W.H. Perera et al. / Phytochemistry xxx (2016) 1e9
7
window in f1 of 0e120 ppm, with 1024 points in f2 and 512
<1% compounds 1 and 2 with the remaining 19% of the extract
being other minor steviol glycoside components such as rebau-
dioside D, dulcoside A, rebaudioside F, rubusoside, rebaudioside B,
steviolbioside, and additional components, respectively. Many
different purification approaches were evaluated and are described
generally here. Commercially available S. rebaudiana leaf extract
(ca 1.5 kg) was dissolved in MeOH or MeOH: H2O (9:1 v/v) at about
200 mg/mL and allowed to crystallize. The crystalline products
were rebaudioside A and stevioside, which accounted for approx-
imately 50% of the starting mass. The enriched mother liquors after
drying were dissolved in hot solution containing EtOAc: MeOH:
H2O: AcOH (100:18:14:0.1 v/v; 200 mg/mL) and the solution
allowed to sit overnight. Two layers were observed and separated.
The lower aqueous layer was dried and the same procedure was
repeated to afford a minor steviol glycosides pool (630 g). This pool
was dissolved in Reb N mobile phase 5 L, [EtOAc: MeOH: H2O: AcOH
(63.1: 19.7: 17.2: 0.1; v/v/v/v)] and loaded into a normal phase
points in f1, 16 transients per increment and relaxation delay of
1
1 s, with multiplicity editing and optimized for a JCH of 146 Hz.
The gHMBC spectra were run on a spectral window in f1 from 0 to
200 ppm with 1024 points in f2 and 512 points in f1, 64 transients
1
per increment and relaxation delay of 1 s, without the JCH filter
n
and optimized for a JCH of 8 Hz. Two gHSQCTOCSY experiments,
with mixing times of 30 and 150 ms, were run on a spectral
window in f1 of 50e120 ppm, with 1024 points in f2 and 512
points in f1 relaxation delay of 1 s and 16 or 64 transients per
increment, correspondingly. Five TOCSY spectra were run with
mixing times of 30, 50, 70, 100 and 150 ms, in either the 1D or
band-selective 2D version, depending on the number of anomeric
protons and their degree of overlap. The ROESY spectra were run
with a mixing time of 200 ms, in 1 k points in both dimensions
and with a relaxation delay of 1 s.
The assignment of the 1H and 13C chemical shifts in the steviol
moiety was made on the cross-peaks seen in the gCOSY, gHSQC and
gHMBC spectra. The stereochemical assignment of the protons was
column 15 i.d. X 135 cm length packed with 37e63 mm flash silica
gel (10 kg). Reb N mobile phase was pumped at 550 mL/min
4 ꢂ 3.78 L forerun and 55 ꢂ 1 L fractions were collected.
The fractions were analyzed by HPLC and pools were selected by
column analysis (Rodenburg et al., 2016; McChesney and
Rodenburg, 2014). Six main pools were obtained from this large
scale chromatography, pool 1: Forerun 4- fraction 3 (12.435 g) rich
in apolar compounds; pool 2: fraction 4e14 (365 g) rebaudiosides
A-C; pool 3: fraction 15e27 (203 g) rebaudiosides E, D, M and
unknown peaks; pool 4 and pool 5: fraction 28e36 (28.01 g)
and fraction 37e49 (14.26 g), rich in rebaudiosides N and O in
different ratios, and pool 6 fraction 50e55 (2.65 g) predominantly
rebaudioside O. Regeneration of the column was performed with
22.7 L of Reb C/MeOH mobile phase (50:50 v/v), [Reb C mobile
phase ¼ 100:18:14:0.1 v/v/v/v of EtOAc: MeOH: H2O: AcOH] and
fractionated in two main regeneration pools, pool 7: 11.4 L (2.43 g)
and pool 8: 11.3 L (7.84 g).
3
made based on the values of JHH estimated from the DQF-gCOSY
3
spectrum or on qualitative estimations of the JCH values from the
gHMBC spectrum. The spectrum optimized for 8 Hz shows large
cross-peaks for the anti-orientation of the proton and carbon three
bonds away.
The collection of 1H and 13C chemical shifts in each sugar moiety
was inferred from the gHSQCTOCSY spectrum at longer mixing
times. The gHSQC spectrum was then used to pair protons and
carbons one-bond away. The order of the methine pairs was
inferred from the cross-peaks in the gHSQCTOCSY spectrum with a
short mixing time. Alternately, the order of the protons in a sugar
starting from the anomeric position was seen in the TOCSY spectra
with increasing mixing time. The TOCSY spectra revealed the
number of large couplings of protons in positions 2e4, which were
used to determine the relative configuration in these positions. For
instance, in a pyranose, three triplets of ca. 9 Hz indicate that all of
the substituents in positions 1e5 are equatorial. All of the hex-
apyranoses displayed this pattern, indicating that they are glucose,
and that they are beta. In the same way, the 6-deoxy hexapyranose
was identified as beta-6-deoxyglucose. The vicinal couplings in the
xylose unit of 1, namely H1eH2, H2eH3, H3eH4 and H4 with the
methylene proton at 3.21, displayed values typical for axial-axial
An aliquot of the pool 3 (200 g) was dissolved in H2O and
further fractionated in a high efficiency reversed-phase column
(7.5 i.d. X 50 cm) packed with 10 mm spherical C18 gel (1.4 kg)
using a four step gradient (6 L each) pumped at 200 mL/min first
with H2O: AcOH (0.1%), followed by CH3CN: H2O: AcOH (10: 90:
0.1); (20: 80: 0.1) and finally with (30: 70: 0.1). The first six liters
were collected in 1 L fractions and the other 18 L in 500 mL
fractions; a total of 15 pools were prepared after column analysis
based on HPLC results, concentrated and dried in a vacuum
oven overnight (Rodenburg et al., 2016; McChesney and
Rodenburg, 2014). Pool 1, fraction 10e12 (1.90 g); pool 2, frac-
tion 13e14 (0.90 g); pool 3, fraction 15 (1.76 g); pool 4b, fraction
16 (15.32 g), pool 5, fraction 17 (11.27 g); pool 6, fraction 18
(7.86 g); pool 7, fraction 19 (5.92 g); pool 8, fraction 20e23
(17.58); pool 9, fraction 24 (2.59 g); pool 10, fraction 25e26
(3.87 g); pool 11, fraction 27 (2.83 g), pool 12, fraction 28 (7.55 g);
pool 13, fraction 29 (5.55 g); pool 14, fraction 30e33 (34.15 g);
pool 15, fraction 34e36 þ regeneration (60.03 g).
couplings on
a pyranose, ca. 9 Hz, proof for the beta-
xylopyranose form. In the arabinose unit of 2, the pyranose ring
was demonstrated by a coupling between H1 and C5. The relative
configuration was established based on vicinal proto-proton cou-
plings, as before. The vicinal couplings of H4 are small, 2e3 Hz,
while H1eH2 and H2eH3 are large. The connectivity of a sugar
residue to another sugar or to positions 19 or 13 in steviol was
revealed by cross peaks of the anomeric protons in each sugar with
carbons outside of that sugar moiety.
3.3. Plant material
Pool 12 (7.55 g) was absorbed in Celite (45 g) and then applied
onto a high efficiency normal-phase column (7.5 i.d. X 50 cm)
packed with 10 mm flash silica gel, pumping MTBE: MeOH: H2O:
The starting material was a partially processed commercially
available leaf extract from Stevia rebaudiana with Lot # SRE50-
14091 purchased from American Mercantile, Memphis, TN, USA.
HPLC comparison of that extract with several other S. rebaudiana
extracts purchased from various sources showed high similarities,
differing only in the relative concentrations of specific glycosides
but not their presence or absence.
AcOH (100: 30: 12.5: 0.01) at 200 mL/min 2 ꢂ 1 L forerun plus
200 mL were collected followed by 10 ꢂ 125 mL fractions and
from 11e50 ꢂ 50 mL fractions. Pools were prepared following the
procedure previously described to afford 11 pools (Rodenburg
et al., 2016; McChesney and Rodenburg, 2014). Compound 1
from pool 2 (1.35 g) was crystallized from MeOH to yield (0.92 g)
after the second crop. Compound 2 from pool 9 was also crystal-
lized from MeOH to yield (0.31 g). Relative peak purities were
determined by HPLC and peak areas for both compounds were
higher than 95%.
3.4. Isolation
The composition of the Stevia rebaudiana extract was approxi-
mately 40% rebaudioside A, 30% stevioside, 10% rebaudioside C, and
Please cite this article in press as: Perera, W.H., et al., Rebaudiosides T and U, minor C-19 xylopyranosyl and arabinopyranosyl steviol glycoside
Perera, Wilmer H.
