ABSTRACT
A cup of coffee is the final product of a complex chain of operations. Wet postharvest processing of coffee is one of these operations, which involves a fermentation that inevitably has to be performed on-farm. During wet coffee processing, the interplay between microbial activities and endogenous bean metabolism results in a specific flavor precursor profile of the green coffee beans. Yet, how specific microbial communities and the changing chemical compositions of the beans determine the flavor of a cup of coffee remains underappreciated. Through a multiphasic approach, the establishment of the microbial communities, as well as their prevalence during wet processing of Coffea arabica, was followed at an experimental farm in Ecuador. Also, the metabolites produced by the microorganisms and those of the coffee bean metabolism were monitored to determine their influence on the green coffee bean metabolite profile over time. The results indicated that lactic acid bacteria were prevalent well before the onset of fermentation and that the fermentation duration entailed shifts in their communities. The fermentation duration also affected the compositions of the beans, so that longer-fermented coffee had more notes that are preferred by consumers. As a consequence, researchers and coffee growers should be aware that the flavor of a cup of coffee is determined before as well as during on-farm processing and that under the right conditions, longer fermentation times can be favorable, although the opposite is often believed.
IMPORTANCE Coffee needs to undergo a long chain of events to transform from coffee cherries to a beverage. The coffee postharvest processing is one of the key phases that convert the freshly harvested cherries into green coffee beans before roasting and brewing. Among multiple existing processing methods, the wet processing has been usually applied for Arabica coffee and produces decent quality of both green coffee beans and the cup of coffee. In the present case study, wet processing was followed by a multiphasic approach through both microbiological and metabolomic analyses. The impacts of each processing step, especially the fermentation duration, were studied in detail. Distinct changes in microbial ecosystems, processing waters, coffee beans, and sensory quality of the brews were found. Thus, through fine-tuning of the parameters in each step, the microbial diversity and endogenous bean metabolism can be altered during coffee postharvest processing and hence provide potential to improve coffee quality.
INTRODUCTION
A cup of coffee is the endpoint of a complex chain of events. This chain includes postharvest processing, roasting, and brewing. Postharvest processing consists of several steps performed on-farm after the coffee cherries have been harvested, and it yields the green coffee beans that can be roasted (1, 2). During this processing, an interplay between microbial activities and endogenous bean metabolism takes place, which results in a specific flavor precursor profile of the green coffee beans (2, 3) Coffee cherries can be processed in different ways (1). Wet processing is usually implemented for Coffea arabica cherries to produce high-grade Arabica coffee.
During wet processing, the harvested coffee cherries are depulped, spontaneously fermented underwater, soaked, and dried (4, 5). The fermentation step aims to remove the mucilage that is firmly attached to the beans. This fermentation is performed by microorganisms that originate from the cherry surfaces, plantation environment, or processing equipment. As processing progresses, microbial communities grow due to variable selective pressures from intrinsic (e.g., coffee cultivar and geography) and extrinsic (e.g., processing, handling, and operational practices) (3, 5–8) factors. How these factors shape these communities remains to be elucidated. Due to this uncertainty, researchers have already tried to standardize the fermentation process by adding selected microbial strains to the fermentation mass, without managing specific operational practices (9–11). Since metabolites of microbial origin (such as organic acids) can be present on the green coffee beans, the mechanisms shaping the coffee ecosystem need to be better understood (3).
As coffee beans are still metabolically active during postharvest processing, they respond to various abiotic stresses, such as those caused by depulping at the start of the processing, anoxic and acidic conditions during underwater fermentation, and drought stress upon drying (12, 13). These stress-related metabolic responses will also change the metabolite composition of the green coffee beans. Coffee bean stress is marked by the evolution of, for instance, γ-aminobutyric acid (GABA) and carbohydrate concentrations (3, 14–16). However, the evolution of such compounds along the entire postharvest processing chain has not been studied extensively.
During roasting, the chemical profiles of the green coffee beans, which are determined not only by cultivar and geography but also by postharvest processing as described above, transform into the characteristic coffee flavor (17–19). Ultimately, gauging the effect of postharvest processing on the coffee cup quality requires sensory analysis by a trained panel. However, reports on the relationship between the sensory data and the fermentation process (postharvest effect) are scarce (7, 20).
Given this complex and interlinked postharvest processing of coffee, an integrated systematic study of the relationship between the coffee processing microbiota, endogenous bean metabolism, operational practices, and cup quality was necessary. Therefore, this study aimed to decode the complete wet processing chain of Arabica coffee under different operational practices, starting from the harvesting of the coffee cherries through on-farm processing until coffee roasting and brewing (Fig. 1). This was tackled through a multiphasic approach, monitoring the coffee microbiota (on-farm microbiological analysis and high-throughput sequencing), the coffee bean composition (meta-metabolomics), and the final quality of the coffee brews (sensory analysis).
(a) Overview of the Arabica coffee postharvest wet processing experiments. The abbreviations of the different samples are indicated as well. The dark blue line represents the standard fermentation process (S), and the light blue line represents the extended fermentation process (E). Sample codes starting with F correspond to fermentation samples, followed by a number corresponding to the fermentation duration in hours. Sample codes starting with SS denote soaking samples after standard fermentation and sample codes starting with SE denote soaking samples after extended fermentation, followed by a number corresponding to the soaking duration in hours. Sample codes starting with DS denote drying samples after standard fermentation and sample codes starting with DE denote drying samples after extended fermentation, followed by a number corresponding to the drying duration in hours. Samples SB and EB denote green coffee bean samples resulting from the standard fermentation process and the extended fermentation process, respectively. (b) Sampling schemes according to the type of analysis. The sampling points for each analysis type are colored based on their location in the processing chain. Gray sampling points denote samples that were not included for analysis.
RESULTS
Microbial community dynamics during postharvest processing.(i) Microbial community dynamics during pooling, depulping, and fermentation. The pooled cherries and depulped beans showed high counts of all microbial groups targeted, namely, lactic acid bacteria (LAB), acetic acid bacteria (AAB), enterobacteria, and yeasts (Fig. 2). Isolate identification and amplicon sequencing of targeted genes of the whole-community DNA affirmed the community members of these groups belong to the Acetobacteraceae (encompassing Acetobacter, Gluconobacter, and Kozakia), enterobacteria, Leuconostoc pseudomesenteroides, Pichia kluyveri (particularly found by amplicon sequencing), Hanseniaspora uvarum, and Candida quercitrusa (Fig. 3 and 4; see also Table S1 in the supplemental material). These high counts persisted once the depulped beans were submerged in the fermentation tank (particularly LAB and enterobacteria). From here on, LAB asserted their quantitative prevalence over other microbial groups during the standard fermentation (S). This prevalence was further developed during the extended fermentation (E) and was retained until the end of fermentation (F). During this phase, a shift from L. pseudomesenteroides to lactobacilli, namely Lactobacillus vaccinostercus, Lactobacillus brevis, and Lactobacillus plantarum, occurred within the LAB communities. Conversely, the counts of enterobacteria and AAB showed a continuous decrease during the standard and extended fermentations. No major shifts occurred within the community compositions of these groups. The yeast counts and communities remained relatively stable during these phases, although Starmerella bacillaris (particularly found through isolate identification) became more pronounced as fermentation progressed and Saccharomycopsis crataegensis was encountered occasionally. Lactococcus lactis was prevalent transiently during the standard fermentation (12 to 24 h). Other species were sporadically encountered, such as Leuconostoc fallax (at 16 h) and Pediococcus pentosaceus (at 48 h). Hence, the initial occurrence of enterobacteria (and to a lesser degree, AAB) and the prevalence of Leuconostoc (accompanied by the transient occurrence of Lactococcus) characterized the standard fermentation. The continued but diminishing prevalence of Leuconostoc and the subsequent prevalence of Lactobacillus characterized the extended fermentation.
Microbial counts during coffee cherry pooling, depulping, fermentation, soaking, and drying. The agar media used for enumeration are plate count agar (PCA) for total bacteria, modified de Man-Rogosa-Sharpe (MRS-S) agar for lactic acid bacteria, violet-red-bile-glucose (VRBG) agar for enterobacteria, yeast glucose (YG) agar for yeasts, and modified deoxycholate-mannitol-sorbitol (mDMS) agar for acetic acid bacteria. Counts of 2.0 log CFU/g indicate counts equal to or below this value. Sample abbreviations are as in the legend for Fig. 1.
Isolate identification during coffee cherry pooling, depulping, fermentation, and soaking. Isolates are grouped into major microbial categories (acetic acid bacteria, lactic acid bacteria, and yeasts). Colors denote the different species identified, and the size of the dots is relative to the number of isolates picked up and identified. The number of isolates picked up and identified at each time point is represented inside each dot. Sample abbreviations are as in the legend for Fig. 1.
Distribution of amplicon sequence variants (ASVs) of the V4 region of the 16S rRNA gene (bacteria) and the internal transcribed spacer (ITS1) region of the fungal ribosomal transcribed unit (yeasts and molds) during coffee cherry pooling, depulping, fermentation, soaking, on the processing apparatus, in the processing environment, and in the postfermentation waters. Sample abbreviations are as in the legend for Fig. 1.
(ii) Microbial community dynamics during soaking. The viability of the microorganisms was retained and/or regained during soaking, since high counts of all microbial groups targeted, particularly LAB, were found after 24 h of soaking (Fig. 2). The relative values of these counts were comparable to those found during fermentation. Leuconostoc pseudomesenteroides and P. kluyveri were important community members, regardless of soaking after the standard or extended fermentations (Fig. 3 and 4). Notably, a higher abundance of Lactobacillus was found by amplicon sequencing when performing soaking after the extended fermentation (reflecting its higher prevalence at the end of the extended fermentation). Minor variations were found for other communities, such as enterobacteria and Lactococcus. Within the yeast communities, P. kluyveri and H. uvarum (particularly found through isolate identification) emerged as prominent members during soaking after both the standard and extended fermentations.
(iii) Microbial community dynamics during drying. During sun drying after the standard and extended fermentations, all microbial groups targeted either decreased in viability, as their counts decreased, or were present at <2.0 log CFU/g (Fig. 2). This was in contrast with that at the end of soaking, i.e., there were relatively higher counts of yeasts during soaking than during drying. The loss of viability was faster and more pronounced after the extended fermentation than after the standard fermentation.
Microbial ecology of the processing apparatus and processing environment.The microbial contamination of the processing apparatus (cherry storage bags, depulper exit shaft, empty fermentation tank, and empty soaking tank) and environmental samples (plantation soil, coffee tree flowers, coffee tree leaves, and fresh cherries from the coffee trees) was generally soil or plant associated but was variable for the different pieces of apparatus analyzed (Fig. 4). Notably, taxa that were found extensively during fermentation (e.g., Leuconostoc, enterobacteria, and AAB) were found in much lower relative abundances on the processing apparatus. These taxa were sporadically encountered in relatively high relative abundances in the coffee phyllosphere (e.g., that of coffee cherries that were attached to the coffee trees or of the coffee leaves). Coffee cherries that were attached to the trees displayed microbial counts that spanned a wide range (see Fig. S1). Microbial groups that were found in high counts during the harvesting-depulping interval were often present at <2.0 log CFU/g when analyzing these coffee cherries (e.g., LAB and enterobacteria). Differences in community compositions of the different samples were elucidated by principal-component analysis (PCA) (see Fig. S2). The fermentation and soaking water samples formed a cluster distinct from the environmental samples. This separation was substantiated by network analysis, through which these environmental samples were disjoined from all other types of samples and were connected with different microbial communities (see Fig. S3). The overall sequencing error rate of all amplicons was 0.04%.
Growth assessment of the epiphytic coffee cherry microbiota.All of the microbial groups followed were able to grow on the appropriate agar media when plating samples from coffee cherries inside the sterile plastic bags used for an imitation of the harvesting-depulping interval of the coffee postharvest processing chain (see Fig. S4). However, the most rapid and most substantial increase (approximately 2.0 log CFU/ml) was that of the LAB communities, which was approximately 14-fold and 5-fold that of the AAB and yeast communities, respectively.
Microbial community dynamics in postfermentation waters.The LAB community profiles of the postfermentation water (PFW; mixture of fermentation water and wash water) samples were similar to those at the end of the standard fermentation (Fig. 4; Fig. S2 and S3). These profiles were characterized by a relatively high prevalence of Leuconostoc, Lactococcus, lactobacilli, enterobacteria, and Pichia. In contrast to the standard fermentation profile, taxa unique to the PFW were found, notably, Clostridium.
Metabolite course in fermentation and soaking waters.(i) Metabolite course in fermentation waters. The nonamino acid compounds quantified in the fermentation water samples (W) were divided into three clusters according to hierarchical clustering analysis of the heatmap data (A1 to A3) based on their profiles during fermentation (Fig. 5). Cluster A1 compounds, represented by sucrose, citric acid, malic acid, and acetaldehyde, reached high concentrations after 12 to 24 h of fermentation and were depleted toward the end (F64). In comparison, cluster A3 compounds, represented by glucose and fructose, also reached the highest concentrations after 12 h of fermentation (F12). Yet, these compounds remained at relatively high concentrations toward the end of the extended fermentation. For example, glucose and fructose concentrations reached 3.6 and 3.1 mg/ml in F64W, respectively, whereas the concomitant sucrose concentration was only 0.025 mg/ml. In contrast, cluster A2 compounds (lactic acid and mannitol) were characterized by their continuous accumulation throughout fermentation. The accumulation of these compounds increased after 16 h of fermentation (F16), resulting in a 2- to 8-fold increase at the end of the extended fermentation (F64). As a result, the most abundant compounds in F64W were lactic acid (8.2 mg/ml), 5-ketogluconic acid (5.7 mg/ml), acetic acid (2.4 mg/ml), mannitol (1.9 mg/ml), ethanol (1.8 mg/ml), and glycerol (0.5 mg/ml). Compounds originating from the coffee plant, such as quinic acid, caffeine, trigonelline, and succinic acid, also displayed small increments in their concentrations. Chlorogenic acids (CGAs) were not found in the fermentation water samples.
Hierarchical clustering analysis and heatmap visualization of selected quantified chemical compound profiles (column dendrogram) in fermentation and soaking waters (FW and SW) (a) and coffee beans (B) (b) along the wet coffee processing chain. The absolute concentrations are summed and displayed on the right; a color key as a measure for the concentrations is also displayed on the right. Sample abbreviations are as in the legend for Fig. 1. dm, dry mass.
The total free amino acid concentrations in the fermentation water samples almost doubled to 1.02 mg/ml from the beginning toward the end of the extended fermentation (Fig. 6). The evolution profile of the amino acids was also divided into two clusters based on hierarchical clustering analysis of the heatmap data (B1 and B2). Similar to cluster A2, the majority of the amino acids grouped in cluster B1 built up during the standard and extended fermentations. Cluster B1 had GABA (0.30 mg/ml), asparagine (0.23 mg/ml), and alanine (0.18 mg/ml) as the most abundant compounds at the end of the extended fermentation (F64). However, certain amino acids of cluster B2 (e.g., arginine, glutamine, valine, leucine, and isoleucine) decreased, especially during the extended fermentation.
Hierarchical clustering analysis and heatmap visualization of free amino acid profiles (column dendrogram) in fermentation and soaking waters (FW and SW) (a) and coffee beans (B) (b) along the wet coffee processing chain. The total free amino acid concentrations are summed and displayed in green on the right; a color key as a measure for the concentrations is also displayed on the right. Sample abbreviations are as in the legend for Fig. 1. dm, dry mass.
More than 100 volatile compounds were identified in the fermentation waters, among which, 50% were esters, 23% alcohols, 12% aldehydes, and 5% terpenes/terpenoids. The total aroma intensity in the fermentation waters increased 5 times in F64 compared to that in F16. The production dynamics of these volatile compounds allowed a grouping into three different clusters based on hierarchical clustering analysis of the heatmap data (C1 to C3) (Fig. 7). At the beginning of the fermentation, the compounds of cluster C3 already built up, e.g., 2,3-butanediol, 2/3-methylbutanal, and ethyl 2/3-methylbutanoate. This was succeeded by the emergence of cluster C2 compounds (especially after 16 h of fermentation; F16), among which, 48% were esters. The major esters, such as 3-methylbutyl acetate, isobutyl acetate, ethyl butanoate, ethyl 2-hydroxypropanoate, and ethyl acetate, reached their highest aroma intensities at F64, with a 10-fold increment compared to that in F16, whereas ethyl hexanoate and isobutyl butanoate increased 26 and 33 times but displayed lower aroma intensities. The build-up of esters was accompanied by the accumulation of the corresponding (higher) alcohols (e.g., ethanol, 3-methyl-1-butanol, 2-methyl-1-butanol, 2-methyl-1-pentanol, and 2-methyl-1-propanol), which also displayed high aroma intensities, as well as organic acids (e.g., acetic acid, propionic acid, butanoic acid, and pentanoic acid), as seen in cluster C2.
Hierarchical clustering analysis and heatmap visualization of nontargeted aroma profiles (column dendrogram) in the fermentation and soaking waters (FW and SW). The aroma intensities normalized to the internal standard are summed and displayed in blue on the right (in arbitrary units [AU]); a color key as a measure for the concentrations is also displayed on the right. Sample abbreviations are as in the legend for Fig. 1.
(ii) Metabolite course in soaking waters. The concentrations of the total nonamino acid metabolites measured decreased to 5.2% of F16W and 1.2% of F64W for the standard (S0W) and extended (SE0W) fermentations, respectively (Fig. 5). After the standard fermentation, simple carbohydrates were still found in clusters A1 and A3 in relatively high concentrations (glucose, 0.27 mg/ml; fructose, 0.13 mg/ml) at the start of soaking (S0), whereas both their concentrations were 0.10 mg/ml in SE0. In both cases, the simple carbohydrates were completely consumed after 24 h of soaking. Concurrently, an accumulation of cluster A2 compounds was found during soaking, and high concentrations were found when extended fermentation took place. Lactic acid (0.70 and 0.68 mg/ml), ethanol (0.14 and 0.24 mg/ml), acetic acid (0.10 and 0.17 mg/ml), and acetaldehyde (0.03 and 0.05 mg/ml) were the major compounds in both S24W and SE24W, respectively, whereas mannitol (0.33 mg/ml) was only abundant in SE24W. In addition, trigonelline and caffeine, which originate from the coffee plant, accumulated at similar concentrations in both S24W and SE24W, with 0.014 and 0.034 mg/ml in S24W and 0.019 and 0.037 mg/ml in SE24W, respectively. The total free amino acid concentrations also increased during soaking, but they were 50 times diluted compared to the corresponding end-points of fermentation (Fig. 6). Whereas asparagine remained as abundant in the soaking waters as in the fermentation waters, glutamic acid and GABA were no longer found in the soaking waters. Instead, the histidine and lysine concentrations were relatively high, albeit still at low levels (5 mg/liter).
Among the 65 volatile compounds found in the soaking waters, 35% were esters, 20% alcohols, 14% ketones, 12% aldehydes, and 9% terpenes/terpenoids (Fig. 7). After the standard fermentation, the total aroma intensity dropped to 27% of F16W after washing and remained at a similar level until the end of soaking. In comparison, the total aroma intensity at SE0W was 16% of F64W and increased to 22% toward the end. The major volatile compounds were still dominated by cluster C2 compounds, as in the fermentation step. The major esters and alcohols remained the same, but high intensities of these compounds were found after the extended fermentation compared to that for the standard one. In comparison, the compounds in cluster C1, such as butyl propanoate, butyl 2-propenoate, butyl butanoate, 1-butanol, and 3-hydroxy-2-butanone, occurred at similar aroma intensities in the soaking waters as in the fermentations. Among the terpenes/terpenoids, linalool (3,7-dimethyl-1,6-octadien-3-ol) remained as the most abundant compound, as in the fermentation water samples, while the other compounds stayed at lower aroma intensities.
(iii) Correlation analysis between the microbiota and volatile compounds. This analysis indicated that esters and alcohols detected in the fermentation and soaking waters, were positively correlated with the taxa Leuconostoc and Lactobacillus and negatively correlated with enterobacteria (see Fig. S5). Some volatiles, in particular, cluster C3 compounds, were positively correlated with the enterobacteria.
(iv) Metabolite course in postfermentation waters. Compounds such as lactic acid, ethanol, fructose, acetic acid, and glucose that were present during fermentation were also abundant in the PFW samples (see Fig. S6). However, butyric acid and propionic acid were found in much higher concentrations during the storage of postfermentation waters, and their concentrations were inversely related to the concentrations of the simple carbohydrates.
Temporal metabolic response of the coffee beans during postharvest processing.All coffee bean samples were rich in an array of coffee bean endogenous compounds, which contained, in decreasing order of concentrations, sucrose, 3-caffeoylquinic acid (3-CQA), caffeine, trigonelline, citric acid, malic acid, and quinic acid. Most of these compounds remained relatively stable during the entire coffee processing chain. In the following paragraphs, only the changes in the compounds targeted are featured (Fig. 5, 6, and 8).
Hierarchical clustering analysis and heatmap visualization of nontargeted aroma profiles (column dendrogram) in the coffee beans (B) during fermentation, soaking, and drying. The aroma intensities normalized to the internal standard are summed and displayed in blue on the right (in arbitrary units [AU]); a color key as a measure for the concentrations is also displayed on the right. Sample abbreviations are as in the legend for Fig. 1.
(i) Metabolite profiling of coffee beans during fermentation. Corresponding to the compounds of cluster A2 in the fermentation waters, most of these compounds were also found at increasing concentrations on the beans (B) (Fig. 5). For example, mannitol and lactic acid concentrations were 3-fold higher, whereas acetic acid, ethanol, and glycerol concentrations were 2-fold higher in F64B than in F16B. After 64 h of fermentation, the concentrations of ethanol, acetic acid, mannitol, and lactic acid in F64B were 4.6, 4.5, 3.6, and 2.2 g/kg, respectively. In addition, although the total simple carbohydrate contents remained relatively stable, glucose and fructose concentrations changed constantly during the fermentation, peaking at F24 and F48. Sucrose concentrations evolved along an out-of-phase pattern compared with that of glucose and fructose, and reached peak concentrations at F12 and F36. Among the proteinogenic amino acids targeted, all but cysteine were found on the beans (Fig. 6). The major amino acids, glutamic acid and asparagine, represented on average 55% of the total free amino acids throughout coffee processing. The concentrations of some amino acids, e.g., glycine, methionine, threonine, and histidine, increased upon fermentation. GABA displayed the highest increase, with a concentration that was 10-fold higher in F64B (353 mg/kg) than in F0B after 64 h of fermentation. The concentrations of glycine, methionine, and histidine also increased, whereas the concentrations of the remaining amino acids remained relatively stable.
More than 170 volatile compounds were detected on the beans along the entire processing chain, among which, 26% were alcohols, 23% esters, 14% aldehydes, 7% terpenes/terpenoids, and 7% furans/furanones (Fig. 8). Among them, 70% were found on the beans during fermentation. At the beginning, the major volatile compounds were mainly aldehydes (e.g., 3-methylbutanal, 3-methyl-2-butenal, methional, and benzaldehyde) and alcohols (e.g., 3-methyl-1-butanol, 5-methyl-1-hexanol, 1-octen-3-ol, and 3-methyl-1-propanol), which belonged to clusters D1 and D3. Some esters were also present, such as ethyl 2-methylbutanoate, 3-methylbutanoate, and acetate 3-methylpentanoate. Other compounds, such as linalool, d-limonene, 2-pentylfuran, and 2-methoxy-3-(2-methylpropyl)pyrazine occurred at relatively high abundances. As fermentation progressed, the total aroma intensities remained similar for the first 16 h of fermentation but doubled after 64 h (F64). As seen in cluster D1, existing and newly appearing alcohols and aldehydes contributed to part of the increment, whereas the production of esters that were also abundant in the fermentation waters contributed to another part. Towards the end of the extended fermentation, more ethyl esters appeared on the beans, such as ethyl acetate, ethyl 2-hydroxypropanoate, ethyl butanoate, ethyl benzene acetate, ethyl hexanoate, 3-methyl-1-butanoyl acetate, and ethyl pentanoate. Among the terpenes/terpenoids, linalool and d-limonene were the major compounds found on the beans, and their aroma intensities increased during fermentation. However, the concentrations of other terpenes/terpenoids (e.g., citral, α-pinene, α-terpineol, and 3-carene) remained relatively stable.
(ii) Metabolite profiling of coffee beans during soaking. After washing, the concentrations of mannitol and lactic acid in S0B and SE0B decreased to 20% of those in F16B and F64B, whereas the concentrations of ethanol (60% and 95%, respectively) and acetic acid (86% and 69%, respectively) decreased to a lower extent in F16B and F64B. Due to the higher concentrations in F64B, these compounds remained 1 to 3 times more abundant in SE0B than in S0B. At the end of soaking, these compounds diffused back into the soaking water after the extended fermentation, but the endpoint SE24B still contained 1.4 times more lactic acid than S24B. The CGA concentrations as well as the concentrations of trigonelline decreased slightly during the soaking step of the extended fermentation but remained relatively stable after the soaking step of the standard fermentation. The GABA concentration in S0B was similar to that in F16B (143 mg/kg) and remained stable during soaking. In contrast, the GABA concentration in SE0B (180 mg/kg) was only half of that in F64B, but it continuously increased to 220 mg/kg during soaking. The total free amino acid concentrations decreased to 90% and 86% of the corresponding fermentation points in S0B and SE0B, respectively, with the decreases in asparagine, aspartic acid, serine, leucine, and methionine concentrations mainly contributing.
A total of 110 volatile compounds were detected in the soaking beans, among which, 33% were esters, 30% alcohols, and 10% aldehydes. The major aroma compounds on the soaking beans remained similar to those during fermentation, but the total aroma intensities dropped to 80% and 65% of the corresponding end points of fermentation in S0B and SE0B, respectively. Compared to that in the standard fermentation, the soaking beans contained a higher abundance of esters (clusters D1 and D2) and reached a two times higher aroma intensity after the extended fermentation. For instance, compounds such as ethyl pentanoate, ethyl-2-hydroxypropanoate, and ethyl hexanoate were 5 to 25 times higher in SE24B than in S24B. Other volatile compounds, such as 2-methyl-1-penten-3-ol, butyrolactone, linalool, 2-methyl-2-butenal, and 2/3-methylbutanal, were also found at higher intensities in SE24B than in S24B.
(iii) Correlation analysis between the microbiota and volatile compounds. This analysis indicated that esters and alcohols (belonging to clusters D1 and D3) in the beans during fermentation and soaking were positively correlated with the taxa Leuconostoc and Lactobacillus and negatively correlated with enterobacteria (in particular, Pantoea, Klebsiella, and Tatumella) (Fig. S5).
(iv) Metabolite profiling of coffee beans during drying. During the drying step (D), the total concentrations of simple carbohydrates, aroma compounds, sugar alcohols, and CGAs dropped slightly for both the standard and extended fermentations, except for some isomers of the CGAs. Citric acid and malic acid concentrations increased in the beans (Fig. 5). The evolution patterns of the glucose and fructose concentrations were similar but out of phase with the concentration pattern of sucrose. The ratios between certain isomers of CGAs also increased during drying, whereas they remained stable during fermentation and soaking. For instance, the ratio of 4,5-diCQA to 5-CQA increased from 0.2 to 0.5, whereas the ratio of 4,5-diCQA to 3,5-diCQA increased from 0.5 to 1.2 during drying in both processes. The resulting green coffee beans finally contained different fat contents, with 11.0% ± 0.2% and 9.5% ± 0.5% in standard- and extended-processed green coffee beans (SB and EB), respectively. EB contained significantly higher concentrations of glucose (4.7 times), fructose (2.5 times), lactic acid (5.0 times), and mannitol (1.4 times) than SB, whereas the concentrations of trigonelline (0.8 times) and glycerol (0.4 times) were significantly lower in EB. The concentrations of many free amino acids, such as GABA, glutamic acid, aspartic acid, proline, valine, tyrosine, and glutamine, built up during the drying step, were accompanied by a decrease of the concentrations of asparagine and alanine. The total amino acid concentrations were higher in EB than in SB. The GABA concentrations continued to increase, and the final concentration in EB (338 mg/kg) was more than 2 times that in SB (144 mg/kg).
During drying, a total of 120 volatile compounds were detected, among which, 26% were alcohols, 17% esters, and 11% aldehydes. The major volatile compounds switched to 1-hexanol, 3-methyl-1-butanol, 1-pentanol, and 2,3-butanediol, as well as hexanal, decane, and butyrolactone. Increases in the aroma intensities of furanones, ketones, and furans appeared once the drying started (green coffee bean smell), as seen in cluster D4, represented by decane, 2-pentanone, 2,6-dimethylpyridine, butyrolactone, undecane, 2-ethyl-5-methyl furan, β-pinene, dihydro-5-methyl-2(3H)-furanone, and 3-pentanone. Cluster D1 compounds were at much lower abundances during the drying step. At the end of drying, the major esters mentioned above were still present and in higher aroma intensities in EB than in SB. The aroma intensities of other volatile compounds, such as linalool (12 times), ethyl-3-methylbutanoate (9 times), and 2,3-butanediol (5 times), were also higher in EB than in SB.
Sensory analysis of coffee beverages.The overall odor intensity and fruity flavor of the coffee beverages were found to be significantly discriminant (P < 0.05) between the coffees brewed from beans subjected to standard and extended fermentations during postharvest processing. Other attributes, including fruity odor, bitterness, acidity, burnt, earthy, rubber, malty, cereal, and cocoa/coffee flavor notes, were perceived as different (0.05 < P < 0.50) (see Fig. S7). Specifically, the coffee from the standard fermentation processing had a higher overall odor intensity, body, and astringency and was more bitter, burnt, earthy, rubber, and cocoa/coffee-like in flavor. The coffee from the extended fermentation processing was perceived as having a more fruity and floral odor and had a higher acidity, fruitiness, and maltiness flavor.
DISCUSSION
Coffee postharvest processing must have an impact on the composition of green coffee beans and hence on the coffee cup quality (1–3). This paper unraveled five major mechanisms for how fermentation duration during wet coffee processing affects the coffee microbiota, green coffee bean composition, and coffee beverage flavor.
(i) First, it showed that the plantation and processing environment and harvesting-depulping interval are important for the microbial evolution during coffee processing. Indeed, the coffee trees and the cherries that they carried were likely the sources of the microorganisms found during coffee processing. Microbial taxa that were highly prevalent during cherry harvesting and pooling (such as Leuconostoc, AAB, and enterobacteria) or during the early stages of fermentation (particularly Leuconostoc) were frequent in the coffee phyllosphere. Taxa that were highly prevalent in the coffee phyllosphere but less so during further processing (such as soil-related taxa, encompassing Bacillales, Rhizobiales, and Gammaproteobacteria) were less suited to the coffee processing conditions. Thus, they were quickly outcompeted by more fastidious taxa, such as Leuconostoc, AAB, and enterobacteria. Hence, the sleeping taxa of the coffee phyllosphere became prevalent microbial members during processing (particularly during fermentation) as a result of changing environmental conditions. These fastidious taxa were occasionally found on the apparatuses used for coffee processing. However, it is unlikely that these taxa originated from these pieces of apparatus, since they were already highly prevalent on the coffee cherries before processing started and these cherries came into contact with the apparatus. Thus, their presence on these pieces of apparatus might be due to remnant DNA of previous cross-contaminations. The contact with substrates was found to be the driving force for the development of microbial communities on apparatus surfaces, as shown before in a brewery context (21). The switch in microbiota at the onset of the coffee processing was reflected in the genetic diversity of its different steps. The genetic diversity of bacterial populations during fermentation and soaking differed significantly from that of the apparatus and environment. Concerning the fungal diversity, the prevalence of Pichia on some pieces of apparatus (depulper, fermentation tank, and soaking tank) pushed the genetic diversity profile closer toward processing (fermentation and soaking), making the distinction between processing and environment (including apparatus) less clear.
The crucial factors facilitating the switch from soil- and plant-related taxa to LAB, AAB, and enterobacteria were the availability of simple carbohydrates (present in the sap exuding from cherries inside the storage bags) and the relatively mild conditions (pH and temperature) experienced on the cherries inside the bags. Before depulping can start, the cherries need to be amassed in sufficient amounts and are therefore pooled in bags awaiting further processing. Inside these bags, the cherries are tightly packed and experience mechanical pressure. This causes the cherries to burst and to leak sap. As the coffee cherry mesocarp is rich in fermentable carbohydrates, such as glucose, fructose, and sucrose (3), the available sap was utilized by the epiphytic coffee cherry microbiota to initiate growth. When simulating the conditions of the harvesting-depulping interval for freshly harvested cherries, yeasts, AAB, and LAB could indeed grow. Yet, none of these microbial groups grew as substantially or as quickly as the LAB. This indicated an aptness of LAB for rapid growth when simple fermentable carbohydrates became available (rapid consumption and acidification) and when external conditions were not too harsh. Moreover, the LAB not only were able to grow in circumstances similar to those experienced in the storage bags but also outgrew other microbial groups during the simulated harvesting-depulping period. This competitive advantage of LAB has been shown before in other food-related niches (22). The combination of the harvesting-depulping interval and the availability of fermentable carbohydrates thus resulted in a prefermentation prevalence of microorganisms, in particular, LAB. Moreover, the microbial counts on cherries inside the bags and on depulped beans before fermentation were higher than reported previously (4, 23, 24).
(ii) Second, microbial community dynamics and intertwined microbial and endogenous activities occurred during fermentation and soaking. Indeed, the evolution of the chemical profiles in the fermentation and soaking waters reflected a dynamic and complex nature of the wet coffee processing chain, mainly as a result of three factors: (i) the constant release of nutrients from mucilage into the fermentation waters, (ii) the microbial activities present in the fermentation and soaking waters, and (iii) a minor exchange of compounds between microbial metabolites and compounds in the coffee beans. Concerning the first factor, the release of compounds from the mucilage provided the starting nutrients for the microorganisms in the fermentation tank. Mucilage is rich in simple carbohydrates, amino acids, caffeine, trigonelline, quinic acid, citric acid, and malic acid, some of which can be used for microbial growth and/or concomitant metabolite production, as shown in the metabolite clusters A1, A2 (partly), and A3 of the present study (3, 18). The increasing concentrations of free amino acids (cluster B1) in the fermentation waters mostly came from the mucilage, with aspartic acid and asparagine as key amino acids. Also, some plant-related volatile compounds, especially terpenes and terpenoids (e.g., linalool and d-limonene) as well as hexanal and hexenol, increased during fermentation. These compounds are related to the plant metabolism (25). The nutrients released were continuously consumed by the microorganisms, resulting in intense microbial activities throughout both the fermentation and soaking steps. At the same time, endogenous plant enzymes could degrade the mucilage macromolecules and accelerate the release of compounds into the fermentation waters.
Concerning the second factor, at the onset of fermentation, the microorganisms consumed sucrose (prior to glucose and fructose), followed by citric acid and malic acid. Correspondingly, the accumulation of microbial metabolites, especially mannitol, lactic acid, ethanol, and acetic acid (cluster A2 compounds), was proportional to the fermentation time. This fast consumption of nutrients at the beginning of fermentation substantiated the concomitant high microbial counts that resulted from the growth spurt of microorganisms during the harvesting-depulping interval. The increasing production of metabolites during the standard and extended fermentations confirmed the continuous growth of LAB as well as a prevalent shift from Leuconostoc to Lactobacillus upon extended fermentation. The metabolite profile was typical for heterofermentative LAB species of Leuconostoc and Lactobacillus. These LAB species, utilizing glucose via the phosphoketolase pathway, preferentially metabolize disaccharides (sucrose) through phosphorylase activity in plant substrates, whereas fructose is preferentially or exclusively reduced to mannitol (22). The delayed consumption of malic acid and citric acid might indicate a shift in metabolism of straightforward substrates (mono- and disaccharides) to a more specialized metabolism (of organic acids, alkaloids, and phenolics), as is also seen during vegetable and fruit fermentations (26, 27). Alternatively, the amino acids arginine and glutamine (cluster B2), followed by valine, isoleucine, and leucine, were consumed toward the end of the extended fermentation. This consumption of branched-chain amino acids was the result of microbial activities (likely LAB, since they were highly prevalent during fermentation). This LAB-mediated breakdown of branched-chain amino acids and its relevance to flavor have been extensively documented for other fermented foods (28, 29). These amino acids are converted via α-ketoacids into aldehydes, alcohols, and/or carboxylic acids. These intermediates, such as 2/3-methylbutanal, 2/3-methylbutanol, 2-methylpropanol, and their corresponding carboxylic acids, were indeed found in the volatile profile in the fermentation waters at increasing concentrations, and they contributed to a high percentage of the aldehydes and alcohols. In combination with these compounds, a low pH in the fermentation tank, and high concentrations of lactic acid, acetic acid, and ethanol, the esters were most probably produced chemically (30). Esters occupied the majority of the volatile compounds toward the end of the extended fermentation (cluster C2 compounds). Finally, GABA could be produced by Lactobacillus or Lactococcus during both standard and extended fermentations. The GABA-producing potential of highly prevalent members of this ecosystem, such as L. brevis, has been widely reported in fermented foods (31, 32). The metabolism of glutamic acid into GABA might have resulted in a competitive advantage through better tolerance toward the acidic stress experienced during extended fermentation (33–35).
Concerning the third factor, minor exchanges of compounds between waters and beans occurred during fermentation and soaking. Compounds present at high concentrations in the fermentation and soaking waters could be adsorbed onto the beans or trapped between the coffee bean endosperm and parchment during fermentation. These compounds were partially retained on the beans after washing, which resulted in a significant drop of the concentrations and aroma intensities of both volatile and nonvolatile compounds on these beans. These compounds were then released into the clean soaking waters due to osmotic differences. For example, monosaccharide concentrations were high at the end of the standard fermentation, and monosaccharides were carried over to the soaking waters, providing the microorganisms with nutrients to grow. Leuconostoc (high prevalence during soaking) could utilize these compounds to generate lactic acid under low acidic stress. In comparison, lactic acid, mannitol, and acetic acid were present at high concentrations at the end of the extended fermentation. These compounds were also carried over into the soaking waters and resulted in a build-up toward the end of soaking. The final concentrations of lactic acid in the soaking waters were similar for both processes, although the origins of the lactic acid were different (microbial versus carry over). In addition, the increase in caffeine and trigonelline concentrations in both fermentation and soaking waters might have been due to minor leakage from the beans. Therefore, a dynamic equilibrium occurred continuously during the fermentation and soaking steps. This equilibrium might be affected by the submerging time, pH, bean-to-water ratio, and environmental temperature. These factors should be considered by coffee farmers in the future.
(iii) Third, the metabolic processes on the beans vary during drying. Compared to the chemical profile changes in the fermentation and soaking waters, the changes in the chemical profiles of the beans were similar but less pronounced during fermentation, soaking, and drying. The metabolites from prolonged microbial activities were still present after soaking and drying (lactic acid and esters). The leakage of caffeine and trigonelline into the waters, however, had no obvious impact on the chemical composition of the beans. When the beans entered the drying step, newly formed compounds, such as pyrazines and furans (cluster D4), due to the high-temperature drying (initiating Maillard reactions), gave the characteristic green coffee bean smell. At the same time, the microbiota present on the beans switched toward a prevalence of more drought-tolerant communities, in particular, yeasts, and the microbial viability decreased concomitantly. This decrease was more pronounced after the extended fermentation due to the prolonged stress.
(iv) Fourth, a significant endogenous bean metabolism within the coffee beans occurred besides microbial activities. As only the mature coffee cherries were harvested, most metabolic events that happened in the hydrated coffee beans were related to seed germination (16). The removal of pulp initiated bean germination and exposed the beans to various abiotic stresses, e.g., anoxia and osmotic stress during fermentation and soaking and dehydration during drying. Whereas GABA (indicator of general stress) built up in the beans at the fermentation and soaking steps, being produced by LAB, its retention percentage was still high at the start of the soaking step. This indicated that the majority of GABA must have been endogenous to the beans. Indeed, GABA has been reported as a signaling compound during hypoxia/anoxia and germination, whereby it can be converted into succinate through the GABA shunt of the plant (36). This corresponded to the rise of succinic acid in the beans during fermentation and soaking. During drying, the production of GABA was prominent in the coffee beans, as has been shown previously (14, 37). As live beans usually switch to an anaerobic metabolism when submerged in water, glycolysis would still occur and result in the production of pyruvate. However, due to the lack of oxygen, pyruvate was converted into ethanol by alcohol dehydrogenase and/or lactic acid by lactate dehydrogenase. Indeed, beans generate more ethanol than lactic acid under anoxia (38). This corroborated the high ethanol and acetic acid concentrations in the coffee beans, even after the washing step following the extended fermentation. The ethanol concentrations remained the same before and after washing, indicating that the ethanol originated from the endogenous bean metabolism. Furthermore, during drying, various osmolytes (such as proline and GABA) accumulated in the beans to increase the osmotic potential of the cells, as plant cells usually do under drought stress (14, 37).
The off-phase evolution profile of sucrose, compared to the profiles of glucose and fructose, might be related to the carbohydrate and lipid remobilization in the live coffee beans. Sucrose is the principal saccharide translocated in plants and seeds. Endogenous invertases can hydrolyze sucrose to glucose and fructose, which can be taken up actively by the endosperm cells prior to utilization in metabolic pathways (39). This would result in the dynamic off-phase evolution of these carbohydrates. At the same time, coffee beans are a rich reservoir of lipids (mainly triglycerides). During bean germination, triglycerides are catabolized through β-oxidation to acetyl coenzyme A (acetyl-CoA) and then enter the glyoxylate cycle for sucrose biosynthesis (39). During the extended fermentation, the beans remained hydrated for a longer time than during the standard fermentation, which might have resulted in an enhanced lipid catabolism, as the total simple carbohydrate concentrations remained similar during fermentation and soaking. This might result in the lower total lipid content reserve of green coffee beans from the extended fermentation than of green coffee beans from the standard fermentation. A high glyoxylate cycle activity has been reported during fermentation, but less so during the drying step of wet coffee processing (16). This might explain the decrease of total simple carbohydrates during the drying step, as the lipid catabolism might not be as active during this step as during fermentation and soaking.
Abundant free amino acids, such as glutamic acid and asparagine, have been reported in beans from different geographical locations (14, 40). These amino acids can be used as transport molecules in the plant to provide nitrogen. The 11S storage protein has been characterized as the major storage protein in coffee beans, which accounts for 45% of the total coffee bean proteins, and is relatively rich in glutamine, glycine, leucine, and glutamic acid. Endogenous protease activity has been shown in green coffee beans (41), which might explain the increase of free amino acids during drying and their higher concentrations after the extended fermentation. Furthermore, the latter process yielded green coffee beans with a lower total CGA content. The prolonged water stress during the extended fermentation might have caused the faster oxidation of polyphenols when the beans were reexposed to oxygen during drying, as reported previously (3). Oxidation of CGAs has already been shown during aerobic incubation of green coffee bean powder. This might be due to endogenous phenol oxidase activity or autooxidation (41). As these reactions were avoided as much as possible during sample preparation (addition of ascorbic acid and EDTA to the bean extracts), it is unlikely that the changes in these compounds were due to these processes. From the above-stated information, it is clear that by changing process parameters or the duration of the processing steps, the endogenous bean metabolism could result in a specific compositional change.
(v) Finally, upon roasting, a series of decisive chemical transformations occur. Both volatiles and nonvolatiles can be precursors of flavor in the final coffee cup. Reducing sugars and amino acids are responsible for the formation of pyrazines through Maillard reactions and Strecker degradation and of furans and furanones through caramelization (42). Such compounds are likely to give rise to the characteristic coffee and toasted notes (43). The use of time of flight mass spectrometry (TOF-MS) in the present study enabled us to find more volatile compounds in the beans at the same time than with commonly applied methods. However, a causal link between the volatile compounds and the microbiota is challenging to draw, since many of these compounds can be of microbial, endogenous bean metabolism, or chemical origin. Moreover, in comparison with previous reports, higher concentrations of esters and lower concentrations of organic acids were found (19, 43, 44). Given the similar extraction tools used, this was likely due to the different extraction protocols and detection techniques applied. Linalool, terpineol, citral, and d-limonene are less volatile than esters or aldehydes and provide foods with citrus and floral notes. Fruity esters (mainly ethyl 3-methylbutanoate) were present in higher concentrations in green coffee beans from the extended fermentation processing. As LAB increase the acidity in the fermentation water, the production of other (off-flavor) compounds might have been minimal. Also, undesirable compounds might have been degraded or favorable ones deposited onto the beans. This might have resulted in the significantly higher fruity notes of the coffee beverages. Higher CGA and trigonelline concentrations in green coffee beans from the standard fermentation processing might contribute to the higher bitterness and astringency of the final coffee cup. In comparison, higher lactic acid concentrations in these green coffee beans might contribute to higher acidity (41).
Consequently, the coffee beverages brewed from beans that were subjected to different fermentation durations were significantly different in specific attributes (in particular, fruity flavors and overall intensity). Crucially, the coffee from beans of the extended fermentation processing did not harbor any quality defects. Some notes that are favored by consumers were even more pronounced in this coffee. Yet, coffee traders and scientists generally believe that such coffee is associated with quality defects. Indeed, the generation of off-flavors (described as fermented notes) and of stinker beans (production of dimethyl sulfide and butyric acid) typically results from coffee beans that ferment too long (45, 46). However, the present study showed that extended fermentation (favoring acidification by LAB) did not necessarily lead to these defects, given the right microorganisms (LAB) that were present and the good farming practices that were applied. The compounds responsible for the defects associated with extended fermentation are often of microbial origin, notably, enterobacterial or clostridial origin (47). Yet, the coffee from beans of extended fermentation processing that came into considerable contact with high concentrations of desirable microbial (LAB) metabolites and showed prolonged endogenous metabolic activity displayed more desirable flavor notes. Therefore, the conditions prevailing during fermentation determined the microbial profile obtained through fermentation. These conditions resulted in a prevalence of LAB and the absence of off-flavor-producing microbial communities and allowed the beans to display extensive endogenous metabolic activity. Effectively, the coffee fermentation ecosystem was shielded from off-flavor production, so that a desirable flavor precursor profile could develop. LAB could aid in this safeguarding through competition for space and nutrients and extensive acidification of the fermenting mass.
When altering the conditions prevailing during fermentation (in casu, the PFW), the emergence of anaerobic taxa (notably, Clostridium) and their metabolites related to off-flavors (notably, butyric acid) prevailed. Thus, it could be that some LAB members of the ecosystem had a protective effect toward coffee quality during fermentation in wet coffee processing. The role of LAB in providing a stable microbial environment in complex ecosystems has been shown previously, for instance, during cocoa bean fermentation (48). Thus, this protective effect translated into a shielding of coffee beans from microbiota producing the typical off-flavors of overfermentation and allowed the beans to display their extensive endogenous metabolic activity. LAB could attain this shielding by rapid nutrient depletion and acidification as well as through competition for nutrients and space. In contrast, when conditions were different, microorganisms producing off-flavors occurred.
In conclusion, the present study monitored the evolution of microbial diversity, metabolites, and bean chemical profiling along the whole wet coffee processing chain and evaluated the sensory quality of the concomitant coffee beverages made. A combination of multiphasic analytical techniques enabled a deeper understanding of each step during coffee processing, which comprised dynamic and complex interactions between microorganisms, coffee beans, and environmental and processing conditions. Notably, prevalent microbial groups were selected before fermentation, in particular, during the pooling of the coffee cherries awaiting depulping. During fermentation, LAB emerged as the prevalent microbial group, and the LAB communities present depended on the fermentation duration. The LAB communities and their metabolites produced shielded the ecosystem from unwanted members (i.e., members associated with quality defects, such as enterobacteria and clostridia) and allowed the coffee beans to display their extensive endogenous metabolism. These dual developments resulted in distinct sensory profiles of coffee beverages produced from beans subjected to standard and extended fermentation processing. Whether the ecology and the metabolite profiles of the coffee beans of the present study exhibited variability or were conservative needs to be confirmed by investigating wet coffee processing in other geographies and with other coffee varieties. In time, knowledge gathered from such studies could help to drive coffee processing toward consistent postharvest processes with controllable outcomes.
MATERIALS AND METHODS
Postharvest wet processing experiments.Coffea arabica L. var. Typica coffee cherries were used for the wet processing experiments carried out at a plantation near Nanegal (Nestlé Ecuador; latitude and longitude coordinates, 0°11'25.8“N and 78°40'41.4”W, respectively; altitude, 1,329 m) in June to July 2015. This research station carries out reproducible coffee wet processing under strictly controlled conditions and according to a standardized protocol. Approximately 300 kg of healthy and mature coffee cherries were handpicked. The cherries were pooled in bags of approximately 50 kg each between harvesting and depulping. After mechanical depulping (UCBE 500; Penagos, Bucaramanga, Colombia), the coffee beans (approximately 150 kg) were submerged in clean water to ferment spontaneously in a concrete tank (1 m by 2 m by 2 m). Half of the beans were fermented for 16 h and then withdrawn (referred to as standard fermentation), while the other half was fermented for 64 h before withdrawal (referred to as extended fermentation). Whereas 16 h is the standard practice for wet fermentation at the local plantation of the present study, this fermentation duration was extended to 64 h, which corresponded to a time point when the decreasing pH became stable, to see the effect of prolonged fermentation on the microbial community dynamics, metabolite profiles, and sensory quality of the coffee produced. After fermentation, both sets of beans were washed, soaked for 24 h, and sun dried, as described previously (3). The temperature and pH of the water were monitored on-line during fermentation and soaking. Samples were taken at specific time points throughout the postharvest processing chain (pooled coffee cherries, coffee beans, fermentation waters, and soaking waters). One part was used for on-site culture-dependent microbiological analysis; the remaining part was immediately frozen at −20°C until further metagenetic and meta-metabolomic analyses. Each sample was given a specific code, as follows (Fig. 1). Pooled coffee cherry and depulped bean samples were denoted PC and DB, respectively. Fermentation samples were denoted F followed by a number (indicating the hours of the fermentation duration) and ended with B for beans or W for fermentation water. Soaking bean samples after standard and extended fermentation were denoted SS and SE, respectively, followed by a number (indicating the hours of the soaking duration) and B. The corresponding soaking water samples were denoted in the same way but ending with W. Drying bean samples after standard and extended fermentation were denoted DS and DE, respectively, followed by a number (indicating the hours of the drying duration) and B. The green coffee beans obtained through standard and extended practices were denoted SB and EB, respectively.
Environmental samples of the plantation soil (approximately 20 g of soil from different places directly under the coffee trees), coffee tree flowers (approximately 20), coffee tree leaves (approximately 10), and fresh cherries (approximately 20 g at eight different times) from different coffee trees were taken as well. Finally, based on the local farm practices, the fermentation water was drained at the end of the fermentation process, after which clean water was added to wash the fermented beans. This mixture of fermentation water and wash water, i.e., the PFW, was collected at the end of the washing step and kept in a clean tank for 24 to 48 h. PFW samples were taken at three separate occasions, denoted PFW1, PFW2, and PFW3.
Selective plating and enumeration.To follow the microbial community dynamics, five microbial groups were targeted by using selective agar media and incubation conditions, which were performed in triplicates. The total aerobic microbiota was enumerated on plate count agar (PCA), LAB on modified de Man-Rogosa-Sharpe agar supplemented with 0.1% (wt/vol) sorbic acid to inhibit fungal growth (MRS-S) (49), AAB on modified deoxycholate-mannitol-sorbitol agar (mDMS) (50), enterobacteria on violet-red-bile-glucose agar (VRBG) (51), and yeasts and molds on yeast-glucose agar (YG) (51). MRS-S and mDMS agar media were supplemented with 0.2% (wt/vol) cycloheximide and 0.005% (wt/vol) amphotericin B to inhibit fungi, and YG agar medium was supplemented with 0.3% (wt/vol) chloramphenicol to inhibit bacteria. All agar plates were incubated aerobically at room temperature for 72 h, except for VRBG (incubated aerobically at room temperature for 24 h) and MRS-S (incubated anaerobically at room temperature for 72 h). Anaerobic incubation in bags was achieved by using AnaeroGen 2.5-liter oxygen scavengers (Thermo Fisher, Waltham, MA). Depending on their origin in the processing chain, samples consisted of cherries, beans, or waters, which were serially diluted in sterile saline (0.85% [wt/vol] NaCl). Starting with 20 g of material, 10-fold dilutions were made and plated in triplicates. For each selective agar medium and time point, an appropriate dilution containing 30 to 300 colonies was used for enumeration. Counts are expressed from triplicate averages as log CFU/ml ± standard deviation. All culture media and their compounds were purchased from Merck (Darmstadt, Germany).
Isolate recovery and identification.After enumeration, 10 to 15 colonies were picked randomly (independent of their appearance, shape, or size) from MRS-S, mDMS, and YG agar media corresponding with high dilutions. These colonies were subsequently recultured in the corresponding liquid medium. After incubation for 24 to 48 h, 1.8 ml was supplemented with glycerol (final concentration of 25% [vol/vol]) and frozen at −20°C until further analysis. Cultures of purified isolates were subjected to DNA extraction, dereplication by (GTG)5-PCR genomic fingerprinting with numerical clustering, and identification of cluster representatives by 16S rRNA gene and internal transcribed spacer (ITS) region sequencing, as described previously (52). When species-level identity could not be resolved for acetic acid bacteria, the dnaK gene was sequenced using the primers dnaK-01-F and dnaK-02-R (53). The accession numbers of the reference sequences used for identification of the isolates can be found in Table S1 in the supplemental material.
Surface swabbing.The surface-associated microbiota of the processing apparatus (cherry storage bags, depulper exit shaft, empty fermentation tank, and empty soaking tank) was assessed by swabbing a surface of 10 cm2 at different angles. Cotton swab tips were wetted with 1 ml of sterile saline before swabbing. Cells attached to the tips were dislodged by sequential washing with 9 ml of saline. The resulting 10-ml cell suspensions were then microcentrifuged (10,000 × g for 10 min at 4°C) before DNA extraction.
DNA extraction and metagenetic analysis.DNA was extracted by either a total DNA extraction protocol, combining enzymatic, chemical, and mechanical cell lysis followed by phenol-chloroform-isoamyl alcohol extraction and column purification, as described previously (3), or by a commercial kit (PowerSoil DNA isolation kit; Mobio, Hilden, Germany), depending on the level of contamination. Total DNA extraction was used for the freshly harvested cherries, postharvest processing samples (pooled cherries, depulped beans, fermenting beans, and soaking beans), and PFW samples. A commercial kit for DNA extraction was used for the processing apparatus and environmental samples (plantation soil, coffee tree flowers, and coffee tree leaves). The V4 hypervariable region of the 16S rRNA gene of bacteria and the ITS1 region of the 26S rRNA gene of yeasts and molds were amplified from the DNA extracts obtained, as described previously (3). A modification of this protocol was the incorporation of a mock community during sequencing (HM-783D; BEI Resources, Manassas, VA). All amplicons generated by the two primer sets were sequenced in parallel. The first six nucleotides of every forward and reverse primer were used to separate the data sets before bioinformatic analysis. Amplicon sequence variants (ASVs) were inferred from the high-throughput amplicon sequencing data by using the dada2 package (version 1.6.0). These ASVs are an alternative for the coarser and less accurate operational taxonomic unit (OTU) clustering approach and can resolve biological differences of as little as one nucleotide (54). The filtering parameters (maxN = 0, truncQ = 2, rm.phix = TRUE, maxEE = 1, and truncLen = 230) were applied before inputting the filtered reads into dada2’s parametric error model. The truncLen parameter was not applied for the ITS1 reads, since the expected sequence length is variable for yeasts and molds. Only ASVs with total abundances of >0.1% are reported. Taxonomy was assigned with the SILVA database (version 128) for the bacterial ASVs and with the UNITE database (version 01.12.2017) for the fungal ASVs. Bacterial ASVs that were not classified to the genus level were resolved by comparing them to the EZBioCloud database (55), and fungal ASVs were resolved by comparing them to the nucleotide database of the National Center for Biotechnological Information (NCBI; Bethesda, MD) with the basic local alignment search tool (BLASTN) algorithm (56).
Growth assessment of the epiphytic coffee cherry microbiota.To assess the potential of the epiphytic coffee cherry microorganisms to grow during the harvesting-depulping interval of the postharvest processing chain, cherries were subjected to conditions mimicking their storage in large bags, which results in exudation of sap. Therefore, approximately 3 kg of freshly harvested cherries were put inside sterile plastic bags to avoid contamination from the environment. A mass of approximately 5 kg was placed on top to simulate the mechanical pressure they undergo during pooling. This mechanical pressure results from common piling of the cherries inside the bags. To follow the evolution of the growth of the main microbial groups, samples were taken and enumerated on MRS-S, mDMS, and YG agar media.
Meta-metabolomic analysis.(i) Sample preparation. Frozen samples of fermentation water plus coffee beans, soaking waters, PFW, and the coffee beans from all aforementioned samples were thawed before use. In the case of the first samples mentioned, fermentation water and coffee beans were separated. All aqueous samples were microcentrifuged (19,400 × g for 15 min at 10°C) prior to analysis. In the case of coffee bean samples, the parchment and silver skin were removed, followed by cooling in liquid nitrogen. The beans were then milled with a coffee grinder (KG49; DeLonghi, Treviso, Italy) to obtain fine powders appropriate for extraction. Three different extraction conditions were applied, including water, 0.01 N hydrogen chloride (Merck), and 40% (vol/vol) methanol (Merck), as described previously (3) with slight modifications, in that 0.2 g of powder was mixed with 5 ml of extraction solvents containing 0.2% (wt/wt) ascorbic acid (Merck) and 0.2% (wt/wt) EDTA (Merck) to inhibit oxidation and enzyme activity, respectively. Each extraction was performed in triplicates.
(ii) Moisture and total fat contents. The moisture content of the coffee beans from all samples was measured by means of an oven method (57). Therefore, grinded bean powder was dried in an oven (T5042; Heraeus, Hanau, Germany) for 24 h until the mass did not vary. The total fat content of the green coffee beans was determined by the Soxhlet method with tert-butyl methyl ether (Acros Organics, Geel, Belgium) at 40°C for 6 h (58).
(iii) Quantification of simple carbohydrates and sugar alcohols. Concentrations of simple carbohydrates (fructose, galactose, glucose, and sucrose) and sugar alcohols (arabitol, erythritol, glycerol, mannitol, myo-inositol, sorbitol, and xylitol) were quantified in triplicates with internal standardization by high-performance anion-exchange chromatography with pulsed amperometric detection (HPAEC-PAD) using an ICS 3000 chromatograph equipped with a CarboPac PA-100 and CarboPac MA-1 column (Dionex, Sunnyvale, CA), respectively (3). The internal standard (IS) solution was prepared by adding 20 mg of rhamnose (Merck) to 500 ml of acetonitrile (Merck). Both fermentation and soaking water samples, as well as the aqueous extracts of coffee beans, were mixed with the IS solution at a 1:3 ratio, microcentrifuged (19,400 × g for 15 min at 10°C), and filtered (Chromafil 0.20-μm polytetrafluoroethylene [PTFE] [in the case of simple carbohydrates] or polyethersulfone [in the case of sugar alcohols] filters; Macherey-Nagel, Düren, Germany) before injecting (10 µl) into the column.
(iv) Quantification of organic acids, alkaloids, and phenolics. The concentrations of organic acids (citric acid, fumaric acid, gluconic acid, isocitric acid, 5-ketogluconic acid, lactic acid, malic acid, oxalic acid, quinic acid, and succinic acid), alkaloids (caffeine and trigonelline), and phenolics (six CGAs [3-CQA, 4-CQA, 5-CQA, 3,4-diCQA, 3,5-diCQA, and 4,5-diCQA], ferulic acid, and caffeic acid) in both fermentation and soaking water samples, as well as coffee bean extracts, were quantified in triplicates with external (organic acids) or internal (alkaloids and phenolics) standardization by ultraperformance liquid chromatography coupled to tandem mass spectrometry (UPLC-MS/MS) using an Acquity system equipped with an HSS T3 column (Waters, Milford, MA), as described previously (3) with minor modifications. The IS solutions for alkaloids and CGAs were 1-ethyl-4-(methoxycarbonyl)pyridinium iodide (0.15 ng/ml; Merck) and rosmarinic acid (1.0 ng/ml; Merck), respectively. Both fermentation and soaking water samples, as well as the extracts of the coffee beans, were mixed with the IS solutions, microcentrifuged (19,400 × g for 15 min at 10°C), and filtered (Chromafil 0.20-μm PTFE filters) before injecting (10 µl) into the column.
(v) Quantification of free amino acids. The proteinogenic amino acids and one nonproteinogenic amino acid (GABA) were quantified in fermentation and soaking water samples, as well as in the acidic extracts of the coffee beans, in triplicates with internal standardization by high-performance liquid chromatography coupled to tandem mass spectrometry (HPLC-MS/MS) using an Alliance 2695 chromatograph equipped with a Micromass Quattro Micro (Waters). The mobile phase, at a flow rate of 1 ml/min, was composed of 1 mM formic acid and 1 mM ammonium formate at pH 4.0 with 5% (vol/vol) acetonitrile (eluant A) and 100% (vol/vol) acetonitrile with 2 mM formic acid (eluant B) (all chemicals from Merck). Two eluant programs were used to achieve good separation for the 21 amino acids on an Astec Chirobiotic T column (15 cm by 4.6 mm, 5-μm particle size; Sigma-Aldrich, St. Louis, MO). The first eluant program was used to quantify aspartic acid, arginine, cysteine, GABA, glutamic acid, glutamine, histidine, and lysine, for which the following gradient was applied: 0.0 to 10.0 min, linear from 40% to 100% eluant A; 10.0 to 26.0 min, isocratic at 100% eluant A; 26.0 to 27.0 min, linear from 100% eluant A to 40% eluant B; and 27.0 to 32.0 min, isocratic at 40% eluant A. The second eluant program was used to quantify all other compounds with a gradient as follows: 0.0 to 12.0 min, isocratic at 20% eluant A; 12.0 to 16.0 min, linear from 20% to 40% eluant A; 16.0 to 20.0 min, linear from 40% to 100% eluant A; 20.0 to 24.5 min, isocratic at 100% eluant A; 24.5 to 25.0 min, linear from 100% to 20% eluant A; and 25.0 to 30.0 min, isocratic at 20% eluant A. l-2-Amino butyric acid (1.2 ng/ml; Merck) was used as the IS. All samples were microcentrifuged (19,400 × g for 15 min at 10°C) and filtered (Chromafil 0.20-μm PTFE filters) before injecting (10 µl) into the column.
(vi) Quantification of short-chain fatty acids and low-molecular-mass volatiles. Short-chain fatty acids (SCFAs; acetic acid, butyric acid, hexanoic acid, isobutyric acid, 3-methylbutyric acid, pentanoic acid, and propionic acid) and low-molecular-mass volatiles (acetaldehyde, ethanol, ethyl acetate, ethyl lactate, and isopentyl acetate) were quantified in fermentation and soaking water samples, as well as coffee bean extracts, in triplicates with external calibration (with inclusion of 1-butanol [Merck] as IS) by gas chromatography coupled to flame ionization detection (GC-FID) using a Focus GC chromatograph (Interscience, Breda, The Netherlands) equipped with a Stabilwax-DA column (Restek, Bellefonte, PA) and a FID-80 detector (Interscience), as described previously (3).
(vii) Volatile profiling via headspace/solid-phase microextraction gas chromatography–time of flight mass spectrometry. Nontargeted volatile profiling was conducted by headspace/solid-phase microextraction coupled to gas chromatography and TOF-MS (HS/SPME-GC-TOF-MS) in triplicates using a Trace 1300 gas chromatograph (Thermo Fisher) equipped with a Stabilwax-MS column (Restek) and coupled to a BenchTOF-HD mass spectrometer (Markes International, Llantrisant, Wales). For analysis of the coffee bean samples, 1.5 g of grinded beans was incubated in a 10-ml screw-top headspace vial at 50°C for 10 min, followed by extraction with agitation at 250 rpm for 45 min using an SPME fiber (DVB/CAR/PDMS, 50/30 µm; Sigma-Aldrich). For analysis of the fermentation and soaking water samples, 2 ml of liquid was incubated at 30°C for 15 min and extracted using the same fiber for 15 min. To each sample, 10 µl of 10 ppm toluene-D8 solution (Sigma-Aldrich) was added, and the vials were placed in a tray cooled to 4°C before analysis. The volatiles from the SPME fiber were thermally desorbed at 260°C at splitless mode and resolved with a fused silica capillary column (Stabilwax-MS, 30 m by 0.25 mm, film thickness of 0.25 μm; Restek) coated with polyethylene glycol. The GC oven temperature was programmed as follows: initially 40°C for 5 min, increased to 130°C at 3°C/min, then increased to 250°C at 8°C/min, and finally held at 250°C for 2 min. Helium gas was used as the carrier gas at a flow rate of 1 ml/min. The TOF was scanned in the m/z range of 35 to 400 with a solvent delay of 2 min. The raw data were deconvoluted by TOF-DS software (Markes), followed by identification of the peaks, which was mainly based on the NIST (National Institute of Standard and Technology, Gaithersburg, MD) library and supported by the Kovats index (59) and standards commercially available. The peak area of each compound identified was normalized to the peak area of the IS. In the case of the bean samples, the peak area was further adjusted according to the moisture content of the coffee beans. Both values were considered values for aroma intensity.
Roasting and sensory evaluation.The green coffee beans from the standard and extended fermentation processing (250 g each) were roasted according to a standard protocol until the color of the roasted beans was consistent, namely, corresponding to a Color Test Neuhaus (CTN) of 90 (Neuhaus Neotec, Ganderkesee, Germany). Coffee beverages were prepared with a Moccamaster coffee machine (Technivorm, Amerongen, The Netherlands) at a ratio of 50 g of roasted beans per liter of water. They were served at 70°C in Nespresso plastic cups (80 ml). A quantitative descriptive analysis (QDA) was applied to measure the intensity of 27 sensory attributes, covering odor, flavor, and texture. The samples were evaluated one at a time for all attributes by 12 trained panelists at the Nestlé Research Center (Vers-chez-les-Blanc, Switzerland).
Statistical analysis.Concerning the microbial identifications, a centered and rotated principal-component analysis (PCA) was performed on the joint V4 and ITS1 ASV covariance in R (version 3.4.2). A network based on the presence/absence of the ASVs with relative abundances of >5% in any sample was created with the Yifan Hu proportional algorithm of Gephi (version 0.9.2). The heatmaps of the meta-metabolomics data were calculated, and these data were clustered using the package massageR and ggplot in RStudio (version 0.99.902). The distance metric was based on Pearson’s correlation coefficient. The subsequent hierarchical clustering was performed based on the average distance between the points in the two clusters. The hierarchical clustering of the metabolite data of the fermentation and soaking water samples was based on the correlation matrix of the fermentation water data, whereas the clustering of the bean samples was based on the correlation matrix of the entire data set. The scaling of each row was performed with the build-in function of heatmap.2. The scaling of the fermentation and soaking water samples was done separately because of the large differences in absolute concentrations. A correlation matrix of the microbial communities (ASV data) and the meta-metabolomics data of the fermentation and soaking steps (processing waters and coffee beans) was constructed based on the Spearman’s rank correlation coefficients. Correlations with a P value of <0.05 were visualized as a heatmap. Analysis of variance (ANOVA) was used for the QDA data of the sensory evaluation. The results were considered statistically different when the P value was <0.05.
Accession number(s).The sequences are available at the European Nucleotide Archive (https://www.ebi.ac.uk/ena) under accession number PRJEB29145 (https://www.ebi.ac.uk/ena/data/view/PRJEB29145).
ACKNOWLEDGMENTS
This work was supported by the Research Council of the Vrije Universiteit Brussel (SRP7 and IOF342 projects), the Hercules foundation (projects UABR09004 and UAB13002), and Nestec S.A., a subsidiary of Nestlé S.A.
We thank Charles Lambot for advice on the experimental design, María Isabel Larrea for helping in the preparation of the materials for the field experiments, Sander Wuyts for the help with the bioinformatics analysis, Dominique Maes for advice on the statistical analysis, and Wim Borremans for technical assistance with the analytical apparatus.
FOOTNOTES
- Received 31 October 2018.
- Accepted 5 January 2019.
- Accepted manuscript posted online 1 February 2019.
Supplemental material for this article may be found at https://doi.org/10.1128/AEM.02635-18.
- Copyright © 2019 American Society for Microbiology.
