Flavor modulation of fermented degummed coffee beans by ultrasound and ultra-high pressure: metabolomic and volatile profiles

Graphical abstract

Abstract

This study established a process in which physical pretreatments were applied after microbial fermentation and degumming, and evaluated how these treatments affected green coffee bean metabolite profiles and roasted-bean flavor. Relative to fermented green coffee beans without physical pretreatment (GF), ultrasound-pretreated fermented green coffee beans (GUS) and ultra-high-pressure-pretreated fermented green coffee beans (GUHP) showed significantly increased extractable contents of total sugars, reducing sugars, polyphenols, and flavonoids (P < 0.05). Metabolomic analysis identified 144 and 656 differential metabolites in GUS and GUHP, respectively, with amino acid metabolism and α-linolenic acid metabolism among the commonly enriched pathways. Compared with green beans, roasting increased the number of compounds of pyrazines and ketones by 13% and 16%, respectively. The two physical pretreatments were associated with different patterns of volatile composition in the roasted beans. GUS showed higher relative abundances of several furans and phenols, whereas GUHP showed greater changes in aldehydes, ketones, acids, and specific pyrazines. The three roasted groups had similar overall cupping scores but differed in the relative distribution of individual sensory attributes. These two physical pretreatments therefore produced different flavor-related compositional and sensory tendencies. Correlation analysis identified associations between volatile compounds and metabolites related to shikimate/phenylpropanoid metabolism, lipid metabolism, and amino acid derivatives. This study provides a theoretical basis for the precise modulation of roasted coffee flavor through the application of microbial fermentation and physical field pretreatment technologies.

1. Introduction

Coffee, along with tea and cocoa, is one of the three major beverages worldwide. It is a globally consumed drink made from roasted coffee beans [1]. According to the 2024 data from the Food and Agriculture Organization of the United Nations (FAO), the global commercial coffee production reached 10.891 billion kilograms in 2022. In the same year, the total export volume of coffee was as high as 8.466 billion kilograms, ranking first in global food export value and generating foreign exchange earnings of 46.397 billion US dollars [2]. The beneficial chemical components and unique sensory characteristics of coffee have always been research hotspots. Its final flavor quality depends not only on the bean variety and origin but is also significantly influenced by its post-harvest processing techniques [3].

In coffee post-harvest processing, fermentation is the core step in the wet processing method. The primary purpose of traditional fermentation is to degrade the mucilage layer (degumming) through microbial activities [4]. This process also significantly affects the composition of flavor precursors [5]. Compared with dry processing, wet-processed coffee generally exhibits more prominent fruity notes and acidity, with weaker bitterness, burnt flavors, and woody note. This difference largely stems from the accumulation of flavor precursor metabolites produced during microbial metabolism in the fermentation process [6]. Studies have shown that enzymes secreted by microorganisms such as yeasts and lactic acid bacteria can not only efficiently degrade coffee pectin but also generate volatile compounds such as alcohols, esters, and aldehydes, which are further converted into the characteristic aroma of coffee during subsequent roasting [7]. fermentation-related metabolites may not be uniformly distributed within coffee beans [8]. Some flavor-related compounds, including organic acids, amino acids, and sugars, may remain on the bean surface, in residual mucilage, or in tissues close to the bean surface [9], [10]. Conventional fermentation therefore provides only limited control over the distribution and subsequent transformation of flavor-related metabolites, restricting the precise regulation of green-bean composition and roasted-coffee flavor.

Non-thermal processing technologies, such as ultrasonic (US) treatment and ultra-high pressure (UHP) treatment, show broad prospects in the food processing field due to their unique advantages in improving food material structures and promoting component transformation [11], [12]. Currently, there have been studies on the effects of US and UHP on coffee extraction or flavor. For example, Ahmed et al. [13] prepared Saudi coffee nanoparticles through US treatment, resulting in a significant increase in caffeine and phenolic content, improved solubility, and better aroma, taste, and flavor in sensory tests compared with conventional coffee, without a noticeable increase in bitterness. However, most existing research on US and UHP in coffee focuses on the analysis of extracts or finished coffee beans [14]. Their application in the intervention stage of green beans after fermentation and degumming but before roasting, as well as a systematic revelation of metabolite changes and flavor formation during this process, remain to be elucidated. The introduction of targeted physical interventions such as US or UHP may open the structural cavities of coffee beans through mechanical effects, promote the penetration of internal substances and the redistribution of surface adherents [15], [16], thereby achieving regulation of metabolite accumulation in green coffee beans and promoting the formation of volatile flavors in roasted beans.

Based on the above background, this study used an optimized mixed-culture fermentation and degumming process as a upstream treatment and subsequently compared fermented coffee beans subjected to US and UHP pretreatment. Untargeted metabolomics was used to characterize the differential metabolite profiles of the green beans, while volatile compound analysis and sensory evaluation were performed after roasting. The objective was to determine the additional and differential effects of US and UHP within a fixed microbial-fermentation background and to identify potential associations between green-bean metabolites and roasted-coffee volatile profiles. This study hopes to provide innovative theoretical foundations and feasible process strategies for integrating biological fermentation and physical field regulation technologies to achieve precise improvement of coffee flavor quality and an increase in product added value.

2. Materials and methods

2.1. Raw materials and reagents

The fresh Catimor coffee cherries (Coffea arabica L. cv. Catimor) were provided by the Coffee Germplasm Nursery of the Institute of Tropical Economic Crops, Yunnan Academy of Agricultural Sciences (China, at an altitude of 1458 m, 25°09′N). Information about the raw materials: They were harvested in November 2023 (the coffee production season in Yunnan, China). Fully ripe cherries were selected (with more than 90% of the pericarp covered in dark red, classified as “ripe” according to the SCAA ripeness standard), and diseased, insect-infested, and unripe cherries were removed. Each batch of raw materials was collected randomly from multiple points in the same plot and then mixed (covering an area of about 0.5 ha). The total sample size was approximately 50 kg. After flotation to remove floating cherries, the sunken cherries were used for the experiments.

Saccharomyces cerevisiae (SC), Lactiplantibacillus plantarum (LP), and Lactobacillus acidophilus (LA) were purchased at the Guangdong Microbial Culture Collection Center (GIM 2.43, GDMCC 1.2685, GDMCC 1.412). Leuconostoc mesenteroides (LM) was purchased from the China Center of Industrial Culture Collection (CICC 21861) (Guangdong, China). Yeast Extract Peptone Dextrose (YPD) medium, de Man, Rogosa and Sharpe (MRS) Broth medium, Plate Count Agar (PCA) medium, and dipotassium hydrogen phosphate were purchased from Solarbio Science & Technology Co., Ltd. (Beijing, China). Pectin, citric acid, and sodium carboxymethyl cellulose were purchased from Shanghai Macklin Biochemical Co., Ltd. (Shanghai, China). D-(+)-Galacturonic acid and 3,5-dinitrosalicylic acid were purchased from Shanghai Yuanye Bio-Technology Co., Ltd. (Shanghai, China).

2.2. Preparation of coffee bean fermentation samples

SC and LM were separately inoculated into YPD medium and MRS broth medium, respectively, and cultured at 30℃ for 24 h. LP and LA were inoculated into MRS broth medium and cultured at 37℃ for 12 h. After three rounds of activation, the cultures were centrifuged at 10,000 rpm for 5 min at 4℃, washed twice with 0.85% sterile saline, and then resuspended in sterile saline for later use.

The fresh coffee beans were cleaned in clean water to remove the outer pericarp. One kilogram of depulped coffee beans was placed in a glass jar. For the microbial fermentation groups, the bacterial suspensions were inoculated at a microbial inoculum size of 10⁷ CFU/g. Unfermented coffee beans were used as the control group (named CK), in the single-strain fermentation groups (LP, LA, LM, SC), a single bacterial suspension was inoculated respectively. In the mixed-strain fermentation groups, for the MP group, LM and LP were inoculated at a ratio of 1:1 (v/v) with a total microbial amount of 10⁷ CFU/g. For the SM group, SC and LM were inoculated at a ratio of 1:1 (v/v); for the SP group, SC and LP were inoculated at a ratio of 1:1 (v/v); and for the SMP group, SC, LM, and LP were inoculated at a ratio of 1:1:1 (the final concentration of each strain was 3.33 × 10⁶ CFU/g).

2.3. Determination of enzyme activities during fermentative degumming

2.3.1. Determination of pectinase activity

The pectinase activity was determined using the 3,5-dinitrosalicylic acid (DNS) method [17]. Take 2 mL of the coffee bean fermentation broth that has been fermented for 24 h, centrifuge it at 10,000 rpm for 5 min at 4℃, and take the supernatant to obtain the crude enzyme solution to be tested. Add 5 mL of pectin substrate to a test tube, incubate it in a 40℃ water bath for 5 min, then add 4 mL of phosphate-citric acid buffer and 1 mL of diluted enzyme solution. For the test group, samples were vortexed thoroughly and incubated in a 40℃ water bath for 30 min. Concurrently, control group samples were heat-inactivated by immersion in a boiling water bath for 5 min. Following incubation, 2 mL aliquots were transferred to fresh tubes, 2 mL of distilled water and 5 mL of DNS reagent were added, mixed, and boiled for 5 min. The reaction mixtures were cooled to room temperature, diluted to a final volume of 25 mL with distilled water, and centrifuged at 3,600 rpm for 8 min. Absorbance of the resulting supernatant was measured at 540 nm. The pectinase activity in the sample was calculated according to the following formula (galacturonic acid standard curve: y = 0.8954x − 0.0297, R² = 0.9971):

$$\mathit{Pectinase\; activity}\text{ (u/g)=}\text{((}{\mathit{A}}_{\mathit{1}}\text{-}{\mathit{A}}_2\text{)×}{\mathit{D}}_{\text{r}}\text{×}\mathit{N}\text{)}/\text{(0.25×}\mathit{k}\text{×}\mathit{t}\text{)}$$

where A₁ is the OD₅₄₀ of the experimental group, A₂ is the OD₅₄₀ of the control group, N is the volume (mL) of diluted enzyme solution used in the enzyme activity assay, D_r is the dilution factor of the supernatant, k is the slope of the galacturonic acid standard curve, t is the reaction time (min).

2.3.2. Determination of cellulase activity

The preparation method of the sample solution to be tested is the same as that of the pectinase sample solution. The determination of cellulase activity was carried out with reference to Ferrari et al. [18] with slight modifications. Take 1.5 mL of carboxymethyl cellulose sodium solution and 0.5 mL of appropriately diluted enzyme solution into a 25 mL test tube. Incubate the mixture in a 40℃ water bath for 30 min. Then, immediately add 1.5 mL of DNS reagent. For the control group, first add 1.5 mL of DNS reagent, followed by 0.5 mL of the enzyme solution to be tested and 1.5 mL of CMC-Na solution. Boil the samples in a boiling water bath for 5 min. After cooling, dilute the solutions to 25 mL. Measure the absorbance values at 540 nm, denoted as A_s for the sample group and A_ck for the control group. $\mathrm{\Delta }$ A = A_s − A_ck according to the standard curve (y = 0.6611x + 0.0047, R² = 0.9972). The cellulase activity in the sample was calculated using the following formula:

$$\mathit{Cellulase}\mathit{activity}(\mathrm{u}/\mathrm{g})=(\mathrm{\Delta }A\times N\times 1000)/(0.5\times 30)$$

where N is the dilution factor, and $\mathrm{\Delta }$A is the glucose content calculated from the standard curve.

2.3.3. Evaluation of the fermentation system

Determination of Degumming Time: (1) Physical Touch Method: Samples were taken every 20 min. Coffee beans were grasped and rubbed by hand. Degumming was considered complete when the beans no longer felt sticky and exhibited a rough, frictional sensation. (2) Auxiliary Indicators: The viscosity of the fermentation broth was measured using an NDJ-5S rotational viscometer at 25℃, and the pH was continuously monitored. The end-point of degumming was confirmed when the viscosity dropped below 30% of the initial value and the pH fell below 4.0. The time required to reach this end-point was recorded as the degumming time.

pH Measurement: Samples were taken from each fermentation group at 24 h. The pH was measured using a digital pH meter (HI99161, HANNA, Italy) after calibration.

Microbial Growth Curve: For each group, 5 g of coffee beans were placed in a sterile centrifuge tube, to which 5 mL of 0.85% sterile saline was added. The mixture was homogenized, and then serially diluted using sterile saline. Appropriate dilutions of the bacterial suspension were spread-plated onto three types of media: PCA, MRS, and PDA. PCA and MRS plates were incubated at 37℃ for 48 h under constant temperature conditions. Chloramphenicol was added to the PDA medium to inhibit bacterial growth, and the plates were incubated at 28℃ for 5 days under constant temperature conditions. The total bacterial count, total lactic acid bacteria count, and total mold and yeast count were determined.

2.3.4. Optimization of fermentation conditions

Using degumming score and pH as evaluation indicators, a single-factor variable approach was employed to optimize the fermentation conditions. One of the following factors was varied while keeping the others constant: inoculum size (10⁵, 10⁶, 10⁷, 10⁸, 10⁹ CFU/g), fermentation temperature (24, 28, 32, 36, 40℃), and fermentation time (24, 32, 40, 48, 56 h). The fixed factors for these experiments were an inoculum size of 10⁷ CFU/g, a fermentation temperature of 32℃, and a fermentation time of 40 h. The degumming score was determined using Table S1.

2.4. Preliminary screening and selection of US and UHP conditions

US pretreatment was conducted under constant temperature conditions at 25℃ for 30 min, and the effects of different US power levels (0, 240, 360, 480, 600 W) on coffee bean composition were systematically investigated. In the study on the influence of UHP treatment on coffee bean components, the effects of varying two key parameters, pressure and time, were systematically examined. With the treatment time fixed at 20 min, the effects of different pressure levels (0, 150, 300, 450, 600 MPa) on component extraction were evaluated. Coffee beans after fermentation and depulping were divided into an UHP group and an US group. The coffee beans in both the UHP and US groups were vacuum-packed in low-density polyethylene bags, with water as the transmission medium for the respective US and UHP treatments. The UHP was applied at 0, 150, 300, 450, and 600 MPa, and the US treatment was performed at power levels of 0, 240, 360, 480, and 600 W, with treatment durations of 0, 10, 20, 30, and 40 min. The effects of these different factors on the total phenol content, flavonoid content, total sugar content, and reducing sugar content of the fermented coffee beans were investigated.

2.5. Determination of physicochemical indices

2.5.1. Determination of polyphenol content

The polyphenol content was determined using the Folin-Ciocalteu method [19]. Approximately 0.25 g of ground coffee powder (80 mesh) was mixed with 5 mL of 85% methanol and subsequently filtered. A standard curve was constructed using gallic acid standard solutions at concentrations of 0.00, 0.02, 0.04, 0.06, 0.08, and 0.10 mg/mL, with distilled water used as a blank. Then, 0.5 mL of either the standard solution or the sample filtrate was mixed with 0.5 mL of Folin-Ciocalteu reagent for 5 min. Following this, 1.0 mL of saturated sodium carbonate solution was added, and the mixture was allowed to react in the dark for 30 min. The absorbance was measured at 760 nm using a spectrophotometer. The standard equation was: y = 4.7767x + 0.0108, R² = 0.9979.

2.5.2. Determination of flavonoid content

The flavonoid content was determined using the aluminum chloride colorimetric assay [20]. A standard curve was prepared using rutin standard solutions (0.00, 0.02, 0.04, 0.06, 0.08, 0.10 mg/mL), with coffee extracts prepared identically to those for polyphenol analysis. Then, 1.0 mL of either the standard or sample solution was mixed with 0.5 mL of 5% NaNO₂ for 6 min. Subsequently, 0.5 mL of 10% Al(NO₃)₃ was added, and the mixture was vortexed for another 6 min. Afterwards, 4 mL of 4% NaOH was added, and the mixture was allowed to stand for 10 min. The absorbance was measured at 510 nm. The standard equation was: y = 0.9148x + 0.0024, R² = 0.9977.

2.5.3. Determination of total sugar content

The total sugar content was determined using the phenol–sulfuric acid method [21]. Approximately 1.0 g of ground coffee powder (80 mesh) was mixed with 20 mL of distilled water and filtered. A standard curve was constructed using glucose standard solutions (0, 0.02, 0.03, 0.04, 0.05, 0.06, 0.07, 0.08 mg/mL). Then, 2 mL of the standard solution, sample filtrate, or distilled water (blank) was mixed with 1.0 mL of 6% phenol solution and 5.0 mL of concentrated sulfuric acid. The mixture was shaken thoroughly, cooled, and allowed to stand at room temperature for 30 min. The optical density was then measured at 490 nm. The standard equation was: y = 9.7055x + 0.0018, R² = 0.9997.

2.5.4. Determination of reducing sugar content

The reducing sugar content was determined using the DNS method [22]. The sample preparation was identical to that for total sugar analysis. A standard curve was prepared using glucose standard solutions (0.0, 0.2, 0.4, 0.6, 0.8, 1.0 mg/mL). Then, 1.0 mL of the standard or sample solution was mixed with 2 mL of DNS reagent and heated in a boiling water bath for 2 min. After cooling under running water, the volume was adjusted to 15 mL with distilled water. The absorbance was measured at 540 nm. The concentration of reducing sugars in the samples was calculated from the standard curve (mg glucose/mL). The standard equation was: y = 0.606x + 0.046, R² = 0.9996.

2.6. Sensory evaluation of coffee beans

Sensory evaluation was conducted according to the Specialty Coffee Association of America (SCAA) cupping protocol [23]. Precisely 8.5 g of freshly ground coffee powder was weighed into a cupping bowl, followed by the addition of 150 mL of hot water at 93℃. The sensory panel consisted of ten trained members (five females and five males) aged between 20 and 35 years. The panel evaluated the coffee samples based on ten attributes: fragrance/aroma, flavor, acidity, body, aftertaste, balance, uniformity, cleanliness, sweetness, and overall impression. The total score for a single sample was the sum of the scores from all ten sensory attributes, with specific criteria for uniformity, cleanliness, and sweetness assigned according to SCAA standards (Table S2).

2.7. Metabolomics analysis

The pretreated green beans, designated as GF, GUS and GUHP. Approximately 20 mg of ground coffee sample was weighed into an EP tube. Then, 1000 µL of a pre-cooled extraction solvent (acetonitrile:methanol:water = 2:2:1, v/v/v) containing isotopically labeled internal standards (octanoic acid and glutamic acid) was added. The mixture was vortexed for 30 s, homogenized by ultrasonication at 35 Hz for 4 min, and then ultrasonicated in an ice-water bath for 5 min. This cycle was repeated three times. Subsequently, the samples were incubated at −40℃ for 1 h and centrifuged at 12,000 rpm for 15 min. The resulting supernatant was transferred to a new glass vial for analysis. Quality control (QC) samples were prepared by pooling equal volumes of supernatant from all samples.

Chromatographic separation was performed using a Vanquish UHPLC system (Thermo Fisher Scientific) equipped with a Phenomenex Kinetex C18 column (2.1 mm × 50 mm, 2.6 μm). The mobile phase consisted of (A) water containing 0.01% acetic acid and (B) a mixture of isopropanol and acetonitrile (1:1, v/v). The autosampler temperature was maintained at 4℃, and the injection volume was 2 µL. Mass spectrometry analysis was conducted on an Orbitrap Exploris 120 mass spectrometer controlled by Xcalibur software (version 4.4, Thermo Fisher Scientific) for both full MS and MS/MS data acquisition. Key parameters were: sheath gas flow rate, 50 arbitrary units (Arb); auxiliary gas flow rate, 15 Arb; capillary temperature, 320℃; full MS resolution, 60,000; MS/MS resolution, 15,000; normalized collision energy (NCE), stepped at 20, 30, and 40 eV; spray voltage, 3.8 kV (positive ion mode) or −3.4 kV (negative ion mode). The raw data files were converted to mzXML format using ProteoWizard software. Metabolite identification was performed using an in-house R package, with searches conducted against the BiotreeDB (V3.0) database. Data analysis was performed on the Lims2 cloud platform (https://biotree.lims2.com). All samples were analyzed in triplicate as independent biological replicates.

2.8. Volatile compounds analysis

For the GF, GUS, and GUHP samples, 500 g portions were taken and roasted using a small-batch drum roaster. After preheating the roaster to 180℃, the green coffee beans were charged into the machine, with the heat input set to level 1.5. Roasting proceeded according to a predefined time–temperature protocol, with the total roast duration controlled within 11 min. The first crack typically occurred around 9 min into the roast, after which roasting continued for approximately 1.5–2 min until the bean temperature reached 203 ± 2℃. Upon completion, the beans were immediately discharged and force-air cooled to room temperature, yielding the medium-roast coffees designated as fermented roasted beans (RF), ultrasonic pretreated roasted beans (RUS), and ultra-high pressure pretreated roasted beans (RUHP).

The volatile flavor compounds in the roasted coffee beans were analyzed using headspace solid-phase microextraction coupled with gas chromatography-mass spectrometry (HS-SPME-GC–MS). For each sample (RF, RUS, RUHP), 2 g of ground beans were placed in a 20 mL headspace vial. A 50/30 μm DVB/CAR/PDMS SPME fiber was exposed to the headspace of the vial, and the volatiles were extracted at 50℃ with agitation for 30 min. After extraction, the fiber was desorbed in the GC inlet at 250℃ for 5 min in splitless mode. GC–MS analysis was performed under the following conditions: a DB-Wax capillary column (30 m × 0.25 mm, 0.25 μm film thickness), carrier gas, helium at a constant flow rate of 1.0 mL/min, injector temperature, 250℃. The oven temperature was programmed as follows: held at 45℃ for 2 min, increased to 180℃ at 3℃/min, then raised to 240℃ at 10℃/min, and finally held at 240℃ for 7 min. MS conditions were: electron ionization (EI) source at 240℃, quadrupole temperature at 150℃, transfer line temperature at 240℃, electron energy, 70 eV, multiplier voltage, 1500 V, mass scan range (m/z), 33–400.

Compound identification combined mass spectral library matching and retention index comparison. (1) Mass spectral identification: The volatile compounds were tentatively identified by comparing their mass spectra with those in the NIST 2018 and Wiley spectral libraries. Compounds with a match factor greater than 800 were considered. The relative content of each compound was calculated using the peak area normalization method via data processing software. All analyses were performed in triplicate.

2.9. Statistical analysis

All data in this study are presented as the mean ± standard deviation of at least three independent replicates (n = 3). Statistical differences among multiple groups were evaluated by one-way analysis of variance (one-way ANOVA), followed by Duncan’s multiple range test for post hoc comparisons using SPSS software (version 23.0). Differences were considered statistically significant at P < 0.05. Data analysis and visualization were conducted on the MetaboAnalyst (https://www.metaboanalyst.ca/), the Lims2 cloud platform (https://biotree.lims2.com/home) and the Metware Cloud platform (https://cloud.metware.cn). Graphs were generated using Origin 2021 and GraphPad Prism 9.

3. Results and discussion

3.1. Determination of microbial degumming process

3.1.1. Screening of degumming microbial consortia

The coffee mucilage layer is primarily composed of pectin (approximately 30%), cellulose (about 8%), and non-cellulosic polysaccharides (approximately 18%). This gelatinous layer hinders the uniform drying of coffee beans and restricts the release of flavor compounds. Therefore, its effective removal is a critical step in coffee processing [24]. This study aimed to screen efficient microbial consortia for degumming through a strategy of combined-culture fermentation.

The pectinase and cellulase activities of different microbial strains after 24 h of fermentation are shown in Fig. 1A. Compared to the CK group, the fermentation system containing LM and SC showed a significant increase in pectinase activity (P < 0.05). Meanwhile, the system containing LM, SC, and LP exhibited significantly higher cellulase activity (P < 0.05). This is attributed to the organic acids produced by microbial metabolism, which lower the system pH. The acidic environment promotes the demethylation of pectin and chelates calcium ions, disrupting the intermolecular structure of pectin and thereby loosening the pectin polysaccharide network [25]. This increases substrate accessibility, creating more favorable conditions for the pectinase and cellulase secreted by the microorganisms, leading to efficient degumming.

Fig. 1.

Effects of multiple factors on enzyme activity, pH dynamics, and microbial growth characteristics during coffee fermentation. (A) Enzyme activities of single-strain fermentation; (B) Enzyme activities of combined-culture fermentation; (C) Effects of inoculum size on the pH and degumming score of the coffee fermentation system; (D) Effects of fermentation time on the pH and degumming score of the coffee fermentation system; (E) Effects of fermentation temperature on the pH and degumming score of the coffee fermentation system; (F) pH variation curves of the natural fermentation and combined fermentation systems; (G) Microbial growth curves in the natural fermentation system; (H) Microbial growth curves in the combined fermentation system. Note: CK, Unfermented coffee beans; LP, Lactiplantibacillus plantarum; LA, Lactobacillus acidophilus; LM, Leuconostoc mesenteroides; SC, Saccharomyces cerevisiae; SP, Co-culture of SC and LP; SM, Co-culture of SC and LM; MP, Co-culture of LM and LP; SMP, Mixed culture of SC, LM and LP. Different uppercase letters (A, B, C) indicate significant differences in pectinase activity among groups (P < 0.05); different lowercase letters (a, b, c) indicate significant differences in cellulase activity among groups (P < 0.05).

Based on the enzymatic profiles of individual strains, SC, LM and LP were selected for co-fermentation. As shown in Fig. 1B, the ternary consortium (SMP group) achieved a pectinase activity of 924.48 ± 80.44 U/mL, which was significantly higher than that of other combinations (P < 0.05). Compared with SC, LM, and LP monocultures, the SMP group showed increases of 31.54%, 38.75%, and 12.08%, respectively. These results demonstrate that under the conditions tested, combined cultivation enhanced pectinase production more effectively than single-strain fermentation. Previous studies have indicated that inoculating coffee fermentation with yeasts or other microorganisms can modulate microbial growth, substrate utilization, organic acid production, and the metabolite profile of coffee beans. Moreover, mixed microbial consortia may improve fermentation efficiency and stability by expanding the range of accessible substrates and forming complementary enzymatic systems [26], [27]. The elevated pectinase activity observed in the SMP group aligns with reports suggesting synergistic interactions in co-culture systems, potentially reflecting a cooperative enhancement effect. Owing to its superior enzymatic performance, the SMP consortium was selected for subsequent process optimization.

3.1.2. Optimization of fermentation process

After determining the optimal microbial consortium, we further optimized the inoculum size, fermentation time, and temperature to establish the best coffee degumming process. As shown in Fig. 1C, with increasing inoculum size, the pH of the fermentation system gradually decreased, while the degumming score first increased and then decreased. The degumming score reached its maximum value (9.33 ± 1.15) when the inoculum size was 10⁷ CFU/g. When the inoculum exceeded this threshold, the degumming score began to decline, possibly due to intensified competition for nutrients caused by excessive microbial density and the inhibitory effect of accumulated metabolic byproducts on microbial activity [28].

Fermentation time also significantly affected the system pH and degumming score (Fig. 1D). The degumming score gradually increased as fermentation proceeded. At 40 h, the system pH dropped to 3.39 ± 0.03, and the degumming score peaked at 9.67 ± 0.58. Beyond 40 h, the increase in the score plateaued. As shown by Lee et al. [29], excessively long fermentation times may lead to the accumulation of undesirable flavor compounds. Therefore, 40 h was determined as the optimal fermentation duration.

Temperature is another key factor influencing microbial activity. As shown in Fig. 1E, with increasing temperature, the system pH first decreased and then increased, reaching its lowest value of 3.35 ± 0.02 at 32℃. The degumming effect first strengthened and then stabilized with rising temperature, achieving the best result at 36℃. This indicates that within the suitable temperature range, microbial metabolism and enzymatic reaction efficiency are highest. When the temperature exceeded 36℃, it likely inhibited microbial activity, preventing further improvement in degumming efficiency.

Based on the results presented in Fig. 1C-E, the degumming score reached its maximum value of 9.33 ± 1.15 at an inoculum size of 10⁷ CFU/g. After 40 h of fermentation, the system pH decreased to 3.39 ± 0.03, and the degumming score attained 9.67 ± 0.58. Optimal degumming performance was achieved at 36℃. Consequently, an inoculum level of 10⁷ CFU/g, a fermentation duration of 40 h, and a temperature of 36℃ were established as the optimal parameters for the fermentation system employed in this study. Zhang et al. [30] reported that fermentation time, initial microbial load, and ambient temperature significantly influence microbial growth, organic acid production, system pH, and the chemical composition of coffee beans. In the present study, by controlling the inoculum size and temperature, a low pH and a high degumming score were achieved within 40 h, indicating that controlled inoculation enhances the consistency of the fermentation process.

3.1.3. Dynamic changes in pH and microbial community during fermentation

To characterize the kinetic profile of the optimized fermentation system, this study compared the pH evolution and microbial growth dynamics between the natural fermentation (NF) group and the SMP consortium fermentation group. As shown in Fig. 1F, the initial pH of both groups ranged from 4.50 to 4.70, followed by a progressive decline. The pH of the NF group dropped to 3.30 at 48 h, whereas the SMP group decreased more rapidly to 3.21 at 40 h, indicating that inoculation with the SMP consortium accelerated the acidification process under the conditions tested.

Microbial enumeration results are presented in Fig. 1G–H. In the NF group, the initial microbial count ranged from 10² to 10⁵ CFU/g, increasing rapidly during the first 16 h, declining slightly between 16 and 24 h, and subsequently stabilizing at approximately 10⁸ CFU/g from 24 to 48 h. lactic acid bacteria and yeasts emerged as the dominant culturable populations in the later stages. In contrast, due to exogenous inoculation, the SMP group maintained an initial count of approximately 10⁶ CFU/g and exhibited rapid growth within the first 32 h. Both yeasts and lactic acid bacteria peaked at 32 h, reaching 3.47 × 10⁸ and 3.58 × 10⁸ CFU/g, respectively, followed by a decline after 40 h.

During wet coffee fermentation, microbial communities dominated by lactic acid bacteria gradually prevail, with growth characterized by the consumption of mucilage sugars, accumulation of organic acids and other metabolites, and concomitant system acidification. Concurrently, fermentation duration, processing methods, and cultivar variety significantly influence microbial succession and the composition of metabolites such as sugars, organic acids, and amino acids, thereby altering the chemical profile and cup quality of green coffee beans [30]. The accelerated acidification and earlier microbial peak observed in the SMP group are consistent with previous reports suggesting that controlled inoculation modulates the microbial succession kinetics in coffee fermentation.

3.2. Preliminary physicochemical screening of US and UHP conditions

In this study, total sugar, reducing sugar, polyphenol, and flavonoid content were used as preliminary physicochemical screening indicators to evaluate the effects of different US and UHP treatment conditions on the composition of microbial-degummed green coffee beans. While these indicators reflect changes in extractable sugars and phenolic compounds following physical treatments, they do not fully represent the accumulation of specific flavor precursors or the ultimate flavor quality of roasted coffee. Therefore, the objective of this section is to identify representative treatment conditions for subsequent comparative analyses of metabolomics, volatile composition, and sensory profiling.

3.2.1. Effects of US treatment on total sugar, reducing sugar, polyphenol, and flavonoid contents in green coffee beans

The effects of different US power levels on the physicochemical parameters of green coffee beans are presented in Fig. 2A. Following US treatment, the measured contents of total sugar, reducing sugar, and polyphenols generally increased, although the response of these indicators to varying power levels was not uniform. Total sugar content exhibited an initial increase followed by a subsequent decrease with rising power, reaching its maximum value of 33.66 ± 1.10 mg/g at 360 W. Conversely, reducing sugar content peaked at 480 W, attaining a value of 10.63 ± 0.16 mg/g. Flavonoid content showed a slight decline after US treatment; however, no statistically significant difference was observed between the 480 W group and the untreated control (P > 0.05). Lee et al. [31] reported that, compared with the untreated Robusta control, roasted coffee prepared from green Robusta beans pretreated by 3% malic acid-assisted sonication (280 W, 1 h) showed marked increases in the relative levels of furfural and 5-methylfurfural, suggesting that organic acid-assisted US pretreatment of green coffee beans may alter the formation and composition of volatile compounds during subsequent roasting. From the perspective of US mechanism, periodic compression and rarefaction within the liquid sound-conducting medium induce acoustic cavitation, generating physical effects such as shock waves and localized shear forces. For packaged, microbially degummed coffee beans, these acoustic-field-related mechanical actions can be transmitted to the samples via the medium and packaging material. This transmission perturbs the surface tissue of the coffee beans and the residual mucilage/pectin network, thereby enhancing the release and apparent extractability of certain soluble components [32], [33]. Changes in reducing sugars may influence substrate availability for subsequent Maillard reactions; however, the final volatile composition ultimately depends on multiple factors, including amino acids, lipids, organic acids, and roasting parameters. At 480 W, the total sugar content was 31.91 ± 0.39 mg/g, showing no statistically significant difference compared with the maximum value recorded for the 360 W treatment (P > 0.05). Meanwhile, this condition maintained a relatively high reducing sugar content without inducing significant flavonoid loss. In contrast, a further increase in power to 600 W failed to yield consistent improvements across all four physicochemical indicators. Based on the overall changes in total sugar, reducing sugar, polyphenol, and flavonoid contents, the power level of 480 W exhibited a relatively balanced physicochemical composition within the investigated range. Therefore, it was selected as the representative US power for subsequent time-gradient screening.

Fig. 2.

Effects of physical pretreatment on the content of soluble components in green coffee beans. (A) Ultrasonic power; (B) Ultrasonic time; (C) Ultra-high pressure; (D) Ultra-high pressure time. Note: different letters (a, b, c, d) indicate significant differences in polyphenol, flavonoid, reducing sugar and total sugar among groups (P < 0.05).

As shown in Fig. 2B, the effects of varying treatment durations at a constant power of 480 W were evaluated. The measured contents of total sugar, reducing sugar, polyphenols, and flavonoids exhibited an overall trend of initially increasing followed by a subsequent decline, maintaining relatively high levels around the 10 min. Further extension of the treatment time resulted in a decrease across multiple indicators. Studies indicate that moderate cavitation facilitates tissue disruption and enhances mass transfer, whereas excessively prolonged treatment does not necessarily further increase the extractability of target compounds [34]. On one hand, sustained cavitation shock may further increase tissue permeability, promoting the migration of certain water-soluble components into the treatment medium. On the other hand, extended US duration is often accompanied by a rise in system temperature and radical-associated sonochemical processes during cavitation bubble collapse, which elevates the risk of oxidation or degradation of sensitive substances such as phenolic compounds [35], [36]. Therefore, 480 W, 10 min were selected as the representative US conditions for subsequent analysis.

3.2.2. Effects of UHP treatment on total sugar, reducing sugar, polyphenol, and flavonoid contents in green coffee beans

The effects of different UHP treatment levels on the contents of total sugar, reducing sugar, polyphenols, and flavonoids in green coffee beans are presented in Fig. 2C. As shown in Fig. 2C, as the pressure increased from 0 MPa to 300 MPa, the measured contents of all four indicators increased, reaching relatively high levels at 300 MPa. However, upon further increasing the pressure to 450 and 600 MPa, all four indicators exhibited varying degrees of decline. Consequently, 300 MPa was selected for subsequent holding time optimization. Studies have demonstrated that moderate pressure can alter the physical state and membrane permeability of plant tissues, and enhance the release or extractability of certain small-molecule components by affecting non-covalent interactions such as hydrogen bonding, hydrophobic effects, and ionic interactions [37], [38].

The influence of varying holding times at 300 MPa was further investigated, with the results shown in Fig. 2D. As the holding time was extended from 0 to 30 min, the contents of total sugar, reducing sugar, polyphenols, and flavonoids generally exhibited an upward trend, peaking at 30 min. When the holding time was further extended to 40 min, several indicators showed a decreasing trend. Appropriately prolonging the holding time may allow the pressure effects to be more fully exerted, thereby increasing the release and measured contents of certain soluble substances. Conversely, excessively long holding times may induce further changes in tissue structure and compositional stability, while also increasing the likelihood of losses in thermolabile compounds. Temperature fluctuations generated during the high-pressure process may also compromise the stability of components such as polyphenols and flavonoids [39].

Based on the comprehensive variations in total sugar, reducing sugar, polyphenol, and flavonoid contents, the conditions of 300 MPa for 30 min were selected as the representative UHP treatment parameters for subsequent metabolomic, volatile component, and sensory profiling analyses.

3.3. Metabolomic characteristics of microbially degummed green coffee beans and the regulatory effects of physical pretreatments

3.3.1. Overall metabolomic characteristics of microbially degummed green coffee beans

The top 20 metabolites in terms of relative content in the fermented green coffee beans are listed in Table S3. Among the metabolites with higher relative rankings, shikimic acid and phenylpropanoids account for a relatively high proportion. This suggests the activation of the phenylpropanoid metabolic pathway in green coffee beans induced by the fermentation process. These metabolites are key products of plant secondary metabolism and serve as direct sources for aroma precursors such as aldehydes and phenols, as well as for antioxidant components like polyphenols and flavonoids, including chlorogenic acid derivatives [40]. Changes in their content have a decisive impact on the final flavor quality and health attributes of coffee. Furthermore, the profile includes metabolites from other categories such as alkaloids, organic nitrogen compounds, fatty acids, and lipids and lipid-like molecules, which together constitute the core components of the metabolome in fermented green coffee beans.

3.3.2. Regulatory effects of US and UHP treatments on the metabolome of microbially degummed green coffee beans

To investigate the effects of US and UHP, both applied post-microbial degumming, on the metabolite profile of green coffee beans, metabolomic technology was employed to analyze the flavor precursor substances in the beans. Principal Component Analysis (PCA) was used to reveal the overall metabolic differences between groups and the variability within specific groups [41]. The PCA results are shown in Fig. 3A. The cumulative contribution rate of the principal components was 51.9% (PC1: 36.6%, PC2: 15.3%), which can reflect the overall variation in the data. Along PC1, the GUHP group showed clear separation from the GF group, indicating that UHP treatment significantly altered the metabolic profile of the coffee beans. In contrast, the separation trend between the GUS group and the GF group on the PCA score plot was relatively weaker. Orthogonal Partial Least Squares-Discriminant Analysis (OPLS-DA) has been proven effective in screening out irrelevant variables and identifying reliable markers for inter-group differences [42]. The OPLS-DA models for GF vs. GUS (Fig. 3B) and GF vs. GUHP (Fig. 3C) both displayed clear separation. The models yielded R²Y values greater than 0.9, with Q² values of 0.763 and 0.875 for the GUS vs. GF and GUHP vs. GF comparisons, respectively, indicating predictive ability. The OPLS-DA results demonstrate that both US and UHP processes substantially affected the metabolites in green coffee beans.

Fig. 3.

Overall regulatory effects of ultrasonic and ultra-high pressure treatments on the metabolome of fermented green coffee beans. (A) PCA; (B) OPLS-DA for GUS vs. GF; (C) OPLS-DA for GUHP vs. GF; (D) Venn; (E) Cluster heatmap.

To clarify the number of metabolic differences induced by US and UHP treatments, and to further elaborate on the differential regulatory effects of the different treatments on coffee bean metabolism, a Venn diagram was used to observe the differences in metabolite numbers between the different comparison groups (GUS vs. GF, GUHP vs. GF).

As shown in Fig. 3D, compared to the GF group, GUS and GUHP shared 301 common differential metabolites. This indicates that despite their different mechanisms of action, both physical treatments can commonly influence a series of fundamental metabolic processes in coffee beans. Among the differential metabolites, the GUS treatment regulated a total of 144, while the GUHP treatment resulted in 656 differential metabolites. This suggests that UHP treatment induced more pronounced changes in the metabolic network of coffee beans. Unlike UHP, which transmits relatively uniformly through samples primarily as hydrostatic pressure, cavitation activity and acoustic energy distribution within an US field are typically spatially heterogeneous. Particularly near interfaces between the transmission medium, packaging materials, and sample tissues, boundary conditions may influence the formation and transmission of cavitation bubble oscillations, local microjets, shock waves, and shear forces [43], [44]. Consequently, US and UHP likely affect the tissue of microbially degummed coffee beans and the release/transformation of their metabolites through physical effects that differ in spatial scale and mode of action. This may account for the distinct ranges of metabolite alterations associated with each treatment. However, as this study did not directly characterize acoustic field distribution, tissue microstructure, or in situ mass transfer processes, this interpretation requires further validation.

To better visualize the expression of differential metabolites from the different physical treatments, a cluster heatmap analysis was further conducted (Fig. 3E). The biological replicates for each sample clustered closely together, confirming good data homogeneity and high reproducibility. Furthermore, there were clear differences in the metabolite expression profiles of green coffee beans obtained through the two physical pretreatment processes, as reflected in the distribution of different colored blocks in the heatmap. In summary, different physical treatment methods all have a regulatory effect on the metabolic profile of fermented green coffee beans, which may further lead to the formation of distinct coffee flavor profiles.

3.3.3. KEGG metabolic pathway analysis

To investigate the underlying mechanisms by which US and UHP regulate the metabolites in fermented coffee beans, differentially expressed metabolites were annotated and analyzed using the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database (https://www.kegg.jp/kegg/pathway.html) [45]. Fig. 4A-B display the top 15 KEGG-enriched metabolic pathways for the different treatment groups. Compared to the GF group, the metabolic pathways activated in the coffee bean samples from the GUS and GUHP groups showed differences. Key regulatory pathways were further screened based on a P-value (P < 0.05).

Fig. 4.

Regulation of metabolic pathways in fermented green coffee beans by ultrasonic and ultra-high pressure treatments. (A-B) KEGG enrichment analysis; (C-D) KEGG pathway analysis.

Among the identified key pathways, the differentially abundant metabolites in the US treated coffee bean samples were primarily enriched in pathways related to alpha-Linolenic acid metabolism, Tyrosine metabolism, and Alanine, aspartate and glutamate metabolism. In contrast, the pathways activated in the UHP treated samples were mainly concentrated in Tyrosine metabolism, Histidine metabolism, Lysine biosynthesis, Arginine and proline metabolism, and alpha-Linolenic acid metabolism. Amino acid metabolism pathways were activated in coffee bean samples subjected to both US and UHP treatments, indicating that the generation and transformation of amino acids are crucial for the flavor alterations in coffee beans regulated by physical pretreatments. Notably, UHP treatment activated more amino acid pathways than US treatment, suggesting that UHP has a more substantial impact on the coffee bean metabolic network. This is may due to its hydrostatic pressure effect, which more comprehensively influences protein conformation, enzyme activity, and cell membrane permeability, thereby triggering metabolic responses involving the biosynthesis and transformation of various amino acids such as histidine, arginine/proline, and lysine [12]. This observation aligns with the result from the Venn diagram analysis, which showed a significantly higher number of differential metabolites in the GUHP group compared to the GUS group.

Furthermore, the alpha-Linolenic acid metabolism pathway was also activated in both pretreatment groups. Alpha-linolenic acid metabolism represents a key pathway for the oxidative degradation of unsaturated fatty acids. Its activation may be related to the promotion of lipid oxidation reactions by the cavitation effect of US and the alteration of membrane permeability by UHP. This reaction is an important precursor pathway for generating volatile flavor compounds such as aldehydes, ketones, and alcohols. The lipoxygenase/cyclooxygenase (LOX/COX) pathway oxidizes α-linolenic acid to hydroperoxides, which are then cleaved to produce aroma substances like C6-C9 aldehydes. The oxidation of unsaturated fatty acids produces lipid hydroperoxides, which are further cleaved to generate various flavor compounds including aldehydes, ketones, and alcohols [46].

Additionally, the KEGG pathway enrichment bubble diagrams (Fig. 4C-D) reveal that both pretreatment methods significantly enriched the Isoquinoline alkaloid biosynthesis pathway. The biosynthesis of isoquinoline alkaloids primarily concerns the metabolism of alkaloid components in coffee. Its activation may be related to the effect of physical treatments on the accumulation or transformation of related alkaloids. Alkaloids are an important category of flavor precursor compounds in coffee, which can degrade during roasting to produce nitrogen-containing heterocyclic compounds such as pyridines and pyrroles that contribute to roasted aromas [51].

In summary, both US and UHP pretreatments can differentially regulate the metabolite composition of coffee beans by activating specific pathways related to amino acid metabolism, lipid oxidation, and alkaloid synthesis. This provides distinct profiles of precursor substances for subsequent flavor formation pathways during roasting, such as the Maillard reaction, Strecker degradation, and lipid oxidation, ultimately shaping the flavor profiles of the differently roasted coffee beans.

3.4. Effects of pretreatments and roasting on volatile profiles and cupping attributes

3.4.1. Overall flavor characteristics of roasted beans

The appearance of coffee beans before and after roasting with different pretreatments is shown in Fig. 5A. To deeply investigate the effects of US and UHP on the volatile flavor compounds of roasted coffee beans, gas chromatography-mass spectrometry (GC–MS) was employed to separate and identify the flavor components. First, the differences in volatile compounds between green and roasted beans were compared (Fig. 5B). After roasting, the number of detected volatile flavor compounds increased from 110 to 129. Based on the proportional distribution of the numbers of detected compounds among different chemical classes, aldehydes, alcohols, acids, alkanes, and esters accounted for smaller proportions after roasting, whereas pyrazines and ketones accounted for larger proportions. The proportion of detected pyrazine compounds increased from 3% in GF to 16% in RF, corresponding to an increase of 13%. This change is consistent with the formation of nitrogen-containing heterocyclic compounds through Maillard-related reactions during coffee roasting. Under high roasting temperatures, amino acids such as aspartic acid and glutamic acid in the coffee beans undergo complex non-enzymatic browning reactions with reducing sugars. Strecker degradation can generate reactive intermediates that participate in the formation of pyrazines and related nitrogen-containing compounds, many of which are associated with nutty and roasted aroma notes [47]. Concurrently, the proportion of detected ketone compounds increased from 6% in GF to 22% in RF, corresponding to an increase of 16%. This compositional change may be associated with thermal reactions involving amino acid degradation and lipid oxidation during roasting [48].

Fig. 5.

Effects of ultrasonic and ultra-high pressure pretreatments on the volatile flavor profile and sensory quality of roasted coffee beans. (A) Appearance of coffee beans; (B) Changes in the relative content of volatile compound categories before and after roasting; (C) PCA of volatile compounds in fermented and physically pretreated coffee beans; (D) Relative content of various categories of volatile compounds in differently pretreated roasted beans (RF, RUS, RUHP); (E) Radar chart of SCAA cupping sensory scores.

3.4.2. Influence of US and UHP pretreatments on the flavor of roasted coffee beans

The PCA results for the volatile flavor profiles of roasted coffee beans subjected to different pretreatments are shown in Fig. 5C. The two principal components explained 24.7% and 42.32% of the flavor variance among the three treatment groups, respectively. The RF, RUS and RUHP groups showed clear separation trends.

The proportional distributions of the detected volatile compounds among different chemical classes in the three roasted-bean groups are shown in Fig. 5D. Phenolic compounds accounted for approximately 3% of the detected compounds in RUS, compared with 2% in RF and 1% in RUHP, indicating a slightly larger proportional representation of phenolic compounds in RUS [49]. This distribution may reflect differences in the relative representation of phenolic volatile compounds following US pretreatment. The larger proportion of detected phenolic compounds in RUS may indicate a greater representation of compounds associated with smoky, spicy, or medicinal aroma notes. The proportions of detected aldehyde compounds were approximately 8%, 6%, and 7% in RF, RUS, and RUHP, respectively. Thus, RUHP showed an intermediate proportional representation of aldehydes between RF and RUS. Aldehydes are commonly associated with floral, fruity, and grassy aroma notes [50]. The aldehydes detected in RUHP may contribute compounds associated with fresh, floral, and fruity aroma notes, although their proportional representation was not higher than that in RF. Ketones accounted for approximately 11%, 13%, and 12% of the detected compounds in RF, RUS, and RUHP, respectively. Compared with RF, the proportions of detected ketones were therefore 2% higher in RUS and 1% higher in RUHP. Ketones contribute creamy, caramel, and fruity aromas [49]. The slightly larger proportional representation of ketones in RUS may indicate a greater representation of compounds associated with caramel-like, creamy, and fruity aroma notes. Acid compounds accounted for approximately 11% of the detected compounds in both RF and RUHP and 10% in RUS. These results indicate that the proportional representation of detected acid compounds was comparable between RUHP and RF and slightly lower in RUS.

To further explore the sensory flavors of the three types of coffee beans, cupping was conducted following the SCAA protocol. The flavor radar chart is shown in Fig. 5E. RUS received the highest overall score (8.1), followed by RUHP (8.0) and RF (7.9). The radar chart shows that the cupping score differences among the three coffees are small. Although the overall scores were similar, differences were observed in the relative distribution of the individual sensory attributes. RUS showed slightly higher scores for uniformity, aftertaste, flavor, and balance, whereas RUHP showed slightly higher scores for aroma, body, sweetness, and acidity. These results suggest that the different pretreatments may have altered the relative balance of individual sensory attributes, possibly in association with their different volatile composition patterns.

The top 20 volatile flavor compounds by relative content are listed in Table 1. Pyrazine compounds are the most abundant volatiles in coffee, dominating the total relative content in all groups and contributing the core nutty and roasted aromas. Furfural accounted for 10.78 ± 5.35% of the relative abundance in RUS and 6.09 ± 0.57% in RUHP, with higher relative abundance in RUS. Because furfural is associated with caramel-like and sweet aroma notes, its higher relative abundance in RUS may partly contribute to differences in the aroma-related compositional tendencies between RUS and RUHP. Furfural, a representative furan compound, primarily contributes caramel-sweet characteristics [51]. Notably, UHP treatment led to a reshaping of the pyrazine composition. Trimethylpyrazine was not detected in RUHP, whereas 2-ethyl-6-methylpyrazine and 2-ethyl-5-methylpyrazine accounted for 2.42 ± 0.24% and 2.04 ± 0.20% of the relative abundance, respectively, and showed significantly higher relative abundances than in RF and RUS. These differences indicate that the selected UHP pretreatment was associated with a distinct relative distribution of pyrazine compounds after roasting. Acetic acid accounted for the highest relative abundance in RUHP (7.39 ± 2.98%) and the lowest proportion in RUS(5.60 ± 5.59%). Acetic acid, a short-chain fatty acid, contributes a sharp, pungent sourness. The relative abundance of acetic acid in RUHP may partly correspond to its slightly higher acidity score in the sensory evaluation.

Table 1.

Relative content of main volatile compounds and odor in coffee bean (TOP 20).

Name	CAS	rt	RF	RUS	RUHP	Odor
			Relative area (%)	Relative area (%)	Relative area (%)
Ketones
2-Propanone, 1-hydroxy-	116–09-6	567.573	2.353 ± 0.009	2.135 ± 0.155	1.583 ± 0.142	pungent sweet caramellic ethereal
2-Propanone, 1-(acetyloxy)-	592–20-1	706.226	2.061 ± 0.030	1.956 ± 0.110	1.512 ± 0.158	fruity buttery dairy nutty
2,3-Pentanedione	600–14-6	1381.59	1.334 ± 0.123	1.481 ± 0.142	0.913 ± 0.165	pungent sweet butter creamy caramel nutty cheese
Aldehydes
Butanal, 2-methyl-	96–17-3	186.191	5.176 ± 1.930	4.044 ± 3.577	3.895 ± 4.083	musty cocoa phenolic coffee nutty malty fermented fatty alcoholic
Acid
Acetic acid	64–19-7	694.066	6.859 ± 1.294	5.602 ± 5.586	7.391 ± 2.984	sharp pungent sour vinegar
Butanoic acid, 3-methyl-	503–74-2	857.172	3.057 ± 0.037	2.769 ± 0.055	2.033 ± 0.147	sour stinky feet sweaty cheese tropical
Furans
Furfural	98–01-1	708.95	8.979 ± 0.393	10.775 ± 5.351	6.092 ± 0.568	sweet woody almond fragrant baked bread
Ethanone, 1-(2-furanyl)-	1192–62-7	741.146	2.580 ± 0.061	2.689 ± 0.189	1.749 ± 0.113	sweet balsam almond cocoa caramel coffee
2-Furanmethanol	98–00-0	851.914	7.525 ± 0.157	7.040 ± 0.323	7.390 ± 2.708	alcoholic chemical musty sweet caramel bread coffee
2-Furanmethanol, acetate	623–17-6	762.144	1.480 ± 0.038	1.666 ± 0.119	0.804 ± 0.067	sweet fruit, banana, radish
Pyrazines
Pyrazine, methyl-	109–08-0	534.833	6.387 ± 0.239	7.212 ± 2.798	5.816 ± 0.639	nutty cocoa roasted chocolate peanut green
Pyrazine, 2,6-dimethyl-	108–50-9	590.462	6.139 ± 0.163	5.025 ± 0.263	5.894 ± 0.540	cocoa roasted nuts roast beef coffee
Pyrazine, 2,5-dimethyl-	123–32-0	584.861	5.729 ± 0.172	6.375 ± 2.98	5.708 ± 0.526	cocoa roasted nuts roast beef woody grass medical
Pyrazine, trimethyl-	14667–55-1	655.784	3.193 ± 0.118	2.723 ± 0.157	N/A	nutty nut skin earthy powdery cocoa baked potato roasted peanut hazelnut musty
Pyrazine, 3-ethyl-2,5-dimethyl-	13360–65-1	702.736	2.949 ± 0.095	2.467 ± 0.119	3.058 ± 0.290	potato cocoa roasted nutty
Pyrazine, ethyl-	13925–00-3	595.933	2.153 ± 0.021	1.80 ± 0.174	2.023 ± 0.173	peanut butter musty nutty woody roasted cocoa
Pyrazine, 2-ethyl-6-methyl-	13925–03-6	639.889	1.959 ± 0.041	1.708 ± 0.108	2.424 ± 0.242	roasted potato
Pyrazine, 2-ethyl-5-methyl-	13360–64-0	645.173	1.656 ± 0.021	1.374 ± 0.061	2.036 ± 0.202	coffee bean nutty grassy roasted
Others
Pyridine	110–86-1	450.559	2.311 ± 0.121	2.062 ± 0.112	1.960 ± 0.225	sickening sour putrid fishy amine
1H-Pyrrole-2-carboxaldehyde	1003–29-8	1094.44	2.547 ± 0.958	1.460 ± 0.053	1.162 ± 0.100	musty beefy coffee

3.4.3. Screening of flavor-differentiating compounds in roasted coffee beans based on OPLS-DA

Volatile compounds differentiating RUS and RUHP from RF were screened using OPLS-DA. As shown in Fig. 6A-B, both RUS and RUHP exhibited significant separation trends from RF, indicating that physical pretreatments have a substantial impact on the volatile compound composition of coffee beans. The OPLS-DA models yielded R²Y values greater than 0.9 and Q² values of 0.946 and 0.949 for RUS vs. RF and RUHP vs. RF, respectively, suggesting that the models adequately fitted the observed data and exhibited a certain degree of predictive ability. Using the screening criteria of VIP > 1, absolute FC > 1.5, and P < 0.05, 12 and 28 different volatile compounds were identified respectively in RUS vs. RF and RUHP vs. RF comparisons (Fig. 6C–D). The number of differential volatile compounds in the RUHP group compared to RF was much higher than that in the RUS group. This result aligns with the trend observed in the aforementioned metabolomics analysis, where the GUHP group had a far greater number of differential metabolites than the GUS group. Together, they indicate that UHP treatment exerts a more intense and extensive regulatory effect on the coffee bean metabolic network and the final synthesis of flavor compounds.

Fig. 6.

Screening of differential volatile flavor compounds and reshaping of the flavor profile in roasted beans pretreated with ultrasonic and ultra-high pressure. (A-B) OPLS-DA score plots and permutation tests for the comparisons of RUS vs. RF and RUHP vs. RF; (C-D) Cluster heatmaps of differential compounds screened based on OPLS-DA (P < 0.05); (E) Radar chart of relative content differences for six key flavor compound categories (pyrazines, furans, aldehydes, ketones, phenols, acids).

3.4.4. Heatmap and flavor radar analysis of flavor-differentiating compounds

The aforementioned differential compounds were analyzed using cluster heatmaps (Fig. 6C-D). Compared to RF, the volatile composition of roasted beans pretreated with UHP changed markedly. Its most significant feature was the substantial promotion of the accumulation of specific acids and ketones, along with an alteration in the composition of specific pyrazine compounds. Specifically, the contents of propanoic acid and 2-hydroxy-3-pentanone were extremely significantly upregulated in the RUHP group. Among these, propanoic acid contributes a cheesy, acidic aroma. Together with the relatively high proportion of acetic acid described in Section 3.4.2, the higher relative abundance of propanoic acid may contribute to the acid-related compositional characteristics of RUHP and may partly correspond to its slightly higher acidity score. 2-Hydroxy-3-pentanone is commonly associated with buttery and creamy aroma notes and may contribute to the corresponding aroma tendency of RUHP. Furthermore, the compositional profile of pyrazines changed significantly. Although compounds like 2-ethyl-6-methylpyrazine increased, the relative proportions of key pyrazines or specific compounds were transformed. This suggests that UHP treatment may affect the relative pathways or product distributions of the Maillard reaction and Strecker degradation, thereby altering the balance of pyrazine-related aroma compounds.

In contrast, US pretreatment exhibited a different regulatory pattern. The volatile-composition changes in RUS were mainly characterized by higher relative abundances of several compounds associated with sweet, roasted, and smoky aroma notes. Furaneol, which imparts caramel-sweet characteristics, and 3-methylphenol, which contributes smoky and medicinal notes, were both extremely significantly upregulated in the RUS group. Unlike the pyrazine composition in RUHP, RUS promoted the generation of pyrazines such as 2-isobutyl-3-methylpyrazine, which may contribute to roasted aroma characteristics. Additionally, regarding acidic compounds, the differential substances in the RUS treatment generally showed inhibitory effects. For example, the contents of nonanoic acid and benzoic acid were significantly downregulated, which is consistent with the result of the decreased total proportion of acids in RUS mentioned in Section 3.4.2.

3.4.5. Flavor radar analysis of differential compounds in pretreated roasted coffee beans

To more intuitively demonstrate the differential shaping of the overall coffee flavor profile by different pretreatments, this study selected six key flavor compound categories decisive for coffee sensory quality. These include: pyrazines (nutty, earthy, roasted aromas), furans (contributing roasted sweetness), aldehydes (presenting floral/fruity sweetness), ketones (imparting buttery, caramel, and fruity notes), phenols (providing smoky flavor), and acids (sourness, cheesy notes).

The total relative content change of differential volatile compounds (VIP > 1, FC > 1.5, P < 0.05) within each category for each group compared to the RF group was calculated. A radar chart was plotted for visual analysis. The results are shown in Fig. 6E. Compared with RF, US and UHP pretreatments produced different patterns in the relative distribution of aroma-related volatile compounds. These compositional differences were accompanied by different emphases in the individual sensory attributes. RUS group showed a clear increase in the phenol dimension, indicating a greater relative contribution of compounds commonly associated with smoky aroma notes. In contrast, RUHP showed greater increases in the acid and aldehyde dimensions, indicating a relatively greater contribution of compounds associated with acidic, floral, and fruity aroma notes. This compositional tendency was also broadly consistent with the slightly higher aroma and acidity subscores of RUHP.

3.5. Correlation analysis between green bean metabolites and roasted bean volatile flavor compounds

3.5.1. Correlation between dfferential metabolites and differential volatile flavor compounds in US treated samples

To explore potential associations between changes in green-bean metabolites and roasted-bean volatile compounds following physical pretreatments, correlation analysis was performed using the differential metabolites and differential volatile compounds identified in the corresponding comparisons (GUS vs. GF, GUHP vs. GF) and differential volatile flavor compounds (RUS vs. RF, RUHP vs. RF). The correlation calculation was based on the average relative content of biological replicates for each group. Metabolite–volatile compound correlated substance pairs with an absolute correlation coefficient |r| > 0.9 and P < 0.05 were retained to construct metabolite-flavor compound regulatory networks (Fig. 7A-B). Fig. 7A shows the correlation network for US treated samples. As displayed, the differential metabolite nodes primarily originated from shikimic acid and phenylpropanoids, polyketides, lipids, and amino acids/peptides. The differential flavor compound nodes were concentrated in furans, phenols, and pyrazines.

Fig. 7.

Association networks between differential metabolites in green beans and differential volatile flavor compounds in roasted beans under physical pretreatments. (A) Pearson correlation network between differential metabolites from ultrasonic treatment (GUS vs. GF) and differential volatile flavor compounds (RUS vs. RF); (B) Pearson correlation network between differential metabolites from ultra-high pressure treatment (GUHP vs. GF) and differential volatile flavor compounds (RUHP vs. RF).

Several differential volatile compounds showed associations with changes in specific green-bean metabolites. Regarding the regulation of furan compounds, furaneol, a key substance imparting caramel sweetness, showed a strong negative correlation (r < -0.9) with metabolites from the shikimic acid pathway, such as benzoic acid, and polyketides like p-toluquinone. This inverse correlation indicates that the relative changes in furaneol occurred in the opposite direction to those of benzoic acid and p-toluquinone under the selected treatment conditions. It suggests a potential association between changes in shikimate/phenylpropanoid- or polyketide-related metabolites and the furan volatile profile. Phenolic aroma substances like 3-methylphenol exhibited strong correlations with various lipid molecules; for instance, a strong negative correlation with the phospholipid PC(14:1(9Z)/P-16:0) was observed, indicating coordinated but inverse changes between this lipid-related metabolite and 3-methylphenol. This association suggests that alterations in lipid-related metabolism may accompany changes in the phenolic volatile profile. In summary, US pretreatment was associated with changes in metabolites related to the shikimate/phenylpropanoid pathway, lipid metabolism, and amino acid metabolism, and these changes co-occurred with variations in several furans, pyrazines, and phenolic volatile compounds. These associations identify candidate metabolite–volatile relationships for further investigation.

3.5.2. Correlation between differential metabolites and differential volatile flavor compounds in UHP treated samples

Based on the correlation analysis (|r| > 0.9) of differential metabolites from UHP treatment and differential volatile flavor compounds from roasted beans, the results are shown in Fig. 7B. Ketone and acid volatile compounds were highly represented among the nodes and connections in the correlation network. The accumulation of ketones is closely associated with changes in various structural metabolites. 2(5H)-Furanone, which imparts a buttery aroma, showed a negative correlation with 7-Keto-8-aminopelargonic acid (containing a fatty acid chain structure), indicating that the two compounds changed in opposite directions under the selected treatment conditions. Concurrently, 2-hydroxy-3-pentanone, which has a creamy, fruity note, showed a strong positive correlation with the benzene ring derivative vanillic acid diethylamide, indicating changes in the relative abundances of the two compounds. This positive correlation suggests a potential association between benzene-ring-containing metabolites and changes in the ketone-related volatile profile [52]. The regulation of pyrazine compounds was significantly correlated with various alkaloids, amino acids, and benzene ring derivatives. 2-Ethyl-6-methylpyrazine showed strong positive correlations with the amino acid derivative tyrosylglutamic acid and the alkaloid normorphine, indicating that their relative abundances changed in parallel under UHP treatment. These associations identify amino acid derivatives and alkaloid-related metabolites as candidates for further investigation.

In summary, UHP treatment was associated with changes in multiple metabolite classes, including lipids, benzene-ring-containing compounds, alkaloids, and organic acid derivatives. These changes co-occurred with variations in several ketone, acid, and pyrazine volatile compounds. The observed correlations provide candidate relationships for future validation.

3.6. Mechanistic insights into the formation of different coffee flavors by US and UHP pretreatments

The aforementioned results indicate that the two physical pretreatment technologies, US and UHP, regulate the metabolic network of fermented green coffee beans through different pathways, consequently forming distinct flavor profiles after roasting. The flavor differences are directly related to the physical effects and metabolic responses triggered by US and UHP. This study confirmed that US and UHP, under appropriate conditions, can effectively promote structural changes in coffee beans, thereby facilitating the dissolution and release of key flavor precursors such as total sugars, reducing sugars, and polyphenols (Section 3.2.1). Sugars are core participants in the Maillard reaction, interacting with amino acids to generate pyrazine and furan compounds. Polyphenols are not only important antioxidant components but can also be transformed or degraded under heat, contributing to coffee's color, astringency, and part of its smoky flavor [53].

The cavitation effect generated by US treatment can disrupt cellular structures [54]. Metabolomic analysis further revealed that the differential metabolites induced by GUS treatment were primarily enriched in pathways such as alpha-linolenic acid metabolism and tyrosine metabolism. Correlation analysis suggests that US treatment likely provides key precursors for the Maillard reaction by affecting amino acid metabolism, thereby contributing to the formation of furan and pyrazine compounds after roasting, enhancing the coffee's classic sweet, roasted, and nutty flavors. In contrast to the localized mechanical disruption caused by US, UHP acts uniformly on biomacromolecules through hydrostatic pressure, preferentially disrupting non-covalent bonds such as hydrogen bonds and hydrophobic interactions. This leads to the dissociation or conformational changes of macromolecular structures in coffee [55], explaining the result that the number of differential metabolites in GUHP was 4.5 times that in GUS. Furthermore, it activated multiple amino acid metabolism pathways, including histidine, arginine/proline, and lysine biosynthesis. Correlation network analysis confirmed that UHP treatment of green coffee beans was associated with alterations in various precursors like lipids, phenylpropanoids, and alkaloids. These were subsequently converted or generated into ketone and acid compounds after roasting, accompanied by a restructuring of the specific pyrazine compounds. These changes may help explain why RUHP showed a relatively greater contribution of acid-, ketone-, and specific pyrazine-related compounds, together with slightly higher aroma, body, sweetness, and acidity subscores than RUS.

Based on the mechanistic differences described above, this study proposes a physical-field flavor regulation strategy for coffee. US pretreatment, by strengthening the supply of classic Maillard reaction precursors, is suitable for product development targeting traditional roasted sweetness (caramel, nutty, roasted) coffee flavors. UHP pretreatment, due to its distinct acidity, is more suitable as a raw material for specialty coffee or milk coffee bases. Although coffee cup evaluation score results showed a slightly higher overall score for RUS, the statistical difference was not significant (P > 0.05), indicating that both technologies can achieve comparable quality levels. The key lies in their selective application based on the target flavor profile.

4. Conclusion

This study demonstrated that US and UHP pretreatments exerted different additional effects on the metabolite composition of fermented green coffee beans and the volatile profiles of the corresponding roasted beans. UHP induced broader metabolomic changes than US, while the two treatments produced different distributions of aroma-related volatile compounds after roasting. Although RUS and RUHP achieved comparable overall cupping scores, RUS showed relatively higher scores for uniformity, aftertaste, flavor, and balance, whereas RUHP showed relatively higher scores for aroma, body, sweetness, and acidity, indicating different sensory attribute tendencies within a similar overall quality level. The incorporation of US or UHP after standardized microbial fermentation therefore provides a feasible processing framework for modulating roasted-coffee flavor and offers a practical basis for selecting downstream physical treatments according to the desired sensory emphasis. However, because non-fermented and physical-treatment-only controls were not included, the independent contributions of microbial fermentation and physical pretreatments, as well as their potential interactions, could not be determined and require further verification.

CRediT authorship contribution statement

Zilin Wang: Writing – original draft, Visualization, Formal analysis. Changyuan Li: Methodology, Formal analysis, Data curation. Yu Zhou: Methodology. Qimin Xie: Formal analysis. Jiahe Dai: Project administration, Investigation. Jia Liu: Writing – review & editing, Funding acquisition. Yang Tian: Writing – review & editing, Supervision, Funding acquisition. Liang Tao: Writing – review & editing, Funding acquisition, Conceptualization.

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 material

The following are the Supplementary data to this article: