantageous meat qualities (Kulyar et al., 2025). Yak meat is an important source of animal protein for plateau herders and is nutritionally notable for its high protein content, low fat and cholesterol levels, and rich mineral profile (Li et al., 2023). Compared with beef from conventional cattle, yak meat exhibits distinctive sensory characteristics and nutritional properties, which contribute to its differentiation from other beef products (Gu et al., 2021). These unique flavor characteristics give yak meat substantial market potential and consumer appeal, attracting growing attention in domestic and international high-end meat markets.

Flavor is a key determinant of meat quality and consumer acceptance. When selecting meat products, consumers consider nutrition and safety but often rely most heavily on sensory attributes such as color, tenderness, and flavor. Meat flavor formation is a complex biochemical process closely linked to the production of volatile compounds through lipid oxidation, amino acid metabolism, carbohydrate degradation, and their interactions (Qiu et al., 2022; Zhou and Shen, 2025b). During cooking or storage, lipid breakdown yields aldehydes, ketones, alcohols, and carboxylic acids. While Maillard reactions generate heterocyclic compounds (e.g., pyrazines, thiophenes, and indoles); together, these compounds form the basis of meat flavor (Yao et al., 2025). Therefore, differences in lipid metabolic patterns and amino acid transformation pathways may contribute to variations in aroma-related characteristics among species, breeds, and rearing systems.

With rising domestic and international demand for yak meat, traditional grazing systems are increasingly unable to meet the needs for scaled production and stable supply. In response, the Qinghai–Tibet Plateau has been improving confined fattening to enhance the production efficiency, reduce fattening cycles, and ensure a stable meat supply. However, many studies and production experiences indicate that, compared with traditional grazing, confinement feeding can adversely affect the sensory quality of yak meat—especially flavor (Zhang et al., 2025). This is mainly attributed to the homogeneity of barn rations, reduced physical activity, and shifts in metabolic patterns, which alter the fat deposition and muscle metabolic pathways and thus influence the formation of volatile flavor compounds. This problem has become one of the major constraints to the high-quality development of the yak meat industry.

In recent years, nutritional regulation and microecological intervention have provided new strategies to address the decline in meat quality under confinement feeding systems (Zhou and Shen, 2025c). Plant-derived bioactive compounds, including polyphenols recovered from fruit-processing materials, have also attracted attention as functional ingredients (Tahir et al., 2024). The intestinal microbiota plays an important role in energy metabolism, lipid deposition, immune regulation, and host metabolic homeostasis (Chen et al., 2025). Recent studies in weaned yaks showed that Tibetan herbal interventions were associated with enhanced antioxidant capacity and changes in gut microbiota or inflammatory mediators (Cidan et al., 2025; Zhu et al., 2025). In a different animal model, probiotic supplementation also affected spermatozoa quality in high-fat-diet-fed male rats (Bahmid, R. and Nurul Sulfi Andini, 2025), suggesting that dietary interventions may be accompanied by systemic physiological responses beyond the gastrointestinal tract.

Probiotic supplementation has been associated with changes in gut microecology, metabolic status, and meat quality-related characteristics in livestock and poultry (Usman et al., 2024; Zhou and Shen, 2025a; Zhang et al., 2023). In ruminants, probiotic supplementation has also been associated with changes in rumen fermentation patterns, nutrient utilization efficiency, muscle lipid composition, and metabolic activity. Therefore, probiotic application may represent a potential nutritional strategy for exploring associations with aroma-related metabolic traits in confined yak production. Previous studies have shown that supplementation with Lactobacillus species improves the tenderness and flavor in pork and chicken (Lin et al., 2025; Wang et al., 2017), while Bacillus subtilis, Lactobacillus plantarum, and Saccharomyces cerevisiae enhance fatty acid deposition and antioxidant capacity (Yang et al., 2025). Lactiplantibacillus plantarum has also been applied for value addition in beef sausage products (Lukman et al., 2025). These findings provide a theoretical foundation for the use of probiotics in improving meat quality. However, the biological basis underlying probiotic-associated changes in muscle quality and flavor formation in livestock remains largely unclear. Therefore, exploring probiotic strains and their mechanisms of action suitable for high-altitude ruminants such as yaks holds both theoretical significance and practical potential.

Enterococcus faecalis is a facultative anaerobic lactic acid bacterium, widely distributed in the intestinal tracts of animals, fermented foods, and natural environments (Witzler et al., 2017). It exhibits strong environmental adaptability and metabolic diversity. In recent years, E. faecalis has attracted increasing attention in animal production owing to its tolerance to gastric acid and bile salts and its potential roles in cellulose degradation, lactic acid production, and secondary metabolism regulation (Shehata et al., 2020; Wang et al., 2019). Previous studies have shown that E. faecalis can improve the rumen microbial community structure, optimize nitrogen metabolism, and promote volatile fatty acid production in ruminants. It can also indirectly influence the quality of meat by modulating fatty acid composition and antioxidant enzyme activity (Mamuad et al., 2019). In addition, E. faecalis has been reported to suppress pathogenic bacteria and enhance host immunity, providing a theoretical basis for its potential use in yak production under the cold, hypoxic conditions of the Qinghai–Tibet Plateau (Mansour et al., 2018). In our previous work, we isolated a potential probiotic strain, E. faecalis JM263, from the intestinal tract of free-grazing yaks in Linzhou County, Tibet, and systematically evaluated its probiotic characteristics. E. faecalis JM263 exhibited tolerance to simulated gastric and intestinal fluids and bile salts, antimicrobial and antioxidant activities, strong fiber-degrading capacity, and favorable effects on in vitro rumen fermentation (Cidan et al., 2025). More recently, a 60-day goat feeding study using dietary E. faecalis JM263 supplementation at 0.1, 0.5, and 1.0 g/kg provided preliminary in vivo evidence of its functional efficacy and tolerability under high-concentrate-diet conditions. E. faecalis JM263 supplementation was associated with improvements in growth-related indices, antioxidant and immune status, intestinal barrier integrity, and hepatic health (Zhou et al., 2026). These previous findings provide a possible biological context in which probiotic-associated host physiological responses may be accompanied by subsequent changes in muscle lipid metabolism and aroma-related characteristics.

Nevertheless, studies on the role and mechanisms of E. faecalis in yak meat quality and flavor formation remain limited. Most existing research has focused on its effects on growth performance, immune function, and intestinal health. While its direct influence on meat flavor formation has not been systematically elucidated. In particular, integrated multi-omics approaches combining electronic nose, flavoromics, and lipidomics technologies have rarely been applied in this context. The electronic nose, an analytical tool mimicking the human olfactory system, can rapidly and objectively evaluate the overall flavor profile of meat and is suitable for macroscopic flavor comparison (Zong et al., 2025). Flavoromics enables qualitative and quantitative analysis of key volatile compounds in meat, revealing the chemical basis of flavor formation (Cheng et al., 2025), whereas lipidomics provides a comprehensive understanding of lipid composition and metabolic networks, elucidating the central role of lipids in flavor generation (Wu et al., 2021). Integrating these three technologies offers a multi-dimensional framework for characterizing aroma-related molecular variations and exploring potential metabolic associations linked to E. faecalis JM263 supplementation in yak meat (Liu et al., 2025).

Therefore, this study systematically investigated the aroma-related changes in the longissimus dorsi muscle of fattening yaks on the Qinghai–Tibet Plateau following dietary supplementation with E. faecalis JM263. The instrumental aroma-related profile was characterized using electronic nose analysis, volatile features were characterized through flavoromics, and the lipidomic profile was characterized via lipidomics. Together, these analyses provide an integrated framework for exploring the associations of E. faecalis JM263 supplementation with aroma-related volatile profiles and lipid-derived aroma precursor metabolism in yak meat. These findings provide new insights into the aroma-related chemical characteristics of yak meat and a theoretical and practical foundation for the application of probiotics in confined yak production and the development of high-quality plateau meat products.

2. Materials and methods

2.1. Ethics statement

All animal experimental procedures were conducted in accordance with the national standard Laboratory Animal – Guideline for Ethical Review of Animal Welfare (Beijing, China). The experimental protocols and animal care were approved by the Animal Protection Committee of the Xizang Academy of Agriculture and Animal Husbandry Sciences (approval No. XKS2024087) and complied with all relevant national and international regulations on animal experimentation.

2.2. Animals and feeding management

This experiment was conducted at the cattle and sheep industrial park experimental base in Jiachuoxiong Township, Shigatse, Tibet, China. The site is located at an altitude of approximately 4000 m (29°17′ N, 88°52′ E) and features a plateau semi-arid climate, characterized by cold and dry winters, cool and humid summers with concentrated rainfall, and large diurnal temperature variations. The mean annual precipitation is about 300 mm, and the average annual temperature is approximately 6.5 °C. Twelve healthy male yaks (280 ± 15.36 kg) were randomly assigned to two groups: a control group (CON) fed a basal diet, and an Enterococcus faecalis group (EFA) supplemented with E. faecalis JM263 at 0.5 g/kg of feed (1 × 10⁸ CFU/g). Before group allocation, each yak was assigned a unique identification number. Random numbers were generated using SPSS software (version 22.0), and the animals were allocated according to the generated random sequence to the CON and EFA groups, with six yaks per group. The E. faecalis JM263 preparation was thoroughly mixed with the basal diet immediately before feeding to ensure uniform distribution. Each yak was housed individually in a separate pen. Animals in both groups were maintained under the same routine farm husbandry and feeding conditions and had ad libitum access to clean water. The feeding trial lasted for 75 days, including a 15-day adaptation period followed by a 60-day experimental period. The ingredients and nutrient composition of the basal diet are provided in Supplementary Table S1.

2.3. Muscle sample collection

All twelve yaks were transported to a commercial slaughterhouse at the end of the fattening period, and the transportation duration was approximately 45 min. Before slaughter, the animals were fasted for 12 h. The yaks were slaughtered according to Halal procedures by a qualified imam. Muscle sample collection was initiated as soon as possible after slaughter using the same standardized procedure for all animals. Longissimus dorsi muscle samples were excised from the left carcass between the 9th and 12th ribs, immediately trimmed of external fat, and transferred directly to liquid nitrogen. The entire procedure from slaughter to cryogenic freezing was completed within approximately 15 min for each animal. No additional analytical sample preparation was performed before cryogenic freezing. The collected samples were subsequently stored at −80 °C until further analysis. The preserved samples were used for electronic nose (E-nose) analysis, GC–MS-based semi-quantitative volatile metabolomics analysis, and LC–MS/MS-based semi-quantitative lipidomics analysis. Each analytical sample corresponded to the longissimus dorsi muscle collected from one individual yak; therefore, the six samples in each group represented six independent biological replicates.

2.4. Electronic nose detection

Bionic olfactory data of muscle samples were collected using a PEN3 electronic nose (AIRSENSE Analytics GmbH, Schwerin, Germany). Approximately 3 g of minced muscle sample was accurately weighed into a 15 mL headspace vial and equilibrated at 25 °C for 30 min. The injection needle was then inserted directly into the sealed vial for measurement. For each biological sample, three independent technical replicate measurements were performed under the same analytical conditions. The mean value of the three technical replicates was used as the electronic nose response value for that individual yak in subsequent statistical analysis. The operating parameters were as follows: sampling time, 1 s per cycle; sensor self-cleaning time, 80 s; sensor zeroing time, 5 s; sample preparation time, 5 s; injection flow rate, 400 mL/min; and analysis sampling duration, 80 s.

2.5. GC–MS-based semi-quantitative volatile metabolomics

2.5.1. Sample preparation

Muscle samples stored at −80 °C were ground into powder under liquid nitrogen. Then, 0.2 g of the powdered sample was immediately transferred into a 20 mL headspace vial (Agilent, Palo Alto, CA, USA) containing 0.2 g of sodium chloride to inhibit any enzymatic activity. For semi-quantitative analysis of volatile compounds, 3-Hexanone-2,2,4,4-d₄ was used as the internal standard. A 20 μL aliquot of the internal-standard solution at 10 μg/mL was added to each vial, corresponding to 0.20 μg of internal standard per sample. The exact sample mass was recorded and used in the semi-quantitative calculation. The internal standard was supplied by CDN/Metware as a customized deuterated standard (internal code: IS-04; Shanghai, China). The vial was sealed with a crimp cap fitted with a polytetrafluoroethylene–silicone headspace septum (Agilent). For GC–MS-based volatile metabolomics, one analytical sample was prepared from each individual yak, resulting in six biological replicates per group. For SPME analysis, each vial was equilibrated at 60 °C for 5 min, after which a 120 μm DVB/CWR/PDMS SPME Arrow (Agilent Technologies, Santa Clara, CA, USA) was exposed to the vial headspace for 15 min at 60 °C. Before sample analysis, the SPME Arrow was conditioned in the GC injection port according to the manufacturer's recommended procedure until a stable blank background was obtained. The same type of SPME Arrow was used throughout the analytical sequence, and each Arrow was used for no more than 30 extraction–desorption cycles, including sample analyses, quality-control injections, and blank runs. Blank vials were analyzed after fiber conditioning and periodically during the sequence to monitor possible carryover or contamination.

2.5.2. GC–MS conditions

After sampling, volatile organic compounds (VOCs) adsorbed on the SPME Arrow were desorbed in the GC injection port of an Agilent 8890 gas chromatograph (Agilent Technologies, Santa Clara, CA, USA) at 250 °C for 5 min. After each desorption step, the SPME Arrow was maintained in the injection port for an additional cleaning/desorption period when necessary to minimize carryover. The absence of obvious residual peaks was confirmed by blank runs before subsequent sample analysis. VOCs were analyzed using an Agilent 8890 gas chromatograph coupled to an Agilent 7000E triple-quadrupole mass spectrometer (Agilent Technologies, Santa Clara, CA, USA), equipped with a 30 m × 0.25 mm × 0.25 μm DB-5MS column (5% phenyl-polymethylsiloxane). Helium was used as the carrier gas at a flow rate of 1.2 mL/min, and the injection port temperature was maintained at 250 °C. The column oven temperature was programmed as follows: 40 °C for 3.5 min, ramped at 10 °C/min to 100 °C, then 7 °C/min to 180 °C, and finally 25 °C/min to 280 °C, held for 5 min. Mass spectra were recorded in electron ionization (EI) mode at 70 eV. The quadrupole mass analyzer, ion source, and transfer line were set at 150 °C, 230 °C, and 280 °C, respectively. VOCs were putatively annotated based on characteristic quantitative and qualitative ions and the available reference information, including molecular weight, molecular formula, CAS number, and NIST reference retention index (NIST_RI). The quantitative and qualitative ions and the corresponding annotation information for each volatile compound are provided in Supplementary Table S4. VOCs were semi-quantified using selected ion monitoring (SIM) mode with reference to the internal standard response.

2.5.3. Data analysis

Unsupervised principal component analysis (PCA) was performed using the prcomp function in R (www.r-project.org). Before PCA, the data were standardized to unit variance. The CV distribution and PCA including QC samples were used to evaluate analytical reproducibility. Hierarchical clustering analysis (HCA) of samples and metabolites, based on Pearson correlation coefficients, was visualized as heatmaps with dendrograms. Pairwise Pearson correlation coefficients (PCCs) among samples were calculated using the cor function in R and displayed as heatmaps. Differential metabolites between the two groups were identified based on VIP > 1, fold change (FC) > 1.5 or < 0.67, and p < 0.05. The semi-quantitative concentration of compound i was calculated as $C_i=\left[\left(V_{\text{IS}}\times C_{\text{IS}}\right)/M\right]\times \left({\mathrm{A}}_{\mathrm{i}}/{\mathrm{A}}_{\text{IS}}\right)\times 10^{-3}$, where${\mathrm{V}}_{\text{IS}}$ is the volume of the internal-standard solution (20 μL),${\mathrm{C}}_{\text{IS}}$ is its concentration (10 μg/mL),$\mathrm{M}$ is the exact sample mass (g), and ${\mathrm{A}}_{\mathrm{i}}$ and ${\mathrm{A}}_{\text{IS}}$ IS  are the peak areas of compound $\mathrm{i}$ and the internal standard, respectively. The final unit of ${\mathrm{C}}_{\mathrm{i}}$ was μg/g of muscle sample. Because compound-specific calibration curves, extraction recoveries, and response factors were not available,${\mathrm{C}}_{\mathrm{i}}$ was considered a semi-quantitative concentration estimate. Relative odor activity values (rOAVs) were calculated to estimate the potential contribution of volatile compounds to the aroma profile. The rOAV of compound i was calculated as rOAVᵢ = Cᵢ/Tᵢ, where Cᵢ represents the internal-standard-assisted semi-quantitative concentration estimate of compound i (μg/g), and Tᵢ represents the reported odor threshold expressed in the same unit (μg/g). Before calculation, all odor thresholds were converted to the same unit as ${\mathrm{C}}_{\mathrm{i}}$ (μg/g), resulting in dimensionless rOAV values. Compounds with thresholds that could not be reliably converted to μg/g were excluded from the rOAV analysis. Therefore, rOAV >1 was used only to screen candidate odor-relevant compounds rather than to confirm aroma activity. Odor threshold values were obtained from a unified threshold dataset compiled from published odor-threshold databases and food flavoromics literature, including Van Gemert's odor threshold compilation and recent rOAV-based aroma studies (Fang et al., 2024; Huang et al., 2022; Xue et al., 2022). When multiple threshold values were available, the value reported in water or the most comparable food matrix was preferentially used. For compounds without reliable published odor thresholds, rOAVs were not calculated and these compounds were not used for key aroma-active compound interpretation. Benjamini–Hochberg false discovery rate (FDR) correction was additionally performed to assess the influence of multiple testing, and both raw p-values and FDR-adjusted q-values are provided in Supplementary Table S4. Because the present study was designed as an exploratory volatilomics analysis with a limited sample size, FDR-adjusted q-values were used to evaluate statistical robustness but were not used as the sole criterion for candidate feature discovery. This combined criterion of VIP value, fold change, and statistical significance was used as an exploratory screening strategy to identify candidate differential volatile metabolites, rather than to define definitive biomarkers. VIP values were extracted from orthogonal partial least squares discriminant analysis (OPLS-DA) models generated using the MetaboAnalystR package in R. Before OPLS-DA, data were log2-transformed and mean-centered. Model performance was evaluated using R²X, R²Y, and Q² values, representing the explained variance of predictors, explained variance of class discrimination, and predictive ability of the model, respectively. A permutation test with 200 iterations was conducted to evaluate model robustness and potential overfitting. OPLS-DA was used as a supervised exploratory tool, and biological interpretation was not based solely on visual group separation or VIP ranking.

2.6. LC–MS/MS-based semi-quantitative lipidomics analysis

2.6.1. Sample preparation and extraction

Muscle samples were removed from −80 °C storage and thawed on ice. After thorough grinding under liquid nitrogen, 20 mg of sample was weighed and mixed with 1 mL of extraction solvent containing internal standards (MTBE: MeOH = 3:1, v/v). The lipid internal standard mixture was used for signal correction and semi-quantitative comparison among samples. The detailed information on the lipid internal standards, including lipid class, internal standard name, CAS number, supplier, catalog number, and concentration, is provided in Supplementary Table S2. Because compound-specific external calibration curves were not constructed for all detected lipid species, the lipidomics data were reported as semi-quantitative values rather than absolute concentrations. For lipidomics analysis, one analytical sample was prepared from each individual yak, resulting in six biological replicates per group. The mixture was vortexed for 15 min, followed by addition of 200 μL ultrapure water and a further 1 min of vortexing. Samples were then centrifuged at 12,000 rpm for 10 min. A 200 μL aliquot of the upper organic phase was collected and evaporated to dryness using a vacuum concentrator. The dried extracts were reconstituted in 200 μL of reconstitution solution (ACN: IPA = 1:1, v/v) for LC-MS/MS analysis. A pooled quality-control (QC) sample was prepared by combining aliquots of the reconstituted extracts from all biological samples. QC samples were injected periodically throughout the analytical sequence, including at the beginning, middle, and end, under the same LC–MS/MS conditions as the biological samples. Analytical stability was evaluated using total-ion-current overlap, Pearson correlations among QC samples, and the coefficient-of-variation distribution of detected lipids. In addition, feature-level CV values among QC samples were calculated to quantitatively assess analytical reproducibility.

2.6.2. HPLC conditions

Lipid extracts were analyzed using an ExionLC™ AD UPLC system coupled to a QTRAP® 6500+ LC–MS/MS system (SCIEX, Framingham, MA, USA). Chromatographic separation was performed on a Thermo Scientific™ Accucore™ C30 column (2.6 μm, 2.1 mm × 100 mm, Thermo Fisher Scientific, Waltham, MA, USA). The mobile phase consisted of solvent A (acetonitrile/water 60:40, v/v, with 0.1% formic acid and 10 mM ammonium formate) and solvent B (acetonitrile/isopropanol 10:90, v/v, with 0.1% formic acid and 10 mM ammonium formate). The gradient program was: 0 min, A/B 80:20; 2 min, 70:30; 4 min, 40:60; 9 min, 15:85; 14 min, 10:90; 15.5 min, 5:95; 17.5 min, switch to 80:20; and 20 min, hold at 80:20. The flow rate was maintained at 0.35 mL/min, the column temperature was 45 °C, and the injection volume was 2 μL. The eluent was directly introduced into a triple quadrupole–linear ion trap (QTRAP) mass spectrometer.

2.6.3. ESI-MS/MS conditions

MS analysis was performed on a QTRAP® 6500+ LC-MS/MS system (SCIEX, Framingham, MA, USA) equipped with a Turbo Ion-Spray ESI source, operating in both positive and negative ion modes and controlled by Analyst 1.6.3 software (Sciex). Source parameters were as follows: turbo spray, source temperature 500 °C; ion spray voltage (IS) 5500 V (positive) and − 4500 V (negative); gas 1 (GS1) 45 psi, gas 2 (GS2) 55 psi, curtain gas (CUR) 35 psi. Instrument tuning and mass calibration were performed in QQQ and LIT modes using 10 μM and 100 μM polypropylene glycol solutions, respectively. QQQ scans were acquired in multiple reaction monitoring (MRM) mode with collision gas (nitrogen) at 5 psi. Declustering potential (DP) and collision energy (CE) for each MRM transition were optimized individually. Specific MRM transitions were monitored according to the metabolites eluting at each time segment. Lipid signals were processed by MetWare (http://www.metware.cn/) using the AB Sciex QTRAP6500 LC–MS/MS platform. Lipid annotations were reported at the molecular species level based on the lipid class and fatty acyl composition supported by the LC–MS/MS data. The sn-position of fatty acyl chains, double-bond positions, and double-bond geometries were not experimentally resolved. Accordingly, fatty acyl chains with undetermined sn-positions were denoted using an underscore (e.g., PE(16,1_17,1)). Semi-quantitative lipid abundance values were calculated based on MRM peak areas and lipid-class-matched internal standards. For lipid species without an identical isotope-labeled standard, signal correction was performed using the internal standard from the same or most closely related lipid class.

2.6.4. Data analysis

Unsupervised principal component analysis (PCA) was performed using the prcomp function in R (www.r-project.org). Prior to PCA, data were standardized to unit variance. Lipid abundance data used for downstream statistical analysis were obtained from internal-standard-assisted semi-quantitative analysis. For data preprocessing, lipid signals were corrected using the corresponding lipid-class-matched internal standards. PCA was performed after unit-variance scaling, whereas data used for OPLS-DA were log2-transformed and mean-centered. Differential metabolites were identified based on VIP > 1, fold change (FC) > 1.5 or < 0.67, and p < 0.05. Benjamini–Hochberg FDR correction was additionally performed to assess the influence of multiple testing, and both raw p-values and FDR-adjusted q-values are provided in Supplementary Table S6. Considering the exploratory nature of the lipidomics analysis and the limited number of animals per group, FDR-adjusted q-values were used as an indicator of multiple-testing robustness rather than as the sole criterion for excluding candidate lipids. This combined VIP–FC–p-value criterion was used to screen candidate differential lipids in an exploratory manner, rather than to establish definitive lipid biomarkers. VIP values were extracted from orthogonal partial least squares discriminant analysis (OPLS-DA) models generated using the MetaboAnalystR package in R. Before OPLS-DA, data were log2-transformed and mean-centered. Model performance was evaluated using R²X, R²Y, and Q² values, representing the explained variance of predictors, explained variance of class discrimination, and predictive ability of the model, respectively. A permutation test with 200 iterations was conducted to evaluate model robustness and potential overfitting. OPLS-DA was used as a supervised exploratory tool, and biological interpretation was not based solely on visual group separation or VIP ranking. Identified lipids were annotated using the KEGG compound database (http://www.kegg.jp/kegg/compound/) and mapped to KEGG pathways (http://www.kegg.jp/kegg/pathway.html) for further pathway analysis. KEGG pathway enrichment analysis was performed using metabolite set enrichment analysis (MSEA), with enrichment significance evaluated using hypergeometric-test p-values. The background set consisted of all detected lipid features successfully annotated to KEGG pathways. The Rich Factor was calculated as the ratio of the number of candidate differential lipids mapped to a given pathway to the total number of detected lipids annotated to that pathway. Pathways were ranked according to the hypergeometric-test p-values.

2.7. Statistical analysis

For the electronic nose data, normality was evaluated using the Shapiro–Wilk test, and homogeneity of variance was assessed using Levene's test. As the sensor-response variables satisfied these assumptions, differences between the CON and EFA groups were assessed using an independent-samples Student's t-test in SPSS software (version 22.0; SPSS Inc., Chicago, IL, USA). For the volatilomics and lipidomics datasets, between-group p-values used for differential-feature screening were calculated using a two-sided Welch's t-test, which does not assume equal variances between groups. Data are presented as the mean ± standard error of the mean (SEM), and p < 0.05 was considered statistically significant. For the electronic nose data, Benjamini–Hochberg correction was additionally performed to control the false discovery rate across the ten sensor responses. Effect sizes were calculated as Cohen's d using pooled standard deviations. A sensitivity power analysis was performed for the electronic nose data. Under the current design with six animals per group, a two-sided Student's t-test, α = 0.05, and 80% power, the minimum detectable standardized effect size was estimated to be Cohen's d ≈ 1.80; for 90% power, the corresponding value was Cohen's d ≈ 2.08. Therefore, the present design was mainly sensitive to large between-group effects, whereas smaller effects may have been underpowered. Pairwise Pearson correlation coefficients among the selected volatile or lipid features were calculated across the 12 biological samples. Mantel tests were performed using ChiPlot (https://www.chiplot.online/) to separately evaluate the associations between the electronic-nose response matrix and the selected volatile or lipid features. Statistical significance was assessed using the permutation-derived raw p-values generated by ChiPlot. No multiple-testing correction was applied to the Mantel-test results because these analyses were conducted as matrix-level exploratory integration analyses rather than multiple independent feature-level hypothesis tests. Therefore, these associations were interpreted as hypothesis-generating exploratory findings rather than confirmatory statistical evidence.

3. Results

3.1. Electronic nose differentiates aroma-related volatile profiles between the two groups of yak meat

Characteristic odor-related responses of the samples were detected by each electronic nose sensor, allowing subtle differences in the instrumental aroma profile of yak meat to be distinguished. A 10-sensor electronic nose system (Table S3) was used to evaluate differences in odor-related sensor responses between the CON and EFA groups. As shown in Fig. 1A and B, the CON and EFA groups exhibited different overall sensor-response patterns. Statistical comparisons presented in Fig. 1C showed that the EFA group exhibited significantly higher sensor responses compared with the CON group: W1C increased by 32.44%, W5C by 21.30%, W1S by 32.63%, W1W by 67.76%, and W2S by 23.58% (p < 0.05). To further evaluate the robustness of these differences under the limited sample size, effect sizes and Benjamini–Hochberg-adjusted p-values were calculated for all ten sensor responses. The five significantly altered sensors remained significant after multiple-testing correction, including W1C, W5C, W1S, W1W, and W2S. These sensors showed large effect sizes, with Cohen's d values of 2.81, 3.13, 2.44, 4.16, and 2.91, respectively. The sensitivity power analysis indicated that the current sample size could detect large between-group effects corresponding to Cohen's d ≈ 1.80 at 80% power. Thus, the major electronic nose differences were supported by large effect sizes, although smaller sensor responses may have been underpowered. Principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) were then performed based on the sensor responses. Both analyses showed visual separation between the two groups, consistent with the differences observed in the individual sensor responses (Fig. 1D and E).

Fig. 1.

Electronic nose analysis of aroma-related response profiles in yak meat from the two groups. (A) and (B) Radar plots of electronic nose responses for the CON and EFA groups, respectively. (C) Sensor response values of the electronic nose for CON and EFA group meat. **p < 0.01, ***p < 0.001. (D) Principal component analysis (PCA) of electronic nose data from CON and EFA groups. (E) Orthogonal partial least squares discriminant analysis (OPLS-DA) of electronic nose data from yak meat in CON and EFA groups.

3.2. Differences in volatile metabolites between the two yak groups

3.2.1. Putative annotation of volatile features

This study aimed to characterize the volatile metabolite profile associated with aroma-related differences in yak meat following E. faecalis JM263 supplementation using a GC–MS analytical platform. A total of 688 volatile features were putatively annotated in both the CON and EFA groups, of which 677 were shared, while 11 were unique to the EFA group (Fig. 2A, Table S4). The relative odor activity value (rOAV) was used as an instrumental screening index to estimate the potential contribution of each volatile compound to the aroma profile. The odor threshold values used for rOAV calculation were obtained from a unified literature-based threshold dataset, and the compound-specific threshold values, odor descriptors, and sample-specific rOAV results are provided in the revised Supplementary Table S4. Compounds without reliable published odor thresholds were not included in the rOAV-based key aroma interpretation. Forty-two candidate odor-relevant volatile features with rOAV >1 were detected in the CON group, and forty-five such features were found in the EFA group (Fig. 2B). Analytical reproducibility was further evaluated using the CV distribution and PCA including QC samples. The QC samples showed a low-CV distribution and clustered closely in the PCA score plot, supporting the overall stability and reproducibility of the GC–MS analytical data (Fig. 2C and D). Orthogonal partial least squares-discriminant analysis (OPLS-DA) was performed as an exploratory supervised multivariate approach to evaluate differences in volatile metabolic profiles between the CON and EFA groups. The score plot showed visual separation between the two groups (Fig. 2E), and the corresponding 200-permutation test was used to evaluate model robustness and potential overfitting (Fig. 2F). The OPLS-DA model showed R²X = 0.729, R²Y = 0.992, and Q² = 0.651. The permutation test indicated that the original model parameters were higher than those obtained from randomly permuted models (R²Y: p = 0.035, 7/200; Q²: p = 0.030, 6/200), supporting the robustness of the model within the exploratory analytical framework. The putatively annotated volatile features in both groups were classified into eleven categories (Fig. 2G and H): aldehydes, ketones, and esters (30.13% vs. 30.23%); hydrocarbons (24.22% vs. 23.84%); alcohols (6.79% vs. 6.83%); amines (8.71% vs. 8.87%); heterocyclic compounds (9.75% vs. 9.88%); terpenoids (5.76% vs. 5.67%); benzene and substituted derivatives (5.47% vs. 5.38%); organic acids and their derivatives (3.69% vs. 3.78%); ethers (2.22% vs. 2.18%); nitrogen compounds (1.48% vs. 1.45%); halogenated hydrocarbons (1.03% vs. 1.16%).

Fig. 2.

Annotation and quality-control assessment of volatile metabolites in yak meat from the two groups. (A) Venn diagram of volatile features putatively annotated in the CON and EFA groups. (B) Scatter plot of relative odor activity values (rOAVs); the x-axis represents the CON and EFA groups, and the y-axis represents the rOAVs of volatile compounds. (C) Coefficient-of-variation (CV) distribution of volatile features in the CON, EFA, and quality-control (QC) samples. The x-axis represents the CV value, and the y-axis represents the cumulative percentage of detected peaks below the corresponding CV threshold. (D) PCA score plot of the CON, EFA, and QC samples. PC1 and PC2 represent the first and second principal components, respectively, and QC denotes quality-control samples. (E) OPLS-DA score plot; the x-axis indicates the predictive component, and the y-axis indicates the orthogonal component. (F) OPLS-DA validation plot based on 200 random permutation tests; the x-axis represents the permuted-model statistics, and the y-axis represents their frequency. Orange bars represent the R²Y values of randomized models, purple bars represent the Q² values of randomized models, and black arrows indicate the R²X, R²Y, and Q² values of the original model. (G, H) Donut charts showing the classification and relative proportions of volatile metabolites in the CON and EFA groups, respectively. Each color represents a different category of volatile compounds, and the sector area indicates the relative proportion of each category. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

3.2.2. Exploratory candidate volatile features in yak muscle between the two groups

Volatile metabolites were screened based on the exploratory criteria of VIP > 1, FC > 1.5 or < 0.67, and nominal p < 0.05. A total of 24 exploratory candidate differential volatile features were screened, including 16 upregulated and 8 downregulated features (Fig. 3A, Fig. S2, Table S5). FDR-adjusted q-values were additionally calculated and are provided in Table S4 to facilitate assessment of multiple-testing robustness. None of these candidate volatile compounds remained statistically significant after FDR correction, and they were therefore interpreted as exploratory candidates. To visualize the significance and expression trends of these metabolites, VIP plots and heatmaps were generated to compare their expression patterns between the two groups. The top 20 exploratory candidate volatile features with the highest VIP values are shown in Fig. 3B. Among them, features putatively annotated as 2-Propyn-1-ol, propionate (VIP = 2.06, downregulated), 1-Cyclopropyl-2-(3,4-dimethoxyphenyl) ethanol (VIP = 1.96, upregulated), and Cyclohexene, 4-(1,5-dimethyl-1,4-hexadienyl)-1-methyl (VIP = 1.89, upregulated) exhibited the highest VIP scores. The 24 exploratory candidate volatile features were further classified into several chemical categories (Fig. 3C), including aldehydes, ketones, and esters; hydrocarbons; alcohols, amines; and terpenoids. As shown in Fig. 3D, E, and S3, aroma-descriptor annotation associated six exploratory candidate volatile features with specific aroma descriptors. Features putatively annotated as decanoic acid, ethyl ester, pentadecane, and γ-dodecalactone were associated with apple, brandy, waxy, fruity, balsamic, floral, and fatty descriptors, whereas features putatively annotated as cyclohexene, 4-(1,5-dimethyl-1,4-hexadienyl)-1-methyl and benzoic acid were associated with waxy, aldehydic, citrus, balsamic, and faint descriptors.

Fig. 3.

Exploratory candidate volatile features in yak meat from the two groups. (A) Volcano plot of differential metabolites. Each point represents a metabolite, where green indicates downregulated, red indicates upregulated, and gray represents non-significant metabolites. (B) OPLS-DA VIP score plot of the top 20 exploratory candidate volatile features ranked by VIP value. (C) Scatter plot of metabolite classification. Upregulated compounds are represented by upward triangles, downregulated compounds by downward triangles, and non-significant compounds by circles. Triangle size corresponds to the VIP value. (D) Flavoromics Sankey diagram. The left column represents literature-annotated aroma descriptors, and the right column represents differential metabolites. Yellow flows indicate upregulated (Up) metabolites, and blue flows indicate downregulated (Down) metabolites. (E) Violin plots showing volatile metabolites associated with aroma descriptors. * indicates p < 0.05, and ** indicates p < 0.01. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

3.3. Overview of lipidomics results in yak meat from the two groups

3.3.1. Lipid annotation in the two groups

Semi-quantitative lipidomic profiling of yak meat from the CON and EFA groups was performed using liquid chromatography–tandem mass spectrometry (LC–MS/MS) to compare the lipid composition in detail. The coefficient of variation (CV) distribution indicated good analytical stability (Fig. 4A). In the PCA score plot including the pooled QC samples, the QC samples clustered closely, supporting the reproducibility of the LC–MS/MS analysis (Fig. 4B). Orthogonal partial least squares-discriminant analysis (OPLS-DA) was performed as an exploratory supervised multivariate approach to evaluate differences in lipidomic profiles between the CON and EFA groups. The score plot indicated visual separation between the two groups (Fig. 4C), and the corresponding 200-permutation test was used to evaluate model robustness and potential overfitting (Fig. 4E). The lipidomic OPLS-DA model showed R²X = 0.628, R²Y = 0.993, and Q² = 0.795. The permutation test demonstrated that the original model parameters were higher than those obtained from randomly permuted models (R²Y: p < 0.005, 0/200; Q²: p = 0.005, 1/200), supporting the robustness of the model within the exploratory analytical framework. These analyses consistently suggested differences in lipid composition and abundance between the two groups. A total of 764 lipid features were annotated in yak meat from both groups (Table S6). As shown in Fig. 4D, these lipid features were classified into six major categories: glycerophospholipids (GP, 501 lipids), glycerolipids (GL, 106 lipids), sphingolipids (SP, 85 lipids), fatty acyls (FA, 63 lipids), and sterol lipids (ST, 7 lipids), and prenol lipids (PR, 2 lipids). This classification highlights the compositional complexity of the meat lipids, with GP and GL being the most abundant classes. Moreover, secondary sub-classification of these lipids (Fig. 4F and G) revealed that TG (12.68% vs. 12.45%), PE (11.37% vs. 11.53%), PE-O (11.37% vs. 11.40%), PI (6.80% vs. 6.82%), and CAR (6.01% vs. 6.16%) were the dominant lipid sub-classes in both the CON and EFA groups.

Fig. 4.

Annotation of lipids in yak meat from the two groups. (A) CV distribution plot for each group; the x-axis represents CV values, and the y-axis represents the cumulative proportion of metabolites below each CV threshold. QC denotes quality control samples. (B) PCA score plot of lipidomic profiles from the CON, EFA, and pooled QC samples. PC1 and PC2 represent the first and second principal components, respectively. (C) OPLS-DA score plot; the x-axis indicates the predictive component, and the y-axis indicates the orthogonal component. (D) Bar plot showing the number of identified lipids in each subclass; colors denote the corresponding lipid categories. (E) OPLS-DA validation plot; the x-axis represents the model's R²Y and Q² values, and the y-axis shows the frequency distribution from 200 random permutation tests. Orange and purple bars represent the R²Y and Q² values of the randomized models, respectively, and black arrows indicate the R²X, R²Y, and Q² values of the original model. (F, G) Donut charts showing the composition of lipid subclasses. Each color represents a lipid subclass, and the sector area corresponds to its relative proportion. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

3.3.2. Differential lipids between yak meat from the two groups

To investigate the association between E. faecalis JM263 supplementation and muscle lipid composition, candidate differential lipids were screened based on the exploratory criteria of VIP > 1, FC > 1.5 or < 0.67, and nominal p < 0.05. A total of 60 candidate differential lipids were identified, including 35 upregulated and 25 downregulated lipids. FDR-adjusted q-values were additionally calculated and are provided in Table S6 to facilitate assessment of multiple-testing robustness. Among these candidates, five lipid molecular species/features—PE(16:1_17:1), PE(17:0_18:1), PA(18:0_18:1), LNAPE(18:1/N-17:0), and PE(16:0_22:4)—remained significant at FDR-adjusted q < 0.05, whereas the remaining candidates were interpreted as exploratory features based on the combined VIP, FC, and nominal p-value criteria. The top 50 differential lipids were visualized in a heatmap (Fig. 5C), showing distinct clustering between the CON and EFA groups, indicating the clear lipid compositional differences. Lipid expression patterns were evaluated by using the OPLS-DA model and visualized through clustering heatmaps and VIP value plots, which highlighted candidate lipid features contributing to the supervised model separation between the two groups. Based on VIP scores from the OPLS-DA model (Fig. 5B and Table S7), the top 20 candidate lipid features were ranked, among which PE (16:1_17:1), PE (17:0_18:1), PA (18:0_18:1), LNAPE (18:1/N-17:0), and Carnitine C14:1 ranked in the top five. VIP ranking was used only for exploratory prioritization and was not considered independent statistical evidence of differential abundance. Among these top-ranked features, PE(16,1_17:1), PE(17,0_18,1), PA(18,0_18,1), and LNAPE(18,1/N-17,0) were additionally supported by FDR-adjusted q < 0.05, whereas Carnitine C14:1 remained an exploratory candidate. These lipids were considered candidate lipid features potentially associated with yak meat aroma-related metabolic differences, although their biological roles require further validation using targeted lipid quantification and larger animal cohorts. Correlation network analysis revealed that glycerophospholipids (GP) and fatty acyls (FA) occupied central positions in the correlation network, displaying strong correlations with multiple lipid species. KEGG pathway enrichment analysis showed that candidate lipids increased in the EFA group were mapped to pathways showing enrichment signals, including biosynthesis of secondary metabolites, glycine, serine and threonine metabolism, and fatty acid metabolism. In contrast, candidate lipids decreased in the EFA group were mapped to pathways showing enrichment signals, including inositol phosphate metabolism, ether lipid metabolism, and teichoic acid biosynthesis.

Fig. 5.

Differential lipids between the CON and EFA groups. (A) Volcano plot of differential lipids, where each point represents a specific lipid; point size indicates VIP value. (B) OPLS-DA VIP score plot of the top 20 candidate lipid features ranked by VIP value. (C) Heatmap showing clustering of differential lipids. The left dendrogram represents the lipid clustering, and the top bar indicates the sample grouping. (D) Correlation network of differential lipids. Node size corresponds to the degree of connectivity; larger nodes indicate higher degrees (more neighbors). Red lines represent positive correlations, blue lines represent negative correlations, and line thickness reflects the absolute value of the correlation coefficient. (E, F) KEGG enrichment plots for upregulated and downregulated differential lipids, respectively. The x-axis represents the RichFactor of each pathway, and the y-axis lists pathway names sorted by p-value. Dot color reflects p-value (red indicates greater significance), and dot size represents the number of enriched differential lipids. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

3.4. Exploratory association analysis of E-nose responses with volatile and lipidomic features

To better understand the aroma-related alterations in yak meat associated with E. faecalis JM263 supplementation and to explore the separate associations of E-nose sensor responses with volatile metabolite features and lipidomic profiles, correlation analyses were performed using 20 key lipids, 24 representative volatile metabolites, and five E-nose sensor responses (W1C, W5C, W1S, W1W, and W2S). As shown in Fig. 6A, Mantel test results revealed exploratory associations between E-nose response values and several volatile features putatively annotated as 2-Propyn-1-ol, propionate (r = 0.66, p = 0.002), 4,6-Octadiyn-3-one, 2-methyl (r = 0.44, p = 0.006), Benzoic acid (r = 0.58, p = 0.006), and 1,5-Cyclooctadiene, 3,4-dimethyl (r = 0.43, p = 0.006). Similarly, E-nose response values showed associations with several lipid species, including PE(17:0_18:1) (r = 0.75, p < 0.001), LNAPE (18:1/N-17:0) (r = 0.77, p < 0.001), and PE(16:0_22:4) (r = 0.78, p < 0.001) (Fig. 6B).

Fig. 6.

Correlation analysis of electronic-nose responses with volatile and lipid features. (A) Associations between electronic-nose responses and selected volatile features. (B) Associations between electronic-nose responses and selected lipid features. In each triangular matrix, the same volatile or lipid variables are displayed on the upper and right axes. The color and size of the inner squares represent the direction and magnitude of Pearson's correlation coefficient (r); orange indicates positive correlations, teal indicates negative correlations, and larger squares indicate stronger absolute correlations. Curved lines represent Mantel-test associations between the electronic-nose response matrix and individual volatile or lipid features. Line colors indicate Mantel-test p-value categories: green, p ≤ 0.001; orange, 0.001 < p ≤ 0.01; teal, 0.01 < p ≤ 0.05; and gray, p > 0.05. Line thickness represents Mantel's r (r > 0.5, 0.25 ≤ r ≤ 0.5, or r < 0.25). Gray circles are connection points and have no independent statistical meaning. Pearson correlations are presented descriptively without significance testing in the figure. Mantel-test p-values are unadjusted, and no multiple-testing correction was applied. The figure was generated using ChiPlot. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

4. Discussion

4.1. Effects of E. faecalis JM263 on the aroma-related volatile profile of yak meat

The E-nose is a widely used technology for detecting volatile flavor compounds and is commonly used in the food, cosmetics, and fermentation industries. The E-nose is highly sensitive to specific functional groups and can analyze and classify samples based on the detected volatile signatures. It provides a rapid and accurate method for identifying odor molecules in meat products, thereby enabling instrumental discrimination of aroma profiles rather than direct assessment of taste, texture, or consumer-perceived sensory quality. Previous studies have reported that dietary supplementation with Lactobacillus plantarum increased the W1S and W2S sensor responses in duck meat, thereby enhancing the flavor richness (Xu et al., 2024). Similarly, supplementation with L. plantarum and L. sakei in chicken diets elevated W5S, W1S, W1W, and W2W sensor responses, promoting the release of various hydrocarbons, methanes, organic sulfur compounds, and sulfur–chlorine volatiles (Tu et al., 2025). In the present study, dietary addition of E. faecalis JM263 significantly increased the E-nose sensor responses of W1C, W5C, W1S, W1W, and W2S in yak meat, indicating marked alterations in instrumental responses associated with volatile compounds (Hu et al., 2025). The increase in the W1C sensor response—which is sensitive to aromatic compounds—suggests an enhanced accumulation of aromatic substances such as benzenes and phenolic derivatives (Zhang et al., 2021). W5C is responsive to short-chain alkane and aromatic compounds, and its increased response further indicates differences in the corresponding volatile signatures between the two groups (Liu et al., 2024). The heightened W1S response, which detects short-chain alkanes such as methane, indicates an altered response to these volatile compounds (Liu et al., 2024). The W1W sensor is sensitive to sulfur-containing compounds, and its increased response suggests altered detection of sulfur-related volatile signatures that may contribute to meat-like aroma characteristics. Meanwhile, the W2S sensor, sensitive to alcohols and organo‑sulfur compounds, detects molecules that may contribute to the overall aroma-related volatile profile (Yu et al., 2022). Collectively, the E-nose results indicate that E. faecalis JM263 supplementation was associated with alterations in the instrumental aroma-related response profile of yak meat and corresponding changes in odor-related volatile signatures. However, because no formal sensory evaluation was performed, these results should not be interpreted as direct evidence of improved sensory quality or consumer-perceived flavor enhancement.

4.2. Effects of E. faecalis JM263 on volatile metabolomics of yak meat

Flavor is a crucial indicator of meat quality and a key determinant of consumer acceptance. Previous research has shown that the incorporation of functional fruit- and vegetable-derived ingredients into a bio-corrected coating can affect meat-product quality characteristics (Giro et al., 2024). Although this post-processing intervention differs from the dietary probiotic strategy used in the present study, it provides complementary evidence that functional interventions can affect meat quality-related attributes. Gas chromatography–mass spectrometry (GC–MS), combined with sensory evaluation, is commonly used to assess product flavor owing to its powerful qualitative and quantitative capabilities (Liu et al., 2020; Shen et al., 2023). When integrated with the relative odor activity value (rOAV), GC–MS provides a complementary approach for exploratory screening of volatile compounds that may be associated with aroma characteristics under specific analytical conditions, but it does not independently confirm their sensory contribution (Li et al., 2024). In recent years, rOAV analysis has been increasingly applied to screen potentially aroma-relevant compounds in various foods, with rOAV >1 generally indicating that a compound may contribute to the overall aroma profile under the given analytical conditions. However, because the present rOAVs were calculated from internal-standard-assisted semi-quantitative concentration estimates and compound-specific extraction efficiencies and detector response factors were unavailable, in addition to variation in odor thresholds across matrices and references, the rOAV values were interpreted as screening indicators of potential aroma relevance rather than definitive evidence of aroma activity. In this study, 42 volatile features with rOAV >1 were detected in the CON group and 45 in the EFA group. Together with the observed changes in the identities and relative abundances of odor-relevant volatile compounds, this result suggests that E. faecalis JM263 supplementation was associated with changes in the instrumental aroma-related volatile profile of yak meat. However, these rOAV-based observations should be considered exploratory indicators of potential aroma relevance rather than direct evidence of sensory impact. However, the numerical increase alone was not considered sufficient evidence of improved sensory quality. The volatile metabolites detected were mainly aldehydes, ketones, esters, hydrocarbons, amines, alcohols, heterocyclic compounds, and terpenoids, consistent with previous reports showing that these compounds play pivotal roles in yak meat flavor characteristics (Zhang et al., 2025). Among the 24 exploratory candidate volatile features, 16 were upregulated and 8 were downregulated. Most upregulated compounds belonged to aldehydes, ketones, esters, hydrocarbons, and terpenoids, suggesting that these chemical classes may represent exploratory volatile features associated with the aroma-related differences observed following E. faecalis JM263 supplementation (Lin et al., 2025; Liu et al., 2022).

Further analysis revealed that volatile features putatively annotated as 1-cyclopropyl-2-(3,4-dimethoxyphenyl)ethanol, cyclohexene, 4-(1,5-dimethyl-1,4-hexadienyl)-1-methyl, and 2-propyn-1-ol, propionate were exploratory candidate volatile features potentially associated with differences in the instrumental aroma-related volatile profile of yak meat. The feature putatively annotated as 2-propyn-1-ol, propionate, which contains a conjugated alkyne–ester structure and has a low odor threshold (∼0.5 ppm), may have originally imparted a green or slightly pungent note to the meat. Its decrease may therefore be associated with a reduced contribution of green or slightly pungent odor notes (Ojha and Roy, 2018). In contrast, the increased abundance of the feature putatively annotated as 1-cyclopropyl-2-(3,4-dimethoxyphenyl) ethanol represented a treatment-associated volatile change that may be related to broader host–microbial metabolic interactions; however, such metabolic crosstalk was not directly evaluated in the present study (Zhang et al., 2025). The increased abundance of the feature putatively annotated as cyclohexene, 4-(1,5-dimethyl-1,4-hexadienyl)-1-methyl, a terpenoid derivative with a low odor threshold, may represent one of the candidate volatile changes associated with the altered aroma-related volatile profile of yak meat (Álvarez-Rodríguez et al., 2022). Although terpenoids are present at low concentrations, their distinct aromatic characteristics may contribute to the overall volatile aroma profile. Together, these changes indicate alterations in the composition of aroma-related volatile compounds. Moreover, aroma-descriptor annotation associated decanoic acid, ethyl ester, Pentadecane, and γ-dodecalactone with apple, brandy, waxy, fruity, balsamic, floral, and fatty descriptors. Generally, ethyl decanoate has been reported to exhibit fruity and sweet odor characteristics, while pentadecane and γ-dodecalactone have been associated with waxy and creamy floral-fruity aroma descriptors, respectively. In the present study, these annotations were considered exploratory indications of potential aroma-related characteristics rather than direct evidence of sensory contributions (D'Auria et al., 2021; Wang et al., 2021). Collectively, these compounds represented exploratory candidate volatile features associated with the observed differences in the aroma-related volatile profile of the EFA group. Overall, these results suggest that E. faecalis JM263 supplementation was associated with changes in selected aroma-related volatile metabolites. However, because no trained sensory panel or consumer evaluation was conducted, these metabolites should be interpreted as candidate volatile features associated with instrumental aroma differences rather than direct evidence of sensory quality improvement.

4.3. Effects of E. faecalis JM263 on lipid metabolism in yak meat

Lipids are major constituents of meat and play vital roles in numerous cellular processes, including membrane integrity, signaling, and energy storage. In meat and meat products, lipids contribute to multiple quality attributes such as appearance, flavor, juiciness, tenderness, and nutritional value (Huang et al., 2023; Nurgul et al., 2025). Among the lipid classes, glycerophospholipids (GP)—the most abundant component (about 50%), in muscle lipids—are key structural constituents of cell membranes (e.g., phosphatidylcholine and phosphatidylethanolamine), and their diversity reflects the complexity of muscle membrane function. Glycerolipids (GL), such as triglycerides and diglycerides are the major energy-storage lipids in muscle, associated with fat deposition, tenderness, and flavor formation (Guo et al., 2024). Consistent with these functions, our lipidomic profiling revealed that GP and GL were the most dominant lipid categories in yak meat. Comparison of lipid subclass composition further showed that the relative abundance of PE, PE-O, PI, and CAR increased in the EFA group. Phospholipids, as essential components of cell membranes, are known to contribute to the overall lipid content in meat (Yang et al., 2024). Upon oxidation, phospholipids generate secondary products including aldehydes, ketones, alcohols, and esters (Wang et al., 2013). These oxidation products can react with proteins and amino acids to form new compounds—such as Schiff bases, imines, and esters—that modify the aroma and overall flavor of meat products (Mahachi et al., 2023).

Candidate lipid screening based on VIP, fold change, and nominal p-value criteria identified several exploratory lipid features associated with E. faecalis JM263 supplementation. Among these candidates, PE(16:1_17:1), PE(17:0_18:1), PA(18:0_18:1), LNAPE(18:1/N-17:0), and PE(16:0_22:4) remained significant after FDR correction, whereas other candidate lipid features, including Carnitine C14:1, were considered exploratory candidates requiring further validation. These alterations were associated with differences in lipid metabolic features related to lipid synthesis and degradation in yak muscle (Qian et al., 2019). PE, the second most abundant phospholipid in mammalian cells (accounting for 15–25% of total lipids), is central to numerous cellular processes due to its unique physicochemical properties—it acts as an antioxidant and assists membrane protein folding (Patel and Witt, 2017). At the molecular-species level, PE(16:1_17:1) and PA(18:0_18:1) contain unsaturated fatty acyl compositions. Because the sn-positions, double-bond positions, and double-bond geometries of these lipid molecular species were not experimentally resolved, specific oxidation pathways or products cannot be assigned to these individual species. Accordingly, their potential relevance to aroma precursor formation should be interpreted only at the level of general changes in unsaturated phospholipid composition rather than specific structure–oxidation relationships (Claes et al., 2023). The exploratory candidate lipid feature Carnitine C14:1 showed an increased abundance in the EFA group and may reflect potential alterations in fatty-acyl carnitine-related metabolism. Acylcarnitines are involved in fatty-acyl transport and are generally related to fatty acid β-oxidation pathways (Humer et al., 2016). However, because β-oxidation flux, fatty acyl-CoA intermediates, downstream oxidation products, and β-oxidation-related enzymes were not directly measured in the present study, Carnitine C14:1 should be interpreted as a candidate lipid feature rather than direct evidence of enhanced fatty acid breakdown or β-oxidation activation (Humer et al., 2016). In contrast, the downregulation of PE(17:0_18:1) and LNAPE(18,1/N-17:0) may reflect redistribution of membrane lipids and activation of degradation pathways. The reduction in LNAPE may reflect alterations in ether- or lysophospholipid-related metabolic processes; however, the corresponding effects on membrane stability and lipid signaling were not directly evaluated in the present study (Leung et al., 2006).

Overall, the lipidomics results suggest that E. faecalis JM263 supplementation was associated with changes in phospholipid remodeling and fatty-acyl-related lipid metabolism. These changes may reflect variation in the availability of lipid-derived aroma precursors, but the present data do not provide direct evidence of enhanced fatty acid β-oxidation or causal conversion of specific lipids into volatile compounds. Because gut microbiota, antioxidant biomarkers, inflammatory mediators, and host metabolic flux were not directly measured in the present study, their potential involvement should be regarded only as possible biological explanations rather than mechanisms demonstrated by the present experiment. Correlation network analysis further indicated that GP and FA occupied central positions within the correlation network, suggesting that these lipid classes were closely associated with the overall lipid compositional differences observed between the two groups. These statistical associations indicate patterns of co-variation among lipid species but do not establish direct biochemical interactions or regulatory relationships. KEGG enrichment analysis indicated that the candidate differential lipids were associated with several metabolic pathways. KEGG enrichment analysis indicated that the candidate differential lipids were associated with several metabolic pathways. Lipids increased in the EFA group were mapped to pathways showing enrichment signals, including biosynthesis of secondary metabolites, glycine, serine and threonine metabolism, and fatty acid metabolism, whereas lipids decreased in the EFA group were mapped to pathways showing enrichment signals, including inositol phosphate metabolism, ether lipid metabolism, and teichoic acid biosynthesis (Qian et al., 2023; Zhang et al., 2023). Because individual enriched pathways may be represented by only a limited number of candidate lipid features, these results were interpreted cautiously as exploratory pathway-level associations rather than direct evidence of pathway activation or altered metabolic flux. Collectively, supplementation with E. faecalis JM263 was associated with changes in phospholipid- and fatty-acyl-related metabolic features. These findings indicate aroma-related lipid metabolic variation but do not establish precursor–product relationships between specific lipids and volatile compounds. However, these findings should be interpreted as exploratory multi-omics associations rather than direct mechanistic proof that probiotic supplementation improves meat sensory quality.

To further explore the associations involving the electronic nose data, two separate exploratory correlation analyses were performed between E-nose responses and volatile features and between E-nose responses and lipid features. The results showed associations between sensor responses and several volatile metabolites, including 2-Propyn-1-ol, propionate, 4,6-Octadiyn-3-one, 2-methyl, Benzoic acid, and 1,5-Cyclooctadiene, 3,4-dimethyl. These findings indicate that these volatile features were associated with variation in the E-nose response profile. Furthermore, the Mantel-test results showed matrix-level associations between electronic nose responses and lipidomic profiles, including lipid features such as PE(17:0_18:1), LNAPE(18:1/N-17:0), and PE(16:0_22:4). Because Mantel tests evaluate similarity between data matrices rather than causal biochemical relationships, these associations were interpreted as hypothesis-generating integrative evidence rather than evidence of direct regulatory links between lipid alterations and volatile compound formation. Taken together, these analyses showed separate exploratory associations of E-nose responses with volatile and lipid profiles, respectively, rather than direct lipid–volatile relationships. These findings, derived from integrated multi-omics analyses, provide exploratory evidence for separate associations of electronic nose responses with volatile metabolites and lipid profiles. However, because the correlation analyses were performed using samples from both treatment groups, the observed associations may partly reflect treatment-associated differences between the CON and EFA groups. Therefore, these correlations should be interpreted as exploratory associations related to E. faecalis JM263-associated metabolic variation rather than intrinsic biological relationships among individual variables. Furthermore, because the present study did not include formal sensory evaluation, these results should not be interpreted as direct evidence that probiotic supplementation enhances meat sensory quality.

The high-altitude plateau environment should also be considered when interpreting the present results. Yaks are highly adapted to the Qinghai–Tibet Plateau, where chronic hypoxia, low temperature, strong ultraviolet radiation, and large diurnal temperature variation can influence energy metabolism, oxidative status, rumen microbial activity, lipid deposition, and muscle metabolic homeostasis (Wang et al., 2025). These physiological adaptations may further affect phospholipid remodeling, fatty acid oxidation, and the generation of lipid-derived volatile compounds. Therefore, the aroma-related metabolic changes observed in the present study should be interpreted within the specific context of confined feeding under high-altitude plateau conditions. Because all animals in this study were raised at the same altitude of approximately 4000 m, altitude was a shared environmental background rather than an independent experimental factor. Thus, the present design cannot distinguish the independent effect of altitude or test the interaction between plateau environment and E. faecalis JM263 supplementation. Future studies involving different altitude gradients, feeding systems, or factorial experimental designs will be required to determine whether plateau environmental factors interact with probiotic supplementation to regulate yak muscle lipid metabolism and aroma-related volatile formation.

Several limitations of this study should be acknowledged. First, the sample size was limited to six yaks per group, and no formal a priori sample size or statistical power calculation was performed before the experiment because prior effect-size estimates for changes associated with E. faecalis JM263 supplementation in yak meat volatile metabolites and lipid profiles were unavailable. The additional sensitivity power analysis based on electronic nose data showed that the current design was mainly capable of detecting large between-group effects, whereas smaller probiotic-associated changes may have been underpowered. Therefore, although the major electronic nose responses were supported by large effect sizes and remained significant after Benjamini–Hochberg correction, the volatilomics and lipidomics findings should be interpreted as exploratory. Second, OPLS-DA was used as a supervised exploratory method to identify discriminant variables, and the observed group separation should not be regarded as independent proof of biological causality. Similarly, OPLS-DA-derived VIP values are model-dependent measures and do not constitute independent statistical evidence for individual features. Although permutation testing was used to assess overall model robustness and potential overfitting, the stability of individual VIP rankings was not independently evaluated by feature-level resampling. Therefore, VIP rankings were used only for exploratory feature prioritization, whereas statistical support for individual features was assessed separately using univariate testing and FDR adjustment. Third, the screened volatile and lipid features should be considered candidate aroma-related features that require further independent validation. In addition, because all yaks were raised under the same high-altitude plateau conditions, the present study could not evaluate the independent effect of altitude or the potential interaction between plateau environment and probiotic supplementation. Although major environmental and procedural factors were minimized through random allocation and standardized feeding, housing, transport, fasting, slaughter, sampling, and storage conditions, residual inter-individual variation in baseline physiological and microbial status could not be fully excluded. Future studies combining larger sample sizes, prospective power calculations, independent validation cohorts, sensory evaluation, and targeted quantification will be necessary to confirm the robustness and reproducibility of the present findings. It should also be noted that the present study was an exploratory animal multi-omics study with a limited sample size. Benjamini–Hochberg FDR-adjusted q-values were calculated for the volatilomics and lipidomics datasets. After FDR correction, none of the 24 candidate volatile metabolites remained significant, whereas five of the 60 candidate lipids remained significant at q < 0.05. Therefore, features identified only by the combined VIP, fold-change, and nominal p-value criteria were interpreted as exploratory candidates rather than definitive biomarkers. Therefore, the volatile metabolites and lipids screened using VIP, fold change, and nominal p-values should be interpreted as candidate aroma-related features rather than definitive biomarkers. Future studies with larger yak cohorts, targeted quantification, and sensory evaluation are needed to further validate these findings.

5. Conclusion

This study identified candidate aroma-related volatile features and lipids associated with instrumental aroma-related variation in yak meat following supplementation with Enterococcus faecalis JM263. The electronic nose effectively discriminated between the two yak meat groups, revealing marked differences in instrumental odor-related response profiles. GC–MS putatively annotated 688 volatile features, of which 24 were screened as candidate differential features. LC–MS/MS identified 764 lipid species, of which 60 were screened as candidate differential lipids, and five lipid features remained significant after FDR correction. Mantel-test analysis revealed separate exploratory associations between electronic nose responses and volatile profiles and between electronic nose responses and lipid profiles, without establishing direct lipid–volatile relationships. Overall, the findings indicate treatment-associated variation in aroma-related volatile profiles and lipid metabolic features rather than a demonstrated mechanistic pathway. Because sensory evaluation was not performed, these findings should not be interpreted as evidence of improved sensory flavor.

CRediT authorship contribution statement

Yanbin Zhu: Writing – original draft, Investigation, Data curation, Conceptualization. Zhuoma Cisang: Visualization, Methodology, Formal analysis, Data curation. Ammara Khan: Writing – original draft, Visualization, Methodology, Formal analysis. Mei Jiang: Visualization, Methodology. Yangji Cidan: Visualization, Methodology. Guifang Liu: Visualization, Software. Fei Gao: Project administration, Investigation. Shijun Wen: Supervision, Data curation. Xinhong Zhou: Writing – review & editing, Supervision, Data curation. Wangdui Basang: Writing – review & editing, Supervision, Project administration, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary material 1

Supplementary material 2

Data availability

Data will be made available on request.