Application Note # LCMS-79
Metabolic Profiling of Different Coffee Types on the
Bruker compact™ QTOF System
Introduction
Metabolomics studies have gained major importance in
food applications. Quality control of food samples can be
established on a large scale via state-of-the-art metabolic
profiling based on statistical data evaluation. However,
the metabolomics workflow puts high demands on an
analytical system in terms of its robustness, dynamic range
and identification capabilities. The Bruker compact QTOF
System combines the robustness of Bruker orthogonalTime-of-Flight instruments with new 10-bit digitizer
technology that allows for compound detection and
identification across an unrivalled dynamic range.
After water and black tea, coffee represents the third
most important beverage with a worldwide consumption
of 4.5 million tons per year [1]. Analytically, it constitutes
a very complex mixture of small molecules which differ in
composition and quantities based on the different coffee
cultivars, cultivation regions and processing procedures.
As a proof of concept study for the compact, we analyzed
13 different types of coffee capsule extracts assigned by
their manufacturer to different intensity categories. The
major analytical task was to correlate high resolution LC-MS
data to the manufacturer’s description via a non-targeted
metabolomics approach.
Authors
Dr. Verena Tellström, Dr. Alexander Harder,
Klaus Meyer, Dr. Aiko Barsch
Bruker Daltonik GmbH, Bremen, Germany
Keywords
Instrumentation and Software
compact
compact
Metabolomics
DataAnalysis
Structure Elucidation
SmartFormula 3D
Coffee
MetFrag
PCA
FragmentationExplorer
ProfileAnalysis
Experimental
Results and Discussion
Capsules of 13 different types of coffee were extracted
using 35 ml of water on a standard coffee capsule machine
(Krups XN 301T Nespresso Pixie). Two replicates of each
type were prepared. A QC sample was generated by
pooling equal volumes of all coffee samples profiled in
this study. Extracts were diluted 1:50 in water prior to
analyzing 5µl of 3 replicates for each extract by UHPLC-MS.
Chromatographic separation was carried out using an
RSLC system (Dionex) with a 50 x 2.1 mm BEH C18,
1.7 um column (Waters), at a flow rate of 0.45 mL/min,
with Solvent A: Water + 0.1% HCOOH and Solvent B:
methanol + 0.1% HCOOH. The following LC gradient
program was used: linear increase from 2% B to 98% B
(over 5 min), constant at 98%B (for 1 min).
Evaluating the compact QTOF performance for the
analysis of large metabolomics sample sets
MS detection was performed using a compact Qq-TOF
mass spectrometer (Bruker Daltonik). The instrument was
operated in ESI positive mode acquiring MS full scan and
auto MS/MS data at a 3 Hz acquisition speed.
ProfileAnalysis 2.1 software (Bruker Daltonik) was used
for statistical data analysis based on features extracted
by the FindMolecularFeatures (FMF) algorithm which
can combine all ions belonging to the same compound
(isotopes, charge states, adducts and common neutral
losses). Further data evaluation was performed using
DataAnalysis 4.1 software (Bruker Daltonik). Molecular
formula determination was carried out by combined
evaluation of mass accuracy, isotopic patterns, adduct and
fragment information using SmartFormula3D software.
The direct link from SmartFormula3D to the open source
MetFrag tool (http://msbi.ipb-halle.de/MetFrag/, [2]) allowed
for the identification of target compounds by means of
database search, in silico fragmentation and matching of
expected and observed fragments. Fragmentation patterns
were further evaluated using the ChemDraw™ based
FragmentationExplorer within DataAnalysis. Compound
identities were confirmed by comparing retention time, MS
and MS/MS spectra to those observed for commercially
available standards.
In metabolomics studies, large numbers of samples are
often analyzed with each consisting of a complex mixture
of small molecules across a wide range of abundances.
In order to monitor the system stability, more and more
researches analyze so-called quality control (QC) samples
before, within and after the real analytical samples. Pools
of all analytical samples were collected for a common
quality control sample since they represent a mixture
similar in complexity and chemical composition. The Base
Peak Chromatograms of 15 QC sample injections spanning
a sequence of ~150 coffee extract injections is shown in
Figure 1 A. Peak shape and intensity remain unchanged
revealing the stability of the compact instrument which is
an important prerequisite for analyzing large metabolomics
data sets.
Applying sophisticated peak picking algorithms is a first
crucial step in data pre-processing and it forms the basis
for statistical analysis of metabolomics datasets. The
FindMolecularFeatures (FMF) peak finder combines ions
belonging to one compound, such as common adducts
(e.g. +Na, +K, +NH4), isotopes and charge states.
Molecular features were extracted using this strategy
within high resolution full scan MS data acquired for
the QC and analytical samples. Based on the extracted
features, a pareto scaled Principal Component Analysis
(PCA) was then calculated. The PCA scores plot shown
in Figure 1 B reveals a close clustering of the QC samples
(red circles) indicating that the data acquisition using the
compact QTOF, and peak picking, worked reliably. This
first quick data quality check typically forms the basis for
starting a deeper evaluation of the biological samples.
The extreme dynamic range of analyte concentrations
in complex biological samples represents one of the
most crucial challenges in metabolomics applications.
To access the dynamic range of the compact QTOF
instrument, which is equipped with a new 10-bit digitizer
technology, we performed a proof of concept study. A
mixture of two small molecules, alanine and caffeine, was
analyzed by direct infusion at an acquisition rate of 1Hz.
As shown in Figure 2 the low concentrated alanine was
detected with an intensity of 46 cts whereas the caffeine
peak had an intensity of 7,155,285 cts. The intensity ratio
between these two values, 7,155,285 / 46 = 1.6 X 10 5,
demonstrates the unique capability of the compact to
detect target compounds on an LC timescale across a
dynamic range > 5 orders of magnitude. Furthermore, the
mass accuracy for both compounds was lower than 1 ppm
thus enabling a reliable compound identification.
Robust instrument performance for large metabolomics data sets
Intens.
x106
PC
PC22
A
Sample
QC
QC Injection No.147
QC
0.50
QC Injection No.77
2.0
Sample
B
~150 Injections
0.5
QC Injection No.1
0.25
1.5
0.0
0.00
1.0
-0.5
-0.25
0.5
-0.50
-1.0
0.0
1
2
3
QC_GA2_01_310.d: BPC +All MS, Masses excluded
4
5
Time [min]
-0.75
QC_GA2_01_321.d: BPC +All MS, Masses excluded
-1.0
-1.0 -0.5
0.0
-0.5
0.5 0.0
1.0
PC0.5
1
1.0
PC 1
Figure 1: A: Overlay of Base Peak Chromatograms of interspersed quality control (QC) samples run during the acquisition of a coffee
metabolomics sample batch demonstrates the stability of the Compact QTOF for analysing large sets of complex samples. B: A PCA scores
plot of coffee metabolomics (green) and QC samples (red) reveals tight clustering of QC samples as an additional proof for reproducibility
during data acquisition.
5 orders of magnitude on LC timescale
Intens.
x106
+MS, 4.2-4.3min #247-256
195.0878
Caffeine
0.6ppm
I = 7 155 285cts
Intens.
6
+MS, 4.2-4.3min #247-256
50
40
4
30
90.0550
Alanine 0.5ppm
I = 46cts
S/N > 8:1
20
10
2
0
217.0696
86
88
90
92
94
96m/z
233.0436
0
0
25
50
75
100
125
150
175
200
225
m/z
Intensity ratio = 7 155 285 / 46 = 1.6 x 10 5 @ 1Hz acquisition speed
Figure 2: A mixture of alanine and caffeine measured at a 1 Hz acquisition rate in a direct infusion experiment demonstrates
>5 orders of magnitude dynamic range with mass accuracies better than 1ppm.
Coffee samples cluster according to the assigned coffee
intensity
For a detailed evaluation of the different coffee types
analysed in this study, the QC samples were removed
from the calculated PCA model. In total, 13 different coffee
extracts, comprising espresso and lungo varieties from
different blends and geographical regions were analysed.
The two “biological” and three technical replicates for each
sample type (highlighted by using the same colour and
symbol) formed clusters in the PCA scores plot as seen in
Figure 3 A. In order to avoid the dominance of the caffeine
content on the clustering results, the corresponding peak
was removed from the presented model. The manufacturer
assigned different intensities to each coffee type (numbers
from 3 (weak) to 10 (strong)). The PCA scores plot clearly
separates coffee extracts described as strong (9 and 10)
which cluster in the negative part of PC1, from those
samples that are assigned as weak (3), and form clusters
on the positive part of PC1. The medium intensity coffee
varieties (4-8) cluster in the centre of the PCA scores plot.
These samples do not show a perfect alignment on PC1
according to the coffee intensity, although a trend can
clearly be seen. Quite likely, the separation of samples is
not only due to characteristics that contribute to intensity,
but also to coffee plant varieties used for blending (data
not shown) and geographical origin. Furthermore, lungo
style samples are recommended by the manufacturer
to be extracted in larger amounts of water compared to
espressos indicating a difference in blending or roasting.
Analytes mainly contributing to the separation of samples
in the PCA scores plot can be accessed from the
corresponding loadings plot (Fig. 3 B). Two compounds
marked X and Y are identified by the loadings plot to
have a high content in strong and weak coffee samples,
respectively. To directly access the relative amount for
both metabolites in all samples, bucket statistics plots are
shown in Fig. 3 C. These plots confirm the high abundance
of compound X in strong coffees and of compound Y in
weak samples.
PCA enables differentiation of samples according to strength coffee
PC 2
0.6
PC 2
A
B
10
0.4
5
0.2
0.2
8
10
6
4
0.0
7
-0.2
4
9
-0.4
4
6
3
3
X
0.0
Y
-0.2
-1.0
-0.5
0.0
0.5
1.0
PC 1
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
PC 1
C
Int.
x105
Bucket: 0.57min : 124.039m/z
9
10
Int.
x106
Bucket: 0.35min : 138.055m/z
3
2.4
10
2.5
7
4
Compound X
4
Compound Y
2.2
6
2.0
8
2.0
4
6
1.5
5
3
4
4
10
1.8
10
3
6
4
5
8
6
7
9
3
1.6
0
10
20
30
40
50
60
70
Analysis
0
10
20
30
40
50
60
70
Analysis
Figure 3: A PCA scores plot reveals separation of samples on PC1 according to coffee intensity assigned by the coffee manufacturer
(numbers from 3 (weak) – 10 (strong)). B: The corresponding PCA loadings plot points to compounds mainly contributing to this
differentiation. The two analytes with highest contribution for strong and weak coffee are marked X and Y, respectively. C: Bucket
statistics plots for two selected loadings visualize the relative abundance of the compounds “X” and “Y” across all samples. These plots
indicate clear abundance trends in strong vs. weak samples and vice versa.
Identification of trigonelline and nicotinic acid
Compound identities were elucidated using the MS/MS
capabilities of the Compact QTOF instrument.
SmartFormula3D readily provides the correct molecular
formula by combining accurate mass and isotopic pattern
information for MS and MS/MS spectra (Fig. 4 A).
Furthermore fragment formulae are easily accessed and can
be used for gaining structural information. For compound
X a unique elemental composition of C6H6NO2 ([M+H]+)
was generated. Via a direct link to MetFrag (Fig. 4 C), an
open source tool for in-silico fragmentation, the molecular
formula can be searched in multiple public databases so
that possible structures can be determined [2]. All hits can
then be fragmented in-silico and the resulting fragment
masses can be matched to the measured fragment ions.
The automated assignment of fragment formulae by
SmartFormula3D enables to match the in-silico predictions
with a “quasi” 0 ppm mass accuracy reducing false positive
hits.
With the optimum score of 1.0, MetFrag suggested
nicotinic acid as the most likely structure corresponding
to compound X. In order to further evaluate the
fragmentation pattern of this analyte, the structure of
nicotinic acid was matched to the MS/MS fragment ions
using the ChemDraw™ based FragmentationExplorer
available within the DataAnalysis™ software. This
tool allows for the annotation and reporting of MS/MS
fragmentation patterns as seen in Fig. 4 D. Ultimately,
the compound identity of nicotinic acid was confirmed by
comparing retention time, accurate mass, isotopic pattern
information and MS/MS fragmentation against the pure
reference standard.
Using the workflow described above, the molecular
formula for compound Y was identified as C7H8NO2
([M+H]+). In-silico fragmentation and comparison to a pure
reference confirmed the analyte to be trigonelline.
Identification of „Compound X“ as nicotinate acid
Intens. 10.
x105
+MS, 0.4-0.6min #65-97
195.0877
A
C
96.0444
1
124.0393
0
x104 10.
-50
0
50
100
150
200
250m/z
+MS2(124.0393), 20.5-30.7eV, 0.4-0.6min #67-99
124.0392
2
1
80.0494
53.0390
0
50
B
96.0443
60
70
80
90
100
110
120
E
m/z
D
Intens.
x10 4
2
QC_aMSMS.d: +MS2(124.0393), 20.5-30.7eV, 0.4-0.6min #67-99
Sample
124.0392
53.0390
2.0
96.0443
OH
1.5
Nicotinic acid Standard
124.0394
N
1.0
80.0496
106.0291
60
70
80
N
80.0494
0.5
53.0394
50
O
124.0392
Nicotinate_aMSMS.d: +MS2(124.0393), 20.5-30.7eV, 0.5-0.5min #79-87
0.50
0.00
N
80.0494
0.75
0.25
+MS2(124.0393), 20.5-30.7eV, 0.4-0.6min #67-99
2.5
1
0
x10 5
1.00
Intens. 1.
x104
90
100
110
53.0390
0.0
120
m/z
50
96.0443
60
70
80
90
100
106.0282
110
OH
120
130m/z
Figure 4: Identification of Compound X: A+B: Based on accurate mass and isotopic pattern information in MS and MS/MS spectra for
unknown compound X SmartFormula3D returned a single molecular formula for the precursor and corresponding formulae for the fragment
ions. C: A direct link to MetFrag [2] enables the ability to send the molecular formula for the precursor and theoretical masses for fragment
ions to the open source web application. MetFrag returned nicotinic acid as most likely structure by matching in-silico fragments with
measured fragment ions. D: Detailed evaluation using the FragmentationExplorer supports the in-silico fragments generated by MetFrag.
E: Verification of the identity of Compound X as nicotinic acid was achieved by comparing retention time, accurate mass and MS/MS spectra
with the pure reference standard.
Conclusion
Many compounds that contribute to the typical flavour
of coffee as a beverage are only formed during the high
temperature treatment of the roasting process. Trigonelline
is one of the major analytes in unprocessed coffee and
is transformed during the roasting process mainly to
pyridine and nicotinic acid [3, 5]. The obtained results
perfectly match this prior knowledge. As seen in figure 5,
trigonelline (compound Y) has a higher content in coffees
classified with strength 3 suggesting a weaker roasting.
In contrast, the trigonelline degradation product nicotinic
acid (compound X) is more pronounced in the samples
described as 9 and 10 in strength indicating a stronger
roasting.
This reliable proposal of compound identities helped
to save analysis time and money spent for purchasing
multiple references in order to confirm the identity of the
target compounds.
Degradation of Trigonelline to Nicotinic acid
N
The compact QTOF provides unrivalled dynamic range
(> 5 orders of magnitude) in combination with mass
accuracy, sensitivity, MS/MS performance and robustness
enabling this instrument to be the tool of choice for
analyzing batches of highly complex metabolomics
samples. Together with sophisticated software for
statistical evaluation of metabolomics datasets, these
features enabled quick pinpointing of relevant compounds
contributing to coffee intensity. Molecular formula
generation for MS as well as MS/MS spectra using
SmartFormula3D together with the in-silico fragmentation
tools MetFrag and FragmentationExplorer generated single
structure candidates for two selected target compounds
characteristic for weak and strong coffee samples.
O
Trigonelline
-COOH
References
-CH3
O
N
OH
N-methylpyridinium
Nicotinic acid
component analysis of high resolution LC-TOF-MS data:
the analysis of the chlorogenic acid fraction in green
coffee beans as a case study. Anal. Methods 2011,
3:144-155
[2] Wolf, S. et al.: In silico fragmentation for computer Figure 5: Schema of trigonelline degradation during coffee roasting
(adapted from [3]). Trigonelline (corresponds to compound Y), a
coffee metabolite, is known to be transformed to nicotinic acid
(corresponds to compound X) proportional to the degree of roasting
[4].
assisted identification of metabolite mass spectra.
BMC Bioinformatics 2010, 11:148
[3] Boettler U. et al : Coffee constituents as modulators of
Nrf2 nuclear translocation and ARE (EpRE)-dependent gene
expression. The Journal of Nutritional Biochemistry 2011,
22 (5): 426-440
[4]http://www.coffeeresearch.org/science/bittermain.htm
[5] Adrian and Frangne: Synthesis and availability of niacin in roasted coffee. Adv Exp Med Biol. 1991, 289: 49-59
For research use only. Not for use in diagnostic procedures.
Bruker Daltonik GmbH
Bruker Daltonics Inc.
Bremen · Germany
Phone +49 (0)421-2205-0
Fax +49 (0)421-2205-103
[email protected]
Billerica, MA · USA
Phone +1 (978) 663-3660
Fax +1 (978) 667-5993
[email protected]
www.bruker.com
Fremont, CA · USA
Phone +1 (510) 683-4300
Fax +1 (510) 490-6586
[email protected]
Bruker Daltonics is continually improving its products and reserves the right
[1] Kuhnert, N. et al.: Scope and limitations of principal N
to change specifications without notice. © Bruker Daltonics 03-2013, LCMS-79, 1817861
OH