H

OH
OH
metabolites
Article
Microbial Transformations of Organically
Fermented Foods
Ruma Raghuvanshi, Allyssa G. Grayson, Isabella Schena, Onyebuchi Amanze, Kezia Suwintono
and Robert A. Quinn *
Department of Biochemistry and Molecular Biology, Michigan State University, East Lansing, MI 48824, USA
* Correspondence: quinnrob@Msu.edu; Tel.: +01-517-353-1426




Received: 15 July 2019; Accepted: 5 August 2019; Published: 10 August 2019


Abstract: Fermenting food is an ancient form of preservation ingrained many in human societies
around the world. Westernized diets have moved away from such practices, but even in these
cultures, fermented foods are seeing a resurgent interested due to their believed health benefits. Here,
we analyze the microbiome and metabolome of organically fermented vegetables, using a salt brine,
which is a common ‘at-home’ method of food fermentation. We found that the natural microbial
fermentation had a strong effect on the food metabolites, where all four foods (beet, carrot, peppers
and radishes) changed through time, with a peak in molecular diversity after 2–3 days and a decrease
in diversity during the final stages of the 4-day process. The microbiome of all foods showed a stark
transition from one that resembled a soil community to one dominated by Enterobacteriaceae, such as
Erwinia spp., within a single day of fermentation and increasing amounts of Lactobacillales through
the fermentation process. With particular attention to plant natural products, we observed significant
transformations of polyphenols, triterpenoids and anthocyanins, but the degree of this metabolism
depended on the food type. Beets, radishes and peppers saw an increase in the abundance of these
compounds as the fermentation proceeded, but carrots saw a decrease through time. This study
showed that organically fermenting vegetables markedly changed their chemistry and microbiology
but resulted in high abundance of Enterobacteriaceae which are not normally considered as probiotics.
The release of beneficial plant specialized metabolites was observed, but this depended on the
fermented vegetable.

Keywords: Fermented Food; metabolomics; GNPS; molecular networking; microbiome

1. Introduction
Fermented foods have been part of the human diet for millennia [1,2]. Though experimentation
with fermented foods began with ancient societies, such as strain cultivation and propagation, more
recently, a resurgence in the science of fermented foods has begun with the explosion of research
into the human microbiome [3,4]. It is well established that diet directly influences the structure and
function of our gut microbiome, which in turn, influences one’s overall health [5]. It has been argued
that those living in a Westernized society have been developing a gut microbial dysbiosis since the
industrial revolution [6]. There is a belief that consumption of probiotic products, such as fermented
foods, may begin to reverse this process and these foods are becoming more recognized for their health
benefits and availability to consumers. However, we still have a relatively poor understanding of the
microbes and microbial processes that exist in fermented foods, particularly pertaining to the chemical
transformations that occur due to natural microbial activity. These natural fermentations are becoming
more popular in westernized countries with many organic farmers and organic food consumers being
avid fermenters of fresh vegetables.

Metabolites 2019, 9, 165; doi:10.3390/metabo9080165 www.mdpi.com/journal/metabolites

Metabolites 2019, 9, 165 2 of 15

Many East Asians societies directly ferment vegetables with either starter cultures or naturally
occurring bacteria and some of the microbial transformations in these foods have been studied. For
example, the molecular and microbial changes in the fermentation of cabbage (Brassica rapa) into
kimchi resulted in increases in the levels of lactic acid, glycerol, pyrotartaric acid, pentanedioic acid,
2-keto-1-gluconic acid, ribonic acid, isocitric acid, and palmitic acid and a general decrease in amino
acids and sugars [7]. The metagenome of kimchi fermentation has also been explored enabling
the identification of microbial species involved and their functional characterization. Lactic acid
bacteria, such as Leuconostoc, Lactobacillus and Weissella, were the primarily genera in fermented kimchi.
These organisms used heterolactic fermentation to produce mannitol, lactate, acetate, and ethanol as
fermentation products [8]. Lactobacilli and yeasts were found to be the primary natural fermenters
of the grain-based African fermented beverage togwa, producing small aldehydes, methyl-butanal,
methy-butanols and other small fermentation products [9]. Naturally, most chemical and microbial
analyses of fermented foods have focused on small molecules from microbial fermentation, particularly
organic acids and other byproducts of microbial metabolism. Less studied, are the larger metabolites
such as plant natural products and how are they modified by microbial fermentation. Many of these
natural products are believed to have unique health benefits, which may be a hidden source of benefit
from these fermented foods.
This study used liquid chromatography-tandem mass spectrometry (LC-MS/MS), molecular
networking on the Global Natural Products Social molecular networking (GNPS) infrastructure, and
16S rRNA gene sequencing to explore the microbial transformations of four organically fermented
vegetables. We paid particular attention to flavonoids and other plant natural products released from
these foods, to determine whether natural microbial food fermentation may make these compounds
more bioavailable by releasing them from the plant prior to digestion.

2. Methods

2.1. Sample Collection

The carrots (Daucus carota), beets (Beta vulgaris), peppers (Capsicum annuum), and radishes
(Raphanus raphanistrum) were purchased from the Michigan State University Student Organic Farm
(East Lansing, MI, USA) Stand and stored at 4 ◦ C for 48 h. Each vegetable was chopped into smaller
pieces and four 0.5g samples of each were collected and stored at −20◦ C prior to fermentation. Thus,
a replicate represented pieces of one individual vegetable from the batch purchased on the same
day from the MSU organic farm. For each vegetable, four 478 mL (one pint) jars were filled 34 with
the chopped vegetables and enough salt water (50 g/L, FisherSci NaCl) to cover them was added.
One jar of each vegetable type was autoclaved for 30 min as a control (day 0 sample was taken prior to
autoclaving) and the other three replicates used for fermentation with no prior sterilization. All of the
jars were sealed and left at room temperature for 4 days (96 h) to ferment. At 24 h intervals after the
first sampling, 500 µL of liquid from each jar was placed into an microcentrifuge tube and filled to
1 mL with the corresponding chopped vegetable. Samples were taken every 24 h both for microbiome
and metabolome analysis and stored at −20 ◦ C until all the samples were collected and then moved
to −80 ◦ C (experimental schematic available in Fig. S1). The pH of each sample was measured after
thawing samples immediately prior to metabolite extraction.

2.2. Extractions
Metabolite extractions were performed by adding 1 mL of LC-MS grade methanol to a
300 µL volume of each sample. The samples were vortexed and then extracted at 4 ◦ C overnight.
After extraction, the samples were spun to pellet particulates and the supernatant was used for
mass spectrometry.
Metabolites 2019, 9, 165 3 of 15

2.3. Mass Spectrometry

A 50 µL aliquot of each sample extract was diluted 1:3 in 50% methanol containing 2 µM phenol
red standard and subjected to LC-MS/MS analysis on a ThermoTM QExactiveTM mass spectrometer
coupled to a Vanquish HPLC system. The mobile phase was an increasing gradient of solution A (100%
LC-MS grade water) to solution B (100% LC-MS grade acetonitrile) both containing 0.1% formic acid.
The stationary phase column was a Waters® Acquity® (Wood Dale, IL, USA) UPLC BEH C18 column,
2.1 mm × 100 mm. The chromatographic program was a 12-min-long run that began with a hold at 5%
solution B for 30 s then an increasing linear gradient of solution B from 5–99% over 8.5 min. This 99%
B solution was then held for 1.5 min followed by a switch to 0% B for the remaining 1.5 min. The full
MS1 mass spectra were collected in positive mode at a resolution of 17,500, an automatic gain control
(AGC) target of 5e5, scan range of 150 to 1500 m/z for the full MS mode (minutes 2–10 of run). The
MS/MS spectra were collected at the same resolution and AGC target with an isolation window of
1.0 m/z, a loop count of 5 and stepping of energies 20, 30 and 40 NCE. The data dependent settings
included a dynamic exclusion of 1.0 s.

2.4. Metabolome Data Analysis

ThermoTM QExactiveTM .raw files were first converted to the centroided .mzXML file format
and uploaded to the GNPS database and MassIVE data repository under MassIVE ID MSV000084067.
Molecular networks were generated on GNPS with the following parameters: parent mass tolerance
of 0.03 Da, fragment mass tolerance of 0.03 Da, 0.65 minimum cosine score, 4 minimum matched
peaks, and minimum cluster size of 2. Library search parameters were the same as the networks.
Feature finding was performed with the mzMine v. 2.0 software with the following parameters: MS1
noise tolerance of 500,000, MS2 noise tolerance of 50,000, and retention time tolerance of 0.2 min.
The chromatograms were deconvoluted, deisotoped, aligned with a retention time tolerance of 0.2 min,
filtered such that a peak must be found in at least 3 samples, and the missing values were imputed
using gap-filling for subsequent statistical analysis. The ThermoTM Compound DiscovererTM software
v3.0.0.294 was also used to identify and quantify known compounds in the untargeted metabolomics
dataset. The ‘untargeted metabolomics workflow’ default parameters were used to identify knowns
with mzCloud MS/MS searching and quantification of feature abundances. The samples were grouped
by vegetable type and day of fermentation for analysis.

2.5. Microbiome Sequencing

The microbiome data was generated according to the protocols developed from the Earth
Microbiome Project with some deviations described below [10] (http://www.earthmicrobiome.org/
protocols-and-standards/). Briefly, genomic DNA was extracted from the combined food and brine
samples of each fermented food using the Qiagen® PowerSoil® (Hilden, Germany) genomic DNA
extraction kit as described in its protocol. Sequencing was performed using Illumina MiSeq available
at Michigan State University genomics core facility. Briefly, using dual indexed Illumina compatible
PCR primers 515f/806r targeted amplicon libraries of the microbial 16S rRNA V4 hypervariable region
were constructed following the methods of [11]. The PCR amplicons were batch normalized using
Invitrogen SequalPrep DNA Normalization Plates and the product was pooled. The pool was QC’d
and quantified using a combination of Qubit dsDNA HS, Agilent 4200 TapeStation HS DNA1000
(Wood Dale, IL, USA) and Kapa Illumina Library Quantification qPCR assays (San Diego, CA, USA).
The pool was loaded onto a Standard Illumina MiSeq v2 flow cell and sequencing was performed
in a 2 × 250bp paired end format using a MiSeq v2 500 cycle reagent cartridge. Complementary to
the 515f/806r sequences custom sequencing and index primers were added to appropriate wells of
the reagent cartridge. Illumina Real Time Analysis (RTA) v1.18.54 was used for base calling and its
output was demultiplexed and converted to FastQ format with Illumina Bcl2fastq v2.19.1. Finally, the
data was processed and analyzed using the Qiita [12] software with the Greengenes database. Alpha-
Metabolites 2018, 8, x FOR PEER REVIEW 4 of 14

Beta-diversity of the metabolomic data was calculated using the Bray-Curtis dissimilarity and
visualized in Principle coordinate analysis space (PCoA) using the EMPeror software [14]. Changes
Metabolites
in 2019, 9, 165
the metabolome through time were analyzed using a random forest (RF) linear regression 4 of 15

approach with the randomForest package in the R statistical software. The percent of data explained
by the time
(Shannon of fermentation
index) was used
and beta-diversity to assessUniFrac)
(weighted how theanalysis
fermentation progressed
were calculated through
using data time and
analyzed
metabolites that changed most significantly through the fermentation were identified
with the Deblur algorithm [13] and taxonomy assignments were performed with the OTU clustering using variable
importance
method (97% plots of this RF regression.
identity).

3. Results
2.6. Statistical Analysis
Beta-diversity
3.1. Overall Metabolome of the metabolomic
Changes in Organic data was calculated
Fermented Foods using the Bray-Curtis dissimilarity and
visualized in Principle coordinate analysis space (PCoA) using the EMPeror software [14]. Changes in
The autoclaved
the metabolome samples
through timehad
were different
analyzed metabolomic signature
using a random than
forest thelinear
(RF) conventionally
regressionfermented
approach
samples and changed comparatively little through time (Figure S2).
with the randomForest package in the R statistical software. The percent of data explained by the The metabolome of time
the
fermented food samples separated primarily based on food type (Figure
of fermentation was used to assess how the fermentation progressed through time and metabolites 1a), reflecting the different
chemical signatures
that changed of the different
most significantly plantthe
through species. Fermentation
fermentation of the foods
were identified usingforvariable
four days induced
importance
large changes
plots of this RFinregression.
the metabolomic signature (Figure 1b,c). An RF linear regression was used to assess
how the metabolomic data reflected changes in the food samples through time (Figure 2a). The
percent of variance explained by time in days from the RF was used as a measure of the degree of
3. Results
change in the fermented food metabolome. The autoclaved samples had no linear relationship with
3.1. Overall
time with theMetabolome Changes
RF regression in Organic
indicating noFermented
progressive Foodschanges were occurring in the metabolome
(TableTheS1). All conventionally
autoclaved samples had treated food types
different howeversignature
metabolomic had strong than metabolite changes through
the conventionally fermented the
four-day fermentation with beet having the largest changes (83.55% variance
samples and changed comparatively little through time (Figure S2). The metabolome of the fermented explained), followed by
pepper (78.62% variance explained) and then carrot and radish (67.98%
food samples separated primarily based on food type (Figure 1a), reflecting the different chemical variance explained). All
vegetables
signatures ofdisplayed a progressive
the different linear
plant species. change in the
Fermentation of metabolomic
the foods for four datadays
except for day
induced 4, where
large changesall
samples were incorrectly classified as day 3. This indicates that the metabolomic
in the metabolomic signature (Figure 1b,c). An RF linear regression was used to assess how the data ceased its linear
progression
metabolomicofdata change after day
reflected four, in
changes likely
the due
foodtosamples
the fermentation
through time process reaching
(Figure 2a). an
Theendpoint
percent or of
avariance
stationary phase. by time in days from the RF was used as a measure of the degree of change in
explained
We used food
the fermented molecular networking
metabolome. on the GNPS
The autoclaved platform
samples had no (gnps.ucsd.edu)
linear relationship to identify
with time unique
with
MS/MS spectra in the database (putative molecules) and their relationships.
the RF regression indicating no progressive changes were occurring in the metabolome (Table By quantifying the mass
S1).
differences between nodes with connected edges we are able to
All conventionally treated food types however had strong metabolite changes through the four-day identify the most common
transformations
fermentation with across
beet the dataset
having [15]. Aschanges
the largest expected, the most
(83.55% common
variance mass shifts
explained), represent
followed by pepper2 Da
(H2), 14 Da
(78.62% (CH2),explained)
variance 16 Da (OH), and18then Da carrot
(H2O),and but radish
we also observed
(67.98% some explained).
variance less common Allbiochemical
vegetables
transformations such as 30 Da, 44 Da, and some larger mass deltas such
displayed a progressive linear change in the metabolomic data except for day 4, where all samples as 132 Da, 146 Da, 162were
Da,
and
incorrectly classified as day 3. This indicates that the metabolomic data ceased its linear progressionon
324 Da (Figure 2b). These larger mass deltas corresponded to the loss of sugar moieties of
different molecules.
change after day four,Alikely
loss of
due132tom/z Da representsprocess
the fermentation a pentose sugar,
reaching an142 m/z a rhamnose
endpoint sugar,
or a stationary 162
phase.
m/z a hexose sugar, and 324 m/z a dihexose.
a) PC2 17.59%
Beet c) PC2 23.96%
Beet
PC2 21.91%
Pepper Day
Carrot 0
Pepper
1
Radish
2

4
PC3 9.71%
PC1 45.92%
PC3 12.89% PC1 26.07%
PC3 7.96% PC1 41.83%
PC2 17.59%

b) Day PC2 25.13%

0 Carrot PC1 24.62%

Radish
1
2
3
4

PC3 12.89% PC1 26.07% PC1 11.41% PC3 10.86%

PC1 43.28% PC1 37.74%

Figure 1. Principle coordinate analysis of a Bray-Curtis distance matrix of metabolomic data

relationships among fermented food samples through time. The overall data relationship based
on (a) different fermented food type and (b) days of fermentation. (c) The isolated analysis of each food
as it is fermented for the 4-day time frame. Samples are colored as a spectrum from white to dark color
corresponding to time for each vegetable. The autoclaved samples in this panel are smaller spheres.
 Figure 1. Principle coordinate analysis of a Bray-Curtis distance matrix of metabolomic data
relationships among fermented food samples through time. The overall data relationship based on
(a) different fermented food type and (b) days of fermentation. (c) The isolated analysis of each food
as it is fermented for the 4-day time frame. Samples are colored as a spectrum from white to dark
color corresponding to time for each vegetable. The autoclaved samples in this panel are smaller
Metabolites 2019, 9, 165 5 of 15
spheres.

Figure
Figure 2.
2. (a)
(a)Scatterplot
Scatterplotof
ofactual
actualvs.
vs.predicted
predicteddays
daysof offermentation
fermentationfrom
fromRF RFregression
regression for
for each
each food
food
type.
type. The
TheRF RFmachine
machinelearning
learningalgorithm
algorithmpredicts
predicts the
the assignment
assignment ofof the
the sample
sample inin days
days based
based on
on the
the
trend in the metabolomic data. The percent variance explained by days in the metabolomic data is
trend in the metabolomic data. The percent variance explained by days in the metabolomic data is
also
also shown.
shown.(b) (b)The
Themost
mostcommon
commontransformations
transformations across
acrossdataset based
dataset based ononmolecular networking
molecular networking in
in GNPS plotted as the pairs (represented by one networking edge and difference in m/z of each pair.
The abundance of each edge represents its occurrence in the network between two MS/MS pairs).
(c) Unique MS/MS spectral richness through the fermentation based on molecular networking in GNPS
and the standard deviations of the replicates.
Metabolites 2019, 9, 165 6 of 15

We used molecular networking on the GNPS platform (gnps.ucsd.edu) to identify unique MS/MS
spectra in the database (putative molecules) and their relationships. By quantifying the mass differences
between nodes with connected edges we are able to identify the most common transformations across
the dataset [15]. As expected, the most common mass shifts represent 2 Da (H2 ), 14 Da (CH2 ), 16 Da
(OH), 18 Da (H2 O), but we also observed some less common biochemical transformations such as
30 Da, 44 Da, and some larger mass deltas such as 132 Da, 146 Da, 162 Da, and 324 Da (Figure 2b).
These larger mass deltas corresponded to the loss of sugar moieties on different molecules. A loss of
132 m/z Da represents a pentose sugar, 142 m/z a rhamnose sugar, 162 m/z a hexose sugar, and 324 m/z
a dihexose.
There were 5392 unique spectra (putatively molecules) identified according to the molecular
networking MS cluster algorithm. There were strong metabolite changes in the fermented foods
through time seen as an initial increase in metabolite richness. All vegetables had a significant increase
in metabolite richness through time except for peppers (Table S2). There was also an observed decrease
at days 2 or 3 depending on the vegetable (Figure 2c). Beats had the largest increase in metabolite
diversity starting at 644 unique spectra and peaking at 876 on day 3, peppers had the largest decrease
in metabolite richness where there was a drop in 100 unique spectra detected between days 2 and 4.
The changes in pH in the brine of each fermented food were also measured. The starting pH of
the brine depended on the vegetable, with beets being highest and pepper lowest (Figure S3). Through
the fermentation all the samples decreased in pH but ended up at approximately the same level at day
4 (mean pH of all vegetables = 3.79, SD = 0.25). This meant that the beets had the largest change in
pH moving over 3 units, whereas peppers had little change moving less than 1 unit over the four-day
fermentation (Figure S3).

3.2. Microbiome Dynamics

Sequencing of 16S rRNA gene amplicons was used to assess the microbial changes during the
organic food fermentations. At the taxonomic level of Order, the major changes through time were
a decrease in Enterobacteriales and increase in Lactobacillales along with changes in some lower
abundance taxa (Figure 3a). Interestingly, the microbiome of the fermented foods before fermentation
were dominated by the order Pseudomonadales, which was rapidly depleted within the first day
of fermentation. Analysis at a lower taxonomic level revealed that changes in microbial diversity
depended on the food type (Figure 3b, Figure S4–S7). Alpha-diversity of the microbiome data did
not follow a notable pattern, except in beets where a noted reduction in alpha-diversity occurred
(Figure 3c). Each fermented food had a unique microbiome at the starting point and through time,
though the fermentation was dominated by particular bacterial groups (Figure 3b,d). OTUs belonging
to the Enterobacteriaceae and Erwinia sp. dominated fermentation of all vegetables, with the same
OTU present in carrot and beet. Lactococcus and Leuconostoc sp. also increased in relative abundance in
the different foods through time. Prior to the fermentation pseudomonads dominated the microbiome
of beet, carrot and radish. These microbes were quickly depleted during the fermentation.
Metabolites 2019, 9, 165 7 of 15

a) #!!"
Beet Carrot Pepper Radish
,-./0
+!"
122345.6507-/7861-/096:722;6<47861-/064/=

*!" 122>?6507=5906/@:722>?6507=59064/=@
122;A-75.6B96:722;A-75.6B64/=
)!" 122?5.9CB7861-/0996:722?5.9CB7861-/0964/=
122D46E7861-/0996:722D46E7861-/0964/=
(!"
122F6GG6507-/7861-/096:722H6C-.7G7C6I64/=
'!" 122345.6507-/7861-/096:722J.9K78964/=
12231-9C7861-/096:72231-9C7GA1/-64/=
&!" 122L/-6507-/7861-/096:722L<0M.74I/0964/=
122345.6507-/7861-/096:722?5.9CB7G7C6I64/=
%!"
122F6GG6507-/7861-/096:7223/07G7C6I64/=
$!" 122F6GG6507-/7861-/096:722N=/<I7G7C6I64/=
122L619449:722L6194464/=
#!"
122L619449:722O61-786194464/=
122F6GG6507-/7861-/096:722PC-/07861-/0964/=
!"
#0# # $ 1 2 % & &3& ' 4
$$%% ' ' #0# # $ 1 2 % & &3& ' 4
$$%% ' ' #0# # $ 1 2 % & &3& ' 4
$$%% ' ' #0# # $1$ $ % 2 3 & '4' '
%%&&
Days

b) #!!"
Beet
#!!"
,-./0
123345-/06789-/0:89/8/1;331<33%&&#%!+
Carrot
123345-/06789-/0:89/8/1;33=8.5/>>81<338?@8-:>:<&&'(*+$
c),-./0
123345-/06789-/0:89/8/1;33<8.5/==81>338?@8-:=:>&&'(*+$
+!" 1233A89:>>89/8/1;33A89:>>@<1<332>/B@<&%(')#)
+!"
123345-/06789-/0:89/8/1;331<33&%)($%!
1233A89:>>89/8/1;33A89:>>@<1<33&&'&%#*
123345-/06789-/0:89/8/1;3340C:5:81<33#!#!##%
'
123345-/06789-/0:89/8/1;331>33%&&#%!+
1233A>/@B6C658B89/8/1;33A>/@B6C658>1>33##!+$'# 5006 7*88,6
*!" *!"
1233DE.:5;6F658G89/8/1;33DE.:5;6F658<1<33$*$'(# 123345-/06789-/0:89/8/1;331>33&%)($%!
123345-/06789-/0:89/8/1;33D8>F65/>>81<33&%*&!'*
123345-/06789-/0:89/8/1;331<33&%(*(!! 123345-/06789-/0:89/8/1;331>33&%+!&#!
)!" )!"
&"# 90::08 ;*/<4)
123345-/06789-/0:89/8/1;331<33*#&&&$
123345-/06789-/0:89/8/1;331<33%$+!+( 1233D-0/E-6969989/8/1;33F89-69699@>1>33##!!+)$
1233H</@G6F658G89/8/1;33H</@G6F658<1<33(&('&+
(!" (!"
1233H</@G6F658G89/8/1;33H</@G6F658<1<33##!+$'#
1233A>/@B6C658B89/8/1;33A>/@B6C658>1>33&&'#!##
1233H</@G6F658G89/8/1;33H</@G6F658<1<33%'*!&$ 1233G/06C 658B89/8/1;331>33&&&'(+$
123345-/06789-/0:89/8/1;33A0/55/0:81<338>5:&%(&&(+
'!" '!"
123345-/06789-/0:89/8/1;3345-/06789-/01<33;/0;6I:8/&&#*#(' 123345-/06789-/0:89/8/1;3340H:5:81>33#!#!##%

&
1233J89-6789:>>89/8/1;33J89-6789:>>@<1<3370/I:<&&%#(!* ()*++,+-.+/01
123345-/06789-/0:89/8/1;331>33+#+($
1233DE.:5;6F 658G89/8/1;33K6I6<E.:5;67:@F 1<33+&#*!%
&!" &!"
123345-/06789-/0:89/8/1;331<33(+#&$% 1233I89:==89/8/1;33I89:==@>1>332=/J@>&%(')#)
123345-/06789-/0:89/8/1;331<33%+&&&*&
1233J/@9656<-6989/8/1;33J/@9656<-691<33&&*$+&& 123345-/06789-/0:89/8/1;33D8=C65/==81>33&%*&!'*
%!" %!"
1233A89:>>89/8/1;33A89:>>@<1<33#)#+$&( 1233A>/@B6C658B89/8/1;33A>/@B6C658>1>33'#%*!*
1233H</@G6F658G89/8/1;33H</@G6F658<1<33'#%*!*

%"#
1233H</@G6F658G89/8/1;33H</@G6F658<1<33208;:$+'!%# 123345-/06789-/0:89/8/1;331>33(+#&$%
$!" $!"
123345-/06789-/0:89/8/1;331<33&%+!&#!
123345-/06789-/0:89/8/1;33D/008-:81<33&%$*#*+ 123345-/06789-/0:89/8/1;331>33*#&&&$
1233H</@G6F658G89/8/1;33H</@G6F658<1<33+$(%)!
1233F/@9656>-6989/8/1;33F/@9656>-691>33&&*$+&&
#!" #!"
123345-/06789-/0:89/8/1;33D/008-:81<33&%(!'##
1233J/@9656<-6989/8/1;33J/@9656<-691<33'&!+&! 123345-/06789-/0:89/8/1;33D/008-:81>33&%(!'##
123345-/06789-/0:89/8/1;3340C:5:81<33&%+&+$(
!"
# 0# # $ 1$ $ % 2% % & 3& & ' 4' '
!"
1233D-0/E-6969989/8/1;33J89-69699@<1<33##!!+)$
1233A89:>>89/8/1;33A89:>>@<1<33$))%&!
0
# # # $ $ $ % % % & 1
123345-/06789-/0:89/8/31;3331<33&&&+*'# 2 3& & ' 4' '
%
123345-/06789-/0:89/8/1;3340H:5:81>33&%+&+$(
123345-/06789-/0:89/8/31;3331>33&&&+*'#

Pepper Radish
#!!" #!!"
,-./0
123345-/06789-/0:89/8/1;331<33&%)($%!
,-./0 $"#
123345/67898:;7;</;/1=3345/67898:;5153320;=>$+'!%#
+!" +!"
123345-/06789-/0:89/8/1;3345-/06789-/01<33;/0;6=:8/&&#*#(' 1233?:-/08@;<-/0>;</;/1=331533&%)($%!

*!" 123345-/06789-/0:89/8/1;3340>:5:81<33&&#)#'*
*!"
123345-/06789-/0:89/8/1;331<33&&'&$+' $
1233A?B>=68@;<-/0;</;/C1=33?B>=68@;<-/0>691533')*+##

1233?:-/08@;<-/0>;</;/1=33?0D>:>;1533#!#!##%
)!" )!"
1233?89:@@89/8/1;33?89:@@A<1<33#)#+$&(
1233E/6<8:85-8<;</;/1=33E/6<8:85-8<1533'&!+&!
123345-/06789-/0:89/8/1;331<33+#+($

!"#
(!" (!"
1233B89-6789:@@89/8/1;33B89-6789:@@A<1<3370/=:<&&%#(!*
123345/67898:;7;</;/1=3345/67898:;51533'#%*!*

1233?:-/08@;<-/0>;</;/1=331533(+#&$%
'!" '!"
1233C</AD6E658D89/8/1;33C</AD6E658<1<33=/065::$)(*()
1233?:-/08@;<-/0>;</;/1=33F/00;->;1533&%(!'##
1233B/A9656<-6989/8/1;33B/A9656<-691<33&&*$+&&
&!" &!" 123345/67898:;7;</;/1=3345/67898:;51533##!+$'#

!
123345-/06789-/0:89/8/1;331<33*#&&&$
1233G/0898:;7;</;/1=331533&&&'(+$
123345-/06789-/0:89/8/1;331<33(+#&$%
%!" %!"

1 2 3' 4
1233?:-/08@;<-/0>;</;/1=331533&%+!&#!

$0
123345-/06789-/0:89/8/1;3340>:5:81<33&%+&+$(

$!" $!"
1233B/A9656<-6989/8/1;33B/A9656<-691<33'&!+&! 1233F-0/H-8<8<<;</;/1=33E;<-8<8<<651533##!!+)$
% & #
2*34
1233?89:@@89/8/1;33?89:@@A<1<33$))%&! 1233E/6<8:85-8<;</;/1=33E/6<8:85-8<1533&&*$+&&
#!" #!"
1233F-0/G-6969989/8/1;33B89-69699A<1<33##!!+)$ 1233?:-/08@;<-/0>;</;/1=331533*#&&&$

123345-/06789-/0:89/8/31;3331<33&&&+*'# 1233?:-/08@;<-/0>;</;/31=3331533&&&+*'#
!" !"
# 0
# # $ 1$ $ % 2% % & 3& & ' 4' ' # 0# # $ 1$
123345-/06789-/0:89/8/1;331<33&%+!&#!
$ % 2% % & 3& & ' 4' '
1233?:-/08@;<-/0>;</;/1=33?0D>:>;1533&%+&+$(

Beet
d)
PC2 (19.72%) PC2 (26.61%)

Beet Carrot
Beet Day
0
Carrot
Pepper PC1 (68.48%) PC1 (54.18%)
1
PC3 (4.91%)
PC3 (8.15%)
Radish 2

PC1 (7.6%)
Pepper PC2 (19.84%)
Radish
3

PC1 (85.72%)
PC1 (47%)
PC1 (3.62%)

PC3 (16.83%)

Figure 3. Microbiome changes in the organically fermented foods. (a) Taxa changes at the Order level
in all fermented food through time, each of the three replicates is shown as its own bar in the bar graph.
(b) Taxa changes at the OTU level in different fermented foods through time. (c) Shannon index of
microbial diversity changes in each fermented food through time. (d) Principle coordinate analysis of
the weighted UniFrac [16] distance of the Deblurred data for the whole microbiome dataset and each
individual vegetable. In the PCoA plots of each individual vegetable, time is denoted as darkening of
color for each representative vegetable according to their representative color scale.
Metabolites 2019, 9, 165 8 of 15

3.3. Changes in Individual Metabolites through Time

GNPS library searching and the ThermoTM Scientific Compound Discoverer® Software were
used to identify
Metabolites 2018, 8, known MS/MS
x FOR PEER REVIEWspectra in the dataset (Figure S9). All compounds in this study were
8 of 14
identified at the level of class 2 according to the metabolomics standards initiative [17]. Sugars were
depletedthrough
depleted throughtimetimeininall
allvegetables.
vegetables. Though
Though the
the exact
exact molecular
molecular structures
structuresofofthese
thesecompounds
compounds
cannot be discerned using our MS/MS methods, a disaccharide and
cannot be discerned using our MS/MS methods, a disaccharide and monosaccharide were monosaccharide wereidentified
identifiedas
as decreasing
decreasing through
through the fermentation
the fermentation (Figure
(Figure 4a).4a).
Plant natural products were of particular
Plant natural products were of particular interest interest to
to this
this study
study due
due to
totheir
theirknown
knownbioactive
bioactiveand and
potential health benefits. Flavonoids, triterpenoids and other small molecules were highlydynamic
potential health benefits. Flavonoids, triterpenoids and other small molecules were highly dynamic
during the fermentation and unique to each vegetable, with some overlap in specific flavonoids
during the fermentation and unique to each vegetable, with some overlap in specific flavonoids across
across vegetables such as quercitin. The aglycone flavonoids chrysin, naringenin, luteolin and
vegetables such as quercitin. The aglycone flavonoids chrysin, naringenin, luteolin and eriodictyol
eriodictyol were identified as well as the glycones rutin, orientin, and glycones of kaempferol and
were identified as well as the glycones rutin, orientin, and glycones of kaempferol and quercitin
quercitin (Figure 4b, Figure S9 level two Metabolomics Standards Initiative [17]). The changes in these
(Figure 4b, Figure S9 level two Metabolomics Standards Initiative [17]). The changes in these molecules
molecules through the fermentation depended on the vegetable. In peppers and radishes, flavonoids
through the fermentation depended on the vegetable. In peppers and radishes, flavonoids increased
increased through time, whereas in carrots there was a general decrease in these compounds.
through time, whereas in carrots there was a general decrease in these compounds. Interestingly in the
Interestingly in the beet fermentation, all flavonoids increased initially in the first few days, but then
beet fermentation, all flavonoids increased initially in the first few days, but then decreased in the latter
decreased in the latter stages indicating that the microbes may have been responsible for the release
stages indicating that the microbes may have been responsible for the release of these compounds from
of these compounds from the plant and their subsequent degradation. Other plant natural products
the plant and
increased their subsequent
in abundance during degradation. Other plant
fermentation including natural
abietic acid,products increased
which rapidly in abundance
increased in beets
during
and peppers, 4-coumaric acid and 18-β-glycyrrhetinic acid in radishes, and the triterpenoids4-coumaric
fermentation including abietic acid, which rapidly increased in beets and peppers, oleanolic
acid
acidand 18-β-glycyrrhetinic
in beets. Ursolic acid inacid in radishes,
carrots, andthe
much like theflavonoids,
triterpenoids oleanolic
decreased inacid in beets.through
abundance Ursolic the
acid
infermentation.
carrots, muchThe likefatty
the flavonoids, decreased in abundance through the fermentation. The
acid linoleic acid also increased in peppers but decreased in carrots (Figure fatty acid
linoleic
4b). acid also increased in peppers but decreased in carrots (Figure 4b).

Figure 4. Cont.
Metabolites 2019, 9, 165 9 of 15
Metabolites 2018, 8, x FOR PEER REVIEW 9 of 14

Figure4.4.Changes
Figure Changesofofunique
uniquemetabolites
metabolites through
through thethe organic
organic food fermentations. (a)
food fermentations. (a)Boxplots
Boxplotsofof
changes
changesininsugar
sugarabundance
abundancethrough
throughthe
thefive-day
five-dayfermentation
fermentationinindifferent
differentvegetables.
vegetables.(b)
(b)Boxplots
Boxplotsof
changes
of changes in plant natural products detected through GNPS and Compound Discoverersearching
in plant natural products detected through GNPS and Compound Discoverer library library
insearching
the four vegetables
in the four fermented
vegetables in this study.
fermented in Putative structures
this study. Putative of the compounds
structures are also shown
of the compounds are
and
alsothe plotsand
shown are the
ordered leftordered
plots are to rightleft
as totriterpenoids, flavonoids
right as triterpenoids, and otherand
flavonoids compounds for each
other compounds
specific
for eachvegetable.
specific vegetable.

3.4. Anthocyanin Dynamics in Radishes

3.4. Anthocyanin Dynamics in Radishes
In the molecular network of all metabolites we discovered a large cluster of spectra containing
In the molecular network of all metabolites we discovered a large cluster of spectra containing
known flavonoids and other compounds, which were highly dynamic through the fermentation.
known flavonoids and other compounds, which were highly dynamic through the fermentation. As
Assuch,
such,wewe further
further investigated
investigated thethe chemistry
chemistry of these
of these spectra
spectra and and propagated
propagated annotations
annotations through
through the
the network to better understand what these spectra were representing and how
network to better understand what these spectra were representing and how the metabolites werethe metabolites
were changing
changing through
through time.time.
ManyMany of these
of these compounds
compounds were glycone
were glycone flavanoids
flavanoids and anthocyanins,
and anthocyanins, such
such as quercitin and cyanidin, which were found in higher abundance in
as quercitin and cyanidin, which were found in higher abundance in the latter days the latter days
ofofthe
thefood
food
Metabolites 2019, 9, 165 10 of 15

fermentation. Propagating the annotations of these complex glycone flavonoids led to the discovery of
a cluster of closely related anthocyanins present in radishes (Figure 5, Figure S8). These molecules
were only detected in day 3 and 4 of the fermentation of this vegetable. These anthocyanins were
putatively identified by MS/MS and parent mass matching and by comparing to known compounds
in the literature [17,18]. We subsequently ran a correlation analysis of one of the most abundant
anthocyanins detected (m/z1005.248, Figure 5) and found that it was highly correlated to Leuconostoc
OTU4482944 (Pearsons r = 0.893) and a Lactococcus OTU1100972 (Pearsons r = 0.775), which are both
known members of the probiotic Order Lactobacillales. Also found in this cluster were chlorogenic
acids and feruloyl-tyramine like compounds. These compounds were much more abundant in the
early days of the fermentation, and highly glycosylated forms of ferulic acid were rapidly depleted
through time.

OH
O
O Node Legend
HO
OH
O Unknown
O
OH O O
HO O cyanidin 3-(feruloyl)diglucoside-5- GNPS Database Hit
O OH OH OH O O
(malonoyl)glucoside
HO O
HO OH
O O+
O OH Pie Legend
O OH OH
OH HO O
O
O O+
O OH
Day 0
O- OH
O OH
OH O-
O
OH
Day 1
O
HO O OH

O O
HO O OH
OH Day 2
Exact Mass: 1210.30
OH cyanidin 3-(p-coumaroyl) Synapyl alcohol
358.149
OH
Radish Anthocyanins O O
Exact Mass: 1034.25 HO O+ diglucoside-5-glucoside Day 3 O
-O 1099.22 HO
O+
1181.3
OH

1211.31
OH
O O
O
O
OH
OH
Day 4 OH
193.086
O OH 1021.24 O O
OH O OH
O 1151.29 O
1197.29 HO OH OH OH Exact Mass: 210.09
OH
O
OH 1167.28 859.212 OH O O OH OH
HO O OH
HO Exact Mass: 919.25 209.081
1035.26
1005.25
O O
Exact Mass: 666.14 HO
OH 667.149 OH
OH OH OH 265.154
O 195.065
-O
-O
O+ O+
O OH 991.255
OH 949.259 919.248 cyanidin 3-sambubiose
O 829.202 O
OH OH OH O OH 251.138
O O 251.138
581.149 O O
OH OH OH Chlorogenic acid HO OH
O O OH OH OH OH
757.218
595.165 Exact Mass: 580.14 621.18 355.102 O
HO OH 535.107 235.144 HO
OH
607.165 O OH
OH O 357.081
Exact Mass: 756.21 OH
421.133 O HO
OH 433.112 OH 353.086
371.097 323.112
O Exact Mass: 354.10
773.212 466.191 433.133
Quercitrin OH
611.16 625.175 OH
339.107
478.191 OH OH 382.112
388.124
HO O 441.102
354.081 Exact Mass: 326.10
463.123 449.107
O 677.311 309.096 471.149
375.091 327.107
OH O OH 565.118 289.091
O 465.102
503.176
551.102 930.25
OH 443.118 506.186
274.074 751.17 1068.28
OH
Exact Mass: 448.10
508.153 551.171 1033.24
1051.26 OH OH
OH 417.174 298.096 1605.4
O OH
HO O O
OH OH
OH O
Isoquercetin OH 536.196
O OH 538.164 OH OH
OH O O O OH
O HO O 252.109 O
OH O OH
HO
O O O
O 339.107
Exact Mass: 550.10
OH O OH HO OH OH
O 489.161
O O OH OH
HO O
OH OH
OH HO O OH
677.207
Exact Mass: 464.10 357.117 OH OH
501.16
Feruloyltyramine 445.134 Exact Mass: 932.26
H 485.165
N 353.148
401.107
369.117
280.118 323.111
O 344.148
OH
3-O-feruloyl-D-quinic acid
O 424.105
342.133 O
314.138
OH HO
OH
Molecular Weight: 313.35 O
389.206
O
HO O
638.244 OH
OH
Exact Mass: 368.11

Figure 5. Molecular network of anthocyanins, flavonoids, ferulic acids and related molecules from
the fermented food LC-MS/MS dataset (anthocyanins found only in radishes). Each node represents a
uniquely clustered MS/MS spectrum and putative metabolite. Nodes are colored using pie charts which
correspond to the spectral count in the different fermentation days according to the legend. Related
metabolites are connected by edges according to their MS/MS cosine score and the width of the edge is
determined by the score. Compounds with a GNPS database hit are highlighted using a red square
behind the node. Putative structures and compounds names are shown when discernable through the
literature or library matches.
Metabolites 2019, 9, 165 11 of 15

4. Discussion
Home food fermentation is seeing a resurgent interest in many western societies due to the
potential for consumption of live probiotic organisms [19,20]. The foods selected for this study were
grown organically to determine the natural process occurring in an ‘in-home’ fermented food, without
any known preservatives or other non-biological chemical additives. We applied modern mass
spectrometry and microbiome bioinformatics approaches to provide a global overview of the changes
occurring in the fermented foods and an untargeted assessment of specific microbial species and
metabolite dynamics. Molecular networking and MS/MS similarity searching applied to fermented
food studies revealed metabolite changes from knowns and unknowns without any a priori biased
targeting of particular molecular groups [21,22].
Principle coordinate analysis plots of metabolome dissimilarity enabled visualization of the
changes through time compared to their autoclaved sterile control samples. The chemistry of all four
foods was significantly different from controls throughout the fermentation, except on the first day
which was sampled prior to incubation. The beta-diversity plots also showed that the foods were highly
different from each other as expected, but also that the degree of change in the foods through time
depended on the vegetable. Carrots, radishes and peppers all had strong progressive changes through
PCoA space reflecting a dynamic metabolome through the entire fermentation process, but beets
changed more rapidly within the first two days and then little movement was observed through PCoA
space in the final three days of fermentation. This indicates that the rate of natural food fermentation
depends on the vegetable and desired flavoring or nutritive value may also depend on timing of
consumption of fermentation arrest. Although a four-day fermentation time is common for ‘in-home’
fermented food recipes, other vegetable fermentation procedures often employ longer ferment times
than that used here. In a kimchi fermentation, for example, similar dynamic changes were observed in
the first 5 to 10 days of fermentation, followed by a shift to a more static state at 20-30 days and then a
final maturation stage at 50 days [7]. Here, the molecular diversity of the foods peaked at three days
(two days for peppers) and then all foods had a slight decrease in metabolite diversity in the final day,
indicating that the microbial process had slowed. If attaining the most diverse metabolite repertoire
of a fermented food is a desired flavor or nutritive point, it may be useful to determine when this
peak occurs within each home fermentation recipe and food type. An RF regression analysis of the
metabolome changes through time also indicated that the process had begun to slow after day three as
the machine-learning based predictions were not able to differentiate day three from day four in all
foods. This untargeted approach to metabolomic profiling of fermented foods provides unique insight
into the overall metabolite changes associated with ‘in home’ salt-based fermentations. Microbial
growth during natural fermentation of brined vegetables has been described to occur in four stages:
initiation, primary fermentation, secondary fermentation, and post-fermentation spoilage [23]. We
saw a similar trend here with an initial delay in metabolome changes followed by a period of rapid
chemical transformation and a final day of relative stasis.
The microbiome profiles of the fermented foods were somewhat surprising. Probiotic lactobacilli
did not come to dominate the fermentation as expected, instead, members of the Enterobacteriaceae
took over as the fermentation progressed. Members of the Lactobacillales (lactic acid bacteria) were
also abundant through the fermentation process, but these were primarily Leuconostoc and Lactococcus
spp., who are not commonly administered as probiotic organisms. Leuconostocaceae are common
constituents of naturally fermented vegetables such as kimchi and can often dominate these microbial
fermentations (8). The probiotic Lactobacillus brevis [24] was detected in both beets and peppers but
was of low relative abundance (<5%). A high abundance of Enterobacteriaceae spp. is not uncommon
in fermented food microbiome profiles, for example, the early days of carrot and cabbage (sauerkraut)
fermentation had an abundance of these organisms whereas lactic acid bacteria came to dominate
at later stages of the fermentation [25,26]. Erwinia was a predominant fermenting organism in all
four vegetables in this study. E. amylovora is a common plant pathogen known to infect a variety of
different plant species. The presence of this bacterium during the fermentation may reflect its ability
Metabolites 2019, 9, 165 12 of 15

to invade plant tissues if compromised, possibly due to chopping prior to the fermentation. It is
also a robust fermenter of sugars and has been explored industrially for its production of desirable
microbial products [27] and has been identified in alcoholic fermentations of different plant parts, such
as pulque (agave) [28]. Prior to the fermentation the microbiome of the vegetables in this study were
dominated by plant-associated pseudomonads, which were quickly depleted after day 1 or 2 when
the Enterobacteriaceae started to dominate. In peppers and radishes, the primary Enterobactericeae
member dominating the fermentation was also present on the vegetable prior to the fermentation
process, in contrast, carrots and beets had low to undetectable amounts of these organisms at time zero.
This is interesting in light of recent characterizations of the vegetable skin microbiome which have
shown that different species of these edible plants have unique microbiomes and this beta-diversity is
driven by the relative abundances of Enterobacteriaceae [29]. The dominance of this family indicates
that organic food fermentation may actually reduce the selection for probiotic organisms such as
lactobacilli that are generally regarded as beneficial to the human gut microbiome. Spontaneous
fermentations are known to be more challenging to control and many industrialized fermented
food producers use starter cultures to directly manipulate fermentation outcomes [30]. Future work
comparing organic and non-organic food fermentations are needed to determine if this microbial
property is unique to organically fermented foods or of these vegetables in particular.
Another aspect of the perceived health benefits of whole foods are plant natural products
such as flavonoids, anthocyanins and triterpenoids. These compounds have been shown to have
a variety of health benefits and consumption of whole foods is a significant dietary source of these
compounds [31–34]. Furthermore, the gut microbiome is known to degrade the glycoside forms
of these compounds producing aglycones that can interact with the gut epithelium and potentially
increase their health benefits [35,36]. As such, we investigated the production of these plant products
in the organically fermented foods through time and found, similarly to the microbial profiles, that
the production of these compounds depended on the vegetable. In beets, peppers and radishes,
many of these natural products increased during the fermentation, including their aglycone forms.
In carrots, however, we observed a reduction in the abundance of specific flavonoids and triterpenoids
as the process progressed. Anthocyanins, complex flavonoid natural products, were particularly
abundant in radishes and showed a progressive increase in their detection through the fermentation.
A Lactococcus sp. and Leuconostoc sp. were highly correlated to their production in radishes. Although
one cannot infer that they are produced by these organisms, this association may further implicate
probiotic mechanisms that exist within the Lactobacillaceae, by releasing plant natural products during
fermentation. Thus, natural fermentation of vegetables increases the production of specific plant
natural products with demonstrated health benefits, but this is dependent on the vegetable consumed.
The microbial diversity observed between the different vegetables fermented in this study may have a
direct influence on the nutritive value of the fermented food.

5. Conclusions
In summary, ‘in-home’ salt-based fermentation of organic vegetables induced drastic changes in
the microbiology and metabolite profiles of these foods. Our results indicate that the presumed nutritive
and probiotic value of this process is highly dependent on the vegetable and microbiome that comes to
dominate the process. Release of beneficial plant natural products into the brine better enabling their
consumption may be a lesser appreciated mechanism behind the nutritive value of fermented food
consumption, particularly in radishes and peppers. However, further studies are needed to determine
whether fermentation of ‘organic’ vegetables preferentially selects for Enterobacteriaceae and if this
is different from conventionally processed vegetables available to consumers. A deeper look into
the microbial process occurring in ‘in-home’ fermented foods is needed to better understand how to
optimize their health benefits. Starter cultures, such as lactobacilli which may have inherent probiotic
properties, have been shown to enable a more controlled fermentation and may greatly enhance the
nutritive value of the fermented food process [28]. Experimentation with this starter cultures on
Metabolites 2019, 9, 165 13 of 15

different vegetables in an ‘in-home’ setting may lead to better guidelines on which foods should be
selected for home fermentation and how to optimize the process to maximize health benefits.

Supplementary Materials: The following are available online at http://www.mdpi.com/2218-1989/9/8/165/s1:

Figure S1: experimental design, Figure S2: Metabolome PCoA, Figure S3: pH changes, Figure S4-S7: Microbiome
data with taxa assignments. Figure S8: Plant natural product network by vegetable. Figure S9: Metabolite
annotation mirror plots. Table S1: Variation explained by RF regressions. Table S2: Pearson corrleations of
metabolite richness and time.
Author Contributions: Collected samples, generated data, analyzed data and wrote paper: R.R., R.A.Q.; Collected
samples and generated data: A.G.G., I.S., O.A., K.S.
Funding: This research received no external funding.
Acknowledgments: The authors would like to acknowledge the Michigan State University RTSF Mass
Spectrometry Core and the MSU organic farm for help with this project.
Conflicts of Interest: The authors declare no conflicts interest.

References
1. Campbell-Platt, G. Fermented foods—A world perspective. Food Res. Int. 1994, 27, 253–257. [CrossRef]
2. Selhub, E.M.; Logan, A.C.; Bested, A.C. Fermented foods, microbiota, and mental health: Ancient practice
meets nutritional psychiatry. J. Physiol. Anthropol. 2014, 33, 2. [CrossRef] [PubMed]
3. Wolfe, B.E.; Dutton, R.J. Fermented foods as experimentally tractable microbial ecosystems. Cell 2015, 161,
49–55. [CrossRef] [PubMed]
4. Daniel, M.; Embriette, H.; Justine W., D.; James T., M.; Antonio, G.; Gail, A.; Alexander A., A.; Bahar, B.;
Caitriona, B.; Yingfeng, C.; et al. American Gut: An Open Platform for Citizen Science Microbiome Research.
mSystems 2018, 3, e00031-18.
5. David, L.A.; Maurice, C.F.; Carmody, R.N.; Gootenberg, D.B.; Button, J.E.; Wolfe, B.E.; Ling, A.V.; Devlin, A.S.;
Varma, Y.; Fischbach, M.A.; et al. Diet rapidly and reproducibly alters the human gut microbiome. Nature
2013, 505, 559–563. [CrossRef]
6. Gilbert, J.; Knight, R.; Blakeslee, S. Dirt is Good: The Advantage of Germs for Your Child’s Developing Immune
System; St. Martin’s Press: New York, NY, USA, 2017.
7. Park, S.E.; Yoo, S.A.; Seo, S.H.; Lee, K.I.; Na, C.S.; Son, H.S. GC–MS based metabolomics approach of Kimchi
for the understanding of Lactobacillus plantarum fermentation characteristics. LWT—Food Sci. Technol. 2016,
68, 313–321. [CrossRef]
8. Jung, J.Y.; Lee, S.H.; Kim, J.M.; Park, M.S.; Bae, J.W.; Hahn, Y.; Madsen, E.L.; Jeon, C.O. Metagenomic analysis
of kimchi, a traditional Korean fermented food. Appl. Environ. Microbiol. 2011, 77, 2264–2274. [CrossRef]
9. Mugula, J.; Nnko, S.A.; Narvhus, J.; Sørhaug, T. Microbiological and fermentation characteristics of togwa,
a Tanzanian fermented food. Int. J. Food Microbiol. 2003, 80, 187–199. [CrossRef]
10. Thompson, L.R.; Sanders, J.G.; McDonald, D.; Amir, A.; Ladau, J.; Locey, K.J.; Prill, R.J.; Tripathi, A.;
Gibbons, S.M.; Ackermann, G.; et al. A communal catalogue reveals Earth’s multiscale microbial diversity.
Nature 2017, 551, 457–463. [CrossRef]
11. Kozich, J.J.; Westcott, S.L.; Baxter, N.T.; Highlander, S.K.; Schloss, P.D. Development of a Dual-Index
Sequencing Strategy and Curation Pipeline for Analyzing Amplicon Sequence Data on the MiSeq Illumina
Sequencing Platform. Appl. Environ. Microbiol. 2013, 79, 5112–5120. [CrossRef]
12. Gonzalez, A.; Navas-Molina, J.A.; Kosciolek, T.; McDonald, D.; Vázquez-Baeza, Y.; Ackermann, G.; DeReus, J.;
Janssen, S.; Swafford, A.D.; Orchanian, S.B.; et al. Qiita: Rapid, web-enabled microbiome meta-analysis.
Nat. Methods 2018, 15, 796–798. [CrossRef] [PubMed]
13. Amir, A.; McDonald, D.; Navas-Molina, J.A.; Kopylova, E.; Morton, J.T.; Xu, Z.Z.; Kightley, E.P.;
Thompson, L.R.; Hyde, E.R.; Gonzalez, A.; et al. Deblur Rapidly Resolves Single-Nucleotide Community
Sequence Patterns. mSystems 2017, 2, 00191-16. [CrossRef] [PubMed]
14. Vázquez-Baeza, Y.; Pirrung, M.; Gonzalez, A.; Knight, R. EMPeror: A tool for visualizing high-throughput
microbial community data. Gigascience 2013, 2, 16. [CrossRef] [PubMed]
Metabolites 2019, 9, 165 14 of 15

15. Hartmann, A.C.; Petras, D.; Quinn, R.A.; Protsyuk, I.; Archer, F.I.; Ransome, E.; Williams, G.J.; Bailey, B.A.;
Vermeij, M.J.A.; Alexandrov, T.; et al. Meta-mass shift chemical profiling of metabolomes from coral reefs.
Proc. Natl. Acad. Sci. USA 2017, 114, 11685–11690. [CrossRef] [PubMed]
16. Lozupone, C.; Knight, R. UniFrac: A New Phylogenetic Method for Comparing Microbial Communities
UniFrac: A New Phylogenetic Method for Comparing Microbial Communities. Appl. Environ. Microbiol.
2005, 71, 8228–8235. [CrossRef] [PubMed]
17. Sumner, L.W.; Amberg, A.; Barrett, D.; Beale, M.H.; Beger, R.; Daykin, C.A.; Fan, T.W.M.; Fiehn, O.;
Goodacre, R.; Griffin, J.L.; et al. Proposed minimum reporting standards for chemical analysis. Metabolomics
2007, 3, 211–221. [CrossRef] [PubMed]
18. Wu, X.; Prior, R.L. Identification and Characterization of Anthocyanins by High-Performance Liquid
Chromatography−Electrospray Ionization−Tandem Mass Spectrometry in Common Foods in the United
States: Vegetables, Nuts, and Grains. J. Agric. Food Chem. 2005, 53, 3101–3113. [CrossRef]
19. Rezac, S.; Kok, C.R.; Heermann, M.; Hutkins, R. Fermented Foods as a Dietary Source of Live Organisms.
Front. Microbiol. 2018, 9, 1785. [CrossRef]
20. Chilton, S.N.; Burton, J.P.; Reid, G. Inclusion of Fermented Foods in Food Guides around the World. Nutrients
2015, 7, 390. [CrossRef]
21. Watrous, J.; Roach, P.; Alexandrov, T.; Heath, B.S.; Yang, J.Y.; Kersten, R.D.; Voort, M.v.d.; Pogliano, K.;
Gross, H.; Raaijmakers, J.M.; et al. Mass spectral molecular networking of living microbial colonies. Proc. Natl.
Acad. Sci. USA 2012, 109, 1743–1752. [CrossRef]
22. Wang, M.; J Carver, J.; V Phelan, V.; M Sanchez, L.; Garg, N.; Peng, Y.; Nguyen, D.D.; Watrous, J.; A Kapono, C.;
Luzzatto-Knaan, T.; et al. Sharing and community curation of mass spectrometry data with Global Natural
Products Social Molecular Networking. Nat. Biotechnol. 2016, 34, 828–837. [CrossRef]
23. Perez Diaz, I.M.; Breidt, F.; Buescher, R.W.; Arroyo-Lopez, F.N.; Jimenez-Diaz, R.; Bautista-Gallego, J.;
Garrido-Fernandez, A.; Yoon, S.; Johanningsmeier, S.D. Fermented and Acidified Vegetables. In Compendium
of Methods for the Microbiological Examination of Foods; American Public Health Association: Washington, DC,
USA, 2015; Volume 51, ISBN 978-0-87553-022-2.
24. Fang, F.; Xu, J.; Li, Q.; Xia, X.; Du, G. Characterization of a Lactobacillus brevis strain with potential oral
probiotic properties. BMC Microbiol. 2018, 18, 221. [CrossRef]
25. Wuyts, S.; Beeck, W.V.; Oerlemans, E.F.M.; Wittouck, S.; Claes, I.J.J.; Boeck, I.D.; Weckx, S.; Lievens, B.;
Vuyst, L.D.; Lebeer, S. Carrot Juice Fermentations as Man-Made Microbial Ecosystems Dominated by Lactic
Acid Bacteria. Appl. Environ. Microbiol. 2018, 84, e00134-18. [CrossRef]
26. Zabat, M.; Sano, W.; Wurster, J.; Cabral, D.; Belenky, P. Microbial Community Analysis of Sauerkraut
Fermentation Reveals a Stable and Rapidly Established Community. Foods 2018, 7, 77. [CrossRef]
27. Tolan, J.S.; Finn, R.K. Fermentation of d-Xylose and l-Arabinose to Ethanol by Erwinia chrysanthemi.
Appl. Environ. Microbiol. 1987, 53, 2033–2038.
28. Tamang, J.P.; Watanabe, K.; Holzapfel, W.H. Review: Diversity of Microorganisms in Global Fermented
Foods and Beverages. Front Microbiol. 2016, 7, 377. [CrossRef]
29. Leff, J.W.; Fierer, N. Bacterial Communities Associated with the Surfaces of Fresh Fruits and Vegetables.
PLoS ONE 2013, 8, e59310. [CrossRef]
30. Leroy, F.; De Vuyst, L. Lactic acid bacteria as functional starter cultures for the food fermentation industry.
Trends Food Sci. Technol. 2004, 15, 67–78. [CrossRef]
31. Jiang, N.; Doseff, A.; Grotewold, E. Flavones: From Biosynthesis to Health Benefits. Plants 2016, 5, 27.
[CrossRef]
32. Yao, L.H.; Jiang, Y.M.; Shi, J.; Tomas-Barberan, F.A.; Datta, N.; Singanusong, R.; Chen, S.S. Flavonoids in food
and their health benefits. Plant. Foods Hum. Nutr. 2004, 59, 113–122. [CrossRef]
33. Pojer, E.; Mattivi, F.; Johnson, D.; Stockley, C.S. The Case for Anthocyanin Consumption to Promote Human
Health: A Review. Compr. Rev. Food Sci. Food Saf. 2013, 12, 483–508. [CrossRef]
34. Shi, J.; Arunasalam, K.; Yeung, D.; Kakuda, Y.; Mittal, G.; Jiang, Y. Saponins from Edible Legumes: Chemistry,
Processing, and Health Benefits. J. Med. Food 2004, 7, 67–78. [CrossRef]
Metabolites 2019, 9, 165 15 of 15

35. Justesen, U.; Arrigoni, E.; Larsen, B.R.; Amado, R. Degradation of Flavonoid Glycosides and Aglycones
During in vitro Fermentation with Human Faecal Flora. LWT—Food Sci. Technol. 2000, 33, 424–430. [CrossRef]
36. Hostetler, G.; Riedl, K.; Cardenas, H.; Diosa-Toro, M.; Arango, D.; Schwartz, S.; Doseff, A.I. Flavone
deglycosylation increases their anti-inflammatory activity and absorption. Mol. Nutr. Food Res. 2012, 56,
558–569. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).