Impact of different frying methods on volatile compounds in tofu

Abstract

To investigate the effects of different frying methods on tofu aroma, sensory evaluation, electronic nose analysis, GC–IMS, and GC–MS combined with chemometric methods were employed to characterize volatile compounds in control, deep-fried, air-fried, and pan-fried samples. Sensory evaluation showed that pan-fried tofu achieved the highest overall acceptability. GC–IMS revealed distinct volatile fingerprints among samples, reflecting marked compositional shifts induced by different frying methods. Furthermore, GC–MS identified 153 volatile compounds, with esters being the predominant group. The integrated analytical approach demonstrated that frying methods significantly affected the generation and transformation of aroma-active compounds, with pan-fried tofu exhibiting the richest and most desirable aroma profile.

Graphical abstract

Highlights

• •Investigated frying effects on tofu aroma using sensory, E-nose, GC-IMS, and GC–MS.
• •Identified 153 volatile compounds; esters were dominant in fried tofu.
• •Pan-fried tofu showed the highest aroma acceptability and OAV values.
• •OPLS-DA revealed distinct aroma profiles among frying methods.
• •Findings support optimized frying strategies for tofu flavor enhancement.

1. Introduction

Tofu, which originated in China and is now widely consumed worldwide, is a staple plant-based protein source known for its nutritional richness and health benefits. It contains bioactive compounds such as isoflavones, saponins, and phytosterols, which have been associated with cardiovascular protection, hormone regulation, and immune enhancement (Lei et al., 2024; Ma et al., 2024). In addition to its functional attributes, tofu is valued for its versatility and mild flavor, which makes it a widely used ingredient in diverse culinary applications (Guo et al., 2018; Wang et al., 2020). However, the relatively bland taste of raw tofu necessitates thermal processing to enhance its sensory appeal. Among various processing methods, frying is widely applied to enhance the texture and aroma of tofu. However, the underlying mechanism of aroma formation—particularly as elucidated by modern analytical techniques such as GC–MS and GC–IMS—remains insufficiently investigated.

Frying is a widely applied thermal treatment.In this process, oil serves as the heat-transfer medium and promotes complex chemical reactions, including the Maillard reaction and lipid oxidation. These reactions generate desirable aromas, colors, and textures (Wang et al., 2021; Zaghi et al., 2019). Common techniques include deep frying, pan frying, and increasingly popular air frying, each imparting distinct sensory characteristics. Deep frying typically produces a crispy texture and intense aroma due to extensive Maillard reactions. This has been demonstrated in studies on duck jerky and fried pork, in which lipid-derived and Maillard reaction volatiles dominated the flavor profile (Pei et al., 2024). In contrast, air frying utilizes hot air convection, resulting in lower fat content while preserving flavor complexity and nutritional quality (Fikry et al., 2024; Zhou et al., 2022). For instance, air-fried onions retained a higher proportion of health-promoting compounds and exhibited reduced levels of polar toxicants compared to deep-fried samples (Cattivelli et al., 2023). Despite the widespread application of these methods, their specific effects on the volatile compound profiles of tofu remain unclear. In particular, little is known about the aroma-active compounds and their sensory implications.

Aroma is a critical factor determining food quality and consumer preference (Hildebrand et al., 2020). In thermally processed foods, aroma formation arises mainly from Maillard reactions, lipid degradation, and their complex interactions (Starowicz & Zieliński, 2019). Traditionally, aroma profiling relies on sensory evaluation, which—while insightful—is subjective and influenced by individual variability among panelists (Munekata et al., 2023). To overcome these limitations, researchers have increasingly adopted advanced analytical approaches for reliable and quantitative assessment of volatile compounds. Among them, electronic nose (E-nose) systems, GC–MS, and GC–IMS are widely used because of their high sensitivity and reproducibility.(Cai et al., 2024; Gu et al., 2021; Yang et al., 2021; Yi et al., 2025). The E-nose, functioning as a bionic sensor array, enables rapid differentiation of overall odor profiles through pattern recognition algorithms. GC–MS allows accurate identification of volatile compounds but requires complex sample pretreatment. In contrast, GC-IMS offers rapid and highly sensitive analysis with minimal sample preparation and has shown increasing application across food, pharmaceutical, and environmental domains (Cai et al., 2024). Given these advancements, this study integrates E-nose, GC–MS, and GC-IMS to systematically investigate the effects of different frying methods on aroma-active compounds in tofu, thereby addressing the current knowledge gap in this field.

Previous studies on the volatile profiles of fried tofu and other thermally processed soy products have mainly relied on a single analytical technique, such as GC–MS or sensory evaluation. These studies focused only on specific compounds or general aroma characteristics. They did not investigate the mechanisms of aroma formation. While these methods provide valuable qualitative insights, they cannot fully capture the dynamic formation and transformation of volatiles during different frying processes. In contrast, the present study adopts an integrated multi-technique approach, combining sensory evaluation, electronic nose, GC-IMS, and GC–MS analyses, supported by chemometric modeling (OPLS–DA and VIP analysis). This integrated strategy enables cross-validation across datasets.It links chemical composition with sensory perception. Together, these connections provide a systematic and comprehensive understanding of how different frying methods shape tofu flavor.

Frying is widely used to improve the sensory appeal of tofu. However, the effects of different frying methods on its aroma-active compounds have not been fully explored. Few studies have combined advanced analytical techniques to profile volatile organic compounds (VOCs) in tofu processed by deep frying, air frying, and pan frying. Therefore, this study aimed to systematically characterize the VOCs and aroma profiles of tofu prepared using different frying methods. This goal was achieved using a combination of sensory evaluation, electronic nose (E-nose) technology, and instrumental analyses. These included gas chromatography–ion mobility spectrometry (GC-IMS), gas chromatography–mass spectrometry (GC–MS), odor activity value (OAV) calculation, and multivariate statistical analysis. The findings are expected to provide new insights into aroma formation in thermally processed plant-based foods. They may also offer theoretical guidance for flavor design, product optimization, and innovation in tofu-based products.

2. Materials and methods

2.1. Chemicals and reagents

Commercial tofu was purchased from a local supermarket (Yonghui, Chengdu, Sichuan, China). Refined soybean oil was obtained from Jinlongyu, Yihai Kerry Arawana Holdings Co., Ltd. (Shanghai, China). The reference substances acetone (CAS: 67–64-1, 0.1 %), 2-propanol (CAS: 67–63-0, 0.05 %), and 1-propanol (CAS: 71–23-8, 0.1 %) were provided by Alpha MOS (Toulouse, France). The internal standard 2-methyl-3-heptanone (CAS: 13019–20-0, 95 %) was purchased from Aladdin Biochemical Technology Co., Ltd. (Shanghai, China). A series of n-alkanes (C4–C8, chromatographic grade) used for retention index (RI) calibration were obtained from Sinopharm Group Beijing Co., Ltd. (Beijing, China).

2.2. Instruments and equipment

The FOX 4000 electronic nose was provided by Alpha MOS (Toulouse, France). The FlavourSpec® gas chromatography–ion mobility spectrometry (GC-IMS) system was obtained from G.A.S. (Dortmund, Germany). The Clarus 680–Clarus SQ8T gas chromatography–mass spectrometry (GC–MS) system was obtained from PerkinElmer (Waltham, MA, USA). It was equipped with an HST40 trap headspace sampler and an Elite-5MS column (30 m × 0.25 mm × 0.25 μm). The KL60-VF506 air fryer was obtained from Joyoung Co., Ltd. (Hangzhou, China).

2.3. Sample processing

2.3.1. Sample pretreatment

Tofu was cut into uniform slices (4 × 4 × 1 cm) with an average weight of 14.00 ± 0.50 g. All frying treatments were terminated when the moisture loss rate reached 35.50 ± 0.50 % (empirical data), ensuring consistency across samples. Four groups were prepared as follows.

Group A (Control): Tofu was used without any thermal treatment.

Group B (Deep-Fried Tofu): A frying pan (20 cm in diameter) was placed on an induction cooker (1400 W), and 20 g of refined soybean oil were heated to 150 °C. Tofu slices were added and flipped every 30 s until the target moisture loss was achieved. The total processing time was 3 min 30 s.

Group C (Air-Fried Tofu): Tofu slices were brushed with refined soybean oil and placed in an air fryer (1400 W) for an initial 5 min. They were then flipped every 30 s and brushed with oil until the desired moisture loss rate was achieved. The total processing time was 18 min 30 s.

Group D (Pan-Fried Tofu): A small amount of refined soybean oil (1.00 g) was added to a preheated frying pan (20 cm, 1400 W). The tofu slices were flipped every 30 s, and an additional 1.00 g of oil was added each time, until the moisture loss rate reached 35.50 ± 0.50 %. The total processing time was 5 min 30 s.

A total of 70 slices were prepared for each group. Sixty were used for sensory evaluation, and the remaining ten were mashed and stored for instrumental analysis.

2.3.2. Sample preparation

Sample Preparation for E-Nose Analysis: The mashed tofu was thoroughly homogenized. Approximately 2 g of the homogenized sample was accurately weighed and transferred into a 10 mL headspace vial. The vial was sealed tightly and stored for subsequent analysis. Thirty vials were prepared for each sample, corresponding to 30 replicates in the analysis.

Sample Preparation for GC-IMS Analysis: The mashed tofu was thoroughly homogenized. Approximately 2 g of the homogenized sample was accurately weighed and transferred into a 20 mL headspace vial. The vial was sealed tightly and stored for subsequent analysis. Three vials were prepared for each sample, corresponding to three replicates in the analysis.

Sample Preparation for GC–MS Analysis: The mashed tofu was thoroughly homogenized. Approximately 2 g of the homogenized sample was accurately weighed and transferred into a 20 mL headspace vial. Simultaneously, 20 μL of the internal standard solution, 2-methyl-3-heptanone (4.08 × 10⁶ μg/L), was added. The vial was sealed tightly and stored for subsequent analysis. Five vials were prepared for each sample, corresponding to five replicates in the analysis.

2.4. Sensory evaluation

Twenty food engineering students (aged 20–25 years) participated as panelists. All had normal olfactory and gustatory functions. Before testing, each participant completed a one-month training program in sensory evaluation. The tests were conducted in a sensory laboratory following the ISO 8589:2007 standard. Conditions were controlled at 20 ± 2 °C and 65 ± 2 % relative humidity.

Panelists evaluated each tofu sample based on three aroma descriptors: caramelized aroma, fatty aroma, and tofu aroma. They employed a 9-point intensity scale (1 = very weak, 9 = very strong) for this assessment. Additionally, overall preference was measured using a 9-point hedonic scale (1 = extremely dislike, 9 = extremely like).

Samples were randomly assigned three-digit codes and presented in a randomized order. Each sample was evaluated in triplicate by each panelist across three separate sessions. Final scores were calculated as the average of the three replicates for each descriptor and preference score.

2.5. Analysis of the electricity-nose

2.5.1. Calibration of the electronic nose instrument

The electronic nose (E-nose) was first run for 8 h before calibration. Physical calibration was then performed. After that, chemical calibration was carried out using standard solutions of acetone, 2-propanol, and 1-propanol provided by Alpha MOS. Sample analysis began only after both calibrations were successfully completed.

2.5.2. Analytical conditions of E-nose

The overall aroma profile was analyzed using an electronic nose (FOX 4000, Alpha MOS, Toulouse, France). Before detection, each vial was incubated at 60 °C for 5 min. Analytical conditions were set as follows: injection volume, 1000 μL; injection rate, 1000 μL/s; carrier gas flow, 150 mL/min; detection time, 120 s; cleaning interval, 180 s; and data acquisition frequency, 1 Hz. Each sample underwent 30 measurements, and the 10 most stable responses were averaged to obtain the final result.

2.5.3. Radar chart and principal component analysis

The raw sensor data were preprocessed using Origin 2022 (OriginLab Corporation, Northampton, MA, USA). Dimensionality reduction was applied to eliminate redundant information, followed by principal component analysis (PCA) to identify differences among samples. Radar plots were constructed based on the averaged sensor responses and visualized using the same software.

2.6. Analysis of GC-IMS

Volatile organic compounds (VOCs) were analyzed using headspace gas chromatography–ion mobility spectrometry (HS–GC–IMS; FlavourSpec®, G.A.S., Dortmund, Germany). Samples were incubated at 60 °C for 30 min with agitation at 500 rpm. A 500 μL aliquot of headspace gas was then injected using a heated syringe maintained at 65 °C. Separation was performed on an MXT-5 column (15 m × 0.53 mm × 1 μm). High-purity nitrogen (≥99.999 %) was used as both the carrier and drift gas. The carrier gas flow rate was programmed as follows: 2 mL/min for the first 2 min, then linearly increased to 100 mL/min over 28 min, giving a total analysis time of 31 min. The drift tube temperature was maintained at 45 °C, with a constant drift gas flow of 150 mL/min. Retention indices (RIs) were calculated using n-ketones (C4–C8) as external standards, and compounds were identified by comparison with the IMS database. Each sample group was analyzed in triplicate.

2.7. Analysis of GC–MS

2.7.1. Aging process

Helium (99.999 % purity) was used as the carrier gas at a constant flow rate of 1 mL/min.Before analysis, the capillary column (Elite-5MS, 30 m × 0.25 mm × 0.25 μm) was conditioned following the manufacturer's procedure (PerkinElmer).

The column temperature program was set as follows: the initial temperature was 50 °C and maintained for 10 min. It was then increased at 3 °C/min to 300 °C and held for 180 min.

2.7.2. Headspace extraction and injection conditions

Samples were incubated at 60 °C for 30 min.The injection needle and transfer line were maintained at 65 °C and 70 °C, respectively.The headspace sampling sequence consisted of four steps: drying for 120 s, equilibration for 10 s, pressurization and depressurization for 120 s, and trap enrichment for 240 s.The enrichment step was repeated four times to ensure adequate collection of volatile compounds.

2.7.3. Gas chromatography conditions

A capillary column (Elite-5MS, 30 m × 0.25 mm × 0.25 μm) was employed. Helium (>99.999 %) served as the carrier gas at a constant flow rate of 1.0 mL/min.The oven temperature program was as follows: initially held at 40 °C for 1 min, then increased to 250 °C at a rate of 5 °C/min, and finally maintained at 250 °C for 2 min.The total run time was 45 min.

2.7.4. Mass spectrometry conditions

The mass spectrometer operated in electron ionization (EI) mode at 70 eV. The ion source temperature was set to 230 °C, and the interface temperature to 280 °C. The electron multiplier voltage was maintained at 1650 V. Mass spectra were recorded over the range m/z 45–450.

2.7.5. Compound identification and quantification

Volatile compounds were identified by comparing their mass spectra with the NIST 2011 spectral library (match factor > 800) and published reference data. Quantification was carried out using the internal standard method. The calculation was based on the following eq. (1):

$${\mathrm{C}}_{\mathrm{i}}=\frac{{\mathrm{C}}_{\text{IS}}\times {\mathrm{V}}_{\text{IS}}\times {\mathrm{A}}_{\mathrm{i}}}{{\mathrm{m}}_{\mathrm{i}}\times {\mathrm{A}}_{\text{IS}}}$$ (1)

In the formula: C_i - concentration of the compound to be measured, μg/kg;

C_IS - concentration of the internal standard，μg/L;

V_IS - volume of the internal standard，L;

A_i - peak area of the compound to be measured;

m_i - weighing of the sample's mass，kg；.

A_IS - peak area of the internal standard compound.

2.7.6. Odor activity value (OAV) calculation

According to Guadagni's odor value concept, compounds that occur at high concentrations and have low odor thresholds contribute strongly to the characteristic aroma of a sample.The odor activity value (OAV) is defined as the ratio between a compound's concentration and its sensory threshold.Threshold data were mainly obtained from the Handbook of Flavor Ingredients (Burdock, 2016) and supplemented with information from the Compilation of Odor Threshold Values in Air and Water (Van Gemert, 2011).A compound is generally considered aroma-active when its OAV is equal to or greater than 1 (Zhu et al., 2020).The OAV was calculated according to Eq. (2):

$$\mathit{OAV}=\frac{{\mathrm{C}}_i}{T}$$ (2)

In the formula:

C_i — concentration of the volatile compound (μg/kg).

T —odor threshold of the compound (μg/kg).

OAV — odor activity value.

2.8. Data processing

Multivariate statistical analyses were performed to evaluate volatile and sensory differences among tofu samples. Principal component analysis (PCA) and radar plots were generated using Origin 2022 (OriginLab Corporation, Northampton, MA, USA).

The Reporter plug-in in the GC–IMS software (FlavourSpec®, G.A.S., Germany) was used to compare spectral variations, and fingerprint spectra were created using the Gallery Plot plug-in.Quantification of volatile organic compounds (VOCs) was conducted in VOCal software (G.A.S., Dortmund, Germany, version 0.4.03).All data are expressed as mean ± standard deviation (SD).One-way analysis of variance (ANOVA) followed by Tukey's HSD test was used to determine significant differences among groups (α = 0.05).Statistical testing was performed in SPSS 26.0 (IBM Corporation, USA).Orthogonal partial least squares discriminant analysis (OPLS–DA) was carried out in SIMCA 18.1 (Umetrics, Umeå, Sweden).

3. Results and discussion

3.1. Sensory analysis

Table 1 summarizes the results of the sensory evaluation of tofu prepared by different frying methods. Significant differences (p < 0.05) were observed among the four groups in both caramelized and fatty aromas. The unprocessed control group (A) showed the lowest scores for these two attributes, indicating limited aroma development due to the absence of heating. In contrast, the deep-fried (B), air-fried (C), and pan-fried (D) tofu samples exhibited significantly stronger fatty aromas. Pan-fried tofu (D) achieved the highest scores for both caramelized and fatty aromas. This result is likely related to the direct contact between the tofu surface and the hot pan, where high temperatures promoted Maillard reactions and caramelization. Regarding tofu aroma—the intrinsic beany note—the control group (A) scored highest, followed by the air-fried (C), pan-fried (D), and deep-fried (B) samples. This indicates that, although the intensity of roasted aromas increased after frying, some panelists preferred the clean and mild aroma of raw tofu. Overall, the fried tofu groups (B, C, and D) received higher overall acceptance scores than the control group (A), with pan-fried tofu (D) rated the most acceptable. Its superior sensory appeal was attributed to its pleasant caramelized and fatty notes, which enhanced the overall flavor perception.

Table 1.

Sensory evaluation of deep-fried, air-fried and pan-fried tofu.

Samples	Roast aroma	Fat aroma	Tofu aroma	Overall receptibility
A	1.00 ± 0.00d	1.00 ± 0.00d	7.50 ± 0.97a	5.30 ± 0.56c
B	4.55 ± 0.67c	5.85 ± 0.58c	5.20 ± 0.51c	5.75 ± 0.90b
C	5.80 ± 0.51b	6.75 ± 0.70b	5.90 ± 0.44b	5.95 ± 0.74b
D	8.30 ± 0.46a	8.05 ± 0.67a	5.80 ± 0.51b	7.60 ± 0.80a

Note: Data are presented as mean ± standard deviation (n = 20). Differences among groups were analyzed by one-way ANOVA followed by Tukey's honestly significant difference (HSD) test at α = 0.05. Values with the same lowercase letters in the same row are not significantly different (p > 0.05). while different letters indicate significant differences (p < 0.05).; Sample codes: A (Control), B (Deep-Fried Tofu), C (Air-Fried Tofu), D (Pan-Fried Tofu).

3.2. Analysis of E-nose

Fig. 1 presents the electronic nose (E-nose) analysis of tofu samples. Radar chart analysis in Fig. 1(A) shows clear distinctions between raw and processed tofu, especially between the unprocessed sample (A) and the fried samples. Significant variations were detected by the TA/2, T40/1, and T40/2 sensors (p < 0.05), indicating differences in the levels of oxidative volatiles generated under different frying conditions. These sensors are particularly sensitive to alcohols and sulfides (He et al., 2025), which are mainly formed through lipid oxidation and Maillard reactions during frying(Zhang et al., 2021). Notably, the volatile profiles of pan-fried and deep-fried tofu were similar, whereas air-fried tofu exhibited more distinct responses. This finding highlights the influence of different heat transfer media and cooking techniques on aroma development(Wang et al., 2023).

Fig. 1.

Analysis of E-nose after different frying methods. Note: A (Control), B (Deep-Fried Tofu), C (Air-Fried Tofu), D (Pan-Fried Tofu).

Fig. 1(B) presents the principal component analysis (PCA) results for tofu samples based on the electronic nose data. The PCA plot clearly shows sample clustering and separation. PCA is a dimensionality reduction technique that retains maximum variance from the original variables to assess differences between samples (Sharma et al., 2023). In this study, Principal Component 1 (PC1) and Principal Component 2 (PC2) explained 85.6 % and 12.2 % of the total variance, respectively, giving a cumulative contribution of 97.8 %.This high cumulative variance indicates that PCA effectively captured the key information from the E-nose data. Typically, a cumulative variance above 80 % is considered sufficient to represent the major characteristics of the dataset(Qiao et al., 2025).

The PCA plot shows that the four samples (A, B, C, and D) are clearly separated, with no overlap within the 95 % confidence interval (p < 0.05). This separation indicates significant differences in volatile profiles among the samples. Sample A is located in the fourth quadrant, whereas sample D appears in the third quadrant on opposite sides of the X-axis, indicating a marked difference between them. In contrast, samples B and C are positioned closer together along the Y-axis of the plot, suggesting smaller variations in their volatile compositions. The PCA results show that the volatile profiles of deep-fried tofu (B) and air-fried tofu (C) are more similar to each other than to pan-fried tofu (D). This similarity arises because both deep-frying and air-frying expose tofu to uniform high-temperature environments, either through hot oil or circulating hot air. These conditions promote lipid oxidation and Maillard reactions, generating comparable volatile compounds. In contrast, pan-fried tofu is heated mainly by direct surface contact, which leads to uneven heat distribution and a distinct aroma profile. The lower heat transfer efficiency during pan-frying further limits volatile formation and alters overall flavor development.

3.3. Analysis of GC-IMS

To further examine how processing methods affect tofu volatiles, the samples were analyzed using gas chromatography–ion mobility spectrometry (GC-IMS). A two-dimensional topographic plot was generated, where each point represents a distinct volatile compound. As shown in Fig. 2(A), most volatiles in samples B, C, and D had retention times within the range of 200–1200 s. A smaller number of compounds appeared around 1600 s. This pattern likely reflects the lower polarity of these substances, which leads to longer retention on non-polar columns compared with more polar compounds (Chen et al., 2018).

Fig. 2.

Analysis of HS-GC-IMS after different reheating methods. Note: A (Control), B (Deep-Fried Tofu), C (Air-Fried Tofu), D (Pan-Fried Tofu). (A). The three-dimensional spectrum of volatile compounds in A, B, C, D; (B). Composition spectrum (top view) and difference spectrum of volatile compounds in (a) 2D qualitative chromatogram of A, (b) 2D qualitative chromatogram of B, (c) 2D qualitative chromatogram of C, (d) 2D qualitative chromatogram of D (e) reference chromatogram of A, (f) Deducition chromatogram of B (g) Deducition chromatogram of C, (h) Deducition chromatogram of D. (C). Fingerprint spectrum of samples gallery plot.

Fig. 2(B) shows the GC-IMS difference comparison map. The control group (A) was used as the reference, and its spectrum was compared with those of samples B, C, and D. Areas in white indicate similar volatile concentrations. Blue regions represent lower concentrations, while red regions indicate higher concentrations compared with the control. Darker colors correspond to greater differences. Between 200 and 400 s, all fried samples (B—D) showed stronger signals than the control, as seen by the darker red zones. This finding suggests that frying promotes the generation of volatile compounds during this stage. From 400 to 600 s, air-fried and pan-fried tofu (C and D) exhibited higher volatile levels, whereas deep-fried tofu (B) showed fewer compounds, possibly due to shorter heating time or more intense Maillard reactions. After 600 s, air-fried tofu continued to release volatiles, while the profiles of deep-fried and pan-fried tofu became similar, indicating stabilization in volatile formation. No significant differences were observed beyond 1000 s.

Fig. 2(C) presents the GC-IMS fingerprint chromatograms of control (A), deep-fried (B), air-fried (C), and pan-fried (D) tofu. These chromatograms provide an intuitive visualization of the volatile composition and highlight differences among processing methods. A total of 115 volatile compounds were detected, of which 18 were unidentified. Based on compound distribution, the chromatogram was divided into five regions (A–E). Regions A and C represent volatiles common to all samples, whereas regions B, D, and E correspond to the characteristic peaks of samples A, C, and D, respectively. Sample A contained 37 characteristic volatile compounds (9 unidentified). Sample B exhibited fewer characteristic peaks, with cumin alcohol as its only marker compound. Sample C included 15 characteristic volatiles (2 unidentified), mainly aldehydes and esters such as pentanal, methyl acetate, 2-methylpropanal, propanal, isobutyl propanoate, ethyl butyrate, thiazole, and methional. Sample D comprised 6 characteristic volatiles (1 unidentified), including (−)-β-pinene, 5-methylfurfural, hexyl butanoate, 4-methylbenzaldehyde, and heptanal. Overall, the results demonstrate that different frying methods significantly alter the volatile profiles of tofu, with air-fried and pan-fried samples showing the most distinctive compound distributions. GC-IMS analysis showed that different cooking methods strongly influenced the volatile profiles of tofu. In air-fried tofu (C), several aldehydes and esters, such as pentanal and ethyl butyrate, were identified. These compounds indicate that lipid oxidation and esterification are key reactions during air-frying. They are derived from the thermal degradation of fatty acids and contribute to the mild fruity and oily aroma of tofu. In contrast, pan-fried tofu (D) contained 5-methylfurfural, a typical product of the Maillard reaction. This compound forms through interactions between amino acids and reducing sugars under high temperatures. The Maillard reaction produces furfural derivatives that create roasted and caramel-like aromas, characteristic of pan-fried tofu. The differences between samples C and D therefore reflect distinct dominant reaction pathways—lipid oxidation in air-frying and Maillard browning in pan-frying—which ultimately lead to divergent flavor profiles (zhang et al., 2021). The four tofu samples showed distinct volatile profiles. Sample A contained the largest number of characteristic volatiles, followed by samples C and D, while sample B had the fewest. These variations in characteristic volatiles correspond to notable differences in odor profiles, suggesting that processing methods strongly alter the aroma characteristics of tofu. Overall, GC-IMS analysis confirms that deep-frying, air-frying, and pan-frying markedly modify the volatile composition and thus the sensory perception of tofu.

3.4. Analysis of GC–MS

GC-IMS highlighted differences in the overall volatile profiles and fingerprint patterns among the tofu samples. To further clarify the volatile characteristics and their sensory contributions, the samples were analyzed using gas chromatography–mass spectrometry (GC–MS). Quantification was conducted with an internal standard method, and odor activity values (OAVs) were calculated to evaluate aroma relevance. Table 2 lists all identified compounds, and Table 3 summarizes their relative proportions and classifications in fried tofu samples.

Table 2.

Qualitative and Quantitative Analysis of Fried Tofu by GC–MS and OAV Analysis.

Category	N0.	CAS	Compounds	The content (μg/kg)				Threshold*	OAV				Sensory description*
				A	B	C	D	(μg/kg)	A	B	C	D
Hydrocarbons	1	124–18-5	decane	38.49 ± 3.16a	ND	ND	ND	3600	<0.1	ND	ND	ND	alkane
	2	1120-21-4	n-hendecane	ND	8.74 ± 0.72a	ND	ND	5600	ND	<0.1	ND	ND	alkane
	3	111–65-9	octane	ND	ND	ND	590.96 ± 82.89a	8000	ND	ND	ND	<0.1	alkane
	4	111–84-2	nonane	17.98 ± 1.45b	264.90 ± 29.93a	ND	ND	12,000	<0.1	<0.1	ND	ND	alkane
	5	1840-42-2	fluoro(trinitro)methane	256.78 ± 19.94b	2996.44 ± 562.56a	ND	ND	ND	ND	ND	ND	ND
	6	142–82-5	heptane	ND	ND	ND	592.89 ± 51.43a	ND	ND	ND	ND	ND	alkane
	7	625–44-5	isobutyl	ND	ND	138.82 ± 9.00a	ND	ND	ND	ND	ND	ND
	8	503–09-3	epifluorohydrin	ND	ND	2520.81 ± 127.05b	4093.59 ± 706.55a	ND	ND	ND	ND	ND
	9	5076-19-7	2,2,3-trimethyl-oxirane	ND	1009.41 ± 107.16b	1757.73 ± 423.12a	ND	ND	ND	ND	ND	ND
	10	2415-72-7	propylcyclopropane	ND	ND	1062.76 ± 156.17a	ND	ND	ND	ND	ND	ND
	11	2213-23-2	2,4-dimethylheptane	ND	15.07 ± 0.63a	ND	ND	ND	ND	ND	ND	ND
	12	62,108–27-4	2,4,6-trimethyl-decane	ND	ND	ND	93.97 ± 7.38a	ND	ND	ND	ND	ND
	13	4312–76-9	1-hydroperoxyhexane	ND	ND	459.52 ± 34.90a	ND	ND	ND	ND	ND	ND
	14	62,108–23-0	2,5,6-trimethyldecane	34.89 ± 2.95a	ND	ND	ND	ND	ND	ND	ND	ND
	15	17,301–23-4	2,6-dimethylundecane	47.88 ± 8.43a	ND	ND	ND	ND	ND	ND	ND	ND
	16	15,869–93-9	3,5-dimethyloctane	42.20 ± 3.88a	ND	ND	ND	ND	ND	ND	ND	ND
	17	17,312–82-2	4,6-dimethylundecane	36.93 ± 3.69a	ND	ND	ND	ND	ND	ND	ND	ND
	18	535–77-3	m-cymene	ND	13.12 ± 3.28b	21.82 ± 3.57a	ND	800	ND	<0.1	<0.1	ND	fresh scent of citrus and woody notes
	19	17,603–76-8	2-methyl-1-octen-3-yne	ND	ND	ND	170.99 ± 71.70a	ND	ND	ND	ND	ND
	20	5989–27-5	d-limonene	ND	190.94 ± 37.46c	251.27 ± 15.17b	297.59 ± 26.05a	45	ND	4.24	5.58	6.61	Citrus flavor
	21	80–56-8	alpha-pinene	ND	10.98 ± 1.49b	11.05 ± 1.11a	ND	100	ND	0.11	0.11	ND	pine, turpentine
	22	100–42-5	styrene	ND	135.31 ± 43.72b	220.00 ± 10.48a	ND	150	ND	0.9	1.47	ND	balsamic, gasoline
	23	586–62-9	?alpha?-terpinolene	ND	7.44 ± 1.04b	ND	48.30 ± 7.27a	200	ND	<0.1	ND	0.24	pine, plastic
	24	99–83-2	alpha-phellandrene	ND	17.20 ± 4.78a	ND	ND	2900	ND	<0.1	ND	ND	turpentine, mint, spice
	25	821–95-4	1-undecene	ND	19.30 ± 10.02a	ND	ND	570,000	ND	<0.1	ND	ND
	26	52,097–85-5	2-propenylidene-cyclobutene	32.79 ± 6.30c	95.32 ± 7.91b	152.21 ± 25.63a	ND	ND	ND	ND	ND	ND
	27	544–25-2	cyclohepta-1,3,5-triene	ND	ND	ND	132.67 ± 17.41a	ND	ND	ND	ND	ND
	28	7642–04-8	cis-2-octene	ND	ND	ND	79.86 ± 11.10a	ND	ND	ND	ND	ND
	29	3404-79-3	2-hexene, 3,5-dimethyl-	ND	ND	ND	86.07 ± 3.18a	ND	ND	ND	ND	ND
	30	5208-49-1	(1r,4r,6 s)-4,7,7-trimethylbicyclo[4.1.0]hept-2-ene	ND	ND	18.32 ± 2.11a	ND	ND	ND	ND	ND	ND
	31	508–32-7	tricyclene	ND	ND	19.16 ± 1.45a	ND	ND	ND	ND	ND	ND
	32	55,976–13-1	(4e)-1,4-undecadiene	ND	ND	ND	39.34 ± 2.45a	ND	ND	ND	ND	ND
	33	61,142–36-7	3-ethyl-2-methyl-1,3-hexadiene	15.09 ± 1.31a	ND	ND	ND	ND	ND	ND	ND	ND
Esters	1	7452–79-1	ethyl 2-methylbutyrate	ND	ND	ND	223.71 ± 80.07a	0.06	ND	ND	ND	3728.5	apple
	2	108–64-5	ethyl 3-methylbutanoate	ND	ND	ND	1221.32 ± 206.67a	0.069	ND	ND	ND	17,700.29	fruit
	3	97–62-1	ethyl 2-methylpropanoate	ND	ND	ND	315.60 ± 160.76a	0.11	ND	ND	ND	2869.09	sweet, rubber
	4	106–32-1	ethyl octanoate	ND	ND	ND	33.84 ± 9.20a	2	ND	ND	ND	16.92	fruit, fat
	5	547–63-7	methyl isobutyrate	ND	ND	ND	4179.49 ± 274.41a	8	ND	ND	ND	522.44	fragrance of apples and pineapples.
	6	556–24-1	methyl 3-methylbutanoate	ND	ND	ND	10,652.61 ± 6222.43a	11	ND	ND	ND	968.42	apple
	7	623–42-7	methyl butanoate	ND	ND	ND	828.93 ± 157.37a	30	ND	ND	ND	27.63	ether, fruit, sweet
	8	115–95-7	linalyl acetate	ND	29.50 ± 1.35a	ND	ND	40	ND	0.74	ND	ND	sweet, fruit
	9	97–63-2	ethyl methacrylate	ND	ND	ND	120.73 ± 3.60a	50	ND	ND	ND	2.41
	10	106–70-7	methyl hexanoate	ND	ND	ND	743.37 ± 166.73a	70	ND	ND	ND	10.62	fruit, fresh, sweet
	11	6622-76-0	methyl tiglate	ND	ND	ND	1203.38 ± 136.24a	130	ND	ND	ND	9.26
	12	’111–11-5	methyl octanoate	ND	ND	ND	191.42 ± 58.53a	200	ND	ND	ND	0.96
	13	80–62-6	methyl methacrylate	ND	ND	ND	381.50 ± 66.61a	860	ND	ND	ND	0.44
	14	79–20-9	methyl acetate	ND	3230.36 ± 480.33b	4184.60 ± 797.17b	12,884.15 ± 3972.50a	5000	ND	0.65	0.84	2.58	Aromatic fragrance
	15	105–37-3	ethyl propanoate	ND	61.85 ± 6.08a	ND	ND	15,100	ND	<0.1	ND	ND	fruit
	16	107–31-3	methyl formate	153.36 ± 6.06b	699.30 ± 511.18b	3557.48 ± 1136.40a	246.33 ± 18.37b	325,000	<0.1	<0.1	<0.1	<0.1
	17	16,714–85-5	pyrulic acid methyl ester	143.81 ± 10.67b	ND	ND	3218.30 ± 415.96a	ND	ND	ND	ND	ND
	18	623–43-8	trans-methyl crotonate(e)	ND	ND	ND	179.83 ± 13.38a	ND	ND	ND	ND	ND
	19	3938–96-3	ethyl methoxyacetate	ND	ND	73.62 ± 6.73a	ND	ND	ND	ND	ND	ND
	20	2155-30-8	methyl (±)-lactate	ND	ND	113.67 ± 11.20a	ND	ND	ND	ND	ND	ND
	21	924–50-5	methyl 3-methyl-2-butenoate	ND	ND	ND	172.86 ± 38.37a	ND	ND	ND	ND	ND
	22	598–58-3	methyl nitrate	ND	ND	ND	1295.41 ± 66.08a	ND	ND	ND	ND	ND
	23	6290-49-9	methyl methoxyacetate	ND	1272.79 ± 116.14a	ND	ND	ND	ND	ND	ND	ND
	24	20,600–96-8	diglycerol tetranitrate	ND	ND	ND	14.41 ± 0.77a	ND	ND	ND	ND	ND
	25	141–78-6	ethyl acetate	ND	3069.20 ± 381.30a	ND	1756.50 ± 109.31b	ND	ND	ND	ND	ND	pineapple
	26	142–92-7	hexyl acetate	ND	28.95 ± 3.94b	68.77 ± 9.07a	ND	ND	ND	ND	ND	ND
	27	1732–00-9	4-octenoic acid methyl ester	ND	ND	ND	66.27 ± 18.98a	ND	ND	ND	ND	ND
	28	101–41-7	methyl benzeneacetate	ND	ND	ND	25.33 ± 1.95a	ND	ND	ND	ND	ND
Ketones	1	431–03-8	2,3-butanedione	ND	ND	1011.16 ± 95.68a	ND	0.18	ND	ND	5617.56	ND	butter
	2	105–42-0	4-methyl-2-hexanone	15.59 ± 1.74a	ND	ND	ND	0.81	19.25	ND	ND	ND	Spicy and fruity scents
	3	502–56-7	5-nonanone	45.04 ± 8.37a	ND	41.98 ± 6.83b	ND	8.2	5.49	ND	5.12	ND	Fatty smell, soapy smell
	4	626–33-5	2-methylheptan-4-one	ND	236.04 ± 13.88a	ND	ND	24	ND	9.84	ND	ND
	5	623–56-3	ethyl isobutyl ketone	ND	33.69 ± 2.83b	43.67 ± 3.87a	ND	40	ND	0.84	1.09	ND
	6	589–63-9	4-octanone	ND	13.88 ± 1.45a	ND	ND	41	ND	0.34	ND	ND
	7	565–80-0	2,4-dimethyl-3-pentanone	58.43 ± 14.84a	37.67 ± 2.34c	ND	54.84 ± 7.76b	80	0.73	0.47	ND	0.69
	8	106–35-4	3-heptanone	57.59 ± 8.95a	ND	ND	34.10 ± 2.31b	80	0.72	ND	ND	0.43	aroma of oil and fat
	9	4643-27-0	2-octen-4-one	10.38 ± 0.85a	ND	ND	ND	200	<0.1	ND	ND	ND
	10	107–87-9	2-pentanone	ND	ND	86.40 ± 13.73a	ND	350	ND	ND	0.25	ND	ether, fruit
	11	18,641–70-8	2,4-dimethyl-3-hexanone	ND	ND	33.08 ± 2.77a	ND	600	ND	ND	<0.1	ND
	12	1888-57-9	3-hexanone,2,5-dimethyl-	685.58 ± 161.77a	25.84 ± 2.53c	294.66 ± 123.39b	97.58 ± 6.84c	ND	ND	ND	ND	ND
	13	5675-22-9	7-(2-methylprop-2-enoxy)-4-phenyl-chromen-2-one	ND	ND	ND	1993.79 ± 855.66a	ND	ND	ND	ND	ND
	14	6137-11-7	4-methyl-3-heptanone	50.76 ± 11.39a	27.27 ± 2.69c	25.76 ± 1.91c	48.78 ± 8.27b	ND	ND	ND	ND	ND
	15	7307-03-1	isobutyrylacetone	16,863.26 ± 5885.27a	ND	ND	ND	ND	ND	ND	ND	ND
	16	18,641–71-9	2,4-dimethylheptan-3-one	354.35 ± 74.91a	ND	ND	ND	ND	ND	ND	ND	ND
	17	7492-38-8	2-methyloctan-4-one	ND	35.59 ± 6.06a	ND	32.16 ± 1.30b	ND	ND	ND	ND	ND
	18	513–86-0	acetoin	ND	ND	49.20 ± 7.26a	ND	14	ND	ND	3.51	ND	butter, cream
Aldehydes	1	2463-53-8	2-nonenal	9.36 ± 0.90a	ND	ND	ND	0.1	93.6	ND	ND	ND	paper
	2	124–13-0	octanal	26.37 ± 0.89a	ND	ND	ND	0.88	29.97	ND	ND	ND	fat, soap, lemon, green
	3	124–19-6	nonanal	93.72 ± 5.17a	11.97 ± 1.94b	ND	ND	3.1	30.23	3.86	ND	ND	fat, citrus, green
	4	505–57-7	2-hexenal	47.81 ± 4.84a	ND	ND	ND	30	1.59	ND	ND	ND	fat, rancid
	5	57,266–86-1	(z)-2-heptenal	49.56 ± 6.56a	ND	ND	ND	56	0.89	ND	ND	ND	Vegetable fragrance, flaxseed oil
	6	100–52-7	benzaldehyde	ND	ND	41.26 ± 7.05a	ND	85	ND	ND	0.49	ND	almond, burnt sugar
	7	66–25-1	hexanal	2577.69 ± 439.71a	127.82 ± 47.15b	291.74 ± 50.76b	ND	230	11.21	0.56	1.27	ND	grass, tallow, fat
	8	2548-87-0	2-octenal (e)	39.42 ± 3.54a	ND	ND	ND	250	0.16	ND	ND	ND	green, nut, fat
	9	111–71-7	heptanal	61.83 ± 11.50a	ND	ND	ND	260	0.24	ND	ND	ND	fat, citrus, rancid
	10	1576–87-0	trans-2-pentenal	52.07 ± 6.72a	ND	ND	ND	2700	<0.1	ND	ND	ND	strawberry, fruit, tomato
	11	15,798–64-8	(e)-but-2-enal	18.12 ± 0.51a	ND	ND	ND	ND	ND	ND	ND	ND
	12	10,312–83-1	methoxyacetaldehyde	ND	1436.02 ± 151.31a	ND	ND	ND	ND	ND	ND	ND
	13	931–96-4	1-methyl-3-cyclohexene-1-carboxaldehyde	27.59 ± 3.37a	ND	ND	ND	ND	ND	ND	ND	ND
Alcohols	1	123–51-3	3-methylbutan-1-ol	ND	59.09 ± 1.59b	136.48 ± 14.60a	ND	6.1	ND	9.69	22.37	ND	whiskey, malt, burnt
	2	111–27-3	n-hexanol	469.31 ± 58.54b	589.02 ± 158.27a	ND	ND	34	13.8	17.32	ND	ND	resin, flower, green
	3	589–98-0	3-octanol	ND	ND	ND	241.52 ± 8.95a	78	ND	ND	ND	3.1
	4	1576–96-1	trans-2-penten-1-ol	135.99 ± 16.84a	ND	ND	ND	89.2	1.52	ND	ND	ND	mushroom
	5	18,720–62-2	2-methylheptan-3-ol	ND	26.67 ± 2.22a	ND	ND	330	ND	<0.1	ND	ND
	6	71–41-0	pentanol	62.69 ± 3.67b	ND	ND	763.76 ± 158.79a	360	0.17	ND	ND	2.12	balsamic
	7	19,550–07-3	2,5-dimethyl-3-hexanol	ND	ND	ND	220.99 ± 64.41a	400	ND	ND	ND	0.55
	8	623–55-2	5-methyl-3-hexanol	ND	27.68 ± 3.98a	ND	ND	410	ND	<0.1	ND	ND
	9	627–59-8	5-methyl-2-hexanol	ND	ND	221.06 ± 29.53a	ND	650	ND		0.34	ND
	10	504–63-2	1,3-propanediol	16.09 ± 2.03a	ND	ND	ND	ND	ND	ND	ND	ND
	11	7731-29-5	trans-4-methylcyclohexanol	657.72 ± 331.78a	ND	104.21 ± 6.02b	ND	ND	ND	ND	ND	ND
	12	10,325–41-4	(2r,3 s)-3-chlorobutan-2-ol	ND	ND	58.28 ± 1.94a	ND	ND	ND	ND	ND	ND
	13	6982–25-8	(±)-2,3-butanediol	ND	ND	95.92 ± 10.50b	611.50 ± 85.51a	ND	ND	ND	ND	ND
	14	67–63-0	2-propanol	ND	ND	288.46 ± 67.32b	1191.71 ± 289.08a	ND	ND	ND	ND	ND
	15	4254-14-2	(r) - 1,2-propanediol	ND	ND	146.01 ± 7.00b	4366.11 ± 768.86a	ND	ND	ND	ND	ND
	16	5454-79-5	cis-3-methylcyclohexanol	ND	96.80 ± 43.10a	ND	ND	ND	ND	ND	ND	ND
	17	69,668–90-2	(z)-3-nonen-2-ol	ND	ND	ND	403.03 ± 23.17a	ND	ND	ND	ND	ND
	18	26,001–58-1	(2z)-2-octen-1-ol	219.13 ± 14.58c	375.41 ± 110.52b	514.17 ± 58.39a	95.79 ± 52.10c	ND	ND	ND	ND	ND
	19	470–82-6	1,8-cineole	ND	28.61 ± 5.57b	ND	33.89 ± 3.28a	26.7	ND	1.07	ND	1.27	mint, sweet
Phenol	1	104–46-1	anethol	8.19 ± 1.09c	15.69 ± 2.66b	ND	20.64 ± 2.88a	57	0.14	0.28	ND	0.36
	2	538–93-2	isobutylbenzene	35.29 ± 6.56a	18.63 ± 1.95b	17.78 ± 0.27c	ND	0.8	44.11	23.29	22.23	ND
	3	104–51-8	butylbenzene	ND	11.71 ± 4.10c	16.38 ± 1.64b	18.60 ± 4.23a	14	ND	0.84	1.17	1.33
	4	103–65-1	n-propylbenzene	ND	81.37 ± 2.77a	79.65 ± 17.87b	60.57 ± 4.94c	19	ND	4.28	4.19	3.19
	5	100–41-4	ethylbenzene	ND	55.16 ± 15.82b	213.70 ± 116.07a	89.72 ± 19.98b	26	ND	2.12	8.22	3.45	Vegetable flavor, nut flavor
	6	620–14-4	3-ethyltoluene	203.71 ± 27.55a	11.91 ± 1.23b	ND	ND	88	2.31	0.14	ND	ND
	7	108–38-3	m-xylene	ND	34.08 ± 9.44b	40.47 ± 3.89a	31.71 ± 3.92c	180	ND	0.19	0.22	0.18	plastic
	8	106–42-3	p-xylene	4.83 ± 0.64a	ND	ND	ND	250	<0.1	ND	ND	ND	Plastic, green, pungent smell
	9	95–47-6	o-xylene	ND	ND	285.74 ± 34.16a	ND	250	ND	ND	1.14	ND	Geranium scent, fragrant scent
	10	611–14-3	2-ethyltoluene	ND	27.10 ± 0.93a	ND	ND	360	ND	<0.1	ND	ND
	11	95–63-6	methyl-p-xylene	ND	50.73 ± 5.70a	ND	49.75 ± 3.54b	590	ND	<0.1	ND	<0.1	plastic
	12	108–67-8	mesitylene	48.95 ± 21.66a	ND	42.60 ± 2.36b	ND	830	<0.1	ND	<0.1	ND
	13	538–68-1	n-amylbenzene	ND	5.97 ± 1.86a	ND	ND	6000	ND	<0.1	ND	ND
	14	526–73-8	1,2,3-trimethylbenzene	213.11 ± 34.51a	ND	ND	ND	ND	ND	ND	ND	ND
	15	74,663–91-5	2,3,4,5,6-pentapropylphenol	ND	ND	ND	526.43 ± 57.19a	ND	ND	ND	ND	ND
	16	106,827–59-2	benzenepropanamide-4-hydroxy-n-(2-phenylethyl)	ND	ND	ND	349.81 ± 15.59a	ND	ND	ND	ND	ND
Amines	1	124–40-3	dimethylamine	ND	ND	166.87 ± 16.97b	1310.22 ± 103.00a	7750	ND	ND	<0.1	0.17
	2	594–39-8	tert-amylamine	824.50 ± 70.58a	ND	ND	ND	ND	ND	ND	ND	ND
	3	687–51-4	h-leu-nh2	ND	ND	3271.83 ± 192.90a	ND	ND	ND	ND	ND	ND
	4	75–12-7	formamide	ND	ND	ND	1680.03 ± 171.22a	ND	ND	ND	ND	ND
	5	42,185–03-5	2-propoxyethanamine	ND	ND	ND	1517.78 ± 1192.54a	ND	ND	ND	ND	ND
	6	10,465–78-8	n,n,n’,n’-tetramethylazodicarboxamide	76.53 ± 8.45a	ND	ND	ND	ND	ND	ND	ND	ND
	7	471–46-5	oxamide	ND	ND	299.38 ± 44.95a	ND	ND	ND	ND	ND	ND
Furan	1	3777-69-3	2-pentylfuran	245.56 ± 12.49d	817.32 ± 286.35b	673.05 ± 33.92c	884.33 ± 108.47a	270	0.91	3.03	2.49	3.28	green bean, butter
Acids	1	64–18-6	formic acid	ND	ND	1543.84 ± 176.68b	4085.25 ± 456.84a	980	ND	ND	1.58	4.17
	2	99–16-1	allantoic acid	ND	3397.19 ± 597.52a	ND	ND	ND	ND	ND	ND	ND
	3	79–33-4	l-lactic acid	ND	ND	84.79 ± 14.02a	ND	ND	ND	ND	ND	ND
	4	4433-63-0	ethylboronic acid	ND	ND	127.75 ± 26.07a	ND	ND	ND	ND	ND	ND
	5	595–98-2	2-hydroxy-2-methyl-propanedioic acid	ND	851.57 ± 72.29b	ND	1464.76 ± 174.04a	ND	ND	ND	ND	ND
	6	144–62-7	oxalic acid	ND	ND	ND	3372.33 ± 100.07a	ND	ND	ND	ND	ND
Ethers	1	115–10-6	dimethyl ether	ND	512.66 ± 37.32b	50.21 ± 3.84b	4358.33 ± 721.25a	303,967	ND	<0.1	<0.1	<0.1	A light and delicate fragrance.
	2	52,195–40-1	(z)-1-methylsulfanylprop-1-ene	ND	ND	316.62 ± 15.19a	ND	ND	ND	ND	ND	ND
	3	628–28-4	n-butyl methyl ether	ND	ND	ND	208.97 ± 23.33a	ND	ND	ND	ND	ND
Others	1	60–34-4	methylhydrazine	ND	7.73 ± 0.16a	ND	ND	1900	ND	<0.1	ND	ND
	2	20,240–70-4	1-allyl-1-methylhydrazine	ND	ND	1922.04 ± 166.54a	ND	ND	ND	ND	ND	ND
	3	42,504–87-0	azaniumyl-(2-methylpropyl)azanium	ND	ND	100.54 ± 24.86a	ND	ND	ND	ND	ND	ND
	4	3811-25-4	clorprenaline	ND	ND	ND	529.89 ± 43.07a	ND	ND	ND	ND	ND
	5	38,870–89-2	methoxyacetyl chloride	ND	ND	208.34 ± 29.57a	ND	ND	ND	ND	ND	ND
	6	13,360–65-1	3-ethyl-2,5-dimethylpyrazine	ND	ND	ND	71.15 ± 16.07a	20	ND	ND	ND	3.56	potato, roast
	7	123–32-0	2,5-dimethylpyrazine	ND	ND	ND	27.39 ± 5.50a	170	ND	ND	ND	0.16	Roasted potato flavor
	8	13,925–03-6	2-ethyl-6-methyl-pyrazine	ND	ND	ND	303.50 ± 8.21a	ND	ND	ND	ND	ND
	9	17,699–16-0	trans-sabinene hydrate, trans-4-thujanol	ND	ND	ND	11.97 ± 0.45a	9200	ND	ND	ND	<0.1	wood, balsamic

Note: ‘ND’ indicates not detected or missing data. Data are presented as mean ± standard deviation (n = 3). Differences among groups were analyzed by one-way ANOVA followed by Tukey's honestly significant difference (HSD) test at α = 0.05. Values with the same lowercase letters in the same row are not significantly different (p > 0.05). while different letters indicate significant differences (p < 0.05). The unit of content is μg/kg. threshold values are from Van Gemert (2018). *Sensory descriptions were characterized based on Burdock (2010) and the Flavornet database (accessed via http://www.flavornet.org/ on 2025-05-17). Sample codes: A (Control), B (Deep-Fried Tofu), C (Air-Fried Tofu), D (Pan-Fried Tofu).

Table 3.

Relative content and types of volatile substances in fried tofu.

category	A		B		C		D
	relative content (%)	type	relative content (%)	type	relative content (%)	type	relative content (%)	type
hydrocarbons	2.08	9	21.49	13	24.05	12	8.05	11
esters	1.18	2	37.69	7	29	5	51.66	22
ketones	72.02	9	1.84	7	5.75	8	2.92	6
aldehydes	11.93	11	7.08	3	1.21	2	0	0
alcohols	6.2	6	5.4	7	5.67	8	10.26	10
phenol	2.04	6	1.4	10	2.53	7	1.48	8
furan	0.97	1	3.68	1	2.44	1	1.14	1
amines	3.58	2	0	0	13.56	3	5.83	3
acids	0	0	19.08	2	6.37	3	11.53	3
ethers	0	0	2.31	1	1.33	2	5.91	2
others	0	0	0.03	1	8.09	3	1.22	4
total	100	46	100	52	100	54	100	70

Note: A (Control), B (Deep-Fried Tofu), C (Air-Fried Tofu), D (Pan-Fried Tofu).

GC–MS analysis identified 153 volatile compounds in total across all tofu samples. The number of detected volatiles varied by processing method: pan-fried tofu (D) contained the most (70), followed by air-fried tofu (C, 54), deep-fried tofu (B, 52), and the control group (A, 46). As shown in Table 3, frying significantly modified the volatile profiles of tofu. The control group (A) was dominated by ketones (72.03 %) and aldehydes (11.93 %), whereas all fried samples contained markedly higher proportions of esters (over 25 %).This result indicates that esterification reactions were likely enhanced during thermal processing, contributing to the formation of fruity and sweet aroma notes.

The compositional differences among samples are mainly associated with the distinct heat transfer mechanisms and heating durations of each frying method (Cao et al., 2020). In deep-frying, tofu is completely immersed in hot oil. This ensures rapid and uniform heat transfer, which accelerates water evaporation but may also cause greater loss of volatile compounds. Air-frying depends on circulating hot air. This promotes lipid oxidation and esterification while maintaining a relatively stable internal structure. In contrast, pan-frying involves slower and less uniform heat transfer because only the surface is in direct contact with the pan. This condition favors the Maillard reaction and the formation of compounds such as 5-methylfurfural. Overall, these results demonstrate that both the heating medium and thermal distribution strongly affect the generation and retention of volatile compounds in tofu.

Air-frying uses circulating hot air as the heating source. Because heat transfer is less efficient than in oil, the cooking time is longer. This extended exposure promotes lipid oxidation, Maillard reactions, and esterification, producing more aroma-active compounds (OAV > 1). In pan-frying, oil serves as the heat transfer medium, but only the tofu surface directly contacts the pan. This causes uneven and slower heating. Localized high temperatures on the contact surface, combined with longer cooking, enhance Maillard browning and generate additional aroma-active volatiles.

Hydrocarbons are mainly generated from the oxidative and radical degradation of unsaturated fatty acids during thermal treatment. They can also form through the thermal decarboxylation of saturated fatty acids (Chun & Ho, 1997; Van der Klis et al., 2011). Hydrocarbons were the most frequently detected compounds (33 in total). The relative contents in samples A, B, C, and D were 2.08 %, 21.49 %, 24.05 %, and 8.05 %, respectively. All fried samples showed markedly higher hydrocarbon levels than the control (A), suggesting that thermal processing facilitates hydrocarbon generation. However, the extent of formation varied among frying methods, with pan-fried tofu (D) exhibiting a lower proportion than deep- and air-fried samples. This trend may be attributed to the addition of oil during frying. At high temperatures, free radical chain reactions promote the formation of fatty acid hydroperoxides. These hydroperoxides subsequently decompose through β-scission reactions, generating numerous small molecular hydrocarbons. Hydrocarbons generally have high odor thresholds; thus, only a few exhibit an odor activity value (OAV) greater than 1.In this study, only two hydrocarbons—d-limonene (with a citrus and minty odor) and styrene (with a sweet, aromatic odor)—had OAVs greater than 1.d-limonene was absent in the control group, implying that it may originate from refined soybean oil (Li et al., 2020).

Alcohols react with carboxylic acids to form esters. When combined with short-chain acids, they typically yield fruity esters, whereas long-chain acids produce esters with slightly greasy notes (De Lima et al., 2018). Consistent with this mechanism, 28 esters—mainly short-chain—were detected in tofu processed by the three frying methods. Among them, 22 esters were identified in pan-fried tofu, and 11 showed odor activity values (OAVs) greater than 1.These included ethyl 2-methylbutyrate, ethyl 3-methylbutanoate, ethyl 2-methylpropanoate, ethyl octanoate, methyl isobutyrate, methyl 3-methylbutanoate, methyl butanoate, ethyl methacrylate, methyl hexanoate, methyl tiglate, and methyl acetate. These compounds impart pleasant fruity aromas reminiscent of apple, pineapple, and grape. Previous studies have reported that esters can mask rancid odors caused by acidic volatiles and improve the overall flavor harmony of fried foods (De Jesus Filho et al., 2021; Tian et al., 2020). Therefore, differences in ester content and variety likely play a key role in sensory perception. Specifically, 2, 7, and 5 esters were identified in the control, deep-fried, and air-fried tofu, respectively. This variation arises from differences in heat-transfer patterns and reaction intensity. During deep-frying, tofu is fully immersed in hot oil, enabling uniform heating and rapid lipid oxidation. Air-frying depends on circulating hot air, while pan-frying involves direct surface contact with a small amount of oil. These heating modes and durations affect esterification between fatty acids and alcohols, leading to distinct ester profiles. The predominance of short-chain fruity esters in pan-fried tofu (D) aligns with its highest sensory scores (Table 1). These esters enhance fruity and sweet notes while masking off-flavors, which explains the superior flavor perception of the pan-fried samples.

Eighteen ketones were detected across all samples. Two compounds—2,5-dimethyl-3-hexanone and 4-methyl-3-heptanone—were shared among samples but decreased in the fried tofu compared with the control. Ketones had the highest relative content (72.03 %) in the control group, with isobutyrylacetone showing a particularly high concentration of 16,863.26 μg/kg. Other ketones with an odor activity value (OAV) greater than 1 included 2,3-butanedione (diacetyl), 4-methyl-2-hexanone, 5-nonanone, 2-methyl-4-heptanone, ethyl isobutyl ketone, and acetoin. Among these, 2,3-butanedione and acetoin are responsible for buttery and creamy aromas. According to previous studies, ketones in fried foods are mainly produced through the degradation of lipid hydroperoxides (Dobarganes et al., 2000).

The control group contained a relatively high proportion of aldehydes (>10 %), while fried samples showed lower aldehyde contents. No aldehydes were shared among all samples. Aldehydes are mainly formed through the auto-oxidation of triglycerides and the oxidative cleavage of C C double bonds in polyunsaturated fatty acids (Yin et al., 2011). Saturated aldehydes ranging from C6 to C12 have low odor thresholds and strongly affect flavor perception. In this study, aldehydes with odor activity values (OAVs) greater than 1—including 2-nonenal, octanal, nonanal, 2-hexenal, and hexanal—were primarily detected in the control group. These compounds contribute oily, fatty, citrus, rancid, and grassy notes to tofu. Previous studies have shown that octanal and nonanal originate from the oxidation of n-9 polyunsaturated fatty acids (Wang et al., 2022), while hexanal is derived from linoleic acid oxidation through the formation of 13-hydroperoxide intermediates (Toldrá & Flores, 1998).

A total of 19 alcohols were identified, including six in the control group and 13 in fried samples (B—D). Detected alcohols included 3-methylbutan-1-ol, 3-octanol, 2-methyl-3-heptanol, 2,5-dimethyl-3-hexanol, 5-methyl-3-hexanol, 5-methyl-2-hexanol, (2R,3S)-3-chlorobutan-2-ol, (±)-2,3-butanediol, 2-propanol, (R)-1,2-propanediol, cis-3-methylcyclohexanol, (Z)-3-nonen-2-ol, and 1,8-cineole. Alcohols are generally formed via the enzymatic oxidation of n-3 and n-6 unsaturated fatty acids or through carbonyl–ammonia (Strecker-type) reactions (Zhang et al., 2021).The overall alcohol content showed minor variation between control, deep-fried, and air-fried tofu. However, in pan-fried tofu, the relative alcohol content exceeded 10 %.Six alcohols—3-methylbutan-1-ol, n-hexanol, 3-octanol, trans-2-penten-1-ol, pentanol, and 1,8-cineole—had OAVs greater than 1, contributing whisky, allspice, mushroom, banana, balsamic, and mint-like aromas.

A total of 16 phenolic compounds were detected, each with a relative content of less than 3 %. Unique to fried tofu were butylbenzene, n-propylbenzene, ethylbenzene, m-xylene, o-xylene, 2-ethyltoluene, methyl-p-xylene, n-amylbenzene, 2,3,4,5,6-pentapropylphenol, 4-hydroxy-n-(2-phenylethyl)-benzenepropanamide. Among these, isobutylbenzene, butylbenzene, n-propylbenzene, ethylbenzene, 3-ethyltoluene, o-xylene had OAV values greater than 1.

2-Pentylfuran was the only furan with an OAV greater than 1 in fried tofu. It contributed green bean and buttery aromas to the samples. Its formation involves lipid oxidation pathways. Specifically, 2-pentylfuran is generated through the β-scission of linoleic acid hydroperoxides, which produces a 1,3-nonadienyl radical. This radical undergoes hydroperoxidation to form 1,3-nonadienyl-1-COOH, followed by intramolecular cyclization to yield 2-oxo-3-pentylcyclopentene. The compound then stabilizes into 2-pentylfuran (Yao et al., 2022).

Volatile organic compounds (VOCs) with an OAV ≥ 1 were identified as key aroma-active substances in the samples. Among them, 17 compounds showed OAVs ≥10, including ethyl 2-methylbutyrate, ethyl 3-methylbutanoate, ethyl 2-methylpropanoate, ethyl octanoate, methyl isobutyrate, methyl 3-methylbutanoate, methyl butanoate, methyl hexanoate, 2,3-butanedione, 4-methyl-2-hexanone, 2-nonenal, octanal, nonanal, hexanal, 3-methylbutan-1-ol, n-hexanol, and isobutylbenzene. In Sample A, 4-methyl-2-hexanone, 2-nonenal, octanal, nonanal, hexanal, n-hexanol, and isobutylbenzene were the major contributors, corresponding to a weaker overall aroma intensity. In Sample B, n-hexanol and isobutylbenzene played dominant roles in flavor perception. For Sample C, 2,3-butanedione, 3-methylbutan-1-ol, and isobutylbenzene were the key contributors. In Sample D, esters such as ethyl 2-methylbutyrate, ethyl 3-methylbutanoate, ethyl 2-methylpropanoate, ethyl octanoate, methyl isobutyrate, methyl 3-methylbutanoate, methyl butanoate, and methyl hexanoate played crucial roles in shaping the fruity and sweet aroma profile. These compounds mainly imparted fruity, buttery, and fatty notes, consistent with the sensory evaluation results.

Although both GC–IMS and GC–MS analyses revealed distinct volatile profiles among the four tofu samples, the distribution trends of characteristic compounds differed between the two techniques. In GC–IMS, the control group (A) exhibited the greatest number of detected volatile peaks. In contrast, pan-fried tofu (D) contained the largest number of identified compounds in GC–MS analysis. This difference arises from the fundamental analytical principles of the two techniques. GC–IMS is particularly sensitive to small, highly volatile, and thermally labile molecules that may be lost during frying. Consequently, it highlights the light volatiles remaining in unheated or mildly processed tofu. GC–MS, on the other hand, is capable of detecting a broader range of volatiles with medium to high boiling points, including lipid oxidation products, Maillard reaction volatiles, and esters formed during frying. Therefore, the two datasets are not contradictory but complementary. Together, they provide a more complete understanding of how different frying methods influence the generation and retention of volatile compounds.

To further validate these chemical distinctions and clarify how volatile compounds differentiate tofu processed by various frying methods, multivariate statistical analyses were performed using OPLS-DA.

3.5. Analysis of multivariate statistical

Flavor compounds with OAVs ≥1 were selected as independent variables, and the three frying methods served as dependent variables. Orthogonal partial least squares discriminant analysis (OPLS-DA) was used to identify volatile compounds that distinguished tofu samples processed by different frying methods (Ping et al., 2024).The OPLS-DA score plot (Fig. 3A) clearly separated the four samples, indicating that the frying method had a strong impact on the overall volatile composition. Samples A (control) and D (pan-fried tofu) were positioned on opposite sides of the X-axis, showing the largest compositional divergence between unheated and pan-fried tofu. In contrast, samples B (deep-fried) and C (air-fried) clustered more closely, suggesting similar chemical bases dominated by lipid-derived volatiles. The model demonstrated excellent reliability (R²X = 0.966, R²Y = 0.992, Q² = 0.972), indicating strong model robustness. These results suggest that frying methods—particularly pan-frying—shift the dominant volatile-formation pathways. Deep-frying primarily promoted lipid oxidation, whereas pan-frying enhanced Maillard-related ester-forming reactions, together reshaping the overall aroma composition.

Fig. 3.

OPLS-DA model of volatile compounds from different frying methods. Note: A (Control), B (Deep-Fried Tofu), C (Air-Fried Tofu), D (Pan-Fried Tofu). Multivariate analysis of volatile compounds from different frying methods based on OPLS-DA model. Note: Note: (a) Scores plot by PLS-DA, R²X = 0.966, R²Y = 0. 992, Q² = 0.972; (b) Inner relation plot by OPLS-DA; (c) VIP scores. Yellow corresponds to compounds with VIP > 1, and pink represents compounds with VIP < 1. (d) cross-validation plot for the OPLS-DA model with 200 calculations in a permutation test: R² = (0.0, 0.337), Q² = (0.0, −1.18). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

The loading plot (Fig. 3B) identifies the major volatile compounds responsible for differentiating tofu samples processed by different frying methods. In the control group (A), aldehydes such as 2-hexenal, octanal, and 2-nonenal showed high loadings, indicating their strong contribution to the unheated sample. GC–MS results confirmed higher relative levels of these oxidation-derived aldehydes, which are known to impart fatty and grassy odors (Zhou et al., 2022).Deep-fried tofu (B) was mainly characterized by 2-methyl-4-heptanone, a typical lipid-oxidation product formed during immersion in hot oil. This finding suggests that intense lipid degradation dominated the volatile formation pathway in deep-frying. Air-fried tofu (C) showed strong loadings for 2,3-butanedione and acetoin. These compounds are associated with buttery and caramel-like aromas produced through moderate Maillard and Strecker reactions under convective heating. Pan-fried tofu (D) was strongly correlated with short-chain esters such as methyl isobutyrate and methyl tiglate. These esters generate pleasant fruity and sweet notes, consistent with the OAV results showing higher aroma intensity in ester-rich samples. Their presence indicates that localized oxidation and esterification at the oil–food interface are promoted during pan-frying, contributing to a more complex and appealing flavor profile. Overall, the results demonstrate that different heat-transfer mechanisms lead to distinct volatile-formation patterns: oxidation dominates in deep-frying, moderate Maillard reactions in air-frying, and low-level oxidation in unheated samples.

The variable importance in projection (VIP) plot (Fig. 3C) highlights volatile compounds (VIP > 1) that contributed most to discriminating tofu samples processed by different frying methods (Ping et al., 2024). Among them, 2-methyl-4-heptanone, 2,3-butanedione, o-xylene, styrene, isoamyl alcohol, and ethyl isobutyl ketone exhibited the highest VIP values. These compounds were the main contributors to the classification accuracy of the OPLS-DA model. Specifically, 2-methyl-4-heptanone and ethyl isobutyl ketone are typical lipid-oxidation products generated under high-temperature oil immersion. Their presence explains the distinct volatile profile observed in deep-fried tofu (B). In air-fried tofu (C), 2,3-butanedione and isoamyl alcohol were dominant. These compounds originate from moderate Maillard and Strecker reactions, producing buttery and caramel-like aromas. Aromatic hydrocarbons such as o-xylene and styrene also showed high VIP scores. This suggests that partial formation or degradation of aromatic rings occurred during air- or pan-frying. Collectively, these high-VIP compounds serve as chemical markers that distinguish the three frying techniques. Their reaction origins—lipid oxidation in deep-frying, Maillard and Strecker chemistry in air-frying, and aromatic transformations under surface heating in pan-frying—explain the clear separation seen in the OPLS-DA score plot.

The reliability of the OPLS-DA model was further verified through a 200-time permutation test (Fig. 3D).The R² intercept (0.337 < 0.4) and the negative Q² intercept confirmed that the model was not overfitted and maintained strong predictive power (Tian et al., 2022). These results indicate that the separation observed in the OPLS-DA score plot reflects genuine chemical differences among frying methods rather than random variation. The model showed high stability and robustness, as supported by the permutation results and the high R²Y (0.992) and Q² (0.972) values. Together, these parameters verify that the identified discriminant compounds—2-methyl-4-heptanone, 2,3-butanedione, isoamyl alcohol, o-xylene, and styrene—are reliable chemical markers for differentiating frying techniques. Overall, this validation strengthens the conclusion that distinct heat-transfer mechanisms induce systematic variations in volatile-formation pathways, ultimately shaping the characteristic aroma profiles of deep-fried, air-fried, and pan-fried tofu.

Overall, the multivariate statistical analysis independently confirmed the results obtained from GC–MS, OAV, and sensory evaluation. The OPLS-DA model effectively distinguished samples according to frying methods based on their volatile profiles. The key discriminant compounds identified—such as 2-methyl-4-heptanone, 2,3-butanedione, isoamyl alcohol, and short-chain esters—correspond well with those highlighted in the GC–MS and OAV analyses. This consistency across analytical approaches reinforces the conclusion that different heat-transfer modes lead to distinct volatile-formation pathways and corresponding sensory characteristics in fried tofu. Therefore, the multivariate model not only validates the analytical findings but also integrates chemical and sensory information, providing an integrated interpretation of the flavor-formation mechanism.

4. Conclusion

This study systematically examined the effects of different frying methods on the aroma profiles of tofu using sensory evaluation, electronic nose, GC–IMS, GC–MS, and multivariate statistical analyses. The results indicated that pan frying was the most effective method for generating desirable aroma compounds. Sensory evaluation confirmed that pan-fried tofu achieved the highest scores in aroma intensity and overall acceptability. GC–MS analysis identified several low-threshold pleasant odorants in pan-fried tofu, resulting in the highest overall OAVs among all samples. OPLS–DA further demonstrated clear separation of pan-fried tofu from other groups based on its volatile profile. Taken together, these findings provide a scientific foundation for optimizing frying processes to enhance the sensory quality, flavor stability, and consumer preference of tofu products.

This study systematically investigated the effects of deep-frying, air-frying, and pan-frying on the formation of volatile flavor compounds in tofu. However, it did not examine other common cooking methods, such as broiling, boiling, or steaming, which may also play significant roles in shaping tofu's flavor characteristics. Additionally, this research focused solely on volatile compounds; non-volatile taste-active components were not considered. Future studies should aim to comprehensively assess the effects of a broader range of cooking techniques and integrate the analysis of both volatile and non-volatile flavor substances to better elucidate the mechanisms underlying flavor formation in tofu products.

CRediT authorship contribution statement

Yu-Wen Yi: Conceptualization, Formal analysis, Funding acquisition, Resources, Writing – original draft, Writing – review & editing. Chun-Yuan Ping: Formal analysis, Investigation, Writing – original draft. Lian He: Conceptualization, Formal analysis, Methodology, Project administration, Software, Validation. Hao Zhang: Data curation, Investigation. Xue-mei Cai: Data curation, Investigation. Jin-Xiang Hu: Project administration, Supervision. Ming-Feng Qiao: Data curation, Visualization, Writing – original draft.

Funding

This research was financially supported by Project funded by the Sichuan Cuisine Industrialization Engineering Research Center of Sichuan Higher Education Institutions, Sichuan Tourism University (GCZX25–10, GCZX-25-25)，the Sichuan Cuisine Development and Research Center of Sichuan Tourism University (Grant No. CC24Z01) and the Key Laboratory of Sichuan Philosophy and Social Sciences for Artificial Intelligence in Sichuan Cuisine (Grant No. CR24Y02, CR25ZD03).

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.

Data availability

Data sharing is not applicable.