Investigation of Varietal Classification Method for Coffee Beans Using Aroma Compounds

Significance of the topic

Coffee quality and authenticity are commercially and sensorially important: Arabica beans command higher value because of superior aroma and acidity, while Robusta is often used as a lower-cost component. Adulteration of Arabica with Robusta has been reported, motivating rapid analytical approaches to verify varietal identity. Volatile aroma profiling by GC-MS/MS coupled with multivariate statistics offers a minimally invasive, high-throughput alternative to laborious chemical extractions and NMR-based markers, enabling routine screening of green or roasted beans for quality control and fraud detection.

Study objectives and overview

This work evaluated whether headspace SPME GC-MS/MS aroma profiling, combined with database-driven compound identification and multivariate/statistical modeling, can reliably discriminate Arabica and Robusta coffee beans. The goals were to: (1) compare unsupervised clustering and PCA results from scan and MRM acquisition modes; (2) identify key aroma compounds distinguishing the two species; and (3) build and validate a classification model (SVM) using eMSTAT Solution for practical discrimination of unknown samples.

Methodology

• Samples: Six commercially roasted coffee brands (four Arabica, two Robusta) milled and sampled as 1 g portions in screw vials; each sample analyzed in triplicate.
• Sample introduction: Headspace solid-phase microextraction (HS-SPME) to concentrate volatiles prior to GC injection.
• Chromatography and detection: Low-polarity capillary column with a 10°C/min oven ramp to shorten run-time while preserving retention stability; splitless injection; helium carrier gas; electron ionization mass spectrometry performed in Scan (m/z 35–400) and targeted MRM modes for sensitivity and reproducibility.
• Data processing: Peak alignment and feature assignment using Signpost MS; compound identification using the Smart Aroma Database (contains MRM transitions and aroma descriptors); statistical analysis including hierarchical clustering, PCA, and model building using eMSTAT Solution (Support Vector Machine classifier).

Used instrumentation

• GC-MS/MS system: Shimadzu GCMS-TQ8040 NX.
• Autosampler: Shimadzu AOC-6000 multifunctional autosampler.
• Column: Low-polarity capillary column (SH-I-5Sil MS, 30 m × 0.25 mm, 0.25 µm film).
• Databases and software: Smart Aroma Database for MRM/compound annotation; Signpost MS for alignment and exploratory analysis; eMSTAT Solution for PCA, visualization and SVM model building; LabSolutions Insight GCMS for chromatogram review.

Main results and discussion

• Feature detection and identification: Scan-mode alignment produced 210 detected features. Switching to targeted MRM using the Smart Aroma Database allowed identification of ~178 aroma compounds, with more than 80% of GC-MS detected features assigned to known aromas. For statistical modeling, 175 compounds with complete data were used.
• Unsupervised separation: Hierarchical clustering and PCA on Scan-mode data already separated Arabica and Robusta into distinct clusters, with Arabica grouping on one side and Robusta on the other, indicating systematic differences in volatile composition.
• MRM-enhanced analysis: PCA of the MRM dataset retained clear separation between species, benefiting from increased sensitivity and reproducibility of targeted acquisition.
• Marker aroma compounds: Arabica samples were enriched in 5-methylfurfural, acetoin, and furaneol acetate — compounds associated with caramelized/roasted and creamy notes consistent with Arabica sensory quality. Robusta displayed higher levels of p-vinylguaiacol, a phenolic causing medicinal/off-flavor notes that may contribute to Robusta’s harsher aroma profile.
• Classification model performance: A leave-one-brand-out validation scheme used three of the four Arabica brands and both Robusta brands for training, then predicted the excluded Arabica brand. The Support Vector Machine model in eMSTAT Solution classified left-out Arabica brands with scores between 85 and 100, indicating robust discrimination potential even with limited brand-level variability.

Practical benefits and applications

• Rapid screening: HS-SPME combined with a short GC ramp and targeted MRM provides a faster workflow than solvent extraction and NMR while maintaining specificity for aroma markers.
• High sensitivity and reproducibility: MRM acquisition improves detection limits and consistency across runs, facilitating quantitative or semi-quantitative comparisons of marker volatiles.
• Accessible data analysis: Integration with curated aroma databases and user-friendly multivariate/statistical software allows laboratories without deep chemometrics expertise to build and deploy discrimination models for QA/QC and fraud detection.
• Qualitative insights: The approach links chromatographic markers to sensory descriptors (via the Smart Aroma Database), helping interpret which volatile classes drive perceived quality differences between species.

Future trends and potential extensions

• Broader sample sets: Expanding to wider geographic origins, harvest years, roast levels, and processing methods would strengthen model generalizability and reveal additional varietal or origin-specific markers.
• Quantitative method development: Calibration with standards could enable absolute quantification of discriminant compounds (e.g., p-vinylguaiacol, 5-methylfurfural) to set threshold criteria for authentication.
• Integration with orthogonal markers: Combining volatile profiling with lipid or non-volatile markers (e.g., 16-O-methylcafestol by NMR or targeted LC-MS) would improve confidence in mixed-sample detection and low-level adulteration scenarios.
• Automation and routine deployment: Streamlined sample handling, library-driven MRM methods, and embedded statistical workflows could enable high-throughput screening in commercial QA/QC laboratories and supply-chain monitoring.
• Machine learning advances: Use of ensemble classifiers, feature selection, and domain adaptation techniques could improve robustness across different roast conditions and instrument platforms.

Conclusions

Headspace SPME GC-MS/MS using a curated aroma database, together with multivariate analysis and SVM classification, can reliably separate Arabica and Robusta coffee beans based on their volatile profiles. Targeted MRM acquisition increased sensitivity and reproducibility, enabling identification of characteristic markers consistent with sensory expectations (Arabica: roasted/creamy volatiles; Robusta: phenolic off-note contributors). The combined experimental and statistical workflow supports rapid, practical authentication suitable for routine laboratory screening.

Reference

1. Gunning Y, et al. Food Chemistry, 2018;248:52–60.