Lipidomic profiling provides insights on Arabica coffee flavor diversity in different postharvest processing methods

Abstract

The processing methods of coffee cherries after harvesting can significantly affect the lipid composition of green coffee beans, thereby influencing their in-cup properties. This study utilized ultra-high performance liquid chromatography-electrospray ionization tandem mass spectrometry (UPLC-ESI-MS/MS) technology to investigate the impact of natural, washed, and honey processing methods on the lipid composition of green coffee beans, accompanied by sensory evaluations. A total of 510 lipids, covering 27 subclasses, were detected. Of these, 150 lipids showed significant differences before and after processing, and 37 lipids were identified as potential biomarkers for distinguishing the three processing methods. Kyoto Encyclopedia of Genes and Genomes (KEGG) metabolic pathway analysis revealed that glycerophospholipid metabolism was the key pathway for the formation of differential lipids among the various processing methods. Significant correlations were observed between lipid composition and the flavor diversity associated with different processing methods. The findings provide a crucial basis for understanding the lipid transformation of coffee during different processing methods and its impact on flavor quality characteristics.

Graphical abstract

Highlights

• •Different coffee processing methods significantly impact lipid profiles in green beans.
• •37 lipid biomarkers distinguish natural, washed, and honey processing methods.
• •Glycerophospholipid metabolism as the key pathway for lipid differences.
• •Lipid differences correlated significantly with flavor diversity from processing methods.

1. Introduction

In recent years, the coffee market has expanded rapidly, with global coffee consumption growing steadily at an annual rate of 3.50 % (Magalhães Júnior et al., 2021). The popularity of coffee is largely due to its complex, pleasant flavor and the stimulating effect of caffeine (Wang et al., 2022). Driven by the “Third wave of coffee”, consumer demand for high-quality coffee has increased, prompting significant changes in the industry and transforming coffee from a traditional commodity to a specialty product that emphasizes both flavor and personalization (Samoggia and Riedel, 2018). Coffee quality is determined by a complex process, with approximately 40 % influenced by field production, 40 % by post-harvest primary processing, and 20 % by secondary processing (Musebe et al., 2007; Hameed et al., 2018). This ratio underscores the critical role of post-harvest processing technology in shaping coffee quality.

Coffee beans, harvested wrapped inside cherries, undergo processing that aims to remove the pectin-rich pulp either before or after drying, resulting in green coffee beans with a moisture content of 10–12 % (Zhai et al., 2024). Common processing methods include natural/dry processing, washed/wet processing and honey/semi-dry processing (Febrianto and Zhu, 2023; Wang et al., 2023; de Melo Pereira et al., 2019). In the natural processing, the whole cherries are directly dried and then peeled to obtain green coffee beans. In the washed processing, the cherries are depulped and then fermented in water. The fermented beans are washed, dried and peeled. The honey processing that merges natural processing and washed processing, and the cherries are directly dried after depulping. These methods influence the microbial dynamics and metabolic activities within the coffee beans, significantly affecting their chemical composition and sensory attributes (Cortés-Macías et al., 2022; van Mullem et al., 2022). The chemical composition of coffee beans, especially lipid compounds, which act as flavor precursors, has an important influence on the flavor and sensory properties of coffee. Recent studies by Banti and Atlaw (2024) and Richard et al. (2020) indicate that while the content of chlorogenic acid, caffeine, and trigonelline remains relatively stable, lipid levels are significantly altered by the choice of processing method.

Lipids are major constituents of coffee beans, predominantly located in the endosperm, and are primarily composed of triacylglycerols (75.2 %) and esters of diterpene alcohols and fatty acids (18.5 %), playing a crucial role in determining coffee quality (Speer and Kölling-Speer, 2006). Among the globally cultivated coffee species, Arabica (Coffea arabica) and Robusta (C. canephora) show notable differences in lipid content. Arabica coffee beans, known for their smooth flavor and high aromatic quality, contain approximately 16.0 % lipids. In contrast, Robusta coffee beans, which are typically stronger and more bitter, have a lower lipid content of about 10.1 % (Dong et al., 2021; Wang et al., 2021). Lipids not only affect the aroma, flavor and body of coffee, but also play a pivotal role in determining the storage stability and quality retention of coffee (Zhu et al., 2023; Pazmiño-Arteaga et al., 2022). Furthermore, green coffee oil, extracted from coffee beans and rich in diterpenoids, fatty acids, and unsaponifiables, is highly valued in the cosmetics industry for its antioxidant and chemopreventive properties (Dong et al., 2023). The impact of processing methods on lipid composition is critical for developing active ingredients in coffee oil, enhancing processing techniques, and improving coffee quality. Despite this, current research has predominantly focused on how fatty acids or certain lipid classes as a group change with different processing methods (Banti and Atlaw, 2024; Richard et al., 2020; da Silva et al., 2022). There is a pressing need for comprehensive lipidomic studies that explore these variations in detail and assess their effects on coffee's final flavor profile.

Lipidomics, a specialized branch of metabolomics, has emerged as a valuable tool in food science for exploring the biochemical basis of flavor, quality differentiation, and product traceability. Recent studies have applied lipidomic profiling to various food systems-including meat, grains, and fermented products-to understand lipid oxidation, precursor formation, and flavor development (Harlina et al., 2023; Shi et al., 2024; Zhang et al., 2024). Lipidomics uses advanced mass spectrometry techniques such as liquid chromatography-mass spectrometry (LC-MS) to analyze the chemical properties of lipids (Li et al., 2020). Although studies have used lipidomics to analyze the composition and species of coffee and track changes in its origin and roasting process(Silva et al., 2020; Silva et al., 2022; Aurum et al., 2022; Zhu et al., 2023; Wang et al., 2025), systematic studies on the effects of different processing methods on the lipid composition of coffee have not yet been carried out. This study combines electrospray ionization (ESI) and tandem mass spectrometry to improve the accuracy and reproducibility of detection by minimizing fragmentation in sample analysis. This approach enables us to comprehensively evaluate the changes in lipids under different processing conditions and reveal their role in the formation of coffee flavor.

We used ultra-high performance liquid chromatography-electrospray ionization tandem mass spectrometry (UPLC-ESI-MS/MS) technology to study the types of lipid molecules in coffee beans and their changes under different processing methods. By combining lipidomics with chemometric methods, we provided a detailed description of how processing methods affect the lipid profile of coffee beans. Additionally, we evaluated the relationship between lipid changes and coffee flavor quality through cupping analysis of samples from each processing method. The results of this study not only fill the gap in lipidomics research in coffee processing, but also provide new insights into how lipid composition may serve as a molecular marker of processing-induced flavor variation.

2. Materials and methods

2.1. Samples preparation

The study was conducted during the 2023/2024 harvest season at the Dehong Tropical Agriculture Research Institute in Yunnan, China, located at 97^o86ʹ10ʺ E, 24^o02ʹ57ʺ N, 796.4 m above sea level. Coffee trees from the Sarchimor series of C. arabica L., planted in 2017 with a density of 2000 trees/ha and a spacing of 2.5 m × 2 m, thrive in the region's subtropical humid monsoon climate, characterized by an average annual temperature of 21 °C and precipitation of 1394.8 mm. The soil, brick red and derived from granite, has a pH of 4.9.

During the harvest period, fully ripe red coffee cherries were picked. In the manual selection process, unripe green cherries, overripe black cherries, and foreign matter were removed, and floating cherries were separated by flotation. Next, approximately 500 g of cherries were taken; the peels were mechanically removed, and pulped fruits were freeze-dried until the moisture content of the green beans reached about 11 % (G600 Moisture Tester, GEHAKA, Brazil), which served as the control sample (CK). The remaining samples underwent three postharvest processing treatments (natural, washed, and honey) following a completely randomized design (CRD). Each treatment was repeated three times, using about 5 kg of coffee cherries each time. The processing flow is outlined in Table 1: (1) Natural processing (N): The selected coffee cherries were placed on a drying bed for natural drying. (2) Washed processing (W): The coffee samples were pulped, soaked, and fermented for 36 h, then washed and dried on a drying bed. (3) Honey processing (H): The coffee samples were placed directly on the drying bed for drying after pulping. Sun-drying was conducted under ambient conditions, with temperatures ranging from 11.5 °C to 24.8 °C, relative humidity of 61.2 ± 8 %, and an average wind speed of 0.81 m s⁻¹. During the drying process, the samples were regularly turned to ensure uniform moisture loss. Due to differences in initial moisture content among the postharvest treatments, the drying duration varied accordingly: N, 19 days; W, 9 days; and H, 10 days. The final moisture content of the shelled beans was controlled to remain between 10 % and 12 %.

Table 1.

Comparison of primary processing methods for preparing green coffee beans.

2.2. Chemicals and reagents

Acetonitrile (ACN), methanol (MeOH), isopropyl alcohol (IPA), dichloromethane (CH₂Cl₂), and methyl tert-butyl ether (MTBE) were sourced from Merck (Darmstadt, Germany). Formic acid (FA) and ammonium formate (AmFA) were acquired from Sigma-Aldrich (St. Louis, MO, USA). Lipid standards (Table S1) were obtained from Sigma-Aldrich (Merck KGaA, Darmstadt, Germany) and Avanti Polar Lipids (Alabaster, AL, USA).

2.3. Lipid extraction

Lipids were extracted using a MeOH and MTBE solvent system. A sample of 20 mg was ground and placed in a 2 ml centrifuge tube. It was homogenized with a 4 mm steel ball at 30 Hz for 20 s using a mixer (Retsch, MM400, Germany). Then, 1 ml of lipid extract (MTBE:MeOH = 3:1, v/v), containing the internal standard mixture (Table S1), was added, and the tube was vortexed for 30 min (Jingxin, MIX-200, China). After adding 300 μL of ultra-pure water, the tube was swirled for 1 min and allowed to rest at 4 °C for 10 min. The mixture was centrifuged at 12,396×g (Eppendorf 5424R, Germany) for 3 min at 4 °C. Subsequently, 400 μL of the supernatant was transferred to a new 1.5 mL tube and concentrated to dryness at −20 °C using a vacuum concentrator (LABCONCO, CentriVap, USA). The dried residue was reconstituted in 200 μL of a lipid complex solution (acetonitrile (ACN):isopropanol (IPA) = 1:1, v/v), vortexed for 3 min, and centrifuged again at −20 °C at 12,396×g for 3 min. Finally, 120 μL of the resulting solution was collected for liquid chromatography-mass spectrometry analysis.

2.4. Lipid analysis

Sample extracts were analyzed using a UPLC-ESI-MS/MS system (UPLC: ExionLC AD; MS: QTRAP® 6500+, MetWare Biotechnology Co., Ltd, Wuhan, China). The UPLC system featured a Thermo Accucore™ C30 column (2.6 μm, 2.1 mm × 100 mm), with a 2 μL injection volume and a column temperature of 45 °C, operating at a flow rate of 0.35 mL/min. Mobile phase A consisted of acetonitrile/water (60/40 v/v), while phase B was acetonitrile/water (10/90 v/v), both with 0.1 % formic acid and 10 mmol/L ammonium formate.

The elution gradient started at an A/B ratio of 80:20 (v/v) at t = 0, transitioning to 70:30 (t = 2 min), 40:60 (t = 4 min), 15:85 (t = 9 min), 10:90 (t = 14 min), and 5:95 (t = 15.5 and t = 17.3 min), before returning to 80:20 (t = 17.5 min) until t = 20.0 min. The effluent was directed to a QTRAP-MS (QTRAP® 6500+) with an ESI Turbo Ion-Spray interface, controlled by Analyst 1.6.3 software, operating in both positive and negative ion modes. The ESI source was set to a temperature of 500 °C and ion spray voltages of 5500 V (positive) and −4500 V (negative). Gas pressures for GS1, GS2, and CUR were set at 45, 55, and 35 psi, respectively.

The instrument was tuned and calibrated with polypropylene glycol solutions (10 μmol/L and 100 μmol/L) for QQQ and LIT modes, respectively. QQQ scans were conducted as MRM experiments with nitrogen collision gas at 5 psi. Individual MRM transitions were optimized for declustering potential (DP) and collision energy (CE). This method enhanced the accuracy and repeatability of quantification by screening precursor ions, inducing ionization and fragmentation, and selecting characteristic fragment ions to minimize interference.

2.5. Lipid profile identification and quantification

Mass spectrometry data were processed using Analyst 1.6.3 software (AB Sciex Pte. Ltd., Singapore). Qualitative analysis utilized the Metware Database (MWDB) from Metware Biotechnology Co., Ltd., based on retention times (RT) and ion data for detected substances.

For lipid quantification, QQQ MS analysis in MRM mode was employed. Chromatographic peaks for each substance were corrected for accuracy before integrating peak areas. Quantitative analysis used the internal standard method, with lipid concentration (C) calculated as follows:

$$C\left(nmol/g\right)=0.001\times \frac{R\times c\times F\times V}{m}$$

Here, R is the ratio of the analyte peak area to the internal standard peak area, F is the correction factor for different substances, c is the internal standard concentration in μmol/L, V is the sample extraction volume in μL, m is the sample weight in grams, and 0.001 is a unit conversion factor.

2.6. Sensory evaluation

Green coffee samples (200 g) were medium roasted according to the Specialty Coffee Association protocol (Specialty Coffee Association, 2018). A panel of three trained tasters with Q-Grader Coffee Certificates evaluated the samples. According to SCA protocol, participants rated 10 sensory attributes (fragrance/aroma, flavor, aftertaste, acidity, body, balance, uniformity, clean cup, sweetness, defects, and overall) on a scale from 6 to 10, with each score in increments of 0.25. Defects were considered negative scores. Participants assigned a total score to each sample based on the ratings of each attribute, and the team's average score reflected the overall quality of the coffee. Additionally, the tasters provided descriptive sensory attributes for each sample based on the Coffee Taster's Flavor Wheel (2016). All procedures and methodologies were approved by the Quality Assurance Department of the Yunnan International Coffee Exchange (YCE) in compliance with its guidelines and regulations, with informed consent obtained from all participating panelists.

2.7. Statistical analysis

Duncan's test of Analysis of Variance (ANOVA) was performed using SPSS 27.0 (IBM Corp, Armonk, USA), with statistical significance set at p < 0.05. Chord plots were generated using Origin 2022, while heatmaps and volcano plots were created with R version 3.5.1. Unsupervised principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) were also performed with R version 3.5.1. A permutation test with 200 permutations was conducted to reduce overfitting. Metabolites showing significant changes between groups were identified using Variable Importance in Projection (VIP) scores (VIP ≥1) and fold changes (FC ≥ 2.0 or FC ≤ 0.5). Differential lipids were annotated using the Kyoto Encyclopedia of Genes and Genomes (KEGG) compound database (https://www.kegg.jp/kegg/compound/) and mapped to the KEGG Pathway database (https://www.kegg.jp/kegg/pathway.html). Pathway analysis was conducted to clarify changes in the metabolic pathways of coffee bean lipids during postharvest processing. Additionally, metabolite set enrichment analysis (MSEA) was utilized to identify significantly modulated pathways, with significance assessed using the P-value from the hypergeometric test.

3. Result and discussion

3.1. Global lipidomic analysis of green coffee beans

The lipid composition of coffee samples, processed by various primary methods, was analyzed using HPLC-ESI-MS/MS (MRM) technology to monitor lipid impacts. We assessed the repeatability of quality control (QC) samples to confirm the accuracy of the lipidomic analysis. The overlap of Total Ion Current (TIC) curves, Pearson correlation, and the distribution of coefficient of variation (CV) values of QC samples demonstrated stable and reliable mass spectrometry signals across different time points (Fig. S1). From the qualitative and quantitative mass spectrometry data, 510 lipid molecules were identified, classified into five major categories: fatty acyls (FA, N = 20, 3.92 %), glycerophospholipids (GP, N = 125, 24.50 %), sphingolipids (SP, N = 37, 7.25 %), glycerolipids (GL, N = 325, 63.72 %), and prenol lipids (PR, N = 3, 0.59 %) (Fig. 1A and B). These findings are detailed in Fig. 1A and Table S2, highlighting that the lipid count in green coffee beans primarily consist of GP, SP, and GL.

Fig. 1.

Composition of lipid profile of green coffee beans. (A) Quantities of lipid categories and subclasses. (B) Percentages of lipid categories and subclasses.

To elucidate the differences in lipid profiles resulting from various primary processing methods, these five categories of lipids were further classified into 27 subclasses (Fig. 1B). Specifically, the FA category has one subclass (free fatty acids [FFA; N = 20; 3.92 %]); the GP category includes 12 subclasses (phosphatidic acid [PA; N = 8; 1.57 %], phosphatidylcholine [PC; N = 19; 3.73 %], phosphatidylethanolamine [PE; N = 32; 6.27 %], phosphatidylglycerol [PG; N = 10; 1.96 %], phosphatidylinositol [PI; N = 17; 3.33 %], phosphatidylserine [PS; N = 8; 1.57 %], lysophosphatidic acid [LPA; N = 5; 0.98 %], lysophosphatidylcholine [LPC; N = 13; 2.55 %], lysophosphatidylethanolamine [LPE; N = 5; 0.98 %], lysophosphatidylglycerol [LPG; N = 2; 0.39 %], lysophosphatidylinositol [LPI; N = 4; 0.78 %], and phosphatidylmethanol [PMeOH; N = 2; 0.39 %]); the SP category contains four subclasses (sphingine [SPH; N = 2; 0.39 %], ceramide [Cer; N = 19; 3.73 %], hexosylceramide [HexCer; N = 4; 0.78 %], and phytoceramide [Cert; N = 12; 2.35 %]); the PR category comprises one subclass (coenzyme Q [CoQ; N = 3; 0.59 %]); and the GL category is made up of nine subclasses (acyldiacylglyceryl glucuronide [ADGGA; N = 2; 0.39 %], diacylglycerol [DG; N = 22; 4.31 %], digalactosyldiacylglycerol [DGDG; N = 6; 1.18 %], diacylglyceryl glucuronide [DGGA; N = 1; 0.20 %], diacylglyceryltrimethylhomoserine [DGTS; N = 9; 1.76 %], monoacylglycerol [MG; N = 4; 0.78 %], monogalactosyldiacylglycerol [MGDG; N = 1; 0.20 %], sulfoquivonosyldiacylglycerol [SQDG; N = 3; 0.59 %], and triacylglycerol [TG; N = 277; 54.31 %]) (Fig. 1B). This detailed subclassification helps elucidate the variations in lipid molecular profiles resulting from different coffee processing techniques.

3.2. Impact of processing methods on coffee bean lipid profiles

Although extensive studies have explored the effects of different processing methods on the chemical composition and flavor characteristics of coffee (Cortés-Macías et al., 2022; Wan et al., 2024; Febrianto and Zhu, 2023), comprehensive information on changes in lipid molecules remains insufficient. This study aims to clarify the specific effects of processing methods on the lipid composition of coffee beans through lipid profile analysis. Fig. 2A shows the differences in lipid composition across the various processing methods. Honey-processed green beans exhibit the highest number of lipid species (506), followed by washed and natural-processed beans (502 species each), while unprocessed coffee beans (CK group) have the fewest lipid species (489). This indicates that the number of lipid species increases following postharvest processing. A total of 27 subclasses were detected in all sample groups, with TG being the most abundant, followed by PE, DG, and FFA. The chord plot revealed that certain lipid subclasses (such as LPA, PA, and MG) decreased after processing, while subclasses such as PE, PS, and DPTS increased, indicating that processing significantly affects the composition of lipid subclasses in coffee beans (p < 0.05). Additionally, subclasses like PA, Cert, and ADGGA showed notable differences across processing methods, further confirming the variations in lipid profiles based on processing techniques.

Fig. 2.

(A) Differences in lipid composition of green coffee beans obtained by different processing methods. Heatmap visualization of the metabolic variations of lipid classes (B) and subclasses (C) between different processing methods. CK, control treatment. N, natural processing; W, washed processing; H, honey processing.

Next, we visualized the changes in lipid content before and after processing, as well as among different processing methods, using heat maps. Fig. 2B illustrates that all processing methods lead to significant reductions in FA and PR content. Coffee seeds are unorthodox seeds, and their germination is an important biological process that occurs during postharvest processing, which is closely related to the processing method of green beans. During germination, metabolic reactions can significantly affect the chemical composition of coffee beans, which in turn influences the in-cup properties of coffee. FA serves as a crucial source of energy during the germination of coffee seeds, while PR acts as a precursor to gibberellins, which play an essential role in breaking seed dormancy and inducing germination. Therefore, the observed reductions in both FA and PR may be linked to germination during processing. Fig. 2C visually highlights the differences in lipid subclasses of coffee beans under various processing methods. Different processing methods induce distinct metabolic processes in coffee beans, resulting in significant differences in lipid profiles. This variation in lipid metabolism directly affects the quality characteristics of coffee and is a key factor behind the sensory differences observed with different processing methods. Consequently, in-depth research on these metabolic processes and their effects on lipid composition will enhance our understanding of coffee flavor characteristics and provide a basis for optimizing processing methods.

3.3. Multivariate statistical analysis

PCA is a method that reduces high-dimensional data to lower dimensions through linear transformation, aiming to retain maximum variance and preserve the characteristic information of the original data (Lever et al., 2017). In this study, PCA was applied to analyze lipidomics data, revealing the distinctive characteristics of each processing method through data simplification. PC1 and PC2 of the PCA model explained 38.02 % and 26.48 % of the total variance, respectively, and no outliers were detected (Fig. 3a). The samples before and after processing were well separated, while the separation between processing methods was relatively poor. This indicates that although different processing methods affect the lipid profile, the differences among them are minor compared to the changes observed before and after processing. The PCA model can only reveal the population structure when the changes within the group are sufficiently smaller than the changes between the groups (Worley and Powers, 2013). Due to the overlap between the processing methods, the subtle differences in lipid composition caused by them are further emphasized.

Fig. 3.

The PCA score plot (A), OPLS-DA score plot (B) and Permutation test of the OPLS-DA model (C) were based on overall lipid molecules in different processing modes. CK, control treatment; N, natural processing; W, washed processing; H, honey processing.

OPLS-DA is a supervised multivariate statistical method that achieves a higher level of overall separation (Wu et al., 2023). In this study, OPLS-DA was used to classify and analyze lipid data further. In the score plot, the distance between parallel samples within the group is short, indicating high repeatability of the data. Different groups can be clearly separated, demonstrating significant differences in the lipid composition among the four sample groups. Specifically, the separation distance between the honey, natural, and washed processed samples and the unprocessed samples gradually increases, indicating a more pronounced effect of processing methods on lipid composition. To further evaluate the effectiveness of the model, we performed a permutation test (Fig. 3c). The prediction parameters of the OPLS-DA model include R²X, R²Y and Q². Among them, R²X and R²Y reflect the model's ability to explain the X and Y matrices, respectively, while Q² represents the model's predictive ability. The closer these three indicators are to 1, the more stable and reliable the model is. A Q² value greater than 0.5 indicates that the model can be considered effective. The test results show that the classification indicators R²X, R²Y and Q² are 0.780, 0.997 and 0.719 respectively, indicating that the OPLS-DA model has good fit and predictive ability.

3.4. Characterization of the key lipids

Due to the “high-dimensional and massive” nature of metabolomics data, it is necessary to combine univariate statistical analyses and multivariate statistical analyses and to examine the data from multiple perspectives to accurately identify differential metabolites (Guo et al., 2022). Based on the OPLS-DA results, the VIP values of the OPLS-DA model were obtained through multivariate analysis, providing an initial screening of differences between samples. Additionally, by integrating P values or fold change (FC) values from univariate analysis, differential metabolites can be further accurately selected. In this study, we set VIP >1 and FC ≥ 2 or FC ≤ 0.5 as the screening criteria to identify lipids with significant differences. The results are visually represented in volcano plots, which facilitate the identification of metabolite differences between samples.

Across the three coffee samples processed by different methods, a total of 150 lipids were found to have changed significantly compared with the unprocessed control group (Table S3; Fig. 4A, B, C). Specifically, in the N vs CK group, 47 lipids were downregulated and 83 were upregulated (Table S3; Fig. 4A); in the W vs CK group, 47 lipids were downregulated and 84 were upregulated (Table S3; Fig. 4B); and in the H vs CK group, 44 lipids were downregulated and 88 were upregulated (Table S3; Fig. 4C). Venn diagram analysis (Fig. 4G) showed that 114 differential lipids were identical across the three sample groups. Coffee beans, as non-orthodox seeds, possess an enzymatic mechanism that initiates germination at the ripe stage. The subsequent processing methods lead to variations in germination processes, which in turn affect the carbohydrate, protein, and lipid composition of green beans, ultimately influencing the sensory properties of roasted coffee beans (Waters et al., 2017; Bytof et al., 2007). Specifically, natural processing formed 10 unique lipids, washed processing formed 7 unique lipids, and honey processing formed 4 unique lipids. These differences highlight the significant impact of processing methods on coffee quality.

Fig. 4.

The differential lipid metabolites in different green coffee samples. Volcano plot of the differential lipids of N vs CK (A), W vs CK (B), H vs CK (C), N vs W (D), H vs N (E) and H vs W (F), the criteria set at VIP ≥1, FC ≥ 2 or ≤ 0.5. Differential lipids Veen diagram of before and after processing (G) and between processing methods (F). CK, control treatment; N, natural processing; W, washed processing; H, honey processing.

We further used volcano plots to analyze the differential lipids across the different processing methods. In the N vs W group, 16 lipids were downregulated and 15 lipids were upregulated (Table S3; Fig. 4D); in the H vs N group, 10 lipids were downregulated and 11 were upregulated (Table S3; Fig. 4E); and in the H vs. W group, 2 lipids were downregulated and 6 were upregulated (Table S3; Fig. 4F). The distribution of differential compounds between groups was visualized with a Venn diagram, revealing a total of 37 differential metabolites across the three groups (Fig. 4H). Among these differential lipids, LPA (16:0) was the only common lipid, showing significant content changes across processing methods. These significantly different lipids could serve as potential biomarkers to distinguish the three coffee processing methods and could be applied in adulteration detection and quality identification.

3.5. Lipid metabolism pathway differences across processing methods

To analyze lipid changes during the postharvest processing of coffee, we mapped the key differential lipids identified in the three processing methods to the KEGG database for metabolic pathway analysis. The results indicated that the three processing methods primarily involved 19 metabolic pathways (Fig. S2), covering Metabolism, Environmental Information Processing, and Cellular Processes. Samples from different processing methods exhibited consistency in the main metabolic pathways, particularly in metabolic pathways, glycerophospholipid metabolism, and biosynthesis of secondary metabolites, with an average relative abundance exceeding 50 %. This suggests that these pathways play a dominant role in coffee postharvest processing. The postharvest processing method of coffee influences the evolution of germination, with lipid metabolism following asynchronous patterns in different processing methods (Waters et al., 2017). Previous studies have demonstrated differences in the expression of specific germination enzymes, such as isocitrate lyase and β-tubulin, highlighting divergent metabolic patterns between natural and washed processing (Bytof et al., 2007; Selmar et al., 2006). In washed processing, the “sprouting” process typically begins on the first day of fermentation, primarily due to the inactivation of inhibitory factors after pulp removal. In contrast, during natural processing, the fruit remains intact, and the osmotic potential of the pulp is maintained for a longer period, with sprouting commencing after several days (Bytof et al., 2007). These differences underscore the significant impact of processing methods on the lipid metabolism of coffee beans.

To further characterize the differences in lipid metabolism under various processing methods, we excluded the 114 common differential lipids from the differential lipid set, retaining 37 specific lipids. We then analyzed the metabolic pathways of these characteristic lipids using MetaboAnalyst 6.0. The results are displayed in bubble charts, where larger and darker bubbles indicate greater significance of pathway enrichment and impact. In natural processing, glycerolipid metabolism, glycerophospholipid metabolism, and sphingolipid metabolism emerge as the most critical differential formation pathways (Fig. 5A). In washed processing, glycerolipid metabolism, glycerophospholipid metabolism, and inositol phosphate metabolism are predominant (Fig. 5B). For honey processing, glycerophospholipid metabolism and sphingolipid metabolism are the key pathways (Fig. 5C). Notably, glycerophospholipid metabolism shows significant pathway enrichment and impact across multiple processing methods, suggesting it has the strongest influence on lipid differential formation during coffee postharvest processing. Traditional analyses focused on total lipid content or abundant lipid classes (Richard et al., 2020; Scholz et al., 2019). In contrast, our use of UPLC-ESI-MS/MS revealed that key processing-related differences are concentrated in low-abundance phospholipids, which play critical roles in membrane dynamics and signal transduction during postharvest metabolic transitions.

Fig. 5.

KEGG enrichment statistics of non-common differential lipids in N (A), W (B), and H (C), where the x-axis shows the rich factor for each pathway, and the y-axis lists the KEGG metabolic pathways. Bubble size and color indicate the number of different lipids and the degree of enrichment, respectively. Glycerophospholipid metabolism of non-common differential lipids in N (D), W (E), and H (F). The numbers correspond to the ID of each lipid subclass. N, natural processing; W, washed processing; H, honey processing.

To further explore lipid transformation in different coffee processing methods, we annotated the KEGG pathways of specific glycerophospholipid metabolism in detail. In natural processing, the main components involved in glycerophospholipid metabolism include LPC (C04230), LPE (C04438), PA (C01194), PG (C00344), and PI (C00416). In washed processing, DG (C00641), PE (C00157), LPI (C03819), and PA play an important role. In honey processing, PE, LPE, PA, and LPI are the main participating components. PA, as a key lipid metabolite, changes significantly during the initial processing of coffee. This phenomenon suggests that PA may play an important physiological and chemical role during the processing stage, affecting the flavor and overall quality of coffee. Therefore, studying the generation and changes of PA is of great significance for optimizing coffee processing technology and improving the quality of the final product. During the seed germination stage, glycerophospholipids play an important role in building cell membranes, supporting cell proliferation and differentiation, regulating signal transduction and metabolism, and promoting energy release and material transport (Cao et al., 2024). The asynchronous germination caused by different processing methods leads to noticeable differences in glycerophospholipid metabolism. Additionally, the anaerobic environment generated by washed processing also affects the metabolic patterns of lipids. These factors together result in differences in glycerophospholipid metabolism across various processing methods.

3.6. Sensory evaluation

Coffee processing is a complex and delicate process, with different methods yielding distinct flavor characteristics (Freitas et al., 2024). Selecting the appropriate processing method can effectively highlight the origin characteristics of coffee beans and their flavor potential. Following the evaluation protocol established by the Specialty Coffee Association (2018), we assessed the coffees processed by the three methods, with the specific results presented in Table S4. In terms of uniformity, clean cup, and sweetness, all processing methods received a perfect score of 10 points, while the scores for other attributes ranged from 7.25 to 7.75 (Table S4, Fig. 6A). All coffees ultimately scored over 80 points, meeting the standards for “specialty coffee” (Specialty Coffee Association, 2018). Notably, the brewing scores for washed and honey methods were significantly higher than those for natural processing (p < 0.05). Among the three processing methods, honey processing coffee exhibited the highest lipid content, which was significantly greater than that of natural processing coffee (p < 0.05). The lipid content of washed processing coffee beans was intermediate between the honey and natural processing methods and did not differ significantly from either group (p > 0.05). The attributes of aroma, flavor, and body followed a consistent trend with lipid content; washed and honey methods scored higher than natural processing, but no significant differences were noted (p > 0.05; Fig. S3; Table S4). Additionally, acidity was significantly different between different methods (p < 0.05), with the highest acidity in washed processing and the lowest acidity in natural processing, which is consistent with the conclusions of other related studies (Cortés-Macías et al., 2022).

Fig. 6.

(A) Scores of various attributes in the cupping test of samples with different processing methods. (B) Sensory descriptors and final score of roasted coffee samples. (C) Correlation analysis between significantly different lipids and cupping attribute scores in different processing methods (Red indicates positive correlation, blue indicates negative correlation, and the value indicates the correlation coefficient.). N, natural processing; W, washed processing; H, honey processing.

Sensory descriptors were rated by professional tasters according to the Coffee Taster's Flavor Wheel (2016). Fig. 6B illustrates the differences in sensory characteristics of coffees processed by various methods. Nutty/cocoa descriptions were observed in all samples, with clear nutty characteristics in further descriptors, among which natural processing showed a particularly obvious almond flavor. Descriptions related to sweet and brow sugar categories were consistent across samples. A deeper analysis showed that honey and maple syrup flavors were particularly identified in natural processing, while caramel was present in washed and honey processed samples. Additionally, descriptions related to the fruity category were present in all samples: natural processing exhibited dried fruit flavors, while washed and honey processing highlighted citrus notes. For sour/fermented descriptors, natural processing displayed alcohol/fermented characteristics, while washed and honey processing emphasized citrus acid. Notably, while honey and washed coffee samples shared similar sensory profiles, their processing methods differed significantly, aligning with findings from other studies (Cortés-Macías et al., 2022). Furthermore, honey processing coffee samples presented a floral descriptor. In recent years, due to the increased environmental burden of wastewater discharge generated by washed processing and seasonal over-limit problems, this method has gradually been eliminated in the large-scale Yunnan coffee production. Honey processing has emerged as a potential alternative for large-scale implementation, which may enhance the overall flavor profile of Yunnan coffee and unlock further flavor potential.

Although sensory evaluation has always been the mainstream evaluation method for assessing coffee quality, its subjectivity and inconsistency have led researchers to explore alternative methods (Romano et al., 2014; Hu et al., 2020). Hu et al. (2020) used the PLS model based on the main chemical components for exploratory prediction of the body score of coffee with different roasting degrees, showing good prediction results. To further explore the relationship between lipid composition and the sensory characteristics of coffee, we conducted a correlation analysis between the differential lipid data and the sensory attributes (Fig. 6C). While the analysis showed significant correlations between specific lipid species and sensory attributes, we acknowledge that these associations do not establish a causal link. Instead, these lipids may serve as sensitive markers reflecting broader metabolic activities during processing that contribute to flavor variation. Among all sensory descriptors, acidity showed the strongest statistical association with lipid composition. Lipidomics offers a new perspective for understanding coffee flavor and quality. Given that the sensory characteristics of coffee are influenced by multiple substances, improving the accuracy of food quality assessments requires the integration of multi-omics methods.

4. Conclusion

In summary, we systematically investigated the effects of three processing methods (natural, washed, and honey) on the lipids of green coffee beans. Through qualitative and quantitative analysis of lipid composition, we identified a total of 510 lipids across 27 subclasses in both the control sample and the green beans processed by the three methods. OPLS-DA analysis revealed significant differences in lipid composition across the processing methods and between pre- and post-processing stages. Specifically, in comparison to the control sample (unprocessed), we identified 130, 131, and 132 significantly different lipids in the dry, wet, and semi-dry methods, respectively, with 114 lipids found in common across these groups. Additionally, we identified 37 significantly different lipids as potential biomarkers for distinguishing between processing methods. Furthermore, glycerophospholipid metabolism was identified as the primary metabolic pathway contributing to lipid differences among the three methods. Notably, significant correlations were identified between differential lipids and sensory attributes. These findings suggest that changes in lipid composition during postharvest processing reflect underlying metabolic shifts that are associated with flavor variation. However, we do not imply a direct causal relationship between specific lipids and sensory outcomes. Further research is needed to elucidate the biochemical mechanisms linking lipid metabolism and flavor development. This study provides a comprehensive insight into lipid metabolism in coffee beans during postharvest processing, offering valuable references for future research on coffee quality assessment and processing optimization.

CRediT authorship contribution statement

Yanbing Wang: Methodology, Validation, Formal analysis, Investigation, Data curation, Visualization, Writing – original draft, Writing – review & editing. Xiaoyuan Wang: Methodology, Validation, Formal analysis, Investigation, Data curation, Visualization, Writing – original draft, Writing – review & editing. Ping Du: Methodology, Writing – review & editing, Project administration, Funding acquisition. Xiaogang Liu: Writing – review & editing, Funding acquisition. Sufang He: Methodology, Writing – review & editing. Lirong Li: Methodology, Writing – review & editing. Xiaoqiong Liu: Methodology, Writing – review & editing, Project administration, Funding acquisition. Zhenjia Chen: Resources, Funding acquisition.

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.

Handling Editor: Professor Aiqian Ye

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Data availability

Data will be made available on request.