Predicting the storage time of green tea by myricetin based on surface-enhanced Raman spectroscopy

Green tea, valued for its fresh flavor and bioactivity, deteriorates during storage due to oxidation and degradation of polyphenols, leading to stale flavor and browning. Specific standards and characteristic compounds indicating deterioration are lacking. Conventional approaches—sensory evaluation and chemical analyses (e.g., GC-MS, HPLC, LC-MS)—can detect changes but are time-consuming, require expertise, and may be subjective. Spectral analysis offers a rapid and convenient alternative; prior work with hyperspectral imaging and near-infrared spectroscopy built models to evaluate storage quality but did not identify indicator compounds. Surface-enhanced Raman spectroscopy (SERS) is nondestructive, fast, objective, and sensitive. Prior SERS studies quantified catechins and monitored wine storage changes, suggesting SERS could identify quality changes during green tea storage. This study investigates sensory quality and chemical compounds in green tea across storage times, acquires SERS spectra using silver nanoparticles, builds models to predict storage time, and identifies characteristic Raman peaks and indicator compounds (notably myricetin) to predict storage time via qualitative and quantitative analyses.

Prior studies show tea quality degrades during storage as polyphenols oxidize and glycosides hydrolyze; flavonol glycosides can form aglycones like myricetin, which accelerates EGCG oxidation and contributes to browning. Traditional analytical methods (GC-MS, HPLC, LC-MS) detect compositional changes but are laborious. Hyperspectral and near-infrared spectroscopy models have been used to assess storage quality but without identifying specific indicator compounds. SERS has demonstrated high sensitivity for flavonoids (e.g., catechin detection at micromolar levels using citrate-capped AgNPs) and has identified storage-related markers in other beverages (e.g., ethyl carbamate in wine). These works motivate using SERS to both discriminate storage time and pinpoint indicative compounds in green tea.

Samples: TPHK green tea (same variety and grade, Anhui, China) harvested 2015–2020; stored at 4 °C. Chemicals: Standards for catechins (C, EC, EGC, GC, ECG, GCG, EGCG), gallic acid (GA), caffeine (CAF), quercetin, kaempferol, myricetin (M) from Sigma-Aldrich; HPLC-grade solvents from Tedia. Sensory evaluation: Conducted per GB/T 23776-2018. Brewed 3.0 g tea in 150 mL boiling water for 4 min; infusion evaluated by a panel of 6 experienced assessors using a 9-point hedonic scale on color, aroma, taste, and overall acceptability. Randomized, blinded presentation; palate rinsing between samples. Ethical approval obtained; informed consent. Chemical analysis: Extraction per GB/T 8304-2013. UPLC (Waters H-Class 2489) with C18 guard and column per Nian method quantified caffeine, GA, catechins (C, EC, EGC, GC, ECG, GCG, EGCG) and myricetin. SERS substrate and data acquisition: Tea-to-water ratio 1:16; soaked at 70 °C for 30 min; filtrate used. Silver nanoparticles (AgNPs) synthesized by microwave-assisted reduction: mix 2 mL 2.0 g/L AgNO3 and 1 mL 10 g/L sodium citrate in 50 mL water; microwave 10 min; cool. Characterization: SEM to assess size/shape (36–52 nm, avg ~43.3 nm) and EDS for elemental composition. Tea infusion mixed 1:1 with AgNPs. Raman spectrometer: Horiba LabRAM HR Evolution; calibrated using Si at 520.7 cm⁻¹; sample deposited on gold-coated glass slides; settings: 10× objective, 200 µm slit, 600 lines/mm grating; 2 s integration per accumulation, total 6 s; spectral range 300–1800 cm⁻¹; 108 spectra collected. Spectral preprocessing: Savitzky–Golay (S-G) smoothing followed by standard normal variate (SNV) to reduce noise and fluorescence. Modeling: PCA used to reduce dimensionality and examine clustering; principal component scores and loadings extracted. Samples split via Kennard–Stone into calibration (72) and prediction (36) sets at 2:1 ratio. Two classifiers built on PCA scores: PCA-LDA and PCA-SVM (RBF kernel). SVM hyperparameters c and g optimized via grid search over 2^−10–2^10 with fivefold cross-validation; performance expressed as correct classification rate (%). Correlation analysis: Pearson correlations between Raman peak intensities (from PCA loadings-identified major peaks) and compound contents. Myricetin quantification by SERS: prepared myricetin standards on gold-coated slides; recorded Raman spectra; built standard curve using 730 cm⁻¹ peak intensity vs concentration; used linear fit y = 0.2227x + 0.0292 (R² = 0.9988) over 0.02–0.32 mg/g to quantify myricetin in samples; validated against UPLC; computed average errors. Statistical analysis: ANOVA with Duncan’s test for sensory; visualizations in Origin; multivariate analyses in MATLAB R2014a.

- Sensory evaluation: Significant differences across years (p < 0.05); all four sensory factors decreased with storage time; overall acceptability dropped sharply in 2015. - Chemical composition trends: With increasing storage time, GA, GC, EGC, EGCG, GCG, and ECG decreased; catechin (C) and myricetin (M) increased; caffeine (CAF) unchanged. Polyphenols oxidation associated with reduced freshness. Pearson correlations showed strong relationships between overall acceptability and compound contents: negative with GA, GC, EGC, EGCG, GCG, ECG (|r| > 0.95, p < 0.01) and positive with C, CAF, M (|r| > 0.95, p < 0.01). - SERS spectral changes: After S-G + SNV preprocessing, average spectra showed increasing intensities at 731.1 and 1320.5 cm⁻¹ and decreasing intensities at 1237.9, 1589, and 1631.3 cm⁻¹ with storage time. Assignments: ~730 cm⁻¹ (polyphenol C–H/benzene ring vibrations), 1237.9 cm⁻¹ (C=C stretch), 1320.5 cm⁻¹ (CH2 bending of carbohydrates), 1589 and 1631.3 cm⁻¹ (benzene ring stretches). Trends explained by catechin oxidation/polymerization, fatty acid oxidation, and polysaccharide degradation. - PCA: First three PCs explained 92.02% variance; samples clustered into two groups (2020–2018 vs 2017–2015). - Classification models: PCA-LDA achieved 80.56% (58/72) calibration and 69.44% (25/36) prediction accuracy. PCA-SVM (optimized, c = 111.4305, g = 0.0010; number of PCs reported as 11) achieved 98.61% (71/72) calibration and 97.22% (35/36) prediction accuracy. - Raman–compound correlations: Twelve major peaks identified (438.673, 732.387, 955.557, 1016.67, 1241.41, 1300.06, 1326.09, 1356.43, 1380.98, 1542.17, 1577.82, 1626.99 cm⁻¹). Peaks at 732.387 and 1326.09 cm⁻¹ correlated positively with M, EC, C and negatively with ECG, GCG, EGCG; peak at 1542.17 cm⁻¹ correlated positively with ECG, GCG, EGCG and negatively with CAF, EC, C. Five peaks (438.673, 1016.67, 1300.06, 1380.98, 1626.99 cm⁻¹) showed little correlation. - Myricetin as indicator: Among common tea polyphenols, a prominent ~730 cm⁻¹ Raman peak was specific to myricetin. The 730 cm⁻¹ peak intensity exhibited a strong linear relationship with myricetin concentration: y = 0.2227x + 0.0292, R² = 0.9988 (0.02–0.32 mg/g). Myricetin concentrations quantified by SERS closely matched UPLC values across years (2020–2015) with maximum average error 5.81%. Myricetin concentration correlated linearly with storage time (R² = 0.9346), increasing with longer storage.

The study demonstrates that SERS captures storage-induced chemical changes in green tea and, when combined with PCA-SVM, accurately predicts storage time, outperforming traditional sensory and standalone chemical analyses in speed and objectivity. Oxidation and hydrolysis of polyphenols during storage drive spectral changes, with increases in the ~730 cm⁻¹ peak and decreases in aromatic ring vibrations consistent with catechin transformation and polymerization. Crucially, the ~730 cm⁻¹ peak is specific to myricetin among tested polyphenols, and its intensity scales with myricetin concentration. Quantitative SERS agreed with UPLC within ~6% error, and myricetin levels showed a strong positive linear correlation with storage duration (R² = 0.9346), validating myricetin as an indicator compound. The PCA-SVM model’s high accuracy (97.22% prediction) indicates that nonlinear classification effectively separates subtle quality differences across years. This integrative approach identifies not only storage time but also mechanistic markers (myricetin) linked to quality deterioration, offering a practical, rapid tool for quality monitoring.

SERS, coupled with a PCA-SVM model, provides a rapid and accurate method to predict green tea storage time, achieving 97.22% prediction accuracy. The Raman peak at ~730 cm⁻¹ is a characteristic signature of myricetin; its intensity correlates linearly with myricetin concentration and increases with storage time. Myricetin concentration itself correlates strongly with storage duration, supporting its role as an indicator compound for storage-related quality change in green tea. Future work should validate and enhance the model with a broader range of green teas and storage conditions and develop portable Raman-based methods for on-site, instant quality assessment.

- Generalizability: Samples were from a single green tea type (TPHK) and stored at 4 °C; broader validation across varieties, origins, processing grades, and storage conditions is needed. - Sample size and years: Limited to 2015–2020 vintages with 108 spectra; larger datasets could further stabilize model performance. - Instrumentation: Bench-top SERS setup; field-deployable, portable Raman systems need development and validation. - Modeling details: Although high accuracy was achieved, model optimization (e.g., selection of PCs) and external validation on independent datasets are necessary to confirm robustness.