Fractionation-assisted NMR-based metabolomics
of Grindelia squarrosa
By Jason McFarlane
May 26, 2017

Abstract
To identify compounds, a nuclear magnetic resonance spectroscopy (NMR)-based
metabolomics method was used on an extract of Grindelia squarrosa. The extract was
separated based on polarity, one of the parameters used to evaluate the potential for a new
bioactive compound to be suitable for oral administration. The seven most polar fractions–out
of eleven total fractions–were analyzed by NMR spectroscopy using an HSQC pulse sequence.
The resulting peaks were queried against the Biological Magnetic Resonance Databank. The
database returned 597 compounds, which were selected to include only those compounds that
had greater than 50% of the peaks matching. After selection, 88 compounds were reported,
including three bioactive compounds: agmatine, 4-guanidinobutyric acid, and dopamine.

Introduction
A large part of modern drug development is based on compounds isolated from natural sources
such as plants and marine animals. Grindelia squarrosa is a plant that has traditionally been
used as an extract of the flowers and leaves to remove mucus from bronchitis patients,1 and
has been shown to have some antimicrobial activity.2 G. squarrosa has a resinous coat to the
flower, presumably as a form of protection from pests; other compounds that have a role in
protecting plants against pests have been shown to have drug activity, such as morphine. 3
These traits make G. squarrosa an potential candidate for novel drug compounds.
Unfortunately, traditional methods of assaying an extract against a specific target–be it a
cancer cell line, receptor or microorganism–and separating the components and “chasing” the
active fraction until the active constituent is isolated has several disadvantages. First of all, the
process is quite inefficient (as many compounds of interest are naturally present only in very
small concentration), requiring multiple assays and very large amounts of sample and solvent to
extract useable concentrations.4 Additionally, testing an extract of unknown composition
against a complex biological target results in a situation that relies more on chance than
scientific insight.5 Finally, as the identity of the active constituent is only identified after the
isolation is done, there is a large rate of rediscovering known compounds.
One method that is gaining popularity in natural product research is metabolomics.
Metabolomics is the study of the secondary metabolites produced by an organism. 6 This
approach is advantageous in drug research as it allows the researcher to quickly discard
compounds that have been previously identified and prioritize those compounds that have
novel structures.5 Metabolomics also allows a library of compounds to be compiled for testing
against multiple targets. Two main techniques are used for metabolomics. Liquid
Chromatography-Mass Spectrometry (LC-MS) is the most common, due to high sensitivity and
ease of automation. However, Nuclear Magnetic Resonance spectroscopy (NMR) is also used;
the advantages of NMR are the wealth of structural information gleaned as well as the ability to
detect any organic compound. In brief, NMR detects the frequency of spinning nuclei in a
magnetic field. That frequency is dependent on the strength of the magnetic field, which in turn
is dependent on the strength of the NMR instrument, the type of nuclei (isotope and element),
and the electrical shielding produced by the electrons surrounding the nucleus. This frequency
produces a diagnostic spectrum of peaks for a particular molecule. For organic molecules,
hydrogens produce a strong signal and are present in all organic molecules. The other nuclei
that can be detected by NMR that is ubiquitous in organic molecules is carbon-13; however,
NMR spectra based on carbon-13 have low sensitivity due to the low natural abundance of
carbon-13, typically about 1%.
One of the most influential works in the natural products field is Lipinski’s Rule of Five. Lipinski’s
Rule of Five, simply put, is a set of guidelines that predict favorable pharmacokinetic properties
of a compound that has potential as an oral drug.7 It was developed in order to offset the
number of compounds discovered through high-throughput screening that, though they show

intrinsic activity, are unable to be absorbed by the body. Lipinski’s Rule of Five states that if a
drug has more than 5 hydrogen bond donors, a molecular weight greater than 500 amu, a logP
greater than 5 (where P is the 1-octanol-water partition coefficient), or more than 10 hydrogen
bond acceptors, it will most likely exhibit poor pharmacokinetic parameters. This allows
potential drug compounds to be eliminated without expensive in vivo testing. Additionally, as
logP is a property that can be determined independent of structure, it can be used as a filter to
increase the presence of potential drug compounds in an extract in a metabolomics driven
approach to natural product discovery.8
One application of Lipinski’s Rule of Five is to use reverse phase column chromatography to
separate compounds based on logP. Camp et al, 2012, tested various solid phase extraction
cartridges to optimize the retention of these components.8 They verified the resulting extract
by high pressure liquid chromatography (HPLC) which correlates logP and retention time.
Building on this method, Grkovic et al, 2014, used solid phase extraction to compile NMR
fingerprints of fractionated natural product extracts.9 They then identified unique fractions
from the 1 dimensional NMR spectra of all of the 220 fractions; when tested, this led to the
discovery of a new agent to explain the cellular mechanism of Parkinson’s disease.
Both solid phase extraction and column chromatography rely on the same principles–polar
compounds are more attracted to the polar mobile phase and non-polar compound are
attracted to the non-polar stationary phase leading to complete retention in the case of solid
phase extraction and separation in the case of column chromatography–and indeed, non-polar
solid phase extraction and reverse phase column chromatography use the same packing
material (C-18 bound to a silica gel support).
Using NMR-based metabolomics, the chemical shifts of the different molecules are used to
query the Biological Magnetic resonance databank (BMRB).10 Unfortunately, due to the large
number of compounds, too many peaks overlap in a 1-dimensional NMR spectrum, so 2dimensional spectra will be obtained to separate this overlap by resolving peaks produced by
hydrogens based on their carbon neighbors.

Methods
The entire above ground portion of G. squarrosa was collected from the northern exposure of
Kenna Cartwright Park in Kamloops, British Columbia. The plant matter was air-dried over a
three-week period, then 0.75 g of leafy matter and 2.97 g flower buds (including both flowers in
bloom and flowers that had not yet opened) were collected for a total mass of 3.72 g plant
matter. This plant matter was ground, then transferred to an Erlenmeyer Flask. 100 mL of ACS
grade hexanes were used to rinse the mortar and pestle used into the Erlenmeyer Flask, after
which the flask was sealed and left to sit for 24 hours.
The contents of the Erlenmeyer were filtered through coarse filter paper and the hexanes
extract discarded. The fibrous plant matter was returned to the Erlenmeyer, and 100 mL of ACS
grade acetone was added. The flask was sealed and the plant material extracted for 24 hours.
The acetone extract was filtered into a 250-mL round bottomed flask (RBF) and evaporated to
dryness by rotary evaporation (rotovap), producing a deep green oil. The plant material was
extracted again with 100 mL of HPLC grade methanol for 24 hours. This final extract was filtered
into a different 250-mL RBF and evaporated to dryness, producing a light green-yellow oil.
The dried acetone extract was resuspended in four 5 mL portions of methanol, which were
transferred to the RBF containing the dried methanol extract before being transferred to a vial.
This vial was stored at 4°C for one week and then dried by nitrogen blow down, producing a
viscous, dark green oil. This oil was resuspended in 10 mL of methanol by ultra-sonication. The
suspension was transferred to a 50-mL RBF. 2 mL of preparative C-18 125 Å 55-105 µm Waters
silica solid phase was added to the RBF and the mixture was rotovapped to produce a thick
green paste, adhering the metabolites in the extract to the surface of the packing material. This
paste was transferred to the surface of a 3 cm x 10 cm C-18 column.
The metabolites were eluted in with a stepwise eluent gradient from 35:65 methanol : water to
100% methanol as outlined in recent research by Camp et al.8 Though the research by Camp et
al was done using a High Pressure Liquid Chromatography column, the eluent composition over
the 20 minute run was mimicked in the steps of the gradients used in this experiment. 100 mL
each of 35, 45, 55, 65, 70, 75, 80 and 90% methanol–with the remainder being made up of
deionized water– was used to elute the compounds from the column. Then 800 mL of 100%
methanol was passed through the column until the dark green band–believed to be
chlorophyll–was eluted.
71-15 mL aliquots were collected. These aliquots were pooled according to Table 1, giving 11
fractions that were evaporated to dryness by rotovap. The fractions with the lowest logP
(fractions 1-7) were resuspended in 0.5 mL methanol-d4 spiked with tetramethylsilane as an
internal standard and centrifuged to remove any insoluble components. Each of the selected
samples were analyzed by NMR spectroscopy, using an HSQC pulse sequence (Number of
Scans=64, Time Domain Data points=2048 hydrogen x 512 carbon-13, hydrogen offset
4.7=ppm, carbon-13 offset=85.0 ppm, Spectral Width=10.9920 ppm hydrogen x 172.9149 ppm

carbon-13). HSQC probes interactions between neighboring hydrogens and carbons on the
same molecule and produces a relatively simple 2-dimensional plot, with the intensity of the
peaks projecting into the 3rd dimension. A list of peaks was extracted from the processed
spectra using the peak picking program integrated into Topspin v2.4. The list of peaks was
queried against the BMRB, which returned a list of compounds. The list of compounds from
each fraction was combined and filtered to remove compounds that had less than 50% of the
expected peaks present in the experimental spectra. As well, any duplicates were removed.
Table 1. The manner in which the aliquots were pooled to produce the fractions later analyzes by NMR spectroscopy.

Fraction number

Subfraction range

Mass (g)

1

1-8

0.2227

2

9-14

0.0138

3

15-20

0.0144

4

21-28

0.0552

5

29-34

0.0233

6

35-45

0.0414

7

46-50

0.0281

8

51-55

0.0222

9

56-60

0.0106

10

61-65

0.0049

11

66-71

0.0099

Results and Discussion
Querying the database produced 597 unique compounds which were filtered to 88 compounds
identified with relative certainty, reported in Table 2. This is greater than the number of
compounds returned by the database before fractionation with 477 compounds, showing that
fractionation allowed for compounds that may have been obscured by other, more
concentrated, compounds in the NMR spectra to be identified. In addition, because not all of
the fractions were analyzed, this increase in compounds between fractionated and crude
extracts is probably much more significant. Compounds of note for their biological activity are:
agmatine (antidiabetic and neuropathic pain reduction11), 4-Guanidinobutyric acid (gastric
lesions inhibitor12), and dopamine (neurotransmitter).
Table 2. The compounds returned from the BMRB database. The peak match is the percentage of expected peaks from each
compound–from the spectrum of the pure compound–that were actually present in the experimental spectra of the mixture.
Only those compounds that had greater than 50% of their peaks present in the NMR spectrum are reported.

Compound
3-Ureidopropionic acid
Acetyl phosphate
Adipic acid
Citrate
cyclohexane
Methanol
Methylmalonic acid
Pyruvic acid
scyllo-Inositol
sulfoacetic acid
thioacetamide
Glutaric acid
Xylitol
Agmatine
D-(+)-Threo-isocitric acid
4-Methyl-2-oxovaleric acid
O-Phospho-L-serine
D-Saccharate
propanol
4-methylvaleric acid
Pimelic acid
isovaleric-acid
suberic acid
2-ketobutyric acid
3-Methyl-2-oxobutinoic acid

Peak Match
1
1
1
1
1
1
1
1
1
1
1
0.88
0.86
0.8
0.8
0.78
0.78
0.75
0.75
0.73
0.73
0.71
0.69
0.67
0.67

D-Citrulline
DL-alpha-Glycerol phosphate
L-Dihydroorotic acid
L-Homoserine
L-Serine
meso-Erythritol
O,O-diethyl thiophosphate
16-hydroxyhexadecanoic acid
N(alpha)-Acetyl-DL-ornithine
(R)-2-hydroxybutyric acid
1-Hexadecanol
DL-Serine
i-Erythritol
Isethionic acid
L-Ascorbate
D-Glucuronate
1,3-Propanediol
2-Aminoethyl dihydrogen phosphate
L-Citrulline
L-Isoleucine
Stearic Acid
Valeric acid
L(+)-Selenomethionine
1-Octanol
DL-2-Aminobutyric acid
N-Acetyl-D-glucosamine 1-phosphate
Nepsilon-Acetyl-L-lysine
L-Arginine L-Glutamate
(R)-Lactate
1,3-Diaminopropane
2-chloroethanol
2-Isopropylmalic acid
2,2-dimethylsuccinic acid
3-butyn-1-ol
3,5-dichlorocatechol
4-Chlorophenol
4-Guanidinobutyric acid
4-Hydroxy-benzoic acid
Ala-Ala
Alanine
alpha-Ketoglutaric acid

0.67
0.67
0.67
0.67
0.67
0.67
0.67
0.62
0.62
0.61
0.61
0.6
0.6
0.6
0.6
0.58
0.57
0.57
0.57
0.57
0.57
0.56
0.55
0.54
0.54
0.53
0.53
0.52
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5

Betaine
Citraconic acid
Creatine
Creatine phosphate
Dihydroxyacetone
DL-threo-beta-Methylaspartate
dopamine
epichlorohydrin
Ethanolamine
Glycogen
Glycolaldehyde
Guaiacol
Isobutyric acid
L-(-) Arabitol
L-Glutamic acid
L-Glutamine
n-acetylglycine
N,N-Dimethylglycine
p-fluorobenzoic acid
Propionic acid
Succinic acid
syringic acid

0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5

Two large issues are inherent in this technique. One is that the BMRB provides no statistical
measurement of the validity of peak assignment. This means that beyond instituting arbitrary
peak match cut offs–as I did at 0.5–there is no way to filter the results for accuracy beyond
physically examining each spectrum individually, which for the total 597 unique compounds
returned by the database would be prohibitively time consuming. The peak match itself is
inadequate as measure of assignment accuracy. The peak match would be skewed towards
compound with simple substructures. For example, citrate–shown in Figure 1–has only two
carbon-hydrogen pairs to produce a signal in the HSQC spectrum. In addition, due to symmetry,
the signal those two pairs produce are at an identical ppm (the units of the spectrum axes). In
order for the BMRB to return a peak match of 1, all it need to detect is the single signal
produced by citrate. This issue is present in many of the compounds shown in Table 2

Figure 1. The structure of citrate. The circled atoms are those that produce a signal in the HSQC spectra.

The second issue that is grounded in the very nature of databases is that no new compounds
can be discovered through simply querying a database. This is because the database is
populated with known compounds. This problem is compounded by small database sizes, as
NMR-based metabolomics is an emerging field. Querying a database that is too small not only
cannot return unknown compounds, but known compounds remain unidentified as well.

Conclusions and Future Work
477 compounds were identified in the crude G. squarrosa extract and 597 compounds were
identified in the fractions of greatest polarity, 1-7. Filtering these compounds by 50% peak
match resulted in 88 compounds, three of which are known to have biological activity.
In order to address the issues with NMR-based metabolomics, reconstructed spectra should be
made from all the identified compounds. This is done by overlaying all of the individual
compounds’ HSQC spectra on one plot, and comparing this plot to the experimental spectra.
Matlab is a matrix manipulation software that can do this sort of reconstruction and overlaying
of compounds.13 The overlay allows for visual confirmation of the BMRB’s peak assignment.
Peaks that are present in the reconstruction, but not the original spectra suggest that the
compounds responsible for those peaks were assigned by the database in error, and should be
excluded from the list of peaks. Peaks that appear in the experimental spectra but not in the
reconstructed spectra are due to novel compounds.
Further investigation should be done on those unassigned peaks. Different NMR pulse
programs probe different interactions in molecules. For example, a COSY spectrum looks at
hydrogens that are separated by up to three bonds. This allows for the substructure of the
molecule to be parsed by “walking” along the neighboring hydrogens. COSY spectra are more
complex than HSQC, and fewer databases are configured in such a way as to allow COSY peaks
to be directly inputted, but identifying an unknown peak in the HSQC spectrum and using the
COSY spectra to see which other peaks belong to the same molecule allows a single molecule to
be queried against a database as if it were a pure compound.

Literature Cited
1.

Stuhr, E. T. The distribution, abundance and uses of wild drug plants in Oregon and
Southern California. Econ. Bot. 1, 57–68 (1947).
2.
Hassan, H. M., Jiang, Z.-H., Asmussen, C., McDonald, E. & Qin, W. Antibacterial activity of
northern Ontario medicinal plant extracts. Can. J. Plant Sci. 94, 417–424 (2013).
3.
Morimoto, S. et al. Morphine metabolism in the opium poppy and its possible
physiological function: biochemical characterization of the morphine metabolite,
bismorphine. J. Biol. Chem. (2001). doi:10.1074/jbc.M107105200
4.
Bendaikha, S., Gadaut, M., Harakat, D. & Magid, A. Acylated flavonol glycosides from the
flower of Elaeagnus angustifolia L. Phytochemistry 103, 129–136 (2014).
5.
Kurita, K. L. & Linington, R. G. Connecting phenotype and chemotype: high-content
discovery strategies for natural products research. J. Nat. Prod. 78, 587–596 (2015).
6.
Dona, A. C. et al. A guide to the identification of metabolites in NMR-based
metabonomics/metabolomics experiments. Comput. Struct. Biotechnol. J. 14, 135–153
(2016).
7.
Lipinski, C. A., Lombardo, F., Dominy, B. W. & Feeney, P. J. Experimental and
computational approaches to estimate solubility and permeability in drug discovery and
development settings. Adv. Drug Deliv. Rev. 23, 3–25 (1997).
8.
Camp, D., Davis, R. A., Campitelli, M., Ebdon, J. & Quinn, R. J. Drug-like Properties:
Guiding Principles for the Design of Natural Product Libraries. J. Nat. Prod. 75, 72–81 (2012).
9.
Grkovic, T. et al. NMR Fingerprints of the Drug-like Natural-Product Space Identify
Iotrochotazine A: A Chemical Probe to Study Parkinson’s Disease. Angew. Chem. Int. Ed. 53,
6070–6074 (2014).
10.
Ulrich, E. L. et al. BioMagResBank. Nucleic Acids Res. 36, D402-408 (2008).
11.
Piletz, J. E. et al. Agmatine: clinical applications after 100 years in translation. Drug
Discov. Today 18, 880–893 (2013).
12.
Hwang, I. Y. & Jeong, C. S. Inhibitory Effects of 4-Guanidinobutyric Acid against Gastric
Lesions. Biomol. Ther. 20, 239–244 (2012).
13.
Bingol, K., Bruschweiler-Li, L., Li, D.-W. & Brüschweiler, R. Customized Metabolomics
Database for the Analysis of NMR 1H–1H TOCSY and 13C–1H HSQC-TOCSY Spectra of
Complex Mixtures. Anal. Chem. 86, 5494–5501 (2014).