Elucidation of key aroma-active compounds and sensory profiles in raw and thermally sterilized milks: Perspective from molecular sensory science

Abstract

Thermal sterilization significantly influences milk aroma, which is a key factor affecting consumer acceptability. This study employed molecular sensory science to characterize aroma profiles and key aroma-active compounds (AACs) of raw and four thermally sterilized milks. High-intensity sterilization enhanced the milky, fatty, and sweet attributes. Eighty volatiles were identified using a self-built milk flavor database, and 45 odorants were sniffed with flavor dilution (FD) factors of 1–64. Twenty-six odorants (FD ≥ 8) were quantified using external standard calibration. To further improve the precision of quantification, eight AACs (FD ≥ 64; odor activity values ≥1) were quantified by stable isotope dilution analysis for the first time. Recombination and omission experiments revealed 16 key AACs, of which 1-octen-3-ol and butyric acid represented key contributors across five milks, octanal, nonanal, γ-nonalactone, and 2(5H)-furanone were key contributors in the sterilized milks. These findings provide a scientific basis for optimizing milk sterilization processes.

Graphical abstract

Highlights

• •A self-built flavor database identified 80 volatiles, with 45 being odorants in the milk samples.
• •Stable isotope dilution assays ensured precise quantification of eight major odorants.
• •Sixteen key aroma-active compounds were identified in raw and thermally sterilized milk samples.
• •Octanal, nonanal, γ-nonalactone, and 2(5H)-furanone were markers for thermally sterilized milk.

1. Introduction

The dairy industry prioritizes both the safety and the characteristic flavor profiles of liquid milk products. As a globally consumed nutritional food, the flavor compounds of milk are primarily generated during thermal processing, a critical factor in determining its sensory quality and consumer acceptance. Milk undergoes various thermal sterilization treatments, such as high-temperature short-time (HTST) and ultra-high temperature (UHT) processing, to extend its shelf life and ensure its safety. Based on the heating mechanism, UHT processing is further categorized into indirect heating UHT (ID-UHT), which utilizes heat exchangers to achieve sterilization via conduction and convection (Deeth & Datta, 2011; Jo et al., 2018), and direct heating methods, including steam infusion UHT (INF-UHT) and steam injection UHT (INJ-UHT) (Eisner, 2021). INF-UHT employs thermally driven phase-change heat transfer, exposing milk as a liquid film or droplets to saturated steam for sterilization, offering advantages such as reduced cooked aroma (Zhao et al., 2023). Conversely, INJ-UHT utilizes kinetic energy to forcefully introduce high-pressure steam directly into the milk stream for rapid heating (Meng, Han, Zhang, et al., 2023). These distinct heating mechanisms cause variations in the temperature distribution and thermal efficiency, which uniquely affect the sensory quality and volatile compound profiles of the milk.

The flavor profiles of milk are shaped by various complex interplays of reactions during thermal sterilization processing, including the Maillard reaction, lipid oxidation, lactose isomerization, and the denaturation of milk fat globule membrane proteins and other proteins (Jo et al., 2018). Previous research has demonstrated that ID-UHT milk typically exhibits a more pronounced “cooked” or “sulfurous” aroma, which is attributed to key volatile compounds such as furfural, 2-heptanone, benzaldehyde, and dimethyl sulfide, et al. (Jo et al., 2018). HTST and INF-UHT milks reportedly possess similar sensory characteristics. Due to less thermal exposure, INF-UHT milk displays a volatile compound profile comparable to that of raw milk, displaying the exclusive presence of 2-nonanone, 2-butanol, and δ-dodecanol (Meng et al., 2024). Compared to INJ-UHT milk, INF-UHT milk exhibits a milder milky aroma and a less pronounced cooked aroma after thermal sterilization (Meng, Han, Yao, et al., 2023; Meng et al., 2024). Furthermore, although INF-UHT milk may display fewer detectable flavor compounds, certain compounds such as 2-undecanone and decanoic acid are present at higher concentrations (Meng, Han, Zhang, et al., 2023). The oxidized aroma of ID-UHT milk is associated with increased concentrations of 2-heptanone, 2-nonanone, and 2-decanone (Xi et al., 2023). Previous studies have compared the impact of different sterilization methods on the sensory attributes and volatile compound profiles of milk. However, minimal research has applied molecular sensory science to investigate the sensory differences and key aroma-active compounds (AACs) in milk from the same source processed by different sterilization techniques.

The consumer acceptance of milk is primarily determined by its flavor profile, which reflects the interaction between the volatile compounds generated during heat processing and human olfactory receptors, with its chemical foundation consisting of hundreds of trace components. Molecular sensory science is a flavor analysis strategy that integrates instrumental analysis and human sensory perception, proposed by Professor Peter Schieberle's team (Steinhaus & Schieberle, 2007). This approach combines gas chromatography-olfactometry-mass spectrometry (GC-O-MS) with aroma dilution analysis (ADA) to elucidate the contribution of individual aroma compounds within complex aroma systems. Then, human olfactory sensitivity is used to identify the odorants that directly influence sensory perception. Finally, the key ACCs are subjected to recombination and omission experiments to validate their role in the overall flavor (Steinhaus & Schieberle, 2007). Wang et al. (Wang et al., 2025) used molecular sensory science to evaluate how sterilization temperature influences the aroma of HTST milk, identifying 13 key AACs with flavor dilution (FD) factors ≥8. Ten of these, including octanal, hexanal, decanal, 2,3-butanedione, butyric acid, and 1-octen-3-ol, significantly shaped the milk's overall aroma profile. Yan et al. (Yan et al., 2024) used molecular sensory science to characterize the aroma profile of pea milk, identifying 10 key AACs, of which hexanal and (E,E)-2,4-decadienal demonstrated the greatest contributions. However, a comparative molecular sensory science study of milk flavor exposed to different thermal sterilization processes remains lacking, highlighting the necessity of investigating the impact of characteristic aroma compounds on the overall flavor profile under different processing methods.

To elucidate the effects of different sterilization methods on the sensory characteristics and key AACs in milk, this study employed the molecular sensory science to evaluate the effects of different sterilization techniques on milk flavor. With the aim of: (i) Sensory evaluation combined with GC–MS were used to compare the sensory attributes and volatile compound profiles of raw milk and four thermally sterilized milks. (ii) GC-O-MS was integrated with ADA to characterize the odorants in the different milk samples, followed by quantification using the external standard (ES) method. (iii) The potential key AACs were quantified using stable isotope dilution assays (SIDA), with FD factors and odor activity values (OAVs) serving as screening criteria. (iv) Finally, the key AACs were validated via aroma recombination and omission experiments. Molecular sensory science facilitates the evaluation of sensory differences among milks subjected to different processing methods from the perspective of aroma characteristics, providing an effective tool for characterizing and comparing milk flavor profiles, while also offering a theoretical basis for optimizing milk processing technologies and improving product quality.

2. Materials and methods

2.1. Milk sample thermal sterilization treatment

This study evaluated the flavor characteristics of raw milk (RAW) and four thermally sterilized milk samples (pasteurization (PAST), ID-UHT, INF-UHT, and INJ-UHT). Fresh raw milk (3.35% fat, 3.94% protein, 4.85% carbohydrates) was sourced from the Junlebao Dairy Farm (Shijiazhuang, Hebei, China), immediately cooled to 4 °C, and transported to a dairy processing facility. All thermally sterilized milk samples were processed on the same day from this single batch of raw milk to eliminate potential variations due to raw material differences or processing dates. The INF-UHT milk, after clarification, was directly sterilized at 155 °C for <1 s using a steam infusion system (SPX, Silkeborg, Denmark), and flash-cooled in a vacuum vessel equipped with a condenser (75–77 °C), followed by downstream homogenization at 40/15 MPa and 76 °C. The other UHT milk underwent homogenization (25/5 bar, 65 °C) before sterilization. Both the ID-UHT and INJ-UHT milk samples were pre-pasteurized at 75 °C for 15 s before UHT processing. The ID-UHT milk was indirectly sterilized in a tubular heat exchanger (MicroThermics, NC, USA) at 145 °C for 4 s, followed by indirect cooling to 25 °C–30 °C. The INJ-UHT milk was sterilized directly at 145 °C for 4 s using a steam injection system (Tetra Pak, Shanghai, China), followed by flash cooling at 75–77 °C and downstream homogenization at 40/15 MPa and 76 °C. Finally, all the samples were cooled to room temperature, aseptically packaged in plastic bottles covered with aluminum foil, and delivered to the laboratory within 24 h (process flow diagram shown in Fig. 1).

Fig. 1.

The flowchart of the thermal sterilization process used for the milk samples.

2.2. Chemicals

The methanol (99.9% purity) was acquired from MREDA Biochem (Beijing, China). The n-Alkanes (C₇-C₃₀) for the linear retention index (RI) calculation and internal standard (IS) 2-methyl-3-heptanone (≥ 99% purity) were purchased from Sigma-Aldrich (St. Louis, USA). The commercial standards for identifying and quantifying the AACs included the following: compounds A4, A12-A13, B5, C1, C2, C9-C10, D2-D3, E3-E5, F1, F8-F10, F13, F15, G1, H1-H4, and I1-I2 were purchased from Macklin Biochem (Shanghai, China). Compounds A9-A10, B2-B3, B6, C3-C4, C12, D1, D7-D8, E1, E6-E8, F2、F7, F11, F12, F17, and G2 were purchased from TCL (Shanghai, China). Compounds A1, A5, A7-A8, A11, C5, C7-C8, C13-C14, C16, D4-D6, F4, F16, and F18 were purchased from Yuanye Bio-Technology (Shanghai, China). Compounds A2, B4, C6, E2, F3, and F5 were purchased from Mreda Biochem (Beijing, China). Compounds A6, F6, and C11 were purchased from Sigma-Aldrich (St. Louis, MO, USA), while compounds A3, A14, B1, C15, E9, G3, and F14 were purchased from Aladdin Chem. (Shanghai, China).

Furthermore, isotopes include [²H₅]-hexanal (95%), [D₁₆]-octanal (95%), [D₁₈]-nonanal (95%), [D₃]-1-octen-3-ol (95%), [D₆]-benzaldehyde (98%), [¹³C]-butyric acid (98%), [¹³C]-octanoic acid (98%), and [¹³C]-decanoic acid (98%) were purchased by ZZBIO Co., Ltd. (Shanghai, China).

2.3. Quantitative descriptive analysis (QDA)

The aroma profiles of raw and four thermally sterilized milk samples were evaluated by 12 trained sensory panelists (4 males, 8 females; aged 20–43 years) with systematic sensory training. The assessors participated voluntarily and provided written informed consent before the experiments (Beijing Technology and Business University Research Ethics Committee, reference number: No. (64) 2024). Based on literature review and expert discussion (Wang et al., 2025; Yu et al., 2024; Z. Zhang et al., 2025), six aroma descriptors were selected to characterize the flavor profiles: milky (acetoin), fatty (2,3-butanedione), cooked (85  °C, 30 min heat-treated milk), oxidized (milk exposed to light for 18 h), sweet (δ-decalactone), and green/grassy (hexanal). The reference compounds were prepared in methanol at concentrations 50 times higher than their odor thresholds in water (Li et al., 2023). The QDA was performed using a six-point scale (0–5, with 0.5 increments). To mitigate sensory fatigue, the panelists cleaned their nasal passages with coffee beans between samples and observed mandatory 1-min rest intervals.

2.4. Extraction of the volatile compounds in the milk samples

The volatile compounds in the milk samples were extracted using solid-phase microextraction-Arrow (SPME-Arrow) fibers (DVB/Carbon WR/PDMS, outer diameter 1.10 mm, stationary phase thickness 120 μm). A mixture containing 5 g of milk, 0.5 g of NaCl, and 0.5 μL of 2-methyl-3-heptanone (0.816 μg/μL in methanol) was accurately weighed into a 20 mL headspace vial, which was then sealed with a PTFE/silicone septum cap. The mixture was vortexed and equilibrated at 45 °C for 20 min. Then, the SPME-Arrow fiber was exposed to the vial headspace for 30 min for volatile compounds adsorption, followed by desorption in the GC inlet at 250 °C for 5 min.

2.5. Odor compound analysis

2.5.1. GC-O-MS analysis

The separation and detection parameters were optimized based on our previous research (Z. Zhang et al., 2025). The volatile compounds were identified and analyzed using a GC-O-MS triple quadrupole system (TQ8040 NX, Shimadzu, Japan) equipped with a multifunctional injector (AOC-6000, Shimadzu, Japan) and an olfactory detection port (ODP3, Gerstel, Germany). An SH-Polar Wax capillary column (60 m × 0.25 mm, 0.25 μm, Shimadzu, Japan) was employed for separation, using helium (99.99%, purity) as carrier gas at 1 mL/min. The chromatographic temperature program included an initial temperature of 40 °C for 5 min, which was increased to 250 °C at 3 °C/min, where it was held for 2 min. The total runtime was 77 min while the solvent delay time was 1.82 min. The MS parameters included electron impact ionization at 70 eV, an ion source temperature of 200 °C, an interface temperature of 250 °C, and multiple reaction monitoring (MRM) mode acquisition in a range of m/z 33–550.

The column effluent was divided equally (1:1 v/v) between the MS and the olfactory detection port. Humidified air was directed to the olfactory port at a flow rate of 45 mL/min to ensure that the nasal mucosa remained moist. Trained panelists (2 females and 1 male with a mean age of 26 years) recorded the retention time (RT), odor descriptors (O), and intensity.

2.5.2. Qualitative analysis of the volatile compounds

The compounds were identified using a method described by Yu et al. (Yu et al., 2023). Four strategies were employed for qualitative verification: MS, RI, O, and comparison with authentic reference standards (STDs). The compounds confirmed via STDs were classified as definitive, while those lacking verification were considered tentative.

2.5.3. Aroma dilution analysis

The ADA was used to evaluate the contribution of the individual compounds to the overall aroma profile. The volatile extracts obtained via SPME-Arrow were serially diluted by adjusting the GC inlet split ratio (1:2ⁿ, n = 0, 1, 2, …) and analyzed using the GC-O-MS system until the panelists could no longer detect odors at the olfactory port. The highest dilution at which an odor was still perceptible was defined as the FD factor for that odorants. Each sample was assessed in triplicate (Han et al., 2024).

2.6. Quantitative analysis

2.6.1. Semi-quantitative analysis based on internal standard

The IS method was utilized for the semi-quantitative analysis of the identified compounds. The relative concentrations were calculated by comparing the peak areas of the identified compounds to that of the IS concentration.

2.6.2. Quantitative analysis using external standard calibration

A five-point ES calibration was used to accurately quantify 26 odorants with FD ≥ 8. Calibration curves were established for each compound to eliminate signal response discrepancies and instrument errors. Preparation of the deodorized milk matrix system was prepared followed the method of Davood et al. (Davood et al., 2022) with minor modifications. A mixture containing 5% (m/v) whey protein concentrate powder (MPC 70, Saishang Dairy Co. Ltd., Ningxia, China) and 4% (m/v) sunflower seed oil (Yihai Kerry Arawana Co. Ltd., China) was pre-homogenized at 10,000 rpm for 5 min using a high-speed disperser. This was followed by secondary homogenization (60 bar primary, 140 bar secondary) using a high-pressure homogenizer (AH-Basic, Haiyi Technology Co. Ltd., China) to obtain the final milk matrix. Calibration curves were generated by plotting the concentration ratios of the target compounds against their corresponding peak area ratios relative to the IS (2-methyl-3-heptanone, 81.6 μg/mL, 0.5 μL). The analysis was performed using GC–MS in MRM mode, as detailed in Section 2.5.1. The target compound concentrations were determined using the corresponding calibration curves (Li et al., 2023).

Eight major AACs with FD ≥ 64 (Table 1) and OAV ≥ 1 (Table 2) were further quantified using SIDA. A dilution series was prepared by combining isotope-labeled standards with their unlabeled analytes in methanol at five different ratios (4:1, 2:1, 1:1, 1:2, 1:4) for GC–MS analysis in MRM mode. Detailed fragment ion information was shown in Table S2. Calibration curves were generated by plotting the peak area ratio (labeled IS to unlabeled analyte) against the corresponding concentration ratio (Li et al., 2023).

Table 1.

Odor-active compounds identified in RAW and thermally sterilized milks by GC-O-MS.

NO	Aroma active compounds	CAS	Aroma description a	FD b					RI c	Identification method d
				RAW	PAST	ID-UHT	INF-UHT	INJ-UHT
Alcohols (8)
1	1-Butanol	71–36-3	Sweat, gasoline	4	1	8	2	16	1118	MS/RI/O/STD
2	1-Pentanol	71–41-0	fruit, greasy	2	–	1	–	4	1224	MS/RI/O/STD
3	1-Hexanol	111–27-3	grassy, green	2	2	4	1	2	1304	MS/RI/O/STD
4	1-Octen-3-ol	3391-86-4	mushroom, green	16	32	64	16	64	1444	MS/RI/O/STD
5	2-Ethyl-1-hexanol	104–76-7	citrus, floral	1	1	1	2	4	1460	MS/RI/O/STD
6	1-Octanol	111–87-5	fresh, fatty	1	1	–	1	–	1558	MS/RI/O/STD
7	1-Octanol, 3,7-dimethyl-	106–21-8	fresh, citrus	8	4	4	8	16	1655	MS/RI/O/STD
8	1-Dodecanol	112–53-8	butter, hay	2	–	1	1	1	1933	MS/RI/O/STD
Aldehydes (6)
9	Hexanal	66–25-1	green, grass	64	32	16	8	16	1058	MS/RI/O/STD
10	Octanal	124–13-0	herbal, fruity	1	4	64	2	64	1267	MS/RI/O/STD
11	Nonanal	124–19-6	fatty, green	1	2	64	1	8	1372	MS/RI/O/STD
12	(E)-2-Octenal	2548-87-0	fatty, grass	32	4	16	4	32	1405	MS/RI/O/STD
13	Benzaldehyde	100–52-7	almond, woody	16	32	64	32	64	1495	MS/RI/O/STD
14	(E)-2-Nonenal	18,829–56-6	bitter melon, fatty	1	1	–	1	–	1512	MS/RI/O/STD
Acids (9)
15	Acetic acid	64–19-7	sour, milky	64	4	8	4	8	1429	MS/RI/O/STD
16	Butyric acid	107–92-6	butter, sweat	64	16	32	16	64	1635	MS/RI/O/STD
17	Hexanoic acid	142–62-1	cream, sour	1	–	–	–	–	1836	MS/RI/O/STD
18	Octanoic acid	124–07-2	stench, fermented	64	32	32	32	64	2036	MS/RI/O/STD
19	Nonanoic acid	112–05-0	leather, cheese	1	1	–	–	–	2175	MS/RI/O/STD
20	Decanoic acid	334–48-5	fatty, rancid	64	8	8	8	8	2239	MS/RI/O/STD
21	Dodecanoic acid	143–07-7	sour, fatty	8	–	–	–	4	2446	MS/RI/O/STD
22	9-Decenoic acid	14,436–32-9	sour, fatty	4	1	1	1	1	2328	MS/RI/O/STD
23	Undecanoic acid	112–37-8	pungent, fruit	32	2	4	1	4	2359	MS/RI/O/STD
Ketones (8)
24	2,3-Butanedione	431–03-8	butter, sweet	4	2	16	2	32	978	MS/RI/O/STD
25	2-Heptanone	110–43-0	fruit, coconut	–	–	8	–	16	1183	MS/RI/O/STD
26	Acetoin	513–86-0	rice, sweet	1	–	1	–	1	1257	MS/RI/O/STD
27	3-Nonanone	925–78-0	fruit	–	–	1	–	1	1362	MS/RI/O/STD
28	2-Undecanone	112–12-9	fatty	–	–	2	–	1	1603	MS/RI/O/STD
29	2(5H)-Furanone	497–23-4	rice, caramel	4	8	32	8	16	1720	MS/RI/O/STD
30	2-Tridecanone	593–08-8	coconut, herbal	–	–	1	–	1	1785	MS/RI/O/STD
31	2-Pentadecanone	2345-28-0	fruit, jasmine	16	2	32	2	8	2021	MS/RI/O/STD
Alkenes (2)
32	Limonene	138–86-3	orange	1	1	–	1	–	1182	MS/RI/O/STD
33	Styrene	100–42-5	metal, cardboard	1	1	2	1	1	1232	MS/RI/O/STD
lactones (9)
34	γ-Butyrolactone	96–48-0	fatty, greasy	4	4	1	4	2	1599	MS/RI/O/STD
35	δ-Hexalactone	823–22-3	fatty, cream	32	8	16	8	16	1801	MS/RI/O/STD
36	δ-Octalactone	698–76-0	greasy, coconut	–	–	1	–	1	1941	MS/RI/O/STD
37	Pantolactone	599–04-2	butter, sweet	16	8	8	8	16	1988	MS/RI/O/STD
38	γ-Nonalactone	104–61-0	fatty, peach, coconut	8	8	16	8	32	1996	MS/RI/O/STD
39	δ-Nonalactone	3301-94-8	nut, fatty, cream	–	–	8	–	16	2018	MS/RI/O/STD
40	δ-Decalactone	705–86-2	butty, peach, coconut	16	8	16	8	32	2161	MS/RI/O/STD
41	γ-Dodecalactone	2305-05-7	peach, fatty	–	2	4	1	32	2334	MS/RI/O/STD
42	δ-Dodecalactone	713–95-1	coconut, cream, peach	16	1	8	–	32	2386	MS/RI/O/STD
Sulfur compounds (2)
43	Dimethyl sulfoxide	67–68-5	caramel, onion	–	1	1	1	1	1551	MS/RI/O/STD
44	Dimethyl sulfone	67–71-0	caramel	–	–	1	–	1	1912	MS/RI/O/STD
Others (2)
45	2,4-Di-tert-butylphenol	96–76-4	plastic, herbal	4	1	2	8	1	2257	MS/RI/O/STD
	Total	Kinds		36	33	39	32	40

^aOdor description obtained by GC-O analysis.

^bFD is the dilution factor; “-”: Not detected.

^cRetention indices obtained by a SH-PolarWax column.

^dThe four identification methods include mass spectra (MS), retention indices (RI), olfactometry (O), and standard compounds (STD).

Table 2.

Quantitative results for odorants (FD ≥ 8) in RAW and thermally sterilized milks by ES.

Aroma active compounds	Linear equation	R2	Threshold a (μg/kg)	Concentration (μg/kg)					OAV					Selected ions b (m/z)
				RAW	PAST	ID-UHT	INF-UHT	INJ-UHT	RAW	PAST	ID-UHT	INF-UHT	INJ-UHT
Alcohols
1-Butanol	y = 0.0004x + 0.0002	0.9930	459.2	273.28 ± 3.39c	81.84 ± 6.39e	328.30 ± 8.66b	105.64 ± 4.88d	852.76 ± 14.62a	<1	<1	<1	<1	1.9	56.00 > 39.00–73.00 > 55.10
1-Octen-3-ol	y = 0.3392x - 0.0235	0.9944	1.5	0.91 ± 0.04bc	0.78 ± 0.01c	1.75 ± 0.04a	1.11 ± 0.01b	1.59 ± 0.05a	<1	<1	1.2	<1	1.1	72.00 > 43.00–72.00 > 57.00
1-Octanol, 3,7-dimethyl-	y = 0.0674x - 0.0012	0.9948	0.79	1.96 ± 0.06b	1.87 ± 0.02c	1.88 ± 0.02c	1.92 ± 0.10b	2.14 ± 0.11a	2.5	2.4	2.4	2.4	2.7	43.10 > 41.10–56.10 > 39.10
Aldehydes
Hexanal	y = 0.1611x - 0.7279	0.9909	5	57.21 ± 0.36a	51.25 ± 0.15b	46.95 ± 0.12d	48.42 ± 0.25c	47.23 ± 0.10d	11.4	10.2	9.4	9.7	9.4	82.00 > 67.00–82.00 > 41.10
Octanal	y = 0.099x - 0.0079	0.9941	0.587	1.31 ± 0.09d	1.56 ± 0.04d	2.99 ± 0.09c	3.68 ± 0.09b	4.87 ± 0.18a	2.2	2.7	5.1	6.3	8.3	84.00 > 69.10–84.00 > 42.00
Nonanal	y = 0.0957x + 0.0012	0.9879	1.1	1.44 ± 0.11c	1.66 ± 0.13c	11.35 ± 0.61a	3.96 ± 0.05b	3.50 ± 0.06b	1.3	1.5	10.3	3.6	3.2	70.00 > 55.10–98.00 > 56.00
(E)-2-Octenal	y = 0.0051x - 0.0041	0.9969	3	11.04 ± 0.19c	9.42 ± 0.07d	16.23 ± 0.84b	15.04 ± 0.51b	22.51 ± 0.16a	3.7	3.1	5.4	5.0	7.5	70.10 > 55.10–70.10 > 42.10
Benzaldehyde	y = 0.1641x - 0.0108	0.9966	24	5.68 ± 0.39d	5.71 ± 7.34c	18.28 ± 4.18b	6.44 ± 9.05b	24.21 ± 8.71a	<1	<1	<1	<1	1.0	77.00 > 51.00–106.00 > 77.00
Acids
Acetic acid	y = 0.0054x - 0.0404	0.9843	10,000	160.37 ± 1.23a	116.21 ± 1.74c	116.97 ± 1.90c	110.80 ± 3.17d	123.10 ± 4.35b	<1	<1	<1	<1	<1	60.00 > 45.00–60.00 > 37.10
Butyric acid	y = 0.0011x - 0.0054	0.9972	240	32,109.86 ± 925.06a	1093.90 ± 212.60e	1261.61 ± 54.18c	1191.22 ± 44.31d	2093.86 ± 82.71b	133.8	4.6	5.3	5.0	8.7	73.00 > 55.00–60.00 > 45.00
Octanoic acid	y = 0.0013x - 0.0069	0.9918	910	16,835.02 ± 596.51a	1036.54 ± 80.16c	950.41 ± 24.82d	876.13 ± 26.47e	1307.51 ± 42.69b	18.5	1.1	1.0	<1	1.4	101.00 > 45.00–101.00 > 57.00
Decanoic acid	y = 0.0017x - 0.0155	0.9846	130	7167.40 ± 119.23a	379.69 ± 1.70c	377.39 ± 17.11c	334.98 ± 2.67d	445.68 ± 19.74b	55.1	2.9	2.9	2.6	3.4	129.00 > 55.10–129.00 > 59.10
Undecanoic acid	y = 0.0001x - 0.0006	0.9904	10	887.38 ± 10.17a	106.93 ± 6.43b	108.74 ± 3.25b	111.20 ± 10.14b	109.71 ± 1.19b	88.7	10.7	10.9	11.1	11.0	143.00 > 87.00–129.00 > 55.00
Dodecanoic acid	y = 0.0016x - 0.0025	0.9967	1200	251.75 ± 5.85a	–	–	–	54.92 ± 1.23b	<1	–	–	–	<1	171.00 > 87.00–200.00 > 87.10
Lactones
δ-Hexalactone	y = 0.0002x - 0.0003	0.9922	100	5887.68 ± 217.91a	351.53 ± 23.86e	773.41 ± 27.73c	425.53 ± 3.18d	1362.03 ± 40.35b	589	3.5	7.7	4.3	13.6	70.00 > 55.10–99.00 > 71.00
γ-Nonalactone	y = 0.0259x - 0.0003	0.9983	7	13.38 ± 0.05b	12.86 ± 0.03c	13.06 ± 0.04bc	12.89 ± 0.02c	14.70 ± 0.06a	1.9	1.8	1.9	1.8	2.1	128.10 > 95.10–100.10 > 72.10
Pantolactone	y = 0.0025x - 0.0005	0.9919	200	253.88 ± 20.62a	9.76 ± 1.27d	10.44 ± 0.19c	9.71 ± 0.33d	14.27 ± 0.74b	1.3	<1	<1	<1	<1	71.10 > 41.10–43.10 > 41.10
δ-Nonalactone	y = 0.0118x- 0.0001	0.9999	2600	–	–	1.33 ± 0.09b	–	3.49 ± 0.19a	–	–	<1	–	<1	99.10 > 43.10–71.10 > 43.10
δ-Decalactone	y = 0.0036x - 0.001	0.9971	66	292.31 ± 0.49b	137.90 ± 5.96d	249.51 ± 9.31c	134.98 ± 7.15d	759.02 ± 13.22a	4.2	2.1	3.8	2.0	11.5	99.00 > 43.10–99.00 > 41.10
γ-Dodecalactone	y = 0.0003x + 0.0004	0.9992	0.43	–	40.15 ± 8.18b	16.62 ± 0.24c	5.45 ± 0.44d	123.18 ± 6.60a	–	93.4	38.6	12.7	286.5	57.10 > 41.10–85.00 > 41.10
δ-Dodecalactone	y = 0.0001x + 0.00006	0.9879	53	2388.25 ± 40.84b	844.16 ± 22.37d	1881.81 ± 21.56c	–	6450.20 ± 72.84a	45.1	15.9	35.5	–	121.7	99.00 > 43.00–99.00 > 41.00
Ketones
2,3-Butanedione	y = 0.0094x + 0.0032	0.9901	14	11.09 ± 1.05c	9.70 ± 1.40d	23.00 ± 1.45b	6.54 ± 0.97e	29.42 ± 1.19a	<1	<1	1.6	<1	2.1	43.00 > 41.00–43.00 > 39.90
2-Heptanone	y = 0.8124x - 0.0355	0.999	140	–	–	43.93 ± 0.46a	–	23.00 ± 0.83b	–	–	<1	–	<1	114.00 > 85.10–114.00 > 71.00
2(5H)-Furanone	y = 0.0015x - 0.0038	0.992	1	28.69 ± 0.22d	30.48 ± 0.94c	42.12 ± 1.12a	28.81 ± 1.03d	33.47 ± 0.51b	28.7	30.5	42.1	28.8	33.5	84.00 > 39.10–55.00 > 53.10
2-Pentadecanone	y = 0.0046x - 0.0002	0.992	1	2.15 ± 0.69b	1.51 ± 0.25c	4.18 ± 0.26a	1.17 ± 0.01d	2.36 ± 0.51b	2.1	1.5	4.2	1.2	2.4	96.10 > 81.00–96.10 > 67.00
Others
2,4-Di-tert-butylphenol	y = 0.0124x - 0.0009	0.991	500	17.96 ± 1.14b	12.61 ± 0.97c	16.77 ± 0.91b	35.38 ± 1.59a	13.28 ± 0.38c	<1	<1	<1	<1	<1	206.00 > 191.20–191.00 > 163.10

^aOdor thresholds in water were obtained from literature. (Burdock, 2016; Gemert, 2003; Han et al., 2024).

^bThe ion pair preceding ‘-’ is quantitative, ‘-’ denotes qualitative ion pairs, the ion before ‘>’ is the parent ion, and the ion after ‘>’ is the daughter ion.

2.7. Determination of the OAVs

The OAVs of the compounds were determined as the ratio of their quantified concentration to their reported odor thresholds in water. Compounds with OAVs ≥1 contributed significantly to the overall aroma profile, while those with OAVs <1 had minimal influence (Li et al., 2023).

2.8. Aroma recombination and omission experiments

Based on the precise quantification results, the AACs (FD ≥ 8 and OAV ≥ 1) were incorporated into the deodorized milk matrix system and equilibrated in a thermostatic oscillator at 10 °C for 2 h to generate a recombinant flavor model. The models were subsequently evaluated as described in Section 2.3.

To evaluate the contribution of AACs, omission models were prepared by removing either the major compound classes or the individual AAC. The panelists used the triangle test methodology to evaluate triads consisting of two identical recombined models and one omission, which were randomly assigned three-digit codes, to determine perceptible differences between these models (Han et al., 2024).

2.9. Statistical analysis

The experimental results were presented as the mean ± standard deviation of three independent replicates. One-way analysis of variance (ANOVA) was performed using SPSS v.26.0 (SPSS, Inc., Chicago, IL, USA), while intergroup differences were evaluated using Duncan's multiple range test (p < 0.05). Microsoft Office 2019 (Microsoft Corp., Redmond, WA, USA), OriginPro 2019b (OriginLab Corp., Washington, MA, USA), and Chiplot (https://www.chiplot.online/) were employed for data organization, basic statistical analysis, and graphical representation, respectively.

3. Results and discussion

3.1. QDA sensory evaluation in raw and four thermally sterilized milk samples

Sensory evaluation is an effective tool for aroma analysis, which involves highly sensitive olfactory detection with a theoretical odor detection limit of approximately 10⁻¹⁹ mol (Rouseff et al., 2009). This study characterized milk samples (RAW, PAST, ID-UHT, INF-UHT, and INJ-UHT) using six aroma attributes (milky, fatty, sweet, cooked, oxidized, and green/grassy) (Fig. 2A). Significant differences were observed for all the attributes except the green/grassy and oxidized attributes (p < 0.05), with the milky, fatty, and sweet attributes exhibiting highly significant differences (p < 0.001). High-intensity sterilization treatment (ID-UHT and INJ-UHT) markedly enhanced the milky (3.67 4.04), fatty (3.38, 3.50), and sweet (2.96, 3.04) attributes compared with RAW, PAST, and INF-UHT milk samples. This could be attributed to the disruption of fat globule membranes during pre-homogenization, which facilitated the release of bound lipids, along with high-intensity sterilization conditions that promoted the Maillard reaction and subsequent flavor compound generation (Y. Liu et al., 2025). However, high-intensity thermal sterilization also significantly increased the cooked attribute in the ID-UHT and INJ-UHT milks (p < 0.05), which was associated with protein denaturation and sulfur compound formation during the Maillard reaction (Y. Liu et al., 2025). Contrarily, the stronger milky and fatty attributes in the RAW milk relative to the PAST and INF-UHT milks suggest that intact phospholipid membranes protect thermal-sensitive volatiles from thermal degradation. Notably, the INF sterilization and post-homogenization during the INF-UHT process preserved the fat globule integrity before thermal treatment. This reduced the flavor loss compared to PAST milk, resulting in an overall aroma profile closer to that of raw milk (Jo et al., 2018). Furthermore, the thermal intensity reduced the characteristics related to freshness. The green/grassy attributes decreased markedly in the ID-UHT (0.79) and INJ-UHT (0.88) milks compared with the RAW milk (1.38), reflecting the degradation or volatilization of compounds such as hexanal at high temperatures. The oxidized attribute exhibited minor variations (1.33–1.88), with ID-UHT milk oxidized attribute being slightly higher (1.88), indicating that intense thermal treatment possibly promoted the oxidized attribute, though its impact was less pronounced than the other attributes.

Fig. 2.

The sensory evaluation of the RAW and thermally sterilized milk samples. (A) The QDA radar plots of the sensory attributes. (C–G) The sensory profiles and aroma recombination models of the RAW, PAST, ID-UHT, INF-UHT, and INJ-UHT milk (Compound A6, B1, B2, B3, B5, C3, C9, C11 were measured by SIDA, while A1, A11, A10, C13, E2, F1, E4, E5, E7, E8, E9, F14, F17 were measured by ES).

3.2. Identification of the volatile compounds in milk samples through GC–MS

The variations in the milk flavor resulting from different sterilization methods are crucial in determining consumer preferences. The content and types of volatile compounds in the thermally sterilized milks were characterized (Fig. 3, Table S1). A total of 80 volatile compounds were identified in nine categories, which included 14 alcohols, 6 aldehydes, 16 acids, 8 esters, 9 lactones, 18 ketones, 3 alkenes, 4 sulfur compounds, and 2 other compounds (Fig. 3A). A compound distribution comparison revealed 76 volatiles in the RAW milk and 78 in the processed samples (Fig. 3B). Several compounds were consistently detected in all five milk samples, including 14 alcohols, 6 aldehydes, 16 acids, 9 lactones, 16 ketones, 3 alkenes, and 2 other compounds. Seven esters were commonly identified in all samples, with hexyl acetate detected exclusively in RAW milk. Dimethyl sulfone was detected in all samples, while dimethyl disulfide, dimethyl trisulfide, and dimethyl sulfoxide were detected in processed milks. Among ketones, 17 ketones were detected overall, with 5-methyl-2-hexanone exclusively present in INJ-UHT milk, while 3-nonanone was identified in processed milks except INJ-UHT milk.

Fig. 3.

(A) The heatmap of the RAW and thermally sterilized milk samples. (B) The accumulated types and concentrations of the volatile compounds in the RAW and thermally sterilized milk samples.

The volatile compound concentration in the milk samples ranged from 103.90 ± 1.59 to 701.11 ± 4.45 μg/L (Fig. 3B), indicating minimal variation in the compound types but significant concentration differences. This was attributed to the establishment of a self-built milk flavor database that utilized standard ion fragments and MRM mode for data acquisition, which significantly increased the sensitivity and resolution for the rapid, accurate analysis of complex food matrices. Acids dominated in both the type and concentration, accounting for 37.46–90.82% of the total volatiles across the treatments, with the highest concentration observed in the RAW milk (636.81 ± 8.35 μg/L). Thermal sterilization markedly reduced the acid content due to lipid degradation, accompanied by conversion into ketones and lactones. The β-oxidation of free fatty acids and the thermal degradation of amino acids promoted ketone formation, while hydroxy acid hydrolysis from triglycerides increased lactone production, enhancing the fatty aroma (Jo et al., 2018). The ID-UHT and INJ-UHT milk samples exhibited higher ketone concentrations than the other samples, indicating that a higher degree of thermal sterilization increased the ketone content. Cadwallader et al. (D. C. Cadwallader et al., 2023) indicated that ketones contributed an oxidized aroma that affected the overall flavor of milk, which confirmed the sensory evaluation results. Lipid oxidation and amino acid transamination promoted aldehyde formation, which enhanced the fatty and green/grass aromas (Elbarbary et al., 2025). A higher sterilization intensity increased the aldehyde concentration. Given their lower thresholds, aldehydes, ketones, lactones, and high acid concentrations were identified as the principal compounds in milk, which was consistent with previous findings (Elbarbary et al., 2025). Furthermore, the ketone and lactone content increased progressively with the sterilization intensity, exhibiting significantly higher levels in the ID-UHT and INJ-UHT samples than in the PAST and INF-UHT samples. In addition to lipid degradation, the Maillard reaction between lactose and protein amino groups during thermal sterilization generated Strecker aldehydes and sulfur-containing compounds, further enriching the volatile profile (Meng et al., 2024).

3.3. Identification of the odorants and FD factors in milk samples via GC-O and ADA

Combining SPME-Arrow/GC-O-MS and ADA enabled the accurate and comprehensive identification of the odorants in the RAW and processed milk samples. The FD value reflects the sensory potency of odorants. A higher FD factor generally indicates a greater potential contribution to the overall flavor profile. A total of 45 odorants were detected in the five milk samples (Table 1), with FD factors ranging from 1 to 64. Specifically, 36 odorants were identified in the RAW milk, 33 in the PAST milk, 39 in the ID-UHT milk, 32 in the INF-UHT milk, and 40 in the INJ-UHT milk. Although a total of 26 odorants were consistently sniffed in all five milk samples, including alcohols, aldehydes, acids, and lactones (Fig. 4A), their concentrations varied. Furthermore, 26 odorants (FD ≥ 8) were identified via ADA (Fig. 4B). Aldehydes, ketones, and lactones, characterized by low odor thresholds and distinct sensory attributes, contributed green, creamy, and buttery aromas. In addition, acids with relatively higher thresholds represented significant contributors in the RAW milk, which were essential for enhancing the overall milky aroma.

Fig. 4.

(A) The Venn diagram of the 45 odorants identified via GC-O in the RAW and thermally sterilized milk samples. (B) The stacked bar plots of the 26 odorants (FD ≥ 8). (C) The heatmap of the 21 AACs (FD ≥ 8 and OAV ≥ 1).

Aldehydes, formed via lipid oxidation or amino acid transamination, play an indispensable role in shaping the aroma profile of milk due to their high concentrations and low odor thresholds, and are especially associated with fatty and green aromas (Dan et al., 2019). The oxidation of specific unsaturated fatty acids serves as key pathways for the generation of these aldehydes. For instance, the oxidation of linoleic acid primarily yields hexanal and (E)-2-octenal, while the degradation of oleic acid is a major source of octanal and nonanal (Frankel, 2012). Of the 26 identified odorants, 5 aldehydes associated with green and fatty attributes were sniffed, including hexanal, octanal, nonanal, (E)-2-octenal, and benzaldehyde. Saturated straight-chain aldehydes (hexanal, octanal, and nonanal) primarily impart a fresh, herbal aroma, while unsaturated aldehydes ((E)-2-octenal and benzaldehyde) contribute fatty aromas. Octanal exhibited the highest FD factor (64) in the ID-UHT and INJ-UHT milk samples, which underwent more intense thermal treatments, contributing herbal and fruity aromas. Nonanal showed the highest FD factor (64) in the ID-UHT milk, presenting fatty and green aromas. Hexanal, (E)-2-octenal, and benzaldehyde consistently demonstrated high FD factors (8–64) across all five milk samples. Hexanal and (E)-2-octenal are lipid-derived aldehydes that are formed via linoleic acid oxidation during fatty acid degradation, contributing fruity and herbal aromas. Benzaldehyde is a Strecker aldehyde formed from the amino acid degradation initiated by free radicals during lipid oxidation. This substance increased at a higher sterilization intensity and imparted almond and caramel aromas (Pripis-Nicolau et al., 2000).

Ketones are naturally present in raw milk and can also form during thermal processing via the β-oxidative decarboxylation of saturated fatty acids or amino acid degradation(Cadwallader & Singh, 2009). Eight ketones were sniffed across the five milk samples, including 2,3-butanedione (buttery, sweet), 2-heptanone (fruity, pudding), acetoin (rice, sweet), 3-nonanone (fruity), 2-undecanone (fatty), 2(5H)-furanone (rice, caramel), 2-tridecanone (caramel), and 2-pentadecanone (fruity). The 2(5H)-furanone was present low odor intensity in RAW milk but increased significantly with increasing sterilization intensity, primarily attributed to Maillard reaction. Its odor intensity was higher in the ID-UHT milk than in the INJ-UHT milk, reflecting localized overheating caused by tubular heat exchange compared with steam injection. The 2,3-butanedione, associated with buttery and creamy aromas, was more abundant in the ID-UHT and INJ-UHT milk samples, likely due to the higher sterilization intensity, consistent with previous findings (Garvey et al., 2020). The acetoin was only sniffed in the ID-UHT and INJ-UHT milks. Acetoin and 2,3-butanedione are interconvertible AACs. Studies have shown that acetoin was produced by lactic acid bacteria fermentation, whereas 2,3-butanedione and related hydrogenation byproducts are produced during this process (J.-M. Liu et al., 2020). Given the absence of lactic acid bacteria in milk, these compounds likely formed by Maillard reaction and lipid oxidation during thermal sterilization (Wang et al., 2025).

Esters in milk are primarily formed through the esterification of short-chain fatty alcohols and free fatty acids. As primary ester fraction constituents, lactones are typically generated through intramolecular cyclization of γ- and δ-hydroxy acids during thermal processing (Collins et al., 2003). Lactones contribute characteristic creamy, sweet, and coconut aromas, enhancing the flavor complexity and sensory quality of dairy products. Nine aroma-active lactones were identified across the five milk samples. Among them, γ-butyrolactone (fatty, greasy), δ-hexalactone (fatty, creamy), pantolactone (buttery), γ-nonalactone (fatty, peach, coconut), and δ-decalactone (buttery, peach, coconut) were consistently detected in all four processed samples, suggesting their potential importance as AACs. Nine lactones were detected in both the ID-UHT and INJ-UHT milk samples, contributing intense fatty and creamy aromas, primarily via the cyclization of γ- and δ-hydroxylated fatty acids during thermal sterilization. Additionally, δ-decalactone and δ-dodecalactone exhibited higher odor intensity in the RAW and PAST milk samples, in agreement with the findings of Schütt et al. (Schütt & Schieberle, 2017).

Acids, primarily medium- and short-chain fatty acids, are widely present in milk and are primarily generated via the hydrolysis of triglycerides by lipases. Due to their relatively high sensory thresholds, their direct influence on the overall flavor profile is limited (Iranmanesh et al., 2018). Nine aroma-active acids were identified across the five milk samples, with notably higher odor intensity in the RAW samples. These included acetic acid (sour, milky), butyric acid (buttery, sweaty), octanoic acid (stench, fermented), decanoic acid (fatty, rancid), dodecanoic acid (sour, fatty), and undecanoic acid (pungent, fruity), all exhibiting FD factors between 8 and 64, highlighting their significant contribution to the flavor profile of milk. Studies have shown that at low concentrations, these acids enhanced the characteristic milky aroma, while at higher levels, they imparted off-flavors such as rancidity or sweaty aromas (Elbarbary et al., 2025). As precursors to aldehydes and ketones, the fatty acids decreased significantly after thermal processing. Butyric acid, octanoic acid, and decanoic acid, which exhibited high FD factors (8–64), contributed notably to the aroma of the processed milks.

Alcohols are primarily derived via aldehyde reduction or enzymatic pathways (e.g., lipoxygenase activity), contribute to green/grassy and earthy attributes, but their high sensory thresholds limit their contribution to the overall flavor profile (S. Zhang et al., 2011). In this study, 1-octen-3-ol and 1-octanol, 3,7-dimethyl were detected in all milk samples. Of these, 1-octen-3-ol, which is formed via lipid oxidation or enzymatic pathways, typically imparts a mushroom aroma and possibly confers a metallic flavor at higher concentrations (Karahadian et al., 1985). The formation of 1-octanol, 3,7-dimethyl is primarily attributed to the metabolic conversion of feed-derived terpene in the bovine body, contributing herbaceous or citrus aroma. Its lipophilic nature facilitates its incorporation into the milk fat globule membrane, which is secreted into the milk (Opdyke, 1974). Preliminary studies suggested that citrus peel supplementation may selectively enhance the fresh flavor profile of milk. Limonene (citrus) was sniffed in the RAW, PAST, and INF-UHT milk samples, in agreement with the previous studies (Solano-Lopez et al., 2005). Additionally, styrene (metal, cardboard) and 2,4-di-tert-butylphenol (plastic, herbal) were sniffed in all the milk samples, primarily due to absorption or migration from packaging materials (D. C. Cadwallader et al., 2023).

3.4. Identification of AACs in milk using OAVs

Although the FD factor may offer initial insight into the odor activity, the effect of the matrix and the release kinetics of the odorants play a crucial role in determining their actual flavor contribution. To objectively evaluate the contribution of the 26 odorants with FD ≥8, the ES method was employed for accurate quantification (Table 2). A total of 21 AACs with OAVs ≥1 were identified, indicating their potential as key AACs in both RAW and thermally sterilized milk samples. Notably, certain compounds exhibited high FD factors but OAVs below 1. The primary reason for this divergence lies in the application of odor thresholds measured in water for OAV calculation, which may not account for the suppression of aroma release caused by milk matrix components such as fats and proteins. Additionally, ADA may overestimate the odor impact because GC-O releases nearly all volatiles for concentrated olfaction, ignoring the synergistic and masking effects produced by interactions between them (Li et al., 2025).

Significant differences were evident between the category distribution, quantity, and concentrations of the 21 potential key AACs (FD ≥ 8 and OAV ≥ 1) in the five milk samples (Fig. 4C). The INJ-UHT milk contained the highest number of potential key AACs (20), followed by the RAW milk (16), the PAST milk (16), ID-UHT milk (18), the INF-UHT milk (14). Twelve AACs consistently exhibited OAV ≥ 1 in all five milk samples, including 1-octanol, 3,7-dimethyl- (2.4–2.7), hexanal (9.4–11.4), octanal (2.2–8.3), nonanal (1.3–10.3), (E)-2-octenal (3.1–7.5), butyric acid (5.0–133.8), decanoic acid (2.6–55.1), undecanoic acid (10.7–88.7), γ-nonalactone (1.8–2.1), δ-decalactone (2.0–11.5), 2(5H)-furanone (28.7–42.1), and 2-pentadecanone (1.2–4.2). These findings agree with Wang et al. (Wang et al., 2025), which identified octanal, hexanal, nonanal, 2,3-butanedione, and butanoic acid as key AACs in PAST milk.

3.5. Quantification of major odor-active flavor compounds by SIDA

Although the ES method is commonly applied for volatile analysis, it is susceptible to matrix effects that can compromise accuracy. SIDA overcomes this limitation by employing isotopically labeled analogs with identical physicochemical properties, which helps eliminate matrix interference and ensure high precision. Therefore, SIDA are particularly suitable for trace-level quantification in complex food systems (Duensing et al., 2024). This study used SIDA to quantify eight major AACs (hexanal, octanal, nonanal, benzaldehyde, 1-octen-3-ol, butyric acid, octanoic acid, and decanoic acid) with FD ≥ 64 and OAV ≥ 1 in milk samples for the first time. RTs and selected ions information for each analyte and its isotope-labeled standard in MRM mode were listed in Table S2. Except for the acids (butyric, octanoic, and decanoic acid), all other target analytes were eluted later than their isotope counterparts, which was consistent with the reverse isotope effect reported by previous studies (Schmarr et al., 2012).

The calibration curves for the 8 isotope-labeled analytes exhibited excellent linearity (R² > 0.99), indicating the high detection sensitivity of the method (Table 3). The calibration curve slopes were used to accurately determine the concentrations of the 8 major AACs. Butyric, octanoic, and decanoic acids consistently exhibited the highest levels, particularly in the RAW milk. Differences were evident between the ES and SIDA quantitative results, which correlated with previous findings (Li et al., 2023). Notably, A approximately fourfold higher concentration of hexanal was observed with SIDA compared to external standard calibration, while minimal differences were evident between the octanal and nonanal levels. These variations may be attributed to multiple factors, one of which was the lower accuracy of the ES quantification method due to its reliance on a simulated milk matrix that may not adequately capture analyte–matrix interactions (Li et al., 2023).

Table 3.

Quantitative results for major AACs (FD ≥ 64 and OAV ≥ 1) in RAW and thermally sterilized milks by SIDA.

Aroma active compounds a	Linear equation	R2	Threshold b (μg/kg)	Concentration (μg/kg)					OAV
				RAW	PAST	ID-UHT	INF-UHT	INJ-UHT	RAW	PAST	ID-UHT	INF-UHT	INJ-UHT
Hexanal	y = 1.5334x −6.4971	0.9920	5	13.85 ± 0.10a	13.00 ± 0.20a	10.56 ± 0.15c	13.64 ± 0.42a	11.15 ± 0.19b	2.8	2.6	2.1	2.7	2.2
Octanal	y = 1.5936x −0.2987	0.9986	0.587	1.24 ± 0.02d	2.36 ± 0.08c	2.51 ± 0.03b	2.90 ± 0.05b	3.79 ± 0.17a	2.1	4.0	4.3	4.9	6.5
Nonanal	y = 8.0182x −0.8643	0.9974	1.1	3.93 ± 0.19d	5.61 ± 0.13b	14.47 ± 0.18a	4.41 ± 0.17c	5.48 ± 0.17b	3.6	5.1	13.2	4.0	5.0
1-Octen-3-ol	y = 2.2028x −6.3536	0.9915	1.5	3.38 ± 0.04c	3.29 ± 0.21d	3.52 ± 0.10bc	4.01 ± 0.04a	3.78 ± 0.04b	2.3	2.2	2.3	2.7	2.5
Benzaldehyde	y = 1.1758x −0.2069	0.9912	24	26.20 ± 1.84d	34.19 ± 0.38c	78.29 ± 2.57b	24.31 ± 0.68e	89.60 ± 3.89a	1.1	1.4	3.3	1.0	3.7
Butyric acid	y = 3.1214x + 0.2978	0.9835	240	3576.31 ± 43.66a	245.29 ± 6.92e	453.85 ± 16.77d	493.19 ± 4.76c	512.14 ± 10.88b	14.9	1.0	1.9	2.1	2.1
Octanoic acid	y = 2.0309x −0.1467	0.9928	910	1750.06 ± 33.65a	1119.83 ± 72.57b	1072.57 ± 7.37e	911.27 ± 39.47d	1051.08 ± 40.31c	1.9	1.2	1.2	1.0	1.2
Decanoic acid	y = 3.3064x + 0.8344	0.9807	130	645.32 ± 25.57a	263.01 ± 7.36b	155.53 ± 2.45d	217.71 ± 31.39c	239.17 ± 1.30bc	5.0	2.0	1.2	1.7	1.8

^aThe main odor active compounds with FD ≥ 64 and OAV ≥ 1.

^bOdor thresholds in water were obtained from literature. (Burdock, 2016; Chen, et al., 2017; Gemert, 2003).

3.6. Aroma recombination and omission experiments

3.6.1. Aroma recombination

Flavor perception is determined not only by the AAC types and concentrations, but also by their interactions (e.g., additive, synergistic, or masking effects). To validate the contribution of key AACs to the overall aroma profile of milk samples, aroma reconstitution models were conducted on five milks, further validating the accuracy of both the qualitative and quantitative analyses. AACs with FD ≥ 8 and OAV ≥ 1 were incorporated into a simulated system at their natural concentrations to form the target model. As shown in Fig. 2(B—F), the aroma reconstitution models successfully reproduced the primary aroma profiles of the five milk samples (91.9%–98.2%), although certain attribute intensities differed. In all the reconstituted samples, the intensities of the milky, fatty, and sweet attributes remained higher than those of other sensory attributes. The RAW reconstitution exhibited higher sweet and green/grassy intensities, with a slightly lower milky intensity (2.79) compared to the original (2.92), while the differences in the other sensory attributes were minimal (Fig. 2B). The PAST reconstitution showed slightly higher intensities for the milky (2.38), fatty (2.58), and cooked (1.42) attributes, but slightly lower intensities were evident for the oxidized and sweet attributes compared to the original (Fig. 2C). The ID-UHT reconstitution showed stronger fatty attributes (3.50), with minor reductions in the cooked, oxidized, and green/grassy attributes compared with the original, while the milky and sweet attributes showed minor differences (Fig. 2D). The INF-UHT reconstitution displayed a stronger fatty attribute (2.13), while the milky, sweet, and oxidized attributes were slightly weaker than in the original (Fig. 2E). The INJ-UHT reconstitution exhibited lower intensities across all sensory attributes compared with the corresponding original, with pronounced fatty, cooked, and oxidized attributes (Fig. 2F).

The observed discrepancies can be attributed to three factors: (i) SPME-arrow relies on partition coefficients, while adsorption selectivity is influenced by compound properties (e.g., molecular weight, boiling point, Henry's law constant, partition coefficient, and matrix affinity), leading to deviations between detected and actual concentrations (Jeleń & Wieczorek, 2023). (ii) The simulated matrix system cannot fully replicate the lipoprotein structure of native milk, while the matrix components in the simulation exert a certain masking effect that alters aroma release (Li et al., 2025). (iii) Aroma perception results from complex interactions among compounds, where synergistic and masking effects modulate sensory outcomes (Han et al., 2024). Despite these differences, the models confirmed that aldehydes, ketones, lactones, and fatty acids with high thresholds and concentrations contributed significantly to the milk aroma and validated the accuracy of the analytical results.

3.6.2. Aroma omission

Omission experiments were conducted based on the reconstitution models to determine the contribution of the AACs. Triangle tests showed that removing either a single AAC or an entire class of AACs (OAV ≥ 1) produced sensory variation. A total of 16 key AACs were identified across the five milk samples (Fig. 5). The elimination of all alcohols, aldehydes, acids, lactones, and ketones exhibited significant sensory differences (p < 0.05). In RAW milk, 7 key AACs exhibit significant differences. Among these, butyric acid, decanoic acid, and undecanoic acid showed extremely significant differences (p < 0.001), while octanoic acid, 1-octen-3-ol, hexanal, and (E)-2-octenal showed significant differences (p < 0.05), indicating the strong contribution of short- and medium-chain fatty acids to the milky aroma in the RAW milk. Similar results were obtained in PAST and INF-UHT models, where the omission of 1-octen-3-ol, hexanal, octanal, nonanal, butyric acid, γ-nonalactone, and 2(5H)-furanone resulted in significant sensory changes (p < 0.05). This was consistent with previous findings on the impact of sterilization temperature on the PAST milk aroma (Wang et al., 2025). The ID-UHT and INJ-UHT omission models contained 13 key AACs (p ≤ 0.05). In the ID-UHT model, the omission of nonanal, butyric acid, δ-decalactone, and 2,3-butanedione produced highly significant differences (p ≤ 0.01). In the INJ-UHT model, δ-decalactone and γ-dodecalactone showed extremely significant differences (p ≤ 0.001), while omitting octanal, (E)-2-octenal, butyric acid, γ-nonalactone, δ-dodecalactone, and 2,3-butanedione resulted in highly significant differences (p ≤ 0.01). These results highlighted the crucial role of lactones and 2,3-butanedione in the milky and fatty attributes of the INJ-UHT milk. Overall, differences in the AAC types and concentrations explained the variations in the aroma profiles of the milk samples. The omission experiments further confirmed the critical role of these key AACs in the milk aroma profiles.

Fig. 5.

The results of the triangle tests in the omission models conducted by 12 panelists. Significance: “***”, extremely significant (p ≤ 0.001); “**”, highly significant (p ≤ 0.01); “*”, significant (p ≤ 0.05); “-”, non-significant or not included in the model.

3.7. Conclusion

This study employed a molecular sensory science approach to systematically compare the aroma profiles and key aroma-active compounds (AACs) in raw milk and four types of thermally sterilized milk. QDA was employed to reveal that while PAST and INF-UHT milk aroma profiles were closer to raw milk, INF-UHT milk still displayed subtle differences in milky and fatty attributes. ID-UHT and INJ-UHT markedly enhanced the milky, fatty, and sweet attributes, increased cooked attribute and reduced green/grass attribute. Based on a self-built milk flavor database, 80 volatile compounds were detected, 45 of which were identified as odorants via GC-O-MS. Through the integration of FD factors, OAVs, and quantitative analysis, 21 potential key AACs were screened. Aroma recombination experiments successfully reproduced 91.9–98.2% of the overall aroma profile, and omission tests ultimately confirmed 16 key AACs. Among them, 1-octen-3-ol and butyric acid were identified as characteristic AACs common to all milk samples; octanal, nonanal, γ-nonalactone, and 2(5H)-furanone were characteristic of thermally sterilized milks, and 2,3-butanedione and δ-decalactone were identified as specific markers for ID-UHT and INJ-UHT milk. This study provides a comprehensive analysis of milk aroma, systematically validating the contributions of key AACs across different processing methods. It deepens the chemical and sensory understanding of milk flavor and establishes a theoretical basis for the precise flavor modulation of dairy products.

It should be acknowledged that the present study was conducted using milk from the same production batch. While this approach effectively controls for raw material variability and isolates the effects of sterilization techniques, it may limit the generalizability of the identified key aroma markers across different milk supplies. Future investigations incorporating samples from multiple origins and production periods are warranted to validate the robustness of these markers. Moreover, to advance a mechanistic understanding of flavor formation, subsequent work should focus on quantifying the reaction kinetics and defining the relationships between precursor and product underlying the generation of key AACs during thermal processing.

CRediT authorship contribution statement

Kunli Xu: Writing – review & editing, Writing – original draft, Software, Methodology, Data curation. Haoying Han: Writing – review & editing, Data curation, Conceptualization. Xiran Wang: Writing – review & editing, Data curation, Conceptualization. Xiaoli Zhang: Methodology, Data curation. Xiaochun Yang: Methodology, Data curation. Xiaoli Lu: Conceptualization. Zhaosheng Han: Conceptualization. Shunyu Wang: Supervision. Yanbo Wang: Supervision. Bei Wang: Writing – review & editing, Supervision, Funding acquisition, Conceptualization.

Ethics statement

The Research Ethics Committee of Beijing Technology and Business University approved the research protocol. Reference number No. (64) 2024.

Funding

This work was supported by the National Natural Science Foundation of China project (No.32572740; Beijing).

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.