Metabolic profiling to detect different forms of beef fraud using rapid evaporative ionisation mass spectrometry (REIMS)

The study addresses the problem of organic meat fraud, where conventional products are misrepresented as organic despite not meeting organic production standards. Organic beef is particularly vulnerable due to premium pricing, ease of relabelling, and the difficulty of verifying production claims. Existing verification relies heavily on certification, with analytical tools like stable isotope ratio analysis (SIRA) providing suggestive but not definitive differentiation between organic and conventional meat; SIRA is also expensive and time-consuming. Regulations in the EU (Council Regulation (EC) No 834/2007) define organic production requirements (e.g., certified organic feedstuffs, ≥60% organic forage, GM feed ban, restricted prophylactic veterinary drug use, and extended withdrawal periods), but analytical confirmation remains challenging. The purpose of this study is to evaluate rapid evaporative ionisation mass spectrometry (REIMS) as a rapid, low-preparation analytical approach to (1) distinguish organic versus conventional beef and (2) verify meat cut identity (neck, rump, shin), using untargeted lipidomic fingerprints (m/z 600–1000) and chemometric models (PCA and PCA-LDA). The overarching hypothesis is that lipidomic fingerprints captured by REIMS contain sufficient information to classify both production system and cut, enabling near real-time fraud detection.

Prior analytical approaches to authenticate organic meat include SIRA, which has shown differences in isotopic composition between organic and conventional Irish beef (e.g., Schmidt et al. reported significant MANOVA differences combining C, N, and supplier; conventional beef exhibited less negative and more variable δ13C than organic). However, seasonal shifts in δ13C and overlapping feeding practices (e.g., organic maize or grass-heavy conventional diets) confound interpretations. δ15N differences are small and can be influenced by fertilizers, supplements, and dietary amino acid composition, further limiting reliability. Fatty acid profiling reflects feeding regimes and has been explored, but no single robust method for organic beef authentication has emerged. Veterinary drug residue testing may indicate non-compliance with organic standards, yet reduced antimicrobial use in conventional systems in Europe diminishes discriminative power. REIMS, an ambient MS technique, has been successfully applied to food authenticity tasks: meat speciation and breed, fish authenticity, detection of additives and adulterants in minced meat, boar-taint detection, and differentiation between organic and conventional poultry. Its advantages include no sample preparation, rapid analysis (seconds), and operation by non-specialists, motivating its evaluation for beef authenticity (both production system and cut verification).

Study design and sampling: 561 beef samples were collected over approximately 13 months (April 2019–November 2020) from abattoirs in the Republic of Ireland, Northern Ireland, and England (ABP Cahir, ABP Newry, ABP Ellesmere, ABP Haverhill). Samples represented both conventional (n=285) and organic (n=276) production systems, and three muscle types/cuts per carcass: neck (supraspinatus), rump (gluteus), and shin (flexor carpi radialis). Breeds included Aberdeen Angus, Limousin, Hereford, and Simmental to reflect commercial diversity. Post-slaughter, carcasses were chilled at 2 °C; sampling occurred three days post-mortem. Each sample (~100–120 g) was divided, labelled, vacuum-packed, and stored at −80 °C; samples were thawed to ~4 °C prior to analysis. Instrumentation and REIMS analysis: A Waters REIMS ion source was mounted on a Waters Xevo G2-XS QTOF mass spectrometer. A bipolar probe connected to an Erbe VIO 500 diathermy generator was used to thermally ablate meat, generating aerosol for ionisation at a heated Kanthal coil adjacent to the MS orifice. A leucine enkephalin lockmass solution (0.1 ng/μL in 2-propanol) was infused via a Waters Acquity UPLC at 0.2 mL/min for accurate mass correction and to aid ionisation. The mass spectral scan rate was 0.5 s/scan; full scan mass range was up to m/z 1200, with downstream chemometrics focused on m/z 600–1000 (lipid region). Daily detector setup used 0.1 ng/μL leucine enkephalin; calibration with 5 mM sodium formate was performed at the start of each day. For each sample, approximately 10 burns (2–5 s each) were performed; no carry-over was observed. Data processing and chemometrics: Data were acquired in MassLynx and processed in Waters Abstract Model Builder (AMX v0.0929.00). Spectra underwent lockmass correction (m/z 574.5615 using leucine enkephalin), background subtraction, and total ion count normalisation. For each sample, replicate spectra were averaged to a single spectrum. Data within m/z 600–1000 were binned at 0.5 Da with a spectral intensity threshold of 2.5. PCA was used for dimensionality reduction followed by LDA (PCA-LDA). Models were built for: (1) cut classification (three-class: neck, rump, shin); (2) production system classification (binary: organic vs conventional); and (3) combined/hierarchical assessments by building per-cut production system models. Model performance was assessed by Leave-20%-Out cross-validation; samples outside a set 50 SD threshold for all classes were treated as outliers. Mislabelled samples identified during exploratory analysis were investigated and removed to build a corrected cut model.

- Cut identification: The initial cut classification model achieved 88.5% Leave-20%-Out cross-validation accuracy (561 spectra; 496 passes, 65 failures, 1 outlier). After correcting suspected mislabelling, the cut model’s cross-validation accuracy improved to 98.2% with clear separation of neck, rump, and shin. - Production system (organic vs conventional): The PCA-LDA model reached 83.9% cross-validation accuracy (470 passes, 90 failures, 1 outlier out of 561). Misclassifications included 53 conventional samples classified as organic and 37 organic samples classified as conventional. Early models with fewer samples achieved <60% but improved to ~84% with increased sample size and curation. - Per-cut production models: Cross-validation accuracies for production system within each cut were 75.6% (shin), 72.7% (neck), and 70.4% (rump). - Lipidomic drivers: Strong lipid fingerprint differences and relative intensities across cuts underpinned near-perfect cut separation. In contrast, organic vs conventional differences were subtler and distributed across many lipid species; no individual biomarkers were identified. - Practical advantages: REIMS enabled simultaneous assessment against multiple models (cut and production system), reducing analysis time and sample numbers. Compared to SIRA, REIMS offered higher accuracy for organic vs conventional differentiation in this study, with minimal preparation, faster analysis (seconds), and lower per-sample cost.

The findings demonstrate that REIMS-derived lipidomic fingerprints can robustly address two core authenticity questions: verifying the cut and inferring the production system. The near-perfect cut classification (>98%) indicates substantial and consistent lipid profile differences among neck, rump, and shin, making REIMS highly suitable for detecting substitution or mislabelling of cuts. Production system classification, while more challenging, achieved ~84% accuracy, showing that many small, diet- and husbandry-related lipid variations collectively provide discriminative power, even across multiple breeds typical of commercial supply chains. The absence of single biomarkers suggests the classification relies on multivariate intensity patterns across numerous glycerophospholipids and sphingolipids. Compared with SIRA, which is susceptible to confounding by overlapping feeding regimes, seasonality, and laborious sample preparation, REIMS offers faster, cheaper, and simultaneously multi-purpose assessments (cut plus production system) suitable for near-line or in-line industrial deployment. The technology’s ability to function without sample preparation and be operated by non-specialists positions it as a practical tool for routine authenticity screening in abattoirs and processing plants.

This study establishes REIMS as a promising, rapid, and practical approach for beef authenticity. It delivers near real-time identification of meat cuts with >98% accuracy and distinguishes organic from conventional beef at ~84% accuracy using untargeted lipidomic fingerprints and PCA-LDA models. These capabilities can deter or detect multiple fraud forms, including organic misrepresentation and cut substitution, within routine industrial workflows. Future work should: (1) expand and diversify sample sets to further improve production system classification; (2) explore targeted or hybrid approaches to identify potential REIMS-detectable biomarkers that complement multivariate signatures; (3) assess higher-value cuts and additional fraud scenarios; and (4) develop more accessible REIMS instrumentation (e.g., ion sources compatible with robust, lower-cost single quadrupole MS) to reduce capital and maintenance barriers for industry adoption.

- Production system separation was weaker than cut separation; accuracy (~84%) depended on large, well-curated datasets and improved with additional sampling. Early models underperformed (<60%). - No single biomarkers for organic vs conventional were identified; discrimination relies on many subtle lipid intensity changes, which may be sensitive to dietary, seasonal, or husbandry variations. - Mislabelled samples affected model performance, necessitating data curation and highlighting operational risks in sample handling. - Instrumentation requirements (high-resolution TOF MS with significant maintenance) presently limit accessibility and scalability; cost and complexity could constrain deployment until more economical platforms are available. - Similarities in veterinary drug use between modern conventional and organic systems, and overlapping feed practices, can reduce biochemical contrast between production systems.