Current Research in Biotechnology 2 (2020) 1–15

Contents lists available at ScienceDirect

Current Research in Biotechnology

journal homepage: www.elsevier.com/locate/crbiot

Temporal shotgun metagenomics of an Ecuadorian coffee fermentation

process highlights the predominance of lactic acid bacteria
⁎
Vasileios Pothakos a,1, Luc De Vuyst a, ,1, Sophia Jiyuan Zhang a, Florac De Bruyn a, Marko Verce a, Julio Torres b,c,
Michael Callanan d, Cyril Moccand d, Stefan Weckx a
a
Research Group of Industrial Microbiology and Food Biotechnology, Faculty of Sciences and Bioengineering Sciences, Vrije Universiteit Brussel, Pleinlaan 2, B-1050 Brussels,
Belgium
b
Nestlé R&D Centre Tours, 101 Avenue Gustave Eiffel, B.P. 49716, 37097 Tours Cedex 2, France
c
Nestlé Ecuador, El Inca, EC 170124, Quito, Ecuador
d
Nestlé Research, Route du Jorat 57, Vers-chez-les-Blancs, CH-1000 Lausanne 26, Switzerland

A R T I C L E I N F O A B S T R A C T

Article history: Temporal shotgun metagenomics of multiple samples of an Arabica wet coffee fermentation process, which was exam-
Received 21 October 2019 ined microbiologically and metabolomically before, was performed to complement these microbiological and bio-
Received in revised form 29 January 2020 chemical data with an in-depth in silico analysis of the structure and functions of the coffee microbiome. The
Accepted 11 February 2020
taxonomic analysis of a massive sequence dataset of ca. 16 Gbp identified >150 microbial species and distinguished
three successive, microbial phases: (i) enterobacteria, acetic acid bacteria, and some yeasts prevailing at the start;
Keywords:
(ii) lactic acid bacteria (LAB) in the second phase; and (iii) acid-tolerant LAB at the end of the fermentation process.
Coffee bean fermentation The functional analysis of prokaryotic genes, based on ca. 138,000 coding sequences, facilitated the gene repertoire
Wet processing overview of this ecosystem, showcasing differences in each fermentation stage. Initial prototrophic characters, plant
Shotgun metagenomics cell wall-degrading activities, and pectinolytic activities were followed by a more auxotrophic and saccharolytic micro-
Taxonomic analysis bial profile, due to an increasing relative abundance of LAB. The almost full genome of 22 bacterial species was recon-
Functional analysis structed and additional coffee-specific features were identified. This temporal metagenomic analysis of a case of
Leuconostoc pseudomesenteroides reproducible wet coffee fermentation processes carried out in a coffee plantation near Nanegal, Ecuador, constitutes
Lactobacillus
a breakthrough for the study of wet coffee processing and the contribution of the growth and activities of different mi-
croorganisms, in particular LAB, to this process.

1. Introduction polysaccharides, such as cellulose), simple carbohydrates (mainly glucose

Coffee is among the most valuable traded commodities worldwide,
exhibiting increasing global production, export, and consumption (ICO, n.
d.). Arabica coffee occupies the largest proportion of the market. This is
due to its superior quality, which is attributed not only to inherent geno-
typic traits of Coffea arabica cultivars (Batista and Chalfoun, 2014;
Sakiyama and Ferrão, 2014) but also to the post-harvest processing applied,
in particular the wet processing method (Avallone et al., 2001; Brando and
Brando, 2014; De Bruyn et al., 2017; Zhang et al., 2019). The implementa-
tion of an underwater fermentation in a concrete tank, during which micro-
organisms deplete the mucilaginous layer of the coffee cherries, which
firmly adheres onto the beans (seeds), is crucial for flavour enhancement
(Brando and Brando, 2014; Velmourougane, 2013; Lee et al., 2015). In par-
ticular, the mucilage contains high concentrations of pectins (next to other
⁎ Corresponding author.
E-mail address: luc.de.vuyst@vub.be. (L. De Vuyst).
1
Equal contribution.
and fructose), citric acid, malic acid, free amino acids, and phenolic com-
pounds, which can be used as nutrients or co-substrates by bacteria and
yeasts (Silva, 2014; Waters et al., 2017). Whereas there are still conflicting
views regarding the contribution of microorganisms in pectin degradation
during coffee processing, the other compounds can be degraded fast by
the microorganisms present, yielding alcohols, aldehydes, esters, ketones,
and organic acids. Such microbial activities result in a pH decline from
above 6.0 to 4.0 or lower in the fermentation water (De Bruyn et al.,
2017; Zhang et al., 2019; Pereira et al., 2015; Feng et al., 2016). However,
the fermentation step can also be detrimental for the coffee cup quality, if it
is not conducted properly, as undesirable microorganisms generate metab-
olites, such as short-chain fatty acids (e.g., butyrate and propionate), con-
tributing to off-flavours (Brando and Brando, 2014; Avallone et al., 2002;
Silva et al., 2000, 2008). Consequently, the chemical composition of
green coffee beans, resulting from wet processing of coffee cherries, is di-
rectly impacted by the diverse microorganisms thriving on the coffee cherry
substrates (Avallone et al., 2002; Batista et al., 2009; Pereira et al., 2017).
The characters of the green coffee beans acquired will be transformed dur-
ing roasting and will ultimately determine the sensorial attributes of a cup

http://dx.doi.org/10.1016/j.crbiot.2020.02.001
2590-2628/© 2020 The Authors. Published by Elsevier B.V. This is an open access
article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
V. Pothakos et al.
of brewed coffee (Waters et al., 2017; Vaughan et al., 2015; Pereira et al.,
2019).
The microbial ecology of a coffee fermentation process can be highly
variable. Enterobacteria, lactic acid bacteria (LAB), acetic acid bacteria
(AAB), bacilli, yeasts, and filamentous fungi can be involved (Waters
et al., 2017; Feng et al., 2016; Pereira et al., 2017; Masoud et al., 2004;
Carvalho Neto et al., 2017; Carvalho et al., 2018). Consequently, findings
concerning the predominant microorganisms are in several cases controver-
sial, as fermentation is attributed to either enterobacteria, LAB, or yeasts, in
turn depending on the respective technological parameters or geographical
conditions of the processes carried out (De Bruyn et al., 2017; Zhang et al.,
2019; Pereira et al., 2017). Although the role of each microbial group is not
yet clear and systematic approaches to study this are lacking, several at-
tempts to apply starter cultures for standardizing the sensorial profiles of
coffee have already been carried out (Pereira et al., 2015, 2014, 2016;
Evangelista et al., 2014; Silva et al., 2013; Velmourougane and
Shanmukhappa, 2008; Carvalho Neto et al., 2018).
A comparative study conducted on both wet and dry coffee processing
under different technological practices has outlined the variations in the
microbial structures along the post-harvest coffee processing chain and
the chemical profiles of the green coffee beans (De Bruyn et al., 2017). A de-
tailed study, encompassing amplicon sequencing and metabolite target
analysis, conducted on a wet coffee processing chain, from the harvested
coffee cherries to the green coffee bean production, in an Ecuadorian
farm has unravelled the importance of LAB during the underwater fermen-
tation step. The importance of LAB was also underlined upon amplicon se-
quencing of Colombian coffee fermentation samples (de Oliveira Junqueira
et al., 2019). To further decipher the pivotal stage of this underwater fer-
mentation during wet coffee processing, the multiphasic approach of the
latter studies lacked a shotgun metagenomic approach, which enables to
map the whole microbial community structure in much more detail and
to predict the functional traits of the microorganisms involved. Whereas
in most cases of food fermentation processes amplicon sequencing is still
applied, the number of shotgun metagenomic analyses increases
(Illeghems et al., 2012; Jung et al., 2011; Salvetti et al., 2016; Wolfe
et al., 2014), albeit that combined taxonomic and functional meta-
pathway reconstruction analyses are rare (Illeghems et al., 2015).
The present study aimed at a shotgun metagenomic analysis of multiple
samples of an underwater fermentation process on a temporal scale, as part
of a wet coffee processing carried out before on an Ecuadorian farm (Zhang
et al., 2019). In particular, it aimed to delineate the structure and function
of its microbial ecology, thereby investigating the shifts of its microbiome,
the source tracking of its microbial members with respect to the natural
habitat of the coffee plantation and processing handlings, as well as identi-
fying the core and transient microbiota. Further, it targeted the mining of
metabolic traits, mechanisms, adaptations, and plausible interactions,
through gene prediction and annotation of the metagenomic reads or
contigs to determine the potential functional role of distinct microbial spe-
cies that increase or decrease in relative abundance and are whether or not
involved in wet coffee processing.
2. Material and methods
2.1. Wet coffee processing experiments and sampling
A large-scale, wet coffee processing experiment was carried out at the
farm of a coffee plantation near Nanegal, Ecuador (Nestlé Ecuador; latitude
and longitude coordinates, 0°11′25.8″N and 78°40′41.4″W; altitude,
1329 m; https://www.google.com/maps/place/0%C2%B011'25.8%22N
+78%C2%B040'41.4%22W/@0.190509,-78.678155,13z/data=!4m2!
3m1!1s0x0:0x0) in June–July 2015, as described before (Zhang et al.,
2019) (Fig. 1). At this research station, coffee fermentations are carried
out in a reproducible way. Ripe coffee cherries belonging to the cultivar
of C. arabica L. var. Typica were selectively handpicked and stored in plastic
bags until an adequate amount of approximately 300 kg was obtained. The
coffee cherries were depulped mechanically (UCBE 500; Penagos,
2
Current Research in Biotechnology 2 (2020) 1–15
Bucaramanga, Colombia), yielding 150 kg of depulped beans that were
then transferred into a clean concrete fermentation tank (1 m × 2 m
× 2 m). Clean water was added to the depulped coffee beans until they
were fully submerged and the whole mass was left to ferment spontane-
ously. After 16 h of fermentation (further referred to as standard fermenta-
tion), half of the fermenting mass was withdrawn and the mucilage-free
beans were consecutively washed, soaked for 24 h, and sun-dried. The re-
maining half was left in the tank until it reached a total fermentation time
of 64 h, which corresponded with a time point that the decreasing pH sta-
bilized (referred to as extended fermentation), before it was further proc-
essed (Fig. 1A). Samples of coffee beans and fermentation water (±
500 g) were withdrawn at six time points (i.e., 0, 8, 16, 24, 36, and 64 h, re-
ferred to as samples F0, F8, F16, F24, F36, and F64, respectively) during the
coffee fermentation process by means of a metal cup equipped with an ex-
tendable shaft, always after the fermenting mass was stirred well. These
time points reflect the start, the middle, and the end of both the standard
fermentation process and the extended fermentation period. The pH of
each sample was measured with a digital pH meter (WTW, Weilheim,
Germany) immediately. Subsequently, all samples were frozen at −20 °C
for transport to Belgium.
2.2. Metagenomic DNA extraction
Total community DNA was extracted from 100 g of fermenting coffee
beans plus 100 g of fermentation water of the six samples taken as a func-
tion of time separately, resulting in six metagenomes, following an in-
house metagenomic DNA extraction protocol (De Bruyn et al., 2017). This
was based on consecutive enzymatic lysis steps coupled to mechanical
and chemical disruption of the microbial cells, targeting filamentous
fungi, yeasts, and bacteria simultaneously.
2.3. Sequencing, sequence data quality control, and pre-processing of the se-
quence reads
The metagenomic DNA of each of the six samples was fragmented into
pieces with an average size of 550 basepairs (bp), using the Covaris M220
device (Covaris, Brighton, UK). PCR-free libraries were created using the
KAPA Hyper Prep Kit (Sopachem, Eke, Belgium) according to the manufac-
turer's instructions, introducing a unique barcode per sample (De Bruyn
et al., 2017). After a 0.4× and 0.65× bead:DNA size selection, using
Agencourt AMPure XP magnetic beads (Analis, Suarlée, Belgium), the size
of the fragments in the final libraries was controlled using an Agilent
2100 Bioanalyzer (Agilent Technologies, Diegem, Belgium). Their concen-
trations were measured using a fluorescence-based method (Qubit 2.0; Life
Technologies, Ghent, Belgium). All libraries with a concentration above
1 ng/μl and a concordant size range were quantified, based on qPCR
using a LightCycler 480 II (Roche Diagnostics, Vilvoorde, Belgium) and
the KAPA Illumina Library Quantification Kit (Sopachem). Subsequently,
all libraries were diluted to a concentration of 2 nM. Per sequencing run,
three libraries were pooled equimolarly prior to denaturation with NaOH,
followed by a dilution to 7 pM and final paired-end (PE) sequencing by
means of an Illumina MiSeq sequencer (Illumina, Eindhoven, The
Netherlands) and a MiSeq Reagent Kit v3 (600-Cycle; Illumina). Library
preparation and sequencing were performed at the Brussels Interuniversity
Genomics High Throughput core facility BRIGHTcore (Jette, Belgium).
The sequence reads obtained were demultiplexed and quality-
processed using Trimmomatic-0.36 (Bolger et al., 2014). In addition,
the sequence reads were subjected to technical sequence removal in pal-
indrome mode (i.e., using Nextera PE adapters), allowing for two mis-
matches. These adapter seeds were extended and clipped if in the case
of PE reads a score of 30 was reached or in the case of single-ended
reads a score of 10 was reached. Further, sequence reads needed to
have an average quality score above 15, scored using a 4-base sliding
window, and leading and trailing bases were trimmed in the case that
the base scores were below 20. Only high-quality sequence reads with
a minimum read length of 50 bases were further considered. Each
V. Pothakos et al. Current Research in Biotechnology 2 (2020) 1–15

Fig. 1. Overview of the large-scale, wet coffee processing experiment carried out at a coffee plantation near Nanegal, Ecuador. A. The fermentation stage, as a pivotal step of
the wet coffee processing method followed. After harvesting and depulping of the mature coffee cherries, the depulped beans covered with a mucilaginous layer were left to
ferment underwater in a large-scale fermentation tank. The six close-ups of the fermentation tank, corresponding with the time points of analysis (indicated as F0, F8, F16,
F24, F36, and F64) show the appearance of the fermenting mass during the standard (16 h) and extended wet coffee fermentation process carried out (64 h). After depletion of
the mucilage, the fermented beans were washed, soaked, sun-dried, and finally dehulled. The green coffee beans from this process are ready for roasting. B. Relative abun-
dance (%) of all microbial genera (bacteria, yeasts, and filamentous fungi) identified (> 0.001%) through alignment of the metagenomic sequences to the non-redundant
protein database (nr) of the National Center of Biotechnology Information (NCBI), using DIAMOND, and through fragment recruitment plots (FRPs). C. Principal coordinates
analysis (PCoA) bi-plot based on Bray-Curtis dissimilarities of the microbial community structures of the six shotgun metagenomes during the course of a standard (16 h) and
extended wet coffee fermentation process (64 h).

3
V. Pothakos et al. Current Research in Biotechnology 2 (2020) 1–15

paired forward and reverse sequence read was merged into a contig, fur- was performed. Fragment recruitment plotting allows to indicate whether
ther referred to as PE contig, with a minimum overlap of 10 nucleotides or not a given species is present by evaluating whether a sufficient amount
(nt), using the PANDAseq assembler software v2.7 (Masella et al., 2012) of metagenomic sequences align to the genome sequence of that given spe-
in a Bio-Linux 8 environment (Field et al., 2006). For the unmerged PE cies, being a reliable indication of the presence of that species. Fragment
sequence reads, for which alignment was not achieved, the forward
reads were retained and the reverse ones were discarded. For the gener-
ation of the final datasets of each metagenome, PE contigs and
unmerged forward reads were imported in the Galaxy platform
(Goecks et al., 2010), combined into one FASTA file, and further re-
ferred to as metagenomic sequences (Supplementary Table S1, Supple-
mentary Fig. S1).

2.4. Taxonomic analysis based on metagenomic sequences

2.4.1. Filtering of coffee tree DNA

Before using various taxonomic analysis tools, metagenomic sequences
that derived from the coffee tree genome were filtered out. Hereto, the
quality-trimmed metagenomic sequences were aligned to the genome se-
quence of Coffea canephora (RefSeq accession number PRJEB4211
(Denoeud et al., 2014)). Although the experiments were performed using
C. arabica, which is an allotetraploid plant and a hybrid between the diploid
C. canephora and Coffea eugenoides, its genome was not publicly available at
the time of analysis. Alignment was performed using the nucleotide BLAST
algorithm (blastn) (Altschul et al., 1990) with the following settings: word
size, 25 nt; minimum alignment identity, 80%; maximum alignment hits
per read, 1; and minimal query coverage, 70%. Consequently, in the follow-
ing text, metagenomic sequences refer to the remaining, non-plant
metagenomic sequences.

2.4.2. Fast taxonomy profiling tools for the classification of prokaryotic

metagenomic sequences
First, the metagenomic sequences were analysed with fast taxonomy
profiling tools to get a rapid and precise overview of the prokaryotic diver-
sity of the coffee microbiome. The taxonomic sequence classifier Kraken is
based on the identification of k-mers in metagenomic sequences and query-
ing them (under default parameters) against a concise genomic library
(MiniKraken DB) (Wood and Salzberg, 2014). This 4 Gb pre-built database
was constructed from complete bacterial, archaeal, and viral genomes in
the National Center for Biotechnology Information (NCBI) RefSeq database
(Pruitt and Tatusova, 2007). The algorithm uses exact alignments and de-
termines lowest common ancestors (LCAs) in a taxonomic tree of all ge-
nomes that contain that k-mer. Additionally, Kaiju was implemented for
the taxonomic analysis, as a second fast taxonomic classification tool,
based on the comparison of translated metagenomic sequences to a refer-
ence database of microbial proteins (Menzel et al., 2015). Taxonomic as-
signment was performed through LCAs of amino acid sequences with
entries in a protein database of annotated bacterial, archaeal, and viral
genomes.

2.4.3. Taxonomic assignment of metagenomic sequences based on sequence

alignment
Second, to have a complementary overview of the prokaryotic diversity
based on an approach using a different algorithm and another database, the
metagenomic sequences were taxonomically assigned by aligning their
translated DNA sequences against NCBI's non-redundant protein database
(nr) using the DIAMOND algorithm (Buchfink et al., 2015). The alignment
outputs were parsed and analysed using MEGAN 6.7 (Huson et al., 2007),
based on a LCA algorithm. The parameters were set as follows: minimum
score, 100; and minimum support, 100. For this intense computational pro-
cess the Tier-2 High Performance Computing cluster Hydra of the VUB-ULB
Shared ICT Services Centre was used.

2.4.4. Fragment recruitment plots

Third, based on the outcome of the aforementioned approaches, for
which the genus level was the lowest reliable taxonomic rank, and to
(caption on next page)
allow taxonomic analysis at species level, fragment recruitment plotting

4
 V. Pothakos et al.
recruitment plotting encompasses the alignment of all metagenomic se-
quences to a custom-made database comprising a set of genome sequences
using the nucleotide BLAST algorithm (blastn). A hit was considered rele-
vant if the sequence identity was at least 80% and the query coverage
was at least 70%. Only the top hit for each metagenomic sequence was
retained. To construct the custom-made database to be used for the frag-
ment recruitment plotting, all available genomes of microbial species be-
longing to the most prevalent genera identified using the approaches
described in Sections 2.4.2 and 2.4.3 were downloaded from the RefSeq
and Whole Genome Shotgun (WGS) databases (Vermote et al., 2018).
One genome per species was selected based on the genome project status
(complete or draft), the source of isolation, the strain information available
(type or reference), and the genome size. The most favourable sequences
were closed genomes of type strains, preferably originating from plant eco-
systems and possessing the largest genome size. The sequences selected (n
= 379) belonged to the microbial groups Acetobacteraceae (n = 24),
Lactobacillaceae (n = 175), Leuconostocaceae (n = 27), Enterococcaceae-
Streptococcaceae (n = 23), Enterobacterales (n = 59), yeasts (n = 45), and
miscellanea (n = 26). The results of these alignments were visualised as
fragment recruitment plots (FRPs) for each distinct microbial group.
2.4.5. Construction of a principal coordinates analysis plot, co-occurrence/co-ex-
clusion plot, and microbial network
The microbial relative abundance data (at genus level) generated
through alignment of the metagenomic sequences against the nr database
and through FRPs were combined and subsequently imported into the
open source statistical package R version 3.2.3 (R Core Team, 2016),
implementing the vegan package (Oksanen et al., 2016). The phyloseq
package was used to construct a Principle Coordinates Analysis (PCoA)
plot based on Bray-Curtis dissimilarities (McMurdie and Holmes, 2013).
The microbial community structure data obtained through the FRPs (at
species level) were imported in the R environment. The sequence relative
abundance data for each species were used to construct a Spearman's
rank-order correlation matrix to evaluate the dependence among the taxa
(R Core Team, 2016). The co-occurrence and co-exclusion relationships be-
tween all microbial species were considered when the correlation was sig-
nificant at a confidence level of 99%. For the microbial network, the
positive Spearman correlations (with a coefficient value >0.7) between spe-
cies were extracted from this matrix and visualised through the open source
platform Gephi version 0.9.1 (Bastian et al., 2009). The betweenness cen-
trality decomposition method was used to quantify the topological impor-
tance of a node (species). The absolute number of positive correlations
with other species defined the size of each node. All edges (connections)
were indicative of the positive interaction between microbial species. The
microbial network was projected in a Fruchterman-Reingold layout.
2.5. Functional analysis of prokaryotic genes
2.5.1. Assembly of metagenomic sequences into contigs
To allow for functional analysis through gene prediction and annota-
tion, an assembly of the metagenomic sequences into contigs was per-
formed through the de novo assembler MEGAHIT version 1.1.0-pre (Li
et al., 2014), which uses succinct de Bruijn graphs (SdBGs). Assembly was
performed independently for each time point, using the meta-sensitive pa-
rameter. Multiple SdBGs were built for different k-mer lengths iteratively,
← Fig. 2. Heat map illustrating the sequence relative abundances of the species
identified in the six shotgun metagenomes during the course of a standard (16 h)
and extended wet coffee fermentation process (64 h). The time points of analysis
are indicated as F0, F8, F16, F24, F36, and F64. The sequence distribution data
were obtained from the taxonomic analysis employing fragment recruitment plots
(FRPs). The average Euclidean measure was used to obtain the distance matrix for
hierarchical clustering of samples and species. The colour bar on the left indicates
the taxonomic order each species was allocated to, whereas the colour key indicates
the absolute number of metagenomic reads recruited in each reference genome.
5
Current Research in Biotechnology 2 (2020) 1–15
starting from a k–min of 21 until a k–max of 99, with increments of 10 nt.
Only contigs with lengths higher than 1 kbp were retained.
2.5.2. Taxonomic assignment of assembled contigs
To allow for connecting genes and functions with species present in the
coffee microbiome, binning and taxonomic assignment of the assembled
metagenomic contigs was performed by aligning them to the customised
database, containing 379 genomes of strains of different microbial species
described in Section 2.4.4 using blastn, with the same parameters as de-
scribed above (Verce et al., n.d.). All assembled metagenomic contigs
from each taxonomic bin were mapped on the respective reference genome,
with the proper orientation, alignment identity, order and coverage, before
a ring image was finally generated.
2.5.3. Determination of the temporal taxonomic shifts
Analysing the temporal shifts of the coffee fermentation process exam-
ined postulated that the microbial species richness was the same for all sam-
pling points but the sequencing depth varied, depending on the changing
relative abundance of each taxon at different time points during coffee fer-
mentation. Hereto, the metagenomic sequences of each time point were
mapped on all assembled metagenomic contigs, regardless of the time
point wherefrom they were assembled. Read alignment was performed by
implementing the Burrows-Wheeler alignment tool, using the MEM algo-
rithm (BWA-MEM) version 0.7.13, which is recommended for more accu-
rate and fast high-quality queries (Li, 2013). After alignment, the SAM
output file was converted into a sorted BAM, using SAMtools v1.2 (Li and
Durbin, 2010), which was subsequently fed to the automated metagenome
binning MetaBAT-v0.30 software (Kang et al., 2015). MetaBAT was imple-
mented to calculate average depths of every assembled contig for each of
the six metagenomes.
2.5.4. Gene prediction and gene annotation
All assembled metagenomic contigs were subjected to gene prediction
and annotation through Prokka (Seemann, 2014) that relies on, among
other tools, Prodigal for gene prediction and the UniProtKB database for
assigning a function to the predicted genes. This resulted in the annotation
of the gene products, which were in addition manually curated through the
KEGG database (Kanehisa et al., 2016), and finally defined as enzymes with
references in KEGG metabolic pathways, non-enzymatic proteins, rRNA
and tRNA genes, or unknown genes labelled as hypothetical proteins.
2.5.5. Temporal gene repertoire heatmap, functional networks, and meta-
pathway reconstruction
Initially, a database was constructed containing all genes identified in
all assembled contigs. The gene database encompassed fields related to
the gene function (Section 2.5.4), the EC number, the identity of the contig
it belonged to, the taxonomic thread of the contig (Section 2.5.2), and the
average contig coverage obtained for each metagenomic sample
(Section 2.5.3), which was subsequently assigned to the genes on the re-
spective contigs.
For the construction of temporal gene repertoire heatmaps, the informa-
tion from the gene database related to specific key functions in the coffee
fermentation process were extracted and imported into the R environment.
The gene entries were plotted after scaling the average contig coverage
values, without applying hierarchical clustering and with the addition of
side colour keys indicating the taxonomy. The functional networks associat-
ing microbial taxa with polysaccharide degradation and carbohydrate up-
take mechanisms were also visualised through Gephi. In this case, a
binary matrix instead of a correlation matrix was constructed, indicating
the presence/absence of concomitant enzymes or uptake mechanisms pre-
dicted by Prokka. The cells of the matrix were populated with values 1 or
0, with 1 indicating the presence of the respective enzyme in the contigs
of a species belonging to that genus or with 0 in the case of its absence.
This binary matrix was then converted into a list of genus-carbohydrate
pairs and was fed into Gephi to generate the networks. Microbial genera
as well as polysaccharide and carbohydrate molecules were represented
V. Pothakos et al.
as nodes; edges corresponded to the presence of enzymes degrading or
importing these substrates, respectively. Additionally, specific features re-
lated with the uptake and metabolism of organic acids, phenolic com-
pounds, as well as the production of polyols were monitored on a
temporal scale.
Finally, all enzymes with reference to KEGG pathways and a signifi-
cance to the coffee fermentation process, as evidenced by former studies
(De Bruyn et al., 2017; Zhang et al., 2019), were combined into a meta-
pathway, corresponding to all metabolic features of the microbial ecosys-
tem collectively. The correlation among specific pathways with the major
microbial groups was also represented as an arc diagram (using the
arcdiagram package in the R environment) (Sanchez, 2014).
3. Results
3.1. General metagenomic sequence processing and classification strategy
Six metagenomes derived from temporal fermentation samples of a
large-scale, wet coffee processing experiment carried out on an
Ecuadorian farm were analysed (Fig. 1). They corresponded with six time
points, encompassing the start, the middle, and the end of both a standard
fermentation process (16 h) and an extended fermentation period (64 h).
Approximately 54,000,000 high-quality metagenomic sequences,
representing a total of 16 Gbp, were obtained from the combined
metagenomes of the six coffee fermentation samples treated during the
present study (Supplementary Fig. S1, Supplementary Table S1). In total,
8,000,000 metagenomic sequences were allocated to C. canephora. Conse-
quently, those sequences were filtered out before other analyses were
performed.
The fast taxonomic profiling tools Kraken and Kaiju provided a rapid
and precise first overview of the prokaryotic diversity in these six
metagenomes (Supplementary Fig. S2, S3). They allowed to classify be-
tween 8.48 and 27.47% and 28.58 and 53.51% of all metagenomic se-
quences, respectively (Supplementary Fig. S2). Subsequently, the
metagenomic sequences were aligned to NCBI's nr database using
DIAMOND, allowing the classification of between 31.11 and 54.63% of
all metagenomic sequences (Supplementary Fig. S2). These alignments
showed a large diversity of microbial genera (bacteria, yeasts, and filamen-
tous fungi), and a dynamic evolution of the microbial ecosystem through-
out the different phases of the underwater fermentation step of the wet
coffee processing carried out. Moreover, higher-organism DNA (plant, in-
sect, and higher animals) was present, indicating environmental contami-
nation from the processing facility and the natural habitat of the coffee
plantation, but this was not dealt with further.
The taxonomic assignment of metagenomic sequences through FRPs at
genus level had very comparable results to the nr database (Fig. 1B and C).
The shift of the communities as a function of time followed the same trend
in the PCoA plot (Fig. 1C). However, the allocation to species rank was not
always accurate when parsing the alignment data with MEGAN. Aligning
the metagenomic sequences to the customised database containing 379
complete and draft genomes, depicted as FRPs, facilitated an optimal
graphical and highly amenable monitoring of the sequence data, allowing
visual inspection and thus more accurate identification to species rank
(Supplementary Fig. S2, S4-S10), of which the details are further described
in Section 3.2. The genomes of the most prevalent species were fully cov-
ered by the metagenomic sequences and with very high alignment identi-
ties (> 95%), confirming their taxonomic assignment. In the cases that
the identities were lower than 95%, the allocation to species level was
more doubtful.
3.2. Temporal taxonomic analysis of the six metagenomes shows that
Leuconostoc pseudomesenteroides is present throughout the underwater fermen-
tation step of a wet coffee processing
Based on the FRPs, Leuconostoc pseudomesenteroides was highly abun-
dant for all sampling points, indicating an adaptation to the coffee cherry
6
Current Research in Biotechnology 2 (2020) 1–15
niche and processing conditions. Nevertheless, based on this FRP methodol-
ogy, three distinct phases could be identified during the underwater fer-
mentation of the wet coffee processing of the present study. In the
beginning of the underwater fermentation (0–8 h), metagenomic sequences
attributed to species allocated to enterobacteria (i.e., Tatumella,
Pectobacterium, Klebsiella, Pantoea, Enterobacter, and Rahnella), AAB
(i.e., Acetobacter and Gluconobacter), Cellulosimicrobium and Frateuria (soil-
related microbial communities), as well as yeast species belonging to the
genera Hanseniaspora, Candida, Pichia, and Wickerhamomyces (Fig. 2),
were prominent and followed an increasing trend, while the relative pro-
portion of metagenomic sequences attributed to Leuc. pseudomesenteroides
decreased. During this stage, the pH of the fermentation medium only
dropped moderately from 5.2 to around 4.5.
During the second phase of the underwater fermentation (8–24 h), the
large majority of metagenomic sequences attributed to members belonging
to the orders Enterobacterales, Micrococcales, Pseudomonadales,
Rhodospirillales, Saccharomycetales, and Xanthomonadales showed a de-
crease in relative abundances. Certain exceptions were found for
enterobacteria, as numerous representatives from the families of Enterobac-
teriaceae (i.e., Raoultella planticola), Erwiniaceae (i.e., Pantoea rwandensis,
Tatumella morbirosei), and Pectobacteriaceae (i.e., Pectobacterium
carotovorum subsp. brasiliense) remained in comparable relative abundance
levels (Fig. 2). Additionally, Cellulosimicrobium funkei and Frateuria aurantia
maintained a high proportion of the metagenomic sequences. The most
characteristic patterns during this phase were the large prevalence of
Leuc. pseudomesenteroides, which encompassed the highest proportion of se-
quences, and the increasing relative abundances of other LAB species, such
as Lactococcus lactis subsp. lactis, Lc. lactis subsp. hordniae, Leuconostoc
fallax, Leuconostoc citreum, Leuconostoc inhae, Leuconostoc mesenteroides,
Weissella spp., Enterococcus spp., and an increasing incidence of Lactobacillus
spp. At this time point, the pH of the fermentation medium remained stable
at approximately 4.0 and no further decrease was measured.
During the third and last phase of the underwater fermentation (24–
64 h), representing the extended fermentation of the coffee mucilage,
the number of Leuc. pseudomesenteroides metagenomic sequences de-
creased after 36 h of fermentation and a significant number of se-
quences (approximately 50%) was allocated to Lactobacillus
vaccinostercus, Lactobacillus brevis, and Lactobacillus plantarum. More-
over, a multitude of lactobacilli was detected as well as representatives
of the genera Weissella and Oenococcus. At this stage, the majority of
non-LAB species found in the former fermentation phases were not
detected.
3.3. Co-occurrence/co-exclusion plot and microbial network
Overall, the clustering of the microbial community composition for the
six metagenomes corroborated the distinction of the different phases
(Fig. 2). The Spearman's rank correlation plot (Fig. 3A) confirmed the neg-
ative interaction between LAB and non-LAB communities that were
characterised by high co-exclusion coefficients. Also, it pointed out the syn-
ergistic pattern of interaction among AAB, enterobacteria, yeasts, and soil-
related microbial communities that concurred in the initial phase of the un-
derwater fermentation.
When the microbial structure of this heterogeneous coffee ecosys-
tem was represented as a network, four major sub-networks could be
identified (Fig. 3B). The first diverse sub-communities of mainly
Gram-negative, soil-related, and fungal species were formed by micro-
bial members that encompassed a large number of metagenomic se-
quences at the onset of the fermentation and mainly decreased or
remained stable further on. The second cluster encompassed mainly
pectinolytic members of enterobacterial species that are also associated
with plant diseases. Lastly, two LAB sub-networks were found, whereby
only Leuconostoc spp. and acid-tolerant lactobacilli, weissellas, and
oenococci aggregated, respectively. The two subspecies of Lc. lactis
also grouped separately from all other taxa.
V. Pothakos et al. Current Research in Biotechnology 2 (2020) 1–15

Fig. 3. Statistical analysis of the metagenomic sequence data. A. Spearman's rank co-occurrence/co-exclusion plot for the most abundant bacterial and fungal lineages (>
0.1% in one metagenome). The Spearman correlation matrix was constructed with 99% confidence level. Co-occurrence is shown by a colour code (dark green, high co-
occurrence; light green, moderate co-occurrence) as well as co-exclusion (dark red, high co-exclusion; light orange, moderate co-exclusion). B. Microbial network
constructed based on the positive Spearman's rank co-occurrence correlations (coefficient values >0.7). All interacting species are represented as nodes and the positive
relationships between them is depicted as an edge. The node size corresponds to the number of positive correlations (i.e., edges) connected to the respective species node.
The colour of the nodes stands for the microbial group the microorganisms belong to (lactic acid bacteria, blue; enterobacteria, magenta; acetic acid bacteria, red; yeasts,
yellow; staphylococci, green; pseudomonads, dark green; cellulosimicrobia, turquoise; stenotrophomonads/frateurias, orange; methylobacteria, purple).

7
V. Pothakos et al.
3.4. Functional analysis of the temporal wet coffee fermentation metagenomes
correlates with the microbial community dynamics
An estimation of the functional gene potential was based on gene pre-
diction and annotation using Prokka on the metagenomic contigs assem-
bled with MEGAHIT, followed by manual curation using the KEGG and
UniProtKB databases. This information was used to predict changes in bac-
terial activities during the coffee fermentation process examined through
statistical analysis and meta-pathway reconstruction.
The assembly of the metagenomic sequences into contigs per dataset re-
sulted in a large number of contigs, i.e., 30,000–60,000 per sample, with an
average length of approximately 2 kbp (Supplementary Table S2, Supple-
mentary Fig. S11). Alignment of the contigs of the six metagenomes to
the reference genomes through FRPs facilitated their taxonomic assign-
ment. In an attempt to reconstruct the genome sequences of bacteria pres-
ent in this ecosystem, the metagenomic contigs for all data sets were
combined. An almost full genome reconstruction of 22 bacterial species
(out of 150) was possible, encompassing AAB, LAB, enterobacteria, and
soil-related bacteria (Supplementary Fig. S12). The presence of taxonomi-
cally related bacterial species with non-sequenced genomes or the occur-
rence of a diverse genotypic pool, due to different strains of the same
species, resulted in low and high alignment identities with the reference ge-
nomes. The characteristic case of the T. morbirosei genome showed the pres-
ence of two distinct genome rings, hinting towards the presence of other
closely related species (e.g., Enterobacter spp. and Acetobacter spp.). How-
ever, for the majority of the bacterial species genomes, the identity to the
references was very high, especially for the LAB members (>95.0%).
Based on the gene prediction using all metagenomic contigs, the genes
were subsequently functionally annotated. These genes encoded enzymes
with references in KEGG metabolic pathways, encoded non-enzymatic pro-
teins, or had an unknown function. After removal of all ribosomal and hy-
pothetical proteins from the datasets, a total of approximately 138,000
genes remained and were manually curated using the KEGG and UniProtKB
databases.
Overall, approximately 25,000 genes, related to cell organization, struc-
ture and regulation, were found for all different bacterial species. In gen-
eral, they were related to transcription/translation, DNA replication/
repair, and cell wall, cell membrane, outer membrane, peptidoglycan,
teichoic acid, and phospholipid biosyntheses, as well as the cell cycle, cell
division, and cell shape. The majority of genes were attributed to
enterobacteria (55.5%), followed by AAB (18.5%), LAB (17.6%), and soil-
related bacteria (8.4%), in accordance with their differences in genome
sizes and not in relative abundances.
3.4.1. Genomic features related to carbohydrate metabolism
First, with respect to polysaccharide degradation, a large number of
genes encoding different enzymes that break down arabinan (i.e., extracel-
lular exo-α-[1 → 5]-L-arabinofuranosidase), arabinogalactan (i.e.,
arabinogalactan endo-β-1,4-galactanase), glucan (i.e., glucan 1,6-α-glucosi-
dase), glucomannan (i.e., fructokinase), pullulan (i.e., pullulanase), xylan
and xylooligosaccharides (i.e., acetylxylan esterase, endo-1,4-β-xylanase,
xylan 1,3-β-xylosidase, xylan α-[1 → 2]-glucuronosidase) were found in
contigs allocated to enterobacteria, LAB, and soil-associated bacterial gen-
era. Genes encoding enzymes specific for other plant cell wall components
were also found. In detail, genes encoding enzymes related to the
depolymerisation of chitin (i.e., chitinase, chitooligosaccharide
deacetylase) and hemicellulose (i.e., β-mannanase/endoglucanase) were
found mainly for soil-related and AAB members, whereas
rhamnogalacturonan (i.e., rhamnogalacturonate lyase) could be broken
down only by enterobacteria. A network plot shows the capacities of the
most prevalent microbial genera releasing several different carbohydrate
residues (Fig. 4A.II).
Alternatively, genes encoding several different carbohydrate uptake
systems were found (Figs. 4B and 5), encompassing
phosphoenolpyruvate-dependent phosphotransferase systems (PTS),
ATP-dependent ATP-binding cassette (ABC) permeases, proton or
8
Current Research in Biotechnology 2 (2020) 1–15
cation gradient-driven transporters, and specific facilitator proteins.
The large majority of genes were allocated to cytoplasmic energy-
coupling components of PTS as well as their integral membrane perme-
ase/sugar phosphotransferase proteins, with different substrate speci-
ficities. In detail, genes encoding glucose-PTS were identified for all
enterobacteria and LAB species. Genes encoding fructose- and
mannitol-PTS were assigned to enterobacteria and lactobacilli. Addi-
tionally, all LAB taxa possessed genes encoding a mannose-PTS,
whereas a gene encoding a hexose 6-phosphate:phosphate antiporter
(uhpT), belonging to the major facilitator transporter family, was
found exclusively in contigs allocated to Leuc. pseudomesenteroides.
Moreover, genes encoding uptake systems for other carbohydrates
were also found, such as those for arabinose (C. funkei, Raoult. planticola,
Pantoea spp., Pectobacterium spp., Tatumella spp., lactococci, and
lactobacilli), lactose (all taxa), maltose (enterobacteria, lactococci, Lb.
plantarum, and Leuc. pseudomesenteroides), and trehalose (all taxa). Like-
wise, genes encoding a ribose system (rbsABC) were found for
Acetobacter orientalis, Gluconobacter spp., enterobacteria, Leuc.
pseudomesenteroides, lactococci, and C. funkei. Genes encoding a xylose
uptake system were found for enterobacteria and C. funkei. Addition-
ally, genes encoding a galactose:H + symporter were identified for
AAB species, Klebsiella spp., Pantoea spp., and Tatumella spp. Genes
encoding a sodium/glucose co-transporter were found for AAB species
as well as several genes for sugar efflux transporters, contrary to few
components of glucose-PTS. Further, genes encoding PTS for lichenan
and N,N′-diacetylchitobiose (lactobacilli, enterobacteria), oligo-β-
mannosides (Leuc. pseudomesenteroides, lactobacilli, and Pantoea spp.),
β-glucosides (lactobacilli, Leuc. pseudomesenteroides, Klebsiella spp.,
Pectobacterium spp., and Tatumella spp.) were identified. Lastly, a gene
encoding sucrose phosphorylase (Leuc. pseudomesenteroides and Leuc.
fallax) and a gene encoding sucrose 6-phosphate hydrolase (Leuc.
pseudomesenteroides, Lc. lactis, and Lactobacillus hordei) were found too.
From the heatmaps representing the gene repertoires in a temporal way
(Fig. 4B), a shift towards an overall saccharolytic character for the micro-
bial ecosystem was indicated, with the prevalence of LAB species after
16 h of coffee fermentation, when all related genomic features were highly
abundant.
3.4.2. Production/consumption of organic acids and degradation of phenolic
compounds
As different organic acids and phenolic compounds are abundant in the
coffee mucilage, serving as co-substrates for energy formation and cofactor
regeneration for the bacteria involved, the genomic features related to their
metabolism were explored. Additionally, these compounds could contrib-
ute to the antimicrobial barrier of the plant tissue and thus potentially select
for tolerant species. Genes encoding a citrate:sodium symporter were found
for Lc. lactis and Leuc. pseudomesenteroides, whereas genes encoding a cit-
rate:malate transporter were found for Enterobacter mori, Pe. carotovorum
subsp. brasiliense, and Raoult. planticola. Genes encoding a malate-2H+:
Na+-lactate antiporter were found for Lactobacillus bifermentans and Lb.
hordei. Moreover, genes encoding a fumarate:malate symporter and an ox-
alate:formate antiporter were identified for AAB and enterobacterial spe-
cies. For all species, genes encoding the enzymes of the Embden-
Meyerhof-Parnas, Entner-Doudoroff, and pentose-phosphate pathways
were found, leading to the production of lactic acid and/or acetic acid
(Fig. 6). Respiring bacteria and lactobacilli contained the genes necessary
to further convert pyruvate into fumarate or succinate, since the genes
encoding the enzymes of the reductive part of the tricarboxylic acid cycle
were present (i.e., malate dehydrogenase, fumarase, and succinate dehy-
drogenase). Genes encoding the enzymes of malolactic fermentation were
related to LAB species.
Concerning the metabolism of aromatic compounds, genes encoding
all enzymes related to the assimilation of quinic acid to produce tyrosine
(through the shikimate pathway) were found. Mainly enterobacteria,
soil-related bacteria, and lactobacilli possessed these genes, whereas
only few of these genes were identified for leuconostocs and lactococci.
V. Pothakos et al. Current Research in Biotechnology 2 (2020) 1–15

Fig. 4. Temporal gene repertoires of the six metagenomes (F0, F8, F16, F24, F36, and F64 correspond with the time points of analysis) of a standard (16 h) and extended wet
coffee fermentation process (64 h), concerning five key functions in the coffee fermentation ecosystem (A.I, Polysaccharide degradation; B, Carbohydrate transport systems
[more details can be found in Fig. 6]; C, Pectinolysis; D, Quinic acid metabolism; and E, TCA cycle). The specific genes for each function were assigned taxonomy (side colour
bar). The average sequencing coverage (Avg. Cov.) of the contigs they belonged to was used to infer the most prominent genes in the six metagenomes. The colour keys are
based on the normalisation of the average contig coverage values obtained for each gene at the six time points. The green bar graph on the right of each heatmap shows the
mean of the six average contig coverage values for each gene. In the functional network (A.II) the capacity to degrade polysaccharides (yellow nodes) is presented for the
major microbial genera found (the nodes represent the microbial groups: acetic acid bacteria, purple; enterobacteria, red; lactic acid bacteria, blue; and soil-related
bacteria, green).

With respect to other phenolic compounds, genes encoding a caffeyl- acid, hence providing ATP with the help of a ferredoxine:NAD+ oxido-
CoA reductase complex (carCDE) were found for Lb. brevis and Lb. reductase, for which the genetic evidence was also found. Similarly, a
vaccinostercus, which allowed to convert caffeic acid into hydrocaffeic gene encoding a phenolic acid decarboxylase (padC), acting on ferulic

9
V. Pothakos et al. Current Research in Biotechnology 2 (2020) 1–15

Fig. 5. Carbohydrate transport systems found in the assembled contigs of all microbial taxa (i.e., phosphoenolpyruvate-dependent phosphotransferase systems, ATP-
dependent ATP-binding cassettes, and facilitator transporter proteins). The fate of each imported molecule is indicated based on the results of the functional analysis (see
Fig. 6). Additionally, a functional network was constructed for each major microbial group found, representing the ability of different genera to take up certain
carbohydrates. The edges connecting the nodes reflect the presence of the respective transport systems. Abbreviations used: All, allose; Ara, arabitol; β-Gls, β-glucoside;
Fru, fructose; Fuc, fucose; Gal, galactose; Glc, glucose; Gls, glucoside; Glt, galactitol; HexP, hexose 6-phospate; Lac, lactose; Mel, melibiose; Mlt, maltose; Mnt, mannitol;
Mns, mannose; Rha, rhamnose; Rib, ribose; Srb, sorbose; Tre, trehalose; Xyl, xylose. The nodes represent the microbial groups (acetic acid bacteria, purple;
enterobacteria, red; lactic acid bacteria, blue; and soil-related bacteria, green), whereas all carbohydrates are represented by yellow nodes.

acid, p-coumaric acid and caffeic acid, and the genes encoding compo-
nents of a phenolic acid decarboxylase complex (bsdBCD) were pre-
dicted in contigs assigned to enterobacteria, Lc. lactis subsp. hordniae,
and Lb. brevis, whereas Lb. plantarum and Lb. vaccinostercus possessed
the genes for both types of decarboxylases.
3.4.3. Production/consumption of polyols
Genes encoding enzymes related to polyol metabolism were identified,
such as erythrose 4-phosphate dehydrogenase, catalyzing the production of
erythritol, which was found for enterobacteria, and mannitol dehydroge-
nase that reduces fructose 6-phosphate into mannitol, which was found in
heterofermentative LAB species. Alternatively, genes encoding the compo-
nents of import proteins for myo-inositol (Gluconobacter spp., LAB species,
and Pantoea spp.), erythritol (Enterobacter spp.), and xylitol (Gluconobacter
spp. and Pantoea spp.) were found.
3.4.4. Bacteriocin production, attachment, motility, stress responses, and other
features
Bacteriocin genes were identified in Lb. plantarum (plantaricin A), Lc.
lactis subsp. lactis (lactococcin A), Lc. lactis subsp. hordniae (subtilosin),
and Acetobacter spp. (colicin and linocin). Additionally, Lb. brevis, Lb. hordei,
Lb. plantarum, Lactobacillus manihotivorans, and Leuc. pseudomesenteroides
possessed genes encoding proteins that attribute immunity to microcin.
Several genes related to attachment and cell aggregation were identified
in the metagenomic contigs of 82 bacterial species. In the case of numerous
enterobacterial taxa, genes related to biofilm formation, surface adhesion,
and cell aggregation were found. In Ent. mori, three genes from the

10
 V. Pothakos et al.

11
Fig. 6. Meta-pathway reconstruction of the most significant pathways and reactions relevant to the coffee fermentation ecosystem based on the enzymes identified in the six shotgun metagenomes during the course of a standard
(16 h) and an extended wet coffee fermentation process (64 h). The arc diagram shows the pathways associated with the major microbial groups identified in the six metagenomes. (1, glucokinase; 2, phosphoglucose isomerase; 3,
phosphofructokinase; 4, aldolase; 5, triosephosphate isomerase; 6, glyceraldehyde 3-phosphate dehydrogenase; 7, 3-phosphoglycerate kinase; 8, phosphoglycerate mutase; 9, enolase; 10, pyruvate kinase; 11, lactate dehydrogenase;
12, glucose 6-phosphate dehydrogenase; 13, 6-phosphogluconate dehydrogenase; 14, ribulose 5-phosphate epimerase; 15, phosphoketolase; 16, phosphotransacetylase; 17, acetaldehyde dehydrogenase; 18, alcohol dehydrogenase;
19, acetate kinase; 20, glucose 6-phosphate dehydrogenase; 21, 6-phosphogluconolactonase; 22, 6-phosphogluconate dehydratase; 23, 2-keto-3-deoxy-6-phosphogluconate aldolase; 24, glycerol dehydratase; 25, 1,3-propanediol
dehydrogenase; 26, glycerol dehydrogenase; 27, dihydroxyacetone kinase; 28, mannitol 1-phosphate dehydrogenase; 29, fructokinase; 30, mannitol dehydrogenase; 31, citrate lyase; 32, oxaloacetate decarboxylase; 33, pyruvate
oxidase; 34, pyruvate-formate lyase; 35, lactoyl-CoA transferase; 36, CoA-dependent lactaldehyde dehydrogenase; 37, lactaldehyde dehydrogenase; 38, α-acetolactate synthase; 39, α-acetolactate decarboxylase; 40, acetoin
reductase; 41, diacetyl reductase; 42, alanine dehydrogenase; 43, pyruvate dehydrogenase; 44, pyruvate carboxylase; 45, citrate synthase; 46, aconitase; 47, isocitrate dehydrogenase; 48, α-ketoglutarate dehydrogenase; 49,
succinyl-CoA synthetase; 50, succinate dehydrogenase; 51, fumarase; 52, malate dehydrogenase; 53, isocitrate lyase; 54, malate synthase; 55, 2-methylcitrate synthase; 56, 2-methylcitrate dehydratase; 57, aconitase; 58, 2-
methylisocitrate lyase; 59, quinate/shikimate dehydrogenase; 60, 3-dehydroquinate dehydratase; 61, shikimate dehydrogenase; 62, shikimate kinase; 63, 5-enolpyruvylshikimate-3-phosphate synthase; 64, chorismate synthase;
65, CoA transferase; 66, caffeyl-CoA reductase; 67, CoA transferase; 68, galactokinase; 69, galactose 1-phosphate uridylyltransferase; 70, glucose isomerase; 71, phospho-β-galactosidase; 72, arabinose isomerase; 73, tagatose 1-
phosphate kinase; 74, tagatose aldolase; 75, arginine deiminase; 76, ornithine transcarbamoylase; 77, carbamate kinase; 78, glutaminase; 79, glutamate decarboxylase; 80, malolactic enzyme; 81, aspartate transaminase).
Current Research in Biotechnology 2 (2020) 1–15
V. Pothakos et al.
pgaABCD locus related to the biosynthesis of a poly-β-1,6-N-acetyl-D-glu-
cosamine (PGA) were found (i.e., a synthase, a N-deacetylase, and an export
protein), serving for the secretion of a biofilm adhesin polysaccharide (Itoh
et al., 2008). Genes from the ecpRABCDE operon, which encodes a common
pilus used for biofilm development and host cell recognition, were pre-
dicted in contigs allocated to Raoult. planticola and members of Pantoea
and Klebsiella. Genes of fimbrial operons, required for adherence to host ep-
ithelial cells and intestinal colonisation, were identified for T. morbirosei
and Pe. carotovorum subsp. brasiliense, among others. For numerous LAB
members, such as Lactobacillus spp., Lactococcus spp., and Leuc.
pseudomesenteroides, PGA-related genes were found. For the latter species,
numerous genes were assigned to other attachment proteins, namely a col-
lagen adhesin and a serine-rich protein.
Concerning motility features, genes associated with flagellum biosyn-
thesis (i.e., rings, hook, and filament) and the function of the flagellar
motor complex were found for 53 species, belonging to enterobacteria,
AAB, and soil-related bacteria. Motility genes were identified for two flag-
ellated LAB species, namely Lactobacillus cacaonum and Lb. hordei. Addi-
tionally, genes with a potential involvement in polysaccharide formation
were found for Lb. plantarum, Lactococcus raffinolactis, Lactobacillus
parabrevis, and Leuc. pseudomesenteroides. The same genes were identified
in Gram-negative species (various AAB and enterobacteria) that contribute
to virulence in some instances.
Numerous genes encoding chaperones, heat shock proteins, cold shock
proteins, carbon starvation proteins, and other stress response mechanisms
were found. Genes encoding specific systems associated with acidic toler-
ance, such as an arginine:agmatine antiporter and a glutamate:γ-
aminobutyrate antiporter, were found for Lc. lactis and only the latter
antiporter for Lb. brevis. Additionally, genes encoding proteins and systems
combatting oxidative stress (i.e., superoxide dismutase, thiol peroxidase,
and thioredoxin:glutathione peroxidase) were found for AAB, LAB,
enterobacteria, and soil-related bacterial taxa. Also, genes encoding cata-
lase were identified in contigs assigned to Lb. plantarum and Lb. brevis.
3.4.5. Biosynthesis of vitamins and amino acids
Genes associated with vitamin biosynthesis, such as biotin, cobalamin
and folate, were related to species of AAB, enterobacteria, and soil-related
bacteria. Genes encoding numerous uptake systems for amino acids and
peptides were identified for all bacterial species, as well as genes encoding
enzymes that convert amino acids. Of particular interest was the glutamate
decarboxylase enzyme that allows the production of γ-aminobutyric acid
(GABA) by the LAB species Lc. lactis, Lb. brevis, and Lb. plantarum.
3.4.6. Meta-pathway reconstruction
All identified pathways and reactions with significance to the coffee fer-
mentation process were used for the reconstruction of a meta-pathway for
this ecosystem (Fig. 6). Therefore, the most significant pathways related
to carbohydrate metabolism and also several secondary metabolic reac-
tions/pathways, summarizing the use of the most abundant substrates of
coffee mucilage, were combined. They could be linked with the different
microbial groups described above.
4. Discussion
The coffee fermentation process dealt with during the present study,
carried out on an Ecuadorian farm, has already been analysed microbiolog-
ically (culture-dependent and amplicon sequencing) and metabolomically
(Zhang et al., 2019). The premise of the present study, analysing six shot-
gun metagenomes on a temporal scale of the same process, postulated
that the microbial species richness was identical for these six metagenomes
and corresponded to the cumulative number of species found throughout
the six ecosystem samples taken as a function of time. Therefore, the fer-
mentation tank could be considered as a closed system, wherein all micro-
bial members always coexisted from the start till the end of the
fermentation and may or may not have been identified in each of the six
metagenomes, depending on their relative abundances and how
12
Current Research in Biotechnology 2 (2020) 1–15
exhaustively each sample was sequenced (McNew and Handel, 2015).
Therefore, the assembly of metagenomic sequences into contigs was per-
formed independently for the six metagenomes, but subsequently the
metagenomic sequences of each sample were mapped to the contigs ob-
tained for all six metagenomes. This facilitated the calculation of the aver-
age contig coverage on a temporal scale and the reconstruction of
bacterial genomes for the species identified, encompassing features from
numerous strains (depending on the size of the genotypic pool in the sam-
ples) and different time points.
Overall, dissecting the coffee fermentation process of the present study
revealed shifts in the microbial communities and variations in the gene rep-
ertoire of the coffee microbiome on a temporal scale and in relation to the
dynamic changes of the fermenting mass. The microbial succession pattern
was initiated by very heterogeneous microbial communities, as shown
culture-dependently before (Zhang et al., 2019), showcasing multiple con-
tamination routes (i.e., the natural environment of the coffee plantation and
processing facility). Typical environmental contaminants were indigenous
to the phyllosphere and rhizosphere (Jackson et al., 2015). Enterobacterial
members (i.e., Enterobacter, Pectobacterium, Klebsiella) known as plant path-
ogens (Toth et al., 2006; Gueule et al., 2015), fungi such as Gibberella spp.
responsible for coffee wilt (Geiser et al., 2005), and AAB that are usually en-
countered in carbohydrate-rich fruit niches (Pothakos et al., 2016), along
with soil-related bacteria, LAB and yeasts, were identified in the early fer-
mentation stages. The gene repertoires of the first shotgun metagenomes
(0 h and 8 h) confirmed the prototrophic nature of the microbial commu-
nity members related to plant cell wall degradation (e.g., cellulolytic and
lignolytic activities), pectinolysis (although endogenous pectinolysis
would take place as well), uptake of cleaved saccharide residues, aerobic
respiration, quinic acid metabolism, virulence, cell signalling, motility,
and chemotaxis, as well as biosynthetic features for amino acids and vita-
mins, reflecting cell and metabolic characteristics of diverse bacteria. At
the same time, a pH of the fermentation mass around 5.0 allowed the pro-
liferation of these more diverse microbial communities. However, the great
majority of the metagenomic sequences was taxonomically allocated to
LAB and only a minor fraction to yeasts. In particular, the relative sequence
abundance of Leuc. pseudomesenteroides was exceptionally high from the be-
ginning of the fermentation, suggesting that it already grew when the coffee
cherries were collected, prior to depulping (De Bruyn et al., 2017). In gen-
eral, LAB constitute a minor fraction of the microbial assemblages on intact
fruit and vegetable habitats, compared to Gram-negative bacteria and
fungi, but their relative abundance increases upon release of nutrients
from surface crevices (Di Cagno et al., 2013). The fast saccharolytic profile
of Leuc. pseudomesenteroides was reflected in the wide range of
polysaccharide-degrading enzymes, identified in its contigs, and the pres-
ence of numerous carbohydrate uptake systems, among which an exclusive
phosphate exchange protein system that facilitates the uptake of phosphor-
ylated hexoses. This element could constitute a privilege for this LAB spe-
cies in the coffee mucilage ecosystem, where damaged plant cells exude
phosphorylated compounds that can be immediately imported in bacterial
cells without consumption of ATP (Kadner, 1995). The same advantageous
uptake mechanism has been described for enteroinvasive pathogens, such
as Escherichia coli and Salmonella enterica, that penetrate mammalian cells
and make use of phosphorylated hexoses (Götz and Goebelt, 2010). At
the end of the standard wet process (16 h), the prevalence of the genus
Leuconostoc, and to a lesser extent Lactococcus, was substantiated and the
adaptation of these LAB to the process was corroborated by their clustering
in the microbial network as distinct subnets. At that time point, the removal
of the mucilage from the coffee beans was complete. This could allow leak-
age of endogenous compounds from the coffee beans into the fermenting
mass and further degradation of plant material of the mucilage, increasing
the viscosity of the fermentation water that may thus establish a lower ox-
ygen influx than before. Extended fermentation (from 24 h to 64 h) resulted
in an increased acidity (pH ~ 4.0) and subsequently a more restricted mi-
crobial diversity. Thus, the later fermentation stages were characterised
by the prevalence of initially sub-dominant, acid-tolerant LAB species
(mainly Lb. vaccinostercus, Lb. brevis, and Lb. plantarum). These findings
V. Pothakos et al.
are in accordance with the outcome of the metagenetics approach previ-
ously employed to analyse the coffee microbiome of the same fermentation
process (Zhang et al., 2019). All this pointed towards the bioprotective role
of the LAB communities, in particular Leuconostoc and Lactobacillus, that es-
tablish a competitive environment, in turn securing the controlled carrying
out of the fermentation process (De Bruyn et al., 2017). Furthermore, the
potential ability to produce bacteriocins could also contribute to the out-
growth of these taxa. However, the prevalence of LAB species resulted in
a community gene profile with reduced biosynthetic capacities and mainly
carbohydrate utilisation systems and polysaccharide degradation. Func-
tional analysis further revealed interesting metabolic features with respect
to the coffee ecosystem, for the main microorganisms identified, such as
GABA production by Lc. lactis, Lb. brevis, and Lb. plantarum, which is related
to bypassing acidic stress (Filannino et al., 2014; Kim et al., 2009; Laroute
et al., 2016), the use of caffeic acid by Lb. brevis and Lb. vaccinostercus
(Bertsch et al., 2013; Filannino et al., 2015), which is related to cofactor re-
generation, as well as the uptake of citric acid, malic acid, and quinic acid
by lactococci and lactobacilli, which are all related to cofactor regeneration,
and usually takes place during vegetable and fruit fermentations (Filannino
et al., 2014; Gänzle, 2015). Further, LAB are involved in the production of
organic acids such as lactic acid, acetic acid and succinic acid, fewer second-
ary metabolism pathways, and auxotrophic features for vitamins and amino
acids. Whereas Pichia may participate in the wet coffee fermentation pro-
cess, yeasts are particularly involved in dry coffee processing, as are species
of AAB (De Bruyn et al., 2017).
In the case of wet processing, for which many variations occur, the fer-
mentation duration, the quality of the water used, the environmental condi-
tions, and the technological steps that precede (e.g., the long storage of the
harvested coffee cherries) can influence the structure of the microbial eco-
system and consequently its function (De Bruyn et al., 2017). At the same
time, the endogenous metabolism of the coffee beans remains in a dynamic
interaction with these environmental stresses (i.e., prolonged anoxia due to
extended underwater submersion and intense acidification), triggering
transcriptional responses inside the coffee bean endosperm (Zhang et al.,
2019; Selmar et al., 2006). In all cases, these synchronous phenomena as
a whole can result in different coffee bean compositions (Zhang et al.,
2019). Therefore, heterogeneous microbial communities attached to or em-
bedded in the coffee cherry layers and also thriving in the fermenting mass
can affect the coffee aroma by degrading undesirable compounds or depos-
iting favourable metabolites onto the beans (Batista and Chalfoun, 2014;
Zhang et al., 2019; Velmourougane, 2013). Labile precursor molecules
react with coffee-specific compounds during roasting, and thus flavour-
impact molecules are generated, which contribute to bitterness, astrin-
gency, and/or acidity (Batista and Chalfoun, 2014).
Summarizing, the temporal shotgun metagenomic analysis of the ex-
tended wet coffee fermentation process of the present study revealed
three distinct microbial phases. These phases were characterised by a suc-
cession of different microbial communities. An initial diverse group of mi-
croorganisms associated with the coffee cherries was taken over by
mainly LAB communities. These LAB communities changed from the
most prevalent Leuc. pseudomesenteroides to acid-tolerant lactobacilli. The
concomitant gene profiles shifted from plant-related microbial activities
to a saccharolytic profile. These different gene profiles underlined the dif-
ferent contributions of each microbial group to wet coffee processing. The
predominance of Leuc. pseudomesenteroides could be corroborated by the
discovery of a unique uptake antiporter system of phosphorylated hexoses
leaking from lysed plant cells, which is the case for exuding mucilage that
provides a competitive advantage for this LAB species. Additionally, a
very diverse inventory of carbohydrate assimilation systems, found in the
LAB species involved and that facilitate their energy yield and utilisation
of organic acids, such as citric acid and malic acid that are abundant in
fruit tissue, could contribute to their overall proliferation. Furthermore,
the ability to use phenolic compounds of the coffee mucilage, such as caffeic
acid, ferulic acid and coumaric acid, could contribute to an energetic advan-
tage for LAB. All these metabolic activities constitute scarcely reported fea-
tures in coffee fermentation ecosystems. The present study thus highlighted
13
Current Research in Biotechnology 2 (2020) 1–15
the bacterial genetic potential of the fermentation phase during wet coffee
processing through a functional shotgun metagenomic analysis. As for any
functional study based on shotgun metagenomics that targets DNA from in-
tact cells, the limitation lies in the uncertainty about the actual gene expres-
sion. Only metatranscriptomics can elucidate actual gene expression in an
ecosystem context. However, this promising methodology also witnesses
substantial technical challenges to obtain reliable data, in particular for
field experiments as described in this study. In conclusion, although the
prevalence of LAB, the high resolution of shotgun metagenomic sequencing
applied on the coffee fermentation metagenomes of the present study ex-
panded our understanding of the microbial diversity of the coffee fermenta-
tion microbiome. This suggests that a very wide spectrum of
microorganisms originating from and adapted to diverse niches, showing
different nutritional requirements, was found in the coffee fermentation
ecosystem. Apparently, the processing conditions and intrinsic factors of
the coffee mucilage will determine the prevailing microbiota, underlining
the importance of good on-farm practices, as has been shown for cocoa
bean fermentation before (De Vuyst and Weckx, 2016). All this knowledge
will finally help to understand the link between the microbial ecology and
wet coffee quality (green coffee beans and coffee cup) and, ultimately, to
use appropriate LAB strains as tractable starter culture communities during
fermentation, thereby steering coffee flavour.
Declaration of competing interest
Authors Julio Torres, Michael Callanan, and Cyril Moccand were
employed by the company Nestlé. The remaining authors declare that the
research was conducted in the absence of any commercial or financial rela-
tionships that could be construed as a potential conflict of interest.
Acknowledgements
The authors would like to acknowledge their financial support from the
Research Council of the Vrije Universiteit Brussel (SRP7 and IOF342 pro-
jects), the Hercules Foundation (grants UABR09004 and UAB13002), and
Nestec S.A., a subsidiary of Nestlé S.A. Part of the computational resources
and services were provided by the Shared ICT Services Centre funded by
the Vrije Universiteit Brussel, the Flemish Supercomputer Center (VSC),
and the Research Foundation – Flanders (FWO-Vlaanderen).
Author contributions
Experiments on the coffee plantation: FDB, VP, JT, and MC. Conceptual-
ization, methodology, and organization of sequencing: SW; investigation,
bioinformatics analysis, and writing original draft: VP; contribution to in-
terpretation and discussion of the scientific findings: VP, SJZ, FDB, MV,
MC, CM, SW, and LDV; review and editing, funding acquisition, resources,
project management, and supervision: LDV.
Data availability
All six metagenomic datasets were submitted to the European Nucleo-
tide Archive of the European Bioinformatics Institute (ENA/EBI) and are ac-
cessible under the study accession number PRJEB24129.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.
org/10.1016/j.crbiot.2020.02.001.
References
Altschul, S.F., Gish, W., Miller, W., Myers, E.W., Lipman, D.J., 1990. Basic local alignment
search tool. J. Mol. Biol. 215, 403–410.
Avallone, S., Guyot, B., Brillouet, J.M., Olguin, E., Guiraud, J.P., 2001. Microbiological and
biochemical study of coffee fermentation. Curr. Microbiol. 42, 252–256.
V. Pothakos et al.
Avallone, S., Brillouet, J.M., Guyot, B., Olguin, E., Guiraud, J.P., 2002. Involvement of
pectinolytic micro-organisms in coffee fermentation. Int. J. Food Sci. Technol. 37,
191–198.
Bastian, M., Heymann S. & Jacomy M. Gephi: an open source software for exploring and ma-
nipulating networks. International AAAI Conference on Weblogs and Social Media
(2009). URL https://gephi.org/.
Batista, L.R. & Chalfoun, S.M. Quality of coffee beans in Cocoa and Coffee Fermentations (eds.
Schwan, R.F. & Fleet, G.H) 477–508 (CRC Press, 2014).
Batista, L.R., et al., 2009. Ochratoxin A in coffee beans (Coffea arabica L.) processed by dry
and wet methods. Food Control 20, 784–790.
Bertsch, J., Parthasarathy, A., Buckel, W., Müller, V., 2013. An electron-bifurcating caffeyl-
CoA reductase. J. Biol. Chem. 288, 11304–11311.
Bolger, A.M., Lohse, M., Usadel, B., 2014. Trimmomatic: a flexible trimmer for Illumina se-
quence data. Bioinformatics 30, 1–7.
Brando, C.H.J. & Brando, M.F. Methods of coffee fermentation and drying in Cocoa and Coffee
Fermentations (eds. Schwan, R.F. & Fleet, G.H) 367–396 (CRC Press, 2014).
Buchfink, B., Xie, C., Huson, D., 2015. Fast and sensitive protein alignment using DIAMOND.
Nature Meth. 12, 59–60.
Carvalho Neto, D.P., et al., 2017. Yeast diversity and physicochemical characteristics associ-
ated with coffee bean fermentation from the Brazilian Cerrado Mineiro region. Fermenta-
tion 3, 11.
Carvalho Neto, D.P., et al., 2018. Efficient coffee beans mucilage layer removal using lactic
acid fermentation in a stirred-tank bioreactor: kinetic. metabolic and sensorial studies.
Food Biosci. 26, 80–87.
Carvalho, D.P.D., Pereira, G.V.M., Soccol, C.R., 2018. High-throughput rRNA gene sequencing
reveals high and complex bacterial diversity associated with Brazilian coffee bean fer-
mentation. Food Technol. Biotechnol. 56, 90–95.
De Bruyn, F., et al., 2017. Exploring the impacts of postharvest processing on the microbiota
and metabolite profiles during green coffee bean production. Appl. Environ. Microbiol.
83 02398–16.
de Oliveira Junqueira, A.C., Pereira, G.V.M., Coral Medina, J.D., Alvear, M.C.R., Rosero, R.,
Carvalho Neto, D.P., Enriquez, H.G., Soccol, C.R. First description of bacterial and fungal
communities in Colombian coffee beans fermentation analysed using Illumina-based
amplicon sequencing. Sci. Rep. 9, 8794 (2019).
De Vuyst, L., Weckx, S., 2016. The cocoa bean fermentation process: from ecosystem analysis
to starter culture development. J. Appl. Microbiol. 121, 5–17.
Denoeud, F., et al., 2014. The coffee genome provides insight into the convergent evolution of
caffeine biosynthesis. Science 345, 1181–1184.
Di Cagno, R., Coda, R., De Angelis, M., Gobbetti, M., 2013. Exploitation of vegetables and
fruits through lactic acid fermentation. Food Microbiol. 33, 1–10.
Evangelista, R.S., et al., 2014. Improvement of coffee beverage quality by using se-
lected yeast strains during the fermentation in dry process. Food Res. Int. 61,
183–195.
Feng, X., et al., 2016. Culture-dependent and -independent methods to investigate the pre-
dominant microorganisms associated with wet processed coffee. Curr. Microbiol. 73,
190–195.
Field, D., et al., 2006. Open software for biologists: from famine to feast. Nat. Biotechnol. 24,
801–803.
Filannino, P., et al., 2014. Metabolic responses of Lactobacillus plantarum strains during fer-
mentation and storage of vegetable and fruit juices. Appl. Environ. Microbiol. 80,
2206–2215.
Filannino, P., Bai, Y., Di Cagno, R., Gobbetti, M., Gänzle, M.G., 2015. Metabolism of phenolic
compounds by Lactobacillus spp. during fermentation of cherry juice and broccoli puree.
Food Microbiol. 4, 272–279.
Gänzle, M.G., 2015. Lactic metabolism revisited: metabolism of lactic acid bacteria in food fer-
mentations and food spoilage. Curr. Opin. Food Sci. 2, 106–117.
Geiser, D.M., Ivey, M.L.L., Hakiza, G., Juba, J.H., Miller, S.A., 2005. Gibberella xylarioides
(anamorph: Fusarium xylarioides), a causative agent of coffee wilt disease in Africa, is a
previously unrecognized member of the G. fujikuroi species complex. Mycologia 97,
191–201.
Goecks, J., Nekrutenko, A., Taylor, J., 2010. Galaxy: a comprehensive approach for
supporting accessible, reproducible, and transparent computational research in the life
sciences. Genome Biol. 11, R86.
Götz, A., Goebelt, W., 2010. Glucose and glucose 6-phosphate as carbon sources in extra- and
intracellular growth of enteroinvasive Escherichia coli and Salmonella enterica. Microbiol.
156, 1176–1187.
Gueule, D., et al., 2015. Pantoea coffeiphila sp. nov., cause of the ‘potato taste’ of Arabica coffee
from the African great lakes region. Int. J. Syst. Evol. Microbiol. 65, 23–29.
Huson, D., Auch, A., Qi, J., Schuster, S., 2007. MEGAN analysis of metagenome data. Genome
Res. 17, 377–386.
ICO, d. International Coffee Organisationhttp://www.ico.org/trade_statistics.asp?section=
Statistics (consulted on July 3, 2018).
Illeghems, K., De Vuyst, L., Papalexandratou, Z., Weckx, S., 2012. Phylogenetic analysis of a
spontaneous cocoa bean fermentation metagenome reveals new insights into its bacterial
and fungal community diversity. PLoS One 7, e38040.
Illeghems, K., Weckx, S., De Vuyst, L., 2015. Applying meta-pathway analyses through
metagenomics to identify the functional properties of the major bacterial communities
of a single spontaneous cocoa bean fermentation process sample. Food Microbiol. 50,
54–63.
Itoh, Y., et al., 2008. Roles of pgaABCD genes in synthesis, modification and export of the
Escherichia coli biofilm adhesin poly-beta-1,6-N-acetyl-D-glucosamine. J. Bacteriol. 190,
3670–3680.
Jackson, C., Stone, B., Tyler, H., 2015. Emerging perspectives on the natural microbiome of
fresh produce vegetables. Agriculture 5, 170–187.
Jung, J.Y., et al., 2011. Metagenomic analysis of kimchi. a traditional Korean fermented food.
Appl. Environ. Microbiol. 77, 2264–2274.
14
Current Research in Biotechnology 2 (2020) 1–15
Kadner, R. Expression of the uhp sugar-phosphate transport system of Escherichia coli in Two-
Component Signal Transduction (eds. Hoch, J. & Silhavy, T.) 263–274 (ASM Press, 1995).
Kanehisa, M., Sato, Y., Kawashima, M., Furumichi, M., Tanabe, M., 2016. KEGG as a ref-
erence resource for gene and protein annotation. Nucleic Acids Res. 4 44 (D1),
D457–D462.
Kang, D.D., Froula, J., Egan, R., Wang, Z., 2015. MetaBAT, an efficient tool for accurately
reconstructing single genomes from complex microbial communities. Peer J. 3, e1165.
Kim, J.Y., Lee, M.Y., Ji, G.E., Lee, Y.S., Hwang, K.T., 2009. Production of γ-aminobutyric acid
in black raspberry juice during fermentation by Lactobacillus brevis GABA100. Int. J. Food
Microbiol. 130, 12–16.
Laroute, V., et al., 2016. GABA production in Lactococcus lactis is enhanced by arginine and co-
addition of malate. Front. Microbiol. 7, 1–11.
Lee, L. W., Cheong, M. W., Curran, P., Yu, B., Liu, S. Q. (2015). Coffee fermentation and flavor
- an intricate and delicate relationship. Food Chem. 185, 182–191 (2015).
Li, H., 2013. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM.
arXiv 1303.3997 [q-bio.GN].
Li, H., Durbin, R., 2010. Fast and accurate long-read alignment with burrows-wheeler trans-
form. Bioinformatics 26, 589–595.
Li, D., Liu, C.M., Luo, R., Sadakane, K., Lam, T.W., 2014. MEGAHIT: an ultra-fast single-node
solution for large and complex metagenomic assembly via succinct de Bruijn graph. Bio-
informatics 31, 1674–1676.
Masella, A.P., Bartram, A.K., Truszkowski, J.M., Brown, D.G., Neufeld, J.D., 2012. PANDAseq:
paired-end assembler for Illumina sequences. BMC Bioinformatics 13, 31.
Masoud, W., Cesar, L.B., Jespersen, L., Jakobsen, M., 2004. Yeast involved in fermentation of
Coffea arabica in East Africa determined by genotyping and by direct denaturating
gradient gel electrophoresis. Yeast 21, 549–556.
McMurdie, P.J., Holmes, S., 2013. Phyloseq: an R package for reproducible interactive analy-
sis and graphics of microbiome census data. PLoS One 8, e61217.
McNew, L.B., Handel, C.M., 2015. Evaluating species richness: biased ecological inference re-
sults from spatial heterogeneity in detection probabilities. Ecol. Appl. 25, 1669–1680.
Menzel, P., Ng, K.L., Krogh, A., 2015. Kaiju: fast and sensitive taxonomic classification for
metagenomics. Nat. Commun. 7, 11257.
Oksanen, J., et al., 2016. vegan: community ecology package. http://CRAN.R-project.org/
package=vegan.
Pereira, G.V.M., et al., 2014. Isolation, selection and evaluation of yeasts for use in fermenta-
tion of coffee beans by the wet process. Int. J. Food Microbiol. 188, 60–66.
Pereira, G.V.M., et al., 2015. Conducting starter culture-controlled fermentations of coffee
beans during on-farm wet processing: growth. metabolic analyses and sensorial effects.
Food Res. Int. 75, 348–356.
Pereira, G.V.M., et al., 2016. Potential of lactic acid bacteria to improve the fermentation and
quality of coffee during on-farm processing. Int. J. Food Sci. Technol. 51, 1689–1695.
Pereira, G.V.M., Soccol, V.T., Brar, S.K., Neto, E., Soccol, C.R., 2017. Microbial ecology and
starter culture technology in coffee processing. Crit. Rev. Food Sci. 57, 2775–2788.
Pereira, G.V., Carvalho Neto, D.P., Magalhães, A.I., Vásquez, Z.S., Medeiros, A.B.P.,
Vandenberghe, L.P.S., et al., 2019. Exploring the impacts of post-harvest processing on
the aroma formation of coffee beans: a review. Food Chem. 272, 441–452.
Pothakos, V., et al., 2016. Acetic acid bacteria in fermented food and beverage ecosystems. In:
Matsushita, K., Toyama, H., Tonouchi, N., Kainuma, A. (Eds.), Acetic Acid Bacteria: Ecol-
ogy and Physiology. Springer, pp. 73–100.
Pruitt, K.D., Tatusova, T., 2007. Maglott, D.R. NCBI reference sequences (RefSeq): a curated
non-redundant sequence database of genomes, transcripts and proteins. Nucleic Acids
Res. 35, 501–504.
R Core Team, 2016. R: A Language and Environment for Statistical Computing. R Foundation
for Statistical Computing, Vienna, Austria https://www.r-project.org.
Sakiyama, N.S. & Ferrão, M.A. Botany and production of coffee in Cocoa and Coffee Fermenta-
tions (eds. Schwan, R.F. & Fleet, G.H) 341–365 (CRC Press, 2014).
Salvetti, E., et al., 2016. Whole-metagenome-sequencing-based community profiles of Vitis vi-
nifera L. cv. Corvina berries withered in two post-harvest conditions. Front. Microbiol. 7,
1–17.
Sanchez, G., 2014. Introduction to the R package arcdiagram. https://github.com/gastonstat/
arcdiagram.
Seemann, T., 2014. Prokka: rapid prokaryotic genome annotation. Bioinformatics 30,
2068–2069.
Selmar, D., Bytof, G., Knopp, S.E., Breitenstein, B., 2006. Germination of coffee seeds and its
significance for coffee quality. Plant Biol. 8, 260–264.
Silva, C.F. Microbial activity during coffee fermentation in Cocoa and Coffee Fermentations
(eds. Schwan, R.F. & Fleet, G.H.) 398–423 (CRC Press, 2014).
Silva, C.F., Schwan, R.F., Dias, E.S., Wheals, A.E., 2000. Microbial diversity during maturation
and natural processing of coffee cherries of Coffea arabica in Brazil. Int. J. Food Microbiol.
60, 251–260.
Silva, C.F., Batista, L.R., Abreu, L.M., Dias, E.S., Schwan, R.F., 2008. Succession of bacterial
and fungal communities during natural coffee (Coffea arabica) fermentation. Food
Microbiol. 25, 951–957.
Silva, C.F., et al., 2013. Evaluation of a potential starter culture for enhance quality of coffee
fermentation. World J. Microbiol. Biotechnol. 29, 235–247.
Toth, I.K., Pritchard, L., Birch, P.R.J., 2006. Comparative genomics reveals what makes an en-
terobacterial plant pathogen. Annu. Rev. Phytopathol. 44, 305–336.
Vaughan, M.J., Mitchell, T., McSpadden Gardener, B.B., 2015. What’s inside that seed we
brew? A new approach to mining the coffee microbiome. Appl. Environ. Microbiol. 81,
6518–6527.
Velmourougane, K., 2013. Impact of natural fermentation on physicochemical, microbiologi-
cal and cup quality characteristics of Arabica and Robusta coffee. Proc. Natl. Acad. Sci.
India Sect. B – Biol. Sci. 83, 233–239.
Velmourougane, K., Shanmukhappa, D.R., 2008. Venkatesh, K. & Prakasan, C. B. & Jayarama.
Use of starter culture in coffee fermentation – effect on demucilisation and cup quality.
Indian Coffee 72, 31–34.
V. Pothakos et al.
Verce, M., De Vuyst, L. & Weckx, S. Shotgun metagenomics of a water kefir fermentation eco-
system reveals a novel Oenococcus species. Front. Microbiol. 10, 479.
Vermote, L., Verce, M., De Vuyst, L., Weckx, S., 2018. Amplicon and shotgun metagenomic
sequencing indicates that microbial ecosystems present in cheese brines reflect environ-
mental inoculation during the cheese production process. Int. Dairy J. 87, 44–53.
Waters, D.M., Arendt, E.K., Moroni, A.V., 2017. Overview on the mechanisms of coffee germi-
nation and fermentation and their significance for coffee and coffee beverage quality.
Crit. Rev. Food Sci. Nutr. 57, 259–274.
15
Current Research in Biotechnology 2 (2020) 1–15
Wolfe, B.E., Button, J.E., Santarelli, M., Dutton, R.J., 2014. Cheese rind communities provide
tractable systems for in situ and in vitro studies of microbial diversity. Cell 158, 422–433.
Wood, D.E., Salzberg, S.L., 2014. Kraken: ultrafast metagenomic sequence classification using
exact alignments. Genome Biol. 15, R46.
Zhang, J.S., De Bruyn, F., Pothakos, V., Torres, J., Moccand, C., Weckx, S., et al., 2019. Follow-
ing coffee production from cherries to cup: microbiological and metabolomic analysis of
wet processing of Coffea arabica. Appl. Environ. Microbiol 85. e-02635-18.