Roasting pretreatment reduces retort odor formation in green tea beverages: Evidence from chemometrics and sensory evaluation

Abstract

Aroma degradation is a pivotal technical challenge restricting the development of tea beverages. Addressing the aroma stability is a pressing issue for the tea beverage industry. In this study, the effect of roasting raw materials on the formation of retort odor in green tea beverages was assessed using chemometrics and sensory evaluation. The results found that roasting could significantly reduce the concentration of specific retort odorants including theaspirane (0.260 → 0 μg/L) and linalool (−31.30 %), while significantly promoting the accumulation of Maillard reaction products including 3-ethyl-2,5-dimethylpyrazine (0 → 0.800 μg/L), 2-acetylfuran (0 → 0.104 μg/L), and norisoprenoids including α-ionone (0 → 0.059 μg/L), β-cyclocitral (28.57 %), β-ionone (82.46 %) (p < 0.05). Sensory evaluation results indicated that moderate roasting (5 min) effectively reduced the intensity of retort odor in green tea beverages while maintaining high overall acceptability. The study provides a theoretical basis for flavor regulation in the tea beverage industry.

Graphical abstract

Highlights

• •Effect of roasting raw materials on retort odor formation was studied.
• •Roasting could significantly reduce the concentration of theaspirane and linalool.
• •Roasting significantly increased Maillard reaction products and norisoprenoids.
• •Moderate roasting could improve the retort odor of green tea beverages.

1. Introduction

Tea (Camellia sinensis (L.) O. Kuntze) is a globally renowned healthy beverage. Notably, the China Food and Nutrition Development Program (2025–2030) has prioritized “tea beverages” as one of the key nutritious foods to be developed for the first time. However, the industrialization of tea beverages faces a critical challenge: flavor instability during sterilization and storage, particularly in green tea beverages. Exposure to light, heat, and oxygen disrupts the balance of aroma compounds, resulting in the development of a distinct off-flavor known as “retort odor” (Kinugasa & Takeo, 1990; Tao et al., 2024; Wang, Gao, et al., 2024; Wang, Tang, et al., 2024). This phenomenon significantly compromises product quality and consumer acceptance, demanding urgent technical solutions to stabilize aroma profiles.

Roasting raw materials are known to influence tea beverage stability, whereas existing studies exhibit critical limitations. For instance, prior work relied on single extraction methods, failing to capture the full spectrum of aroma compounds and even lacked aroma-related data (Chen & Zhang, 2001; Fu et al., 2020; Yuan et al., 2010). Some studies focused broadly on “aroma stability” but overlooked the specific formation and modulation of retort odorants – the primary drivers of flavor deterioration (Liang, 2003; Fu et al., 2020). In addition, though roasting treatments (e.g., Fu et al., 2020) and roasted compounds (e.g., pyrazines and pyrroles in Tao et al., 2024) showed potential to improve off-flavors overall, the causal relationship between raw material roasting intensity and retort odor suppression remains unestablished.

To address these gaps, this study integrated comprehensive aroma extraction (solid-phase microextraction (SPME) and solid-phase extraction (SPE) coupled with gas chromatography–mass spectrometry (GC–MS)) and chemometric analysis — a powerful tool for identifying critical flavor markers in tea beverages (Wang et al., 2023; Wang, Gao, et al., 2024; Wang, Tang, et al., 2024) — to systematically investigate how the roasting level of green tea raw materials controlled the formation of retort odor and overall aroma profile. The objectives of this study were: (1) to clarify the changes in aroma compounds of unroasted green tea beverages during simulated storage processes, (2) to determine the effects of roasting treatment on aroma compounds in simulated stored beverages, and (3) to preliminarily speculate on the potential mechanisms by which roasting treatment improves steamed aroma. The study provides important scientific support for companies to select suitable raw materials and improve the aroma deterioration of tea beverages during shelf life.

2. Materials and methods

2.1. Chemicals and regents

Ferrous sulfate heptahydrate (analytically pure, purity ≥99 %), potassium sodium tartrate (analytically pure, purity ≥99 %), disodium hydrogen phosphate (analytically pure), and potassium dihydrogen phosphate (analytically pure, purity ≥99.5 %) were purchased from Shanghai Yuanye Biotechnology Co. (Shanghai, China). Pentane (High Performance Liquid Chromatography, HPLC grade), methyl tert-butyl ether (MTBE, HPLC grade) and C₇-C₄₀ n-alkanes standard were purchased from Shanghai Anpel Experimental Technology Co. (Shanghai, China). Ethanol (purity ≥99.7 %) was purchased from Sinopharm Chemical Reagent Co. (Shanghai, China). Sodium chloride (NaCl, purity ≥99.5 %) and anhydrous sodium sulfate (purity ≥99 %) were purchased from Shanghai Aladdin Biochemical Technology Co. (Shanghai, China). Ethyl decanoate (purity ≥98 %, GC grade) was obtained from TCI Industrial Development Co. (Shanghai, China). Pure water was purchased from Hangzhou Wahaha Group Co. (Hangzhou, China).

2.2. Sample preparations

2.2.1. Raw materials (green teas with different roasting levels)

The raw materials used in this study were sourced from Longyou, Zhejiang Province, and were harvested and processed in September 2023. Fresh tea leaves (one bud and three leaves standard) were processed as follows: withering (tank, 25 °C, 20 h), hot-air fixation (150 °C, 4 min), rolling (10 min of light pressure, 10 min of heavy pressure, and 10 min of light pressure), and drying (105 °C, 15 min → 75 °C to constant weight).

Once the green tea was cooled to room temperature, it was divided into three groups. The first group remained untreated and served as the control. For the second group, the green tea was manually stir-fried at the pot temperature of 230 °C for 5 min and a slight roasted aroma was developed. For the third group, the green tea was manually stir-fried at the pot temperature of 230 °C for 15 min and an obvious roasted aroma was developed. The three green teas with different roasting levels were designated as ck, 5, and 15 min, and kept at 4 °C until use.

2.2.2. Preparation of tea beverages

The green tea beverages were performed based on the experimental procedure of Wang, Gao, et al. (2024) and Wang, Tang, et al. (2024). First, 5.00 g of green tea prepared in Section 2.2.1 was accurately weighed into a 250 mL glass conical flask, and 225 mL of pure water at 70 °C was added to the flask. Then, the flask was immersed into a 70 °C water bath for 15 min and manually shaken every 5 min. After the extraction, the tea infusion was filtered through 300-mesh gauze into a 250 mL volumetric flask while still hot. The tea infusion was rapidly cooled down to about 25 °C in an ice-water bath, followed by volume fixation (250 mL). According to the national standard (GB/T 21733–2008 Tea Drinks), the mass concentration (mg/L) and solid concentration (brix) of tea polyphenols (TP) in the tea infusion were measured, and the tea infusion was diluted to a TP concentration of 500 mg/L. Next, the tea infusion was sterilized at 90 °C for 20 s. Finally, the tea infusion was hot-filled into aseptic bottles and sealed. The bottled tea beverages were allowed to cool naturally for subsequent storage tests.

2.2.3. Heat treatment simulated storage test

The tea beverages prepared in Section 2.2.2 were subjected to water bath treatment at 95 °C for 0 (control), 40 and 160 min, respectively. Following the treatment, the samples were promptly removed from the water bath and cooled to about 25 °C using an ice-water bath. Each treatment was performed in triplicate. The treated samples were then stored in a refrigerator at −18 °C for subsequent analysis.

2.3. Extraction of aroma substances

2.3.1. SPME

The procedures were slightly modified from the experimental method described by Wang, Gao, et al. (2024) and Wang, Tang, et al. (2024). An accurate volume of 30 mL of each sample was placed in a 100 mL headspace bottle. Then, 10 μL of ethyl decanoate (internal standard (IS), diluted with ethanol to 5 mg/L), 6.0 g of NaCl, and a magnetic stirrer were sequentially added to the bottle. The vial was tightly capped and equilibrated in a water bath at 45 °C for 15 min. Subsequently, a divinylbenzene/carboxylate/polydimethylsiloxane (50/30 μm DVB/CAR/PDMS) coated fiber was exposed to the headspace of the extraction vial for 45 min at the same temperatrue for volatile extraction. After extraction, the fiber was inserted into the GC–MS injector port and desorbed at 250 °C for 5 min.

2.3.2. SPE

The extraction was conducted according to the experimental steps described by Wang, Gao, et al. (2024) and Wang, Tang, et al. (2024) and Xu et al. (2023). First, 10 μL of ethyl decanoate (IS, diluted with ethanol to 1000 mg/L) was added to 300 mL of the sample. The sample was then extracted three times using 60 mL of a pentane/MTBE (v/v = 4:1) mixture. The extracted organic layer was dried over anhydrous sodium sulfate and concentrated to approximately 1 mL using a nitrogen concentrator at a flow rate of 1 mL/min at 30 °C. Next, an SPE column (Supelclean TM LC-Si, 6 mL tube, 500 mg filler) was activated with 6 mL of eluent (a mixture of pentane and MTBE, v/v = 1:1). The concentrate was loaded onto the SPE column and eluted with 10 mL of the same eluent. The sample was then concentrated to approximately 50 μL by nitrogen purging at a flow rate of 0.6 mL/min at 30 °C. The aroma profile of the sample was assessed by sniffing a drop of the concentrate added to a test strip and comparing it to the original sample's odor profile to ensure no distortion of the aroma. Finally, 3 μL of the SPE-extracted aroma concentrate was manually injected using a 10 μL Agilent syringe into an inlet (250 °C) equipped with a quartz cotton non-split liner tube.

2.4. GC–MS analysis

The volatile compounds in the samples were analyzed using an Agilent 8890 GC system coupled with an Agilent HP 5977 GC/MSD ion trap mass spectrometer, equipped with a DB-5MS column (30 m × 250 μm × 0.25 μm) for the separation of aroma compounds. The initial temperature of the column was set at 40 °C, with the injection port temperature at 250 °C and a flow rate of 1.2 mL/min. The system operated at a pressure of 9.147 psi in splitless mode, using high-purity helium (99.999 %) as the carrier gas. The GC temperature program was as follows: held at 40 °C for 2 min, then increased to 170 °C at 3 °C/min and held for 2 min, subsequently raised to 250 °C at 10 °C/min, and finally held at 250 °C for 2 min. For MS detection, the mass spectrometer was operated in electron ionization mode at 70 eV, with a mass scan range of 40–400 m/z and an ion source temperature of 230 °C (Wang, Gao, et al., 2024; Wang, Tang, et al., 2024).

2.5. Identification of aroma substances

The volatile metabolites in the samples were identified using two complementary approaches: (1) matching mass spectra against the National Institute of Standards and Technology Library (NIST, 98 L) and (2) utilizing the MassHunter Workstation Unknowns Analysis software (version B.07.01/internal version 7.1.524.0, Agilent Technologies, Inc., 2008). To enhance identification confidence and conpensate for potential chromatographic retention time shifts, the linear retention index (RI) for each compound was calculated. This was achieved by analyzing a homologous series of n-alkanes (C₇-C₄₀) under identical chromatographic conditions, following the established method described by Alasalvar et al. (2012). Compounds were tentatively identified based on spectral similarity (match factor > [80 %] combined with a RI tolerance of ± [20] units relative to literature or database values). For quantification, an internal standard (IS) method was employed. A known amount of a suitable IS (ethyl decanoate) was added to each sample prior to analysis. The concentration of each identified volatile component in the samples (expressed as μg/L) was calculated by multiplying the ratio of the peak area of the volatile metabolite to the peak area of the internal standard by the concentration of the internal standard (μg/L).

2.6. Validation by sensory evaluation

Sensory evaluation was conducted by ten professionally trained panelists from Tea Research Institute, Chinese Academy of Agricultural Sciences (Hangzhou, China), comprising six females and four males. Prior to participating in the sensory test, the panelists were fully informed of the test content, requirements, and risks, and provided their informed consent. Before the evaluation, 50 mL of the prepared tea beverage (Section 2.2) was accurately weighed and transferred to a mug with lid, which was then kept in a 35 °C water bath for 20 min to ensure the overall flavor of the sample reached a consistent and stable state (Wang, Gao, et al., 2024; Wang, Tang, et al., 2024). The panelists followed the methods described in the “Methods for Sensory Evaluation of Tea” (GB/T 23776-2018) to score the retort and roasted odors and the acceptability of tea beverages. The evaluation criteria were as follows: 0–2: “very weak”, 2–4: “weak”, 4–6: “neutral”, 6–8: “strong”, 8–10: “very strong” (Wang et al., 2022; Wang et al., 2023).

2.7. Statistical analysis

One-way analysis of the variance (ANOVA) was performed using IBM SPSS statistical software (version 25.0, SPSS Inc., USA), with a significance level of p < 0.05 (Duncan's test) considered statistically significant. Partial least squares-discriminant analysis (PLS-DA) was conducted using SIMCA (version 14.1, Umetrics AB, Sweden). Data visualization plots were created using Origin (version 2021, Origin Lab Inc., USA) and Adobe Illustrator (version 2021, Adobe Systems Inc., USA). All experiments were replicated three times to ensure data reliability.

3. Results and discussion

3.1. Changes of aroma substances during high temperature simulated storage in tea beverage made from the unroasted raw material

3.1.1. Aroma substance profile of unheated tea beverage using SPME and SPE

In this study, SPME and SPE were combined to extract aroma compounds from the target samples. Analysis of the aroma substance profile of tea beverage made from the unroasted raw material revealed that 69 and 128 aroma compounds were detected in the SPME and SPE extracts, respectively, with total aroma concentrations of 25.524 and 273.612 μg/L, respectively (Fig. 1).

Fig. 1.

Aroma substance profile of tea beverages with different heat treatments. “un-ck” indicates an unroasted and unheated sample, “un-40” indicates an unroasted sample heated for 40 min, “un-160” indicates an unroasted sample heated for 160 min. For SPME, 1–10 denote hydrocarbons, pyrroles and derivatives, esters, ketones, alcohols, aldehydes, heteroxides, acids, sulfurous compounds, and phenols and derivatives, respectively. For SPE, 1–11 denote hydrocarbons, pyrroles and derivatives, esters, ketones, alcohols, aldehydes, heteroxides, acids, lactones, phenols and derivatives, and sulfurous compounds, respectively.

In the SPME extract, hydrocarbons were the most frequently detected compounds (18), followed by alcohols (14), aldehydes (13), ketones (11), and esters (10). Heteroxides (2) and phenol and derivatives (1) were less frequently detected. In terms of concentrations, alcohols had the highest total concentration (15.217 μg/L), followed by aldehydes (4.006 μg/L) and phenols and derivatives (3.283 μg/L). Ketones and hydrocarbons were present at approximately 1 μg/L, while esters and heteroxides were below 1 μg/L (Table S1). In contrast, the SPE extract contained higher numbers of alcohols (31), hydrocarbons (28), and esters (24). Acids (19), ketones (16), aldehydes (15), pyrroles and derivatives (12) were also detected, along with smaller quantities of lactones (6), phenols and derivatives (3), and heteroxides (2). Acids and alcohols were the most prominent by concentrations, which were 92.965 and 73.524 μg/L, respectively. Ketones, hydrocarbons, esters, and lactones were present at 34.715, 23.886, 13.374, and 12.925 μg/L, respectively. Phenols and derivatives, aldehydes, and pyrroles were around or below 10 μg/L. Trace heteroxides were also detected (Table S2). These results indicated that SPME had a stronger affinity for alcohols, while SPE showed a greater affinity for acids and alcohols. The combined use of these two extraction methods facilitated a more comprehensive extraction of aroma compounds from the samples. This methodological approach has been successfully applied in prior studies for aroma extraction, specifically in Biluochun tea infusion (Wang, Gao, et al., 2024; Wang, Tang, et al., 2024) and Longjing tea beverages (Tao et al., 2024). The results of this study further support this conclusion, but the final conclusion still needs to be rigorously verified in combination with the experimental data in the following sections.

3.1.2. Overall changes in aroma substance profile of tea beverage after high temperature simulated storage

By simulating storage with heat treatment for the unroasted tea beverage, it was found that the aroma profiles of unheated and heated tea beverage samples had obvious differences (Fig. 1). By comparing the differences among the three treatment groups, the following trends were observed.

For SPME samples, the total number of aroma compounds showed an initial increase followed by a decrease as the heat treatment time was prolonged. The number of aroma compounds changed from 69 (ck) to 99 (40-min treatment) and then reduced to 83 (160-min treatment). In contrast, the total aroma concentrations increased with the treatment time, changing from 25.524 μg/L (ck) to 31.166 μg/L (40-min treatment) and finally to 37.578 μg/L (160-min treatment). In terms of specific aroma categories, the number of hydrocarbons, esters, and alcohols gradually increased, while the number of ketones and aldehydes remained relatively unchanged. In terms of concentrations, compared with the control group, hydrocarbons, aldehydes, alcohols, and phenols and derivatives in the 160-min treatment group increased to 188.822 %, 186.151 %, 131.316 %, and 124.900 %, respectively (Table S1). The concentrations of ketones, esters, and heteroxides also showed an increasing trend with prolonged treatment time.

For SPE samples, the total number of aroma compounds gradually decreased. The key aroma categories contributing to this change were alcohols (31 → 27 → 25, ck → 40-min treatment → 160-min treatment), hydrocarbons (28 → 25 → 21), esters (24 → 28 → 19), aldehydes (15 → 11 → 12), and pyrroles and derivatives (12 → 11 → 7). Additionally, the total aroma concentrations decreased significantly, following the trend of 273.612 → 208.997 → 147.895 μg/L. The key aroma categories responsible for the change were acids, ketones, hydrocarbons, and alcohols. Compared with ck, after heat treatment, the total concentration changes of these aroma categories reached −66.437 %, −51.544 %, −40.897 %, and − 16.847 %, respectively (Table S2). It was worth noting that acids were undetectable in SPME samples but predominated in SPE samples. In SPME, only three types of acids were detected, with concentrations ranging from 0 to 0.099 μg/L with heat treatment (Table S1), while in SPE, the detected concentrations of acids ranged from 31.201 to 92.965 μg/L (Table S2). This indicated the SPE method showed a significantly stronger preference for acids than the SPME method. Based on the above results, the difference in adsorption preference between the SPME and SPE methods was one of the key factors causing the difference in their key aromas. In a previous study on the aroma of green tea beverages (Tao et al., 2024), the differences between the two methods were also pointed out.

3.1.3. Screening for potential differential substances

To clarify the key contributors to the aroma profile differences, a supervised PLS-DA classification model was constructed for screening, as shown in Fig. 2.

Fig. 2.

Chemometric analysis of aroma substances in tea beverages with different heat treatments. a-d denote the chemometric analysis of SPME samples. The scores scatterplot (a; R²X (cum) = 0.858, R²Y (cum) = 0.991, Q² (cum) = 0.979), loading scatterplot (b), 200 permutations plot (c), and 3D VIP scatterplot (d). e-h denote the chemometric analysis of SPE samples. The scores scatterplot (e; R²X (cum) = 0.979, R²Y (cum) = 0.994, Q² (cum) = 0.985), loading scatterplot (f), 200 permutations plot (g), 3D VIP scatterplot (h).

Under SPME conditions, the samples from three treatments were distinctly classified into three groups (Fig. 2a). Substances near the periphery were identified as key contributors to group differences in the subsequent loading scatterplot (Fig. 2b). The model's reliability was validated via 200 permutations, meeting the criteria that the intercept of of R² with the y-axis > 0 and the intercept of Q² with the y-axis < 0 (Fig. 2c), with model parameters R²X (cum) = 0.858, R²Y (cum) = 0.991, and Q² (cum) = 0.979. Based on variable importance in the projection (VIP) > 1, 25 key potential differential substances were screened, as shown in Fig. 2d and Table S3. Of these aroma substances, the concentrations of linalool, nonanal, 2,6-di-tert-butylbenzoquinone, benzaldehyde, phenylacetaldehyde, 2,4-di-tert-butylphenol, theaspirane, geraniol, α-terpineol, 2-methylbutyraldehyde, octanol, and myrcene gradually increased with prolonged treatment time. In addition to the key retort odorants reported by Wang, Gao, et al. (2024) and Wang, Tang, et al. (2024), several other aroma substances also were identified. Research has found that the content of linalool, which imparts a floral aroma, increased significantly by 144 % after 22 days of black tea fermentation, presumably due to extracellular enzymes secreted by D. hansenii promoting the release of linalool from aromatic precursors (Du et al., 2022; Huang et al., 2025). In tea beverage systems without enzyme involvement, the increase in both linalool and geraniol content may be attributed to the non-enzymatic hydrolysis of their corresponding glycoside precursors during thermal processing (Kumazawa & Masuda, 2001). Previously, 2,6-di-tert-butylbenzoquinone was observed in Longjing tea from the Yuezhou production area (Wang et al., 2018), Baojing Huangjincha 1 (Huang et al., 2020) and Congou black tea (Liu et al., 2021) from different production areas. Mao (2018) found that the content of 2,6-di-tert-butylbenzoquinone gradually increased with the roasting degree of large-leaf yellow tea. Kim et al. (2007) observed an increase in 2,4-di-tert-butylphenol content when green tea infusion were heated at 85 or 95 °C. Liu (2014) also detected this compound in the fresh scent-flavor green tea infusion stored for 7 days. Our results provided additional evidence that heat treatment promoted the production of the compound. The content of 2-methylbutyraldehyde was found to be upregulated in Yinghong Gongfu black tea during drying (Ye et al., 2021) and shaken black tea (Xue et al., 2022). Dou (2007) detected octanol in both green tea raw material and beverage, but its content was significantly lower in the raw material than in the beverage, possibly due to chemical changes during tea beverage preparation. Research has found that as the roasting temperature increased, the content of myrcene gradually increased (Mao, 2018). Myrcene was one of the key contributors to the chestnut aroma in green tea (Zhu et al., 2018), imparting a fruity and herbal scent of green tea (Xiao et al., 2024).

Subsequent PLS-DA modeling analysis of SPE samples using the same methodology revealed clear differentiation between samples from different treatment groups. The model parameters were R²X (cum) = 0.979, R²Y (cum) = 0.994, and Q² (cum) = 0.985, with the intercept of of R² with the y-axis > 0 and the intercept of Q² with the y-axis < 0, indicating no overfitting and satisfying reliability. Based on the criteria of VIP > 1, 34 potential differential substances were screened (Fig. 2e-h, Table S3). Among them, tetrahydrofurfuryl alcohol and 2,3-dihydrobenzofuran showed a significant increase with prolonged treatment time. In contrast, certain acids, such as n-hexadecanoic acid, octadecanoic acid, oleic acid, and myristic acid, along with 7,9-di-tert-butyl-1-oxaspiro[4.5]deca-6,9-diene-2,8-dione, significantly decreased. Tetrahydrofurfuryl alcohol exhibits an oily caramel aroma with coffee and nutty notes at low concentrations. 2,3-Dihydrobenzofuran has a chemical tar-like, phenolic, and smoky odor. Fatty acids like n-hexadecanoic acid, octadecanoic acid, oleic acid, and myristic acid are closely related to tea aroma quality and serve as important precursors for tea aroma compounds (Wang et al., 2019). Their decrease during heat treatment may be attributed to the fact that the rate of hydrolysis of lipids such as triglycerides or oxidation of aldehydes to fatty acids was less than the rate of conversion of fatty acids to fatty alcohols and fatty aldehydes (Wang et al., 2021).

3.2. Effect of roasting raw materials on aroma substances after simulated storage

To determine whether roasting pretreatmentcould improve the retort odor in tea beverages after simulated storage, three raw materials with different roasting levels were prepared and then used for tea beverage production. A total of 187 compounds were detected in SPME samples among the three tea beverages (Table S1 & Fig. S1), including 43 hydrocarbons, 34 pyrroles and derivatives, 28 ketones, 27 esters, 22 alcohols, 19 aldehydes, 9 heteroxides, 3 acids, 1 sulfurous compound and 1 phenol and derivative. In SPE samples, 250 aroma compounds were identified in the samples (Table S2 & Fig. S2), comprising 41 hydrocarbons, 38 pyrroles and derivatives, 35 esters, 29 ketones, 41 alcohols, 19 aldehydes, 6 heteroxides, 26 acids, 6 lactones, 6 phenols and derivatives, and 3 sulfurous compounds.

A supervised PLS-DA model was applied to clarify the effect of roasting raw materials on tea beverage aroma during heat treatment. In SPME samples, the model distinctly distinguished samples from different treatment groups (Fig. 3a). Aroma compounds around the periphery in the loading scatterplot (Fig. 3b) were considered key contributors to group differences. Model reliability was confirmed with R²X (cum) = 0.89, R²Y (cum) = 0.999, Q² (cum) = 0.993 (Fig. 3c). A total of 69 potential contributors were screened using the criteria of VIP > 1.5 (Fig. 3d, Table S4). In SPE samples, a similar PLS-DA model was established (R²X (cum) = 0.956, R²Y (cum) = 0.987, Q² (cum) = 0.961, Fig. 3e). The control and roasted samples were clearly separated along the x-axis, while the distinction between the 5-min and 15-min roasted groups was less pronounced but showed a shift from the positive to negative y-axis direction with increased roasting levels. The compounds around the periphery (Fig. 3f) were regarded as potential contributors. After model validation (Fig. 3g), 55 potential contributors were identified based on the criteria of VIP > 1 (Fig. 3h, Table S4). To better understand how the roasting level of raw materials affected the aroma substance profiles of tea beverages during heat treatment, 21 aroma compounds (SPME, Table 1) and 25 aroma compounds (SPE, Table 2) that showed significant variation across roasting reatment groups were selected for further analysis.

Fig. 3.

Chemometric analysis of aroma substances in tea beverages prepared with different roasting levels of raw materials. a-d denote the chemometric analysis of SPME samples. The scores scatterplot (a; R²X (cum) = 0.89, R²Y (cum) = 0.999, Q² (cum) = 0.993), loading scatterplot (b), 200 permutations plot (c), and 3D VIP scatterplot (d)). e-h denote the chemometric analysis of SPE samples. The scores scatterplot (e; R²X (cum) = 0.956, R²Y (cum) = 0.987, Q² (cum) = 0.961), loading scatterplot (f), 200 permutations plot (g), 3D VIP scatterplot (h).

Table 1.

Changes in key aroma compounds of unheated and heated tea beverages prepared with different roasting levels of raw materials (SPME conditions).

Compounds	Odor descriptiona	Odor threshold(μg/L)b	Concentrations(μg/L)c
			un-ck	un-40	un-160	5-ck	5–40	5–160	15-ck	15–40	15–160
1-Ethyl-1H-pyrrole	Burning	10,000	0.000	0.029	0.076	1.163	1.183	1.349	4.146	4.620	4.338
3-Furaldehyde	Fruity, floral	–	0.000	0.000	0.000	0.000	0.049	0.000	0.193	0.090	0.160
Benzyl alcohol	Fruity, sweet	2546.21	0.127	0.146	0.182	0.098	0.093	0.108	0.080	0.099	0.140
α-Ionone	Violet	3.78	0.024	0.000	0.000	0.032	0.000	0.000	0.073	0.039	0.059
cis-Linalool oxide (furanoid)	Sweet floral, green, fruity	100	0.400	0.364	0.363	0.449	0.395	0.493	0.707	0.609	0.594
1-Ethyl-1H-pyrrole-2-carbaldehyde	Roasted	–	0.000	0.104	0.000	0.402	0.386	0.442	1.556	2.111	1.969
3-Methylbutanal	Malt	1.1	0.494	0.780	0.000	0.749	1.176	1.469	0.683	0.999	1.322
Theaspirane	Herbal, sweet, woody	4.85	0.069	0.255	0.260	0.062	0.067	0.090	0.021	0.041	0.000
β-Cyclocitral	Mint-like	3	0.106	0.092	0.098	0.111	0.106	0.133	0.145	0.131	0.126
trans-Linalool oxide (furanoid)	Sweet floral, citrus, fruity	190	0.458	0.414	0.414	0.510	0.436	0.532	0.614	0.571	0.529
Furfuryl methyl ether	–	–	0.000	0.000	0.000	0.000	0.000	0.000	0.110	0.037	0.000
Phenylacetaldehyde	Floral, honey	6.3	0.372	0.607	1.058	0.653	1.048	1.293	0.704	2.345	2.073
Acetophenone	Foxglove, bitter almond, rubber	65	0.000	0.061	0.101	0.056	0.000	0.115	0.085	0.111	0.178
2-Acetylfuran	Coffee-like	0.0150252	0.000	0.000	0.000	0.003	0.034	0.040	0.075	0.000	0.104
3-Methylfuran	–	–	0.000	0.000	0.000	0.043	0.050	0.061	0.000	0.096	0.190
Pentanal	Green, fatty, moldy	12	0.154	0.120	0.083	0.108	0.139	0.000	0.000	0.081	0.000
Cedrol	Sweet fruity, cedar	500	0.257	0.335	0.447	0.322	0.574	0.378	0.535	0.302	0.497
β-ionone	Cedar, violet	8.4	0.112	0.072	0.057	0.121	0.084	0.073	0.215	0.089	0.104
4,5-Dimethyl-2-isopropyloxazole	–	–	0.000	0.000	0.000	0.257	0.289	0.288	1.081	0.616	0.781
2,2,4-Trimethyl-1,3-pentanediol diisobutyrate	Moldy	–	0.020	0.027	0.017	0.022	0.000	0.009	0.137	0.008	0.000
1-(2-Furanylmethyl)-1H-pyrrole	Vegetable, green	100	0.000	0.000	0.000	0.000	0.000	0.000	0.549	0.337	0.550

^aReference: Fenaroli's Handbook of Flavor Ingredients (sixth edition).

^bReference: Compendium of Olfactory Thresholds of Compounds (second edition of the original book).

^cDenotes concentration of different treatments, where ck denotes unroasted and unheated, un-40 denotes unroasted and heated 40 min, un-160 denotes unroasted and heated 160 min, 5-ck denotes roasted 5 min and unheated, 5–40 denotes roasted 5 min and heated 40 min, 5–160 denotes roasted 5 min and heated 160 min, 15-ck denotes roasted 15 min and unheated, 15–40 denotes roasted 15 min and heated 40 min, 15–160 denotes roasted 15 min and heated 160 min. The same as below.

Table 2.

Changes in key aroma compounds of unheated and heated tea beverages prepared with different roasting levels of raw materials (SPE conditions).

Compounds	Odor descriptiona	Odor threshold (μg/L)b	Concentrations (μg/L)
			un-ck	un-40	un-160	5-ck	5–40	5–160	15-ck	15–40	15–160
Tetrahydrofurfuryl alcohol	Caramel, coffee, nuts	–	12.603	10.173	29.322	0.000	0.000	0.197	0.515	0.000	0.211
n-Hexadecanoic acid	Waxy, fatty	–	35.162	23.570	8.851	0.000	0.370	1.151	3.881	1.047	3.356
Styrene	Sweet, floral	65	1.681	1.168	1.037	0.000	4.706	6.276	6.320	6.414	8.575
Octadecanoic acid	Fatty	20,000	30.773	16.833	6.605	3.377	0.317	1.016	1.115	0.642	2.925
1-Ethyl-1H-pyrrole-2-carbaldehyde	Roasted	–	0.202	0.169	0.091	0.776	0.000	0.000	2.318	1.478	1.727
7,9-Di-tert-butyl-1-oxaspiro[4,5]deca-6,9-diene-2,8-dione	–	–	22.152	13.592	8.333	6.293	1.050	2.287	4.462	3.042	5.058
2,3-Dihydrobenzofuran	Roasted	–	0.000	3.252	3.786	0.000	1.100	3.360	3.284	2.112	4.254
Benzyl alcohol	Fruity, sweet	2546.21	13.260	15.749	8.571	4.291	2.507	3.410	2.560	2.920	2.733
Benzeneacetic acid	Sweet, honey	12,000	5.067	5.157	6.053	0.000	0.000	0.000	0.000	0.000	0.000
7-Methoxy-2H-1-benzopyran-2-one	Sweet, creamy	–	9.730	5.562	3.823	2.088	0.000	1.015	0.959	1.046	1.571
2,4-Di-tert-Butylphenol	Phenolic, leather	500	9.450	6.558	4.022	4.626	1.318	3.425	2.832	3.834	4.608
1H-Pyrrole-2-carboxaldehyde	–	65,000	0.000	0.000	0.106	0.242	0.162	0.319	1.128	0.625	0.762
Linalool	Citrus, floral	0.22	4.990	4.166	2.453	1.787	0.349	1.670	0.000	0.483	1.690
2-Ethyl-1-hexanol	Elegant, fruity	25,482.2	5.077	5.755	2.578	1.970	0.495	0.906	1.207	1.188	1.089
3-Ethyl-2,5-dimethylpyrazine	Earthy, potato	8.6	0.000	0.000	0.000	0.574	0.287	0.344	1.045	0.701	0.800
2-Ethyl-5-methylpyrazine	Roasted, coffee	16	0.000	0.000	0.000	0.873	0.253	0.568	1.390	0.985	0.684
Benzene	Gasoline-like	111,000	1.115	1.763	2.131	0.000	0.000	0.000	0.000	0.000	0.986
(Z)-2-Methyl-5-(1-propenyl)-pyrazine	–	–	0.000	0.000	0.000	0.000	0.000	0.000	0.504	0.322	0.000
3-Methylindole	Mothball, feces, ripe fruits	0.13	0.000	0.000	0.000	0.000	0.000	0.000	0.306	0.160	0.210
cis-Linalool oxide (pyranoid)	Sweet floral, citrus, fruity	3000	3.162	3.232	1.477	1.144	0.566	0.717	0.659	0.634	0.580
2,6-Diethylpyrazine	Sweet	6	0.000	0.000	0.000	0.000	0.000	0.063	0.277	0.200	0.180
cis-Linalool oxide (furanoid)	Sweet floral, green, fruity	100	1.747	1.887	0.964	0.873	0.419	0.594	0.875	0.000	0.791
trans-Linalool oxide (furanoid)	Sweet floral, citrus, fruity	190	2.636	2.798	1.345	1.220	0.559	0.767	1.067	0.792	0.870
Nonanal	Citrus, soap	1.1	1.292	2.319	0.711	0.984	0.000	0.343	0.000	0.000	0.595
2-Hexyl-1-decanol	–	–	0.000	0.000	0.000	0.000	0.000	0.246	0.302	0.000	0.291

^aReference: Fenaroli's Handbook of Flavor Ingredients (sixth edition).

^bReference: Compendium of Olfactory Thresholds of Compounds (second edition of the original book).

For SPME samples, 1-ethyl-1H-pyrrole (un-ck → 5-ck → 15-ck: 0 → 1.163 → 4.146 μg/L), 3-furaldehyde (0 → 0 → 0.193 μg/L), α-ionone (0.024 → 0.032 → 0.073 μg/L), cis-linalool oxide (furanoid, 0.400 → 0.449 → 0.707 μg/L), 1-ethyl-1H-pyrrole-2-carbaldehyde (0 → 0.402 → 1.556 μg/L), β-cyclocitral (0.106 → 0.111 → 0.145 μg/L), trans-linalool oxide (furanoid, 0.458 → 0.510 → 0.614 μg/L)), furfuryl methyl ether (0 → 0 → 0.110 μg/L), phenylacetaldehyde (0.372 → 0.653 → 0.704 μg/L), acetophenone (0 → 0.056 → 0.085 μg/L), 2-acetylfuran (0 → 0.003 → 0.075 μg/L), cedrol (0.257 → 0.322 → 0.535 μg/L), β-ionone (0.112 → 0.121 → 0.215 μg/L), and 1-(2-furanylmethyl)-1H-pyrrole (0 → 0 → 0.549 μg/L) gradually increased with the roasting level (Table 1). Part of our results were consistent with previous studies. Tao et al. (2024) found that roasting raw materials of Longjing tea led to a significant increase in the content of phenylacetaldehyde, which was 0.03 μg/L before roasting and 1.7 μg/L after roasting. Fu et al. (2020) found a significant increase in the concentration of 1-(2-furanylmethyl)-1H-pyrrole in beverages prepared from raw teas (p < 0.05) after roasting treatments of steamed and pan-fired green tea raw materials and a significant positive correlation between the content of this substance and the degree of roasting in steamed autumn green tea beverages (p < 0.05). In contrast, benzyl alcohol (0.127 → 0.098 → 0.080 μg/L), theaspirane (0.069 → 0.062 → 0.021 μg/L), and pentanal (0.154 → 0.108 → 0 μg/L) decreased with prolonged roasting treatment. In addition, as the roasting degree increased, the content of 3-methylbutanal, 3-methylfuran, and 4,5-dimethyl-2-isopropyloxazole initially increased and then decreased. Prolonged heat treatment significantly deteriorated the aroma of tea beverage (Table 1), which was consistent with previous finding (Fu et al., 2020). In samples heat-treated for 160 min, the concentrations of some roasted compounds (such as 1-ethyl-1H-pyrrole (un-160 → 15–160: 0.076 → 4.338 μg/L, rate of change: 5607.89 %), 1-ethyl-1H-pyrrole-2-carbaldehyde (0 → 1.969 μg/L) and 2-acetylfuran (0 → 0.104 μg/L)), some floral compounds (such as phenylacetaldehyde (1.058 → 2.073 μg/L, 95.94 %), β-ionone (0.057 → 0.104 μg/L, 82.46 %), cis-linalool oxide (furanoid, 0.363 → 0.594 μg/L, 63.64 %), trans-linalool oxide (furanoid, 0.414 → 0.529 μg/L, 27.28 %), 3-furaldehyde (0 → 0.160 μg/L), α-ionone (0 → 0.059 μg/L)), as well as other compounds (such as acetophenone (0.101 → 0.178 μg/L, 76.24 %), β-cyclocitral (0.098 → 0.126 μg/L, 28.57 %), 4,5-dimethyl-2-isopropyloxazole (0 → 0.781 μg/L), 1-(2-furanylmethyl)-1H-pyrrole (0 → 0.550 μg/L), and 3-methylfuran (0 → 0.190 μg/L) gradually increased after roasting. While the concentrations of theaspirane, pentanal and 2,2,4-trimethyl-1,3-pentanediol diisobutyrate decreased to below the detection limit after heating the heavy roasted (15 min) samples for 160 min. Benzyl alcohol initially decreased and then increased, while 3-methylbutanal showed the opposite trend.

For SPE samples, the concentrations of octadecanoic acid (un-ck → 5-ck → 15-ck: 30.773 → 3.377 → 1.115 μg/L), 7,9-di-tert-butyl-1-oxaspiro[4.5]deca-6,9-diene-2,8-dione (22.152 → 6.293 → 4.462 μg/L), benzyl alcohol (13.260 → 4.291 → 2.560 μg/L), benzeneacetic acid (5.067 → 0 → 0 μg/L), 7-methoxy-2H-1-benzopyran-2-one (9.730 → 2.088 → 0.959 μg/L), 2,4-di-tert-butylphenol (9.450 → 4.626 → 2.832 μg/L), linalool (4.990 → 1.787 → 0 μg/L), 2-ethyl-1-hexanol (5.077 → 1.970 → 1.207 μg/L), benzene (1.175 → 0 → 0 μg/L), cis-linalool oxide (pyranoid, 3.162 → 1.144 → 0.659 μg/L), cis-linalool oxide (furanoid, 1.747 → 0.873 → 0.875 μg/L), and trans-linalool oxide (furanoid, 2.636 → 1.220 → 1.067 μg/L) gradually declined with increasing roasting levels (Table 2). A previous study found that the flavor dilution (FD) values of benzeneacetic acid (floral) and 7-methoxy-2H-1-benzopyran-2-one (sweet) decreased in roasted Longjing tea beverages (Tao et al., 2024), which was similar to the results of the present study. It was important to note that their results showed an increase in the FD value of linalool (Tao et al., 2024), which was inconsistent with our results and might be related to the source of tea and the degree of roasting. Whereas the concentrations of 1-ethyl-1H-pyrrole-2-carbaldehyde (0.202 → 0.776 → 2.318 μg/L), 2,3-dihydrobenzofuran (0 → 0 → 3.284 μg/L), 1H-pyrrole-2-carboxaldehyde (0 → 0.242 → 1.128 μg/L), 3-ethyl-2,5-dimethylpyrazine (0 → 0.574 → 1.045 μg/L), 2-ethyl-5-methylpyrazine (0 → 0.873 → 1.390 μg/L), (Z)-2-methyl-5-(1-propenyl)-pyrazine (0 → 0 → 0.504 μg/L), 3-methylindole (0 → 0 → 0.306 μg/L), 2,6-diethylpyrazine (0 → 0 → 0.277 μg/L), 2-hexyl-1-decanol (0 → 0 → 0.302 μg/L) increased along with the roasting level. It was found that the content of 3-methylindole in the roasted Longjing tea beverages ranged from undetectable to a significant increase in FD value (Tao et al., 2024). It was worth mentioning that tetrahydrofurfuryl alcohol, n-hexadecanoic acid, and styrene showed a decrease followed by an increase with deeper roasting. In samples heat-treated for 160 min, the concentrations of styrene (un-160 → 15–160: 1.037 → 8.575 μg/L, rate of change: 726.90 %), 2-ethyl-5-methylpyrazine (0 → 0.684 μg/L), 3-ethyl-2,5-dimethylpyrazine (0 → 0.800 μg/L), 2,6-diethylpyrazine (0 → 0.180 μg/L), 3-methylindole (0 → 0.210 μg/L), 1H-pyrrole-2-carboxaldehyde (0.106 → 0.762 μg/L), and 2-hexyl-1-decanol (0 → 0.291 μg/L) increased after roasting intervention. Whereas the concentrations of tetrahydrofurfuryl alcohol (29.322 → 0.211 μg/L, −99.28 %), benzyl alcohol (8.571 → 2.733 μg/L, −68.11 %), benzeneacetic acid (6.053 → 0 μg/L), and linalool (2.453 → 1.690 μg/L, −31.30 %) significantly decreased. In addition, n-hexadecanoic acid, octadecanoic acid, 1-ethyl-1H-pyrrole-2-carbaldehyde, 7,9-di-tert-butyl-1-oxaspiro[4.5]deca-6,9-diene-2,8-dione, 2,3-dihydrobenzofuran, 7-methoxy-2H-1-benzopyran-2-one, 2,4-di-tert-butylphenol, linalool, 2-ethyl-1-hexanol, benzene, cis-linalool oxide (pyranoid), cis-linalool oxide (furanoid), trans-linalool oxide (furanoid) showed a decrease followed by an increase. Previous studies found that 7-methoxy-2H-1-benzopyran-2-one did not change much before and after sterilization with or without roasting (Tao et al., 2024), which was inconsistent with the results of the present study and might be related to the tea source and the degree of roasting.

Combining the threshold values (Table 1, Table 2), it could be found that distinct cvoncentration trends for of specific aroma compounds under roasting and thermal treatment. The concentrations of 2-acetylfuran, phenylacetaldehyde, 3-ethyl-2,5-dimethylpyrazine, 2,6-diethylpyrazine, 3-methylindole, α-ionone, β-cyclocitral, and β-ionone significantly increased with roasting intensity amd further increased after thermal treatment. Conversely, the concentrations of theaspirane and linalool decreased with both roasting intensity and subsequent thermal treatment. Based on this, it could be concluded that during the simulated storage process of green tea beverages subjected to heat treatment, roasting the raw materials could reduce the concentration of several specific retort odorants (such as theaspirane and linalool) in the beverage, while promoting the accumulation of Maillard reaction products (MRPs, such as phenylacetaldehyde,2-acetylfuran,3-ethyl-2,5-dimethylpyrazine,3-methylindole,2,6-diethylpyrazine) and norisoprenoids (such as α-ionone,β-cyclocitral,β-ionone).

Subsequently, an intuitive sensory assessment trial (Table S5) was conducted to confirm whether the treatment attenuated the perception of retort odor in the samples. The results showed that heat treatment increased retort odor and decreased aroma acceptability. However, roasting treatment markedly reduced retort odor, with moderate roasting (5 min) showing the optimal effect. This suggested that moderate roasting not only enhanced aroma acceptability but also improved retort odor, while excessive roasting (15 min) may negatively impact aroma acceptability. Thus, moderate roasting was more effective at reducing the retort odor of tea beverage after simulated storage.

3.3. Potential mechanisms of roasting raw materials to reduce retort odor

The above findings revealed a significant difference in the aroma substances between roasted and unroasted samples after simulated storage. Fig. 4 illustrated roasting-induced changes in key aroma substances (VIP > 1, low odor thresholds). Combined with Table S4, it was found that theaspirane (VIP: 1.584) and linalool (VIP: 1.643) were significantly reduced (p < 0.05) in tea beverages with shelf-life after roasting treatment. Plus, several MRPs (3-ethyl-2,5-dimethylpyrazine (earthy, potato; VIP: 1.527), 2-acetylfuran (coffee-like; VIP: 1.303)), and norisoprenoids (α-ionone (violet; VIP: 1.776), and β-cyclocitral (mint-like; VIP: 1.508), β-ionone (cedar, violet; VIP: 1.103)) became more prominent (p < 0.05) under the same conditions. The key substances induced by roasting caused the herbal aroma in the samples with high intensity retort odor to weaken, while the violet, mint, potato, and roasted properties were enhanced. The synergistic effects of multiple key compounds collectively regulated the aroma characteristics of green tea beverages during their shelf life. Therefore, it was speculated that roasting might induce perceptual interactions among aroma compounds in tea beverages—a phenomenon extensively documented in flavor chemistry research. For instance, Jin et al. (2023) observed a significant masking effect between methyl salicylate and hexanal, as well as between methyl salicylate and benzaldehyde. In contrast, linalool and benzaldehyde, linalool and hexanal, and hexanal and benzaldehyde exhibited superimposed effects. Future research will be conducted to explore the role of perceptual interactions in reducing the retort odor of tea beverage by roasting the raw materials.

Fig. 4.

Concentrations of key aroma substances under different roasing treatments. UBUH, unroasted and unheated; B1UH, roasted 5 min, unheated; B2UH, roasted 15 min, unheated; UBH1, unroasted, heated 40 min; B1H1, roasted 5 min, heated 40 min; B2H1, roasted 15 min, heated 40 min; UBH2, unroasted, heated 160 min; B1H2, roasted 5 min, heated 160 min; B2H2, roasted 15 min, heated 160 min. Different letters on the columns indicate significant differences (Duncan's test, p < 0.05) within different roasting groups (UBUH/B1UH/B2UH, UBH1/B1H1/B2H1, UBH2/B1H2/B2H2).

4. Conclusions

This study evaluated the effect of roasting levels of raw materials on reducing the formation of retort odor in green tea beverages. Results showed that roasting treatment could significantly reduce the concentration of specific retort odorants such as theaspirane (herbal, sweet, woody; 0.260 → 0 μg/L) and linalool (citrus, floral; 2.453 → 1.690 μg/L, −31.30 %), but at the same time significantly promoted the accumulation of MRPs such as 3-ethyl-2,5-dimethylpyrazine (earthy, potato; 0 → 0.800 μg/L), 2-acetylfuran (coffee-like; 0 → 0.104 μg/L), and norisoprenoids such as α-ionone (violet; 0 → 0.059 μg/L), β-cyclocitral (mint-like; 0.098 → 0.126 μg/L, 28.57 %), β-ionone (cedar, violet; 0.057 → 0.104 μg/L, 82.46 %). The dominant role of the above-mentioned key substances weakened the herbal aroma while enhancing the characteristics of violet, mint, potato, and roasting in the samples with high intensity retort odor, thereby effectively regulated the aroma quality of green tea beverages during their shelf life. Sensory evaluation analysis results indicated that moderate roasting (5 min) could effectively reduce the retort odor in green tea beverages while maintaining high overall acceptability. Nevertheless, the study has not thoroughly investigate the interactions between roasted substances and key retort odor compounds. Future research will focus on verifying interaction types (synergistic, additive, masking, or no effect) based on aroma recognition thresholds, intensity, and contribution. This study will provide better theoretical guidance for targeted flavor quality regulation of tea beverages.

CRediT authorship contribution statement

Jie-Qiong Wang: Writing – original draft, Software, Methodology, Investigation, Formal analysis, Data curation. Ying Gao: Writing – review & editing, Supervision, Funding acquisition. Jian-Xin Chen: Validation. Fang Wang: Validation. Yuan-Yuan Ma: Data curation. Zhi-Hui Feng: Writing – review & editing. Jun-Feng Yin: Supervision. Liang Zeng: Supervision. Weibiao Zhou: Supervision. Yong-Quan Xu: Writing – review & editing, Supervision, Project administration, Funding acquisition.

Ethics statement

Prior to the sensory tests in this study, panelists from the Tea Research Institute of the Chinese Academy of Agricultural Sciences (Hangzhou, China) were fully informed about the test specifics, requirements, and risks, and signed informed consent forms. Additionally, the test was authorized for sensory evaluation by this institution.

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

Data availability

Data will be made available on request.