Solvent selection 1739

Dominik Bergs1 Research Article

Juliane Merz1
Axel Delp2 A Standard Procedure for the Selection of
Matthias Joehnck2
Georg Martin3 Solvents for Natural Plant Extraction in the
Gerhard Schembecker1 Early Stages of Process Development
1
Laboratory of Plant and
Process Design, Department The standard procedure for the selection of solvents for natural plant extraction
of Biochemical and Chemical is a hybrid approach of solubility estimation and experimental investigations to
Engineering, Technische reduce the time of solvent screening to a minimum. The focus is on the selection
Universität Dortmund, of a suitable solvent for natural product extraction in the early stages of process
Dortmund, Germany. development when thermodynamic data are not yet available and solid-liquid
2
Merck Millipore, Darmstadt,
extraction is still an alternative to other techniques. The standard procedure is
Germany. implemented in Microsoft® Excel™, provides the necessary information, creates
3
experimental test plans, performs calculations and supports the user by struc-
LANXESS Deutschland GmbH, tured descriptions to select the most desirable solvent for extraction. In this study,
Leverkusen, Germany. its application to the natural products rebaudioside A from Stevia rebaudiana
Bertoni leaves, caffeine from guaraná seeds, and artemisinin from Artemisia an-
nua leaves is demonstrated. The standard procedure predicted solvents that were
found in the literature, plus solvents with even better extraction performance
than the ones described in the literature.

Keywords: Artemisinin, Natural products, Solvent selection, Standard procedure, Steviol

glycosides
Received: April 25, 2013; revised: May 31, 2013; accepted: July 29, 2013
DOI: 10.1002/ceat.201300276

1 Introduction extract. In solvent extraction, the main task is to find a suitable

solvent to extract the target compound from the plant materi-
Processing natural products is gaining importance in the food, al. Therefore, solvent or solubility screening is necessary. There
cosmetics, and pharmaceutical industries. Especially, the end are two main approaches to determine the solubility of a target
consumer has a growing interest in natural products with a compound in different solvents: solubility prediction methods
sustainable organic origin. For example, the market for ex- on the one hand and experiment-based solubility estimations
tracts from plants grew from US$ 0.26 billion in 1993 to on the other hand. Several empirical and semi-empirical equa-
US$ 1.99 billion in 2008, which is an annual market growth of tions for solubility prediction can be found in the literature.
about 15 % [1]. The target compound can be located in any These equations require a huge number of experimental solu-
part of a plant. Depending on region and seasonal fluctuation, bility data to determine model coefficients, which can have a
its concentration can vary from 0.1 to 35 wt %. physical character [5, 6] or may depend on functional group
Common plant extraction methods for pharmaceutical and contributions and molecular structure [7, 8]. Bart and co-
food processing are supercritical fluid extraction, hydrodistilla- authors [1] and Kassing et al. [9] give a comprehensive scien-
tion, steam distillation, cold pressing, and solvent extraction tific literature review covering solubility estimation and mod-
[1–3], whereas liquid CO2 dominates the market [4]. This arti- eling for plant-based extraction. A more qualitative method is
cle focuses on solvent extraction, also known as solid-liquid the structured selection based on the Robbins chart of solute-
extraction, of plant material to generate a primary crude liquid solvent group interaction, which is a guide to rank general
classes of solvents. Here, a large number of equilibrium data
have been analyzed for a qualitative comparison of activity
– coefficients of functional groups from a natural product, with
Correspondence: Prof. Dr.-Ing. G. Schembecker ([Link]
activity coefficients of solvent classes [10, 11].
@[Link]), Laboratory of Plant and Process Design, Depart- The most common situation is that thermodynamic data for
ment of Biochemical and Chemical Engineering, Technische Universität new natural products are either not available or insufficient.
Dortmund, Emil-Figge-Straße 70, 44227 Dortmund, Germany. Furthermore, the estimation of solubility to describe extrac-

Chem. Eng. Technol. 2013, 36, No. 10, 1739–1748 © 2013 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim [Link]
1740 G. Schembecker et al.

tion can only be a rough approximation. Most of the methods mation about a suitable solvent. These selected solvents are
have been developed in the field of pharmaceuticals and poly- then investigated experimentally with special extraction condi-
mers and are therefore of limited applicability in the field of tions regarding the extraction performance via yield, selectivity
natural product extraction. Swelling behavior, selectivity of of extraction, and solvent costs. The special extraction condi-
extraction, solvent viscosity, surface tension, and cellular local- tions are developed to render solvents with different extraction
ization of the target compound have an impact on the extrac- behaviors comparable, and criteria for the evaluation and
tion efficiency [1]. Experiment-based solvent selection can take comparison of the experimental results are introduced. The
these effects into consideration. Beside trial-and-error experi- theory of the standard procedure is presented in Sect. 2. The
ments, which require much time and material, a structured, standard procedure is tested and validated for the extraction of
experiment-based selection of liquid solvents for a typical sol- rebaudioside A from leaves of Stevia rebaudiana Bertoni, for
id-liquid extraction process for plant materials is addressed by the extraction of caffeine from guaraná seeds, and for the
Sarker and co-authors [12]. Here, a broad unspecific polarity extraction of artemisinin from leaves of Artemisia annua. The
screening is executed using the solvents water, ethanol, and results are presented and discussed in Sect. 3.
methanol as polar components, ethyl acetate and dichloro-
methane as medium-polarity components, and n-hexane, pet-
rol ether, and chloroform as nonpolar components. 2 Methodology
In the field of nutraceutical products, ethanol, water, or a
mixture of the two is often chosen without any further investi- The standard procedure for the systematic selection of a sol-
gation of extraction efficiency because of consumer acceptance vent for extraction consists of six steps. In Fig. 1, the six main
[11, 13]. A combination of rigorous modeling and design of steps are depicted. The following section introduces the theory,
experiments for process development of plant-based extrac- the estimation and calculation methods, and the experiments
tions with high accuracy is addressed by Kassing et al. [14]. for the standard procedure. Before starting to use it, a robust
Here, botanical investigations on plant tissue structure are analytical method must be available for the target compound
used to consider plant-based effects on the extraction efficien- and the main side components to evaluate the experimental
cy, but this approach can only be used for the prediction and results.
optimization of process parameters when a solvent has already
been selected. Pfennig [13] published a knowledge-based de-
sign method for solid extraction based on laboratory-scale
experiments. Here, the pre-selection of solvents is based on a
database that contains only information about the legislation
of solvents in different fields of application, exposure limits,
and solvent prices. Furthermore, the octanol/water partition
coefficient of the solvents as an indicator of polarity is used,
which is insufficiently precise for natural products. We also
think that a hybrid approach of heuristics, solubility estima-
tion methods, and laboratory-scale experiments is the method
of choice to find a suitable solvent in solid-liquid extraction
because solubility estimation methods are of limited applic-
ability in the field of natural products. The objective is to select
suitable solvents for natural product extraction in the early
stages of process development when thermodynamic data is Figure 1. Guideline of the standard procedure for the selection
not yet available and solid-liquid extraction is still an alterna- of solvents for natural plant extraction in the early stages of pro-
tive to other techniques. Therefore, a standard procedure is cess development.
developed in this work using heuristic methods, a combination
of three solubility prediction methods, and standardized labo- The major goal of the standard procedure is to find desirable
ratory-scale experiments. Therefore, the standard method is solvents to extract a natural target compound from plant
not meant to replace existing methods for modeling and esti- material with high yield, high selectivity, and low costs, in a
mation of solubility which are certainly more precise. The short time. The standard procedure helps the user to deter-
standard procedure is implemented in a Microsoft® Excel™ mine the necessary properties to gain information about the
tool and provides the necessary information, creates test plans, target compound. In addition, a comprehensive solvent data-
performs calculations and supports the user by structured base is implemented with more than 100 solvents and their
descriptions to obtain results with little effort regarding time physical and chemical properties, Hildebrand and Hansen
and costs. Ten solvents with the highest predicted solubility are parameters, and 1-octanol/water partition coefficients. The
selected via heuristics and solubility estimation, from a data- standard procedure can handle low amounts of information in
base of over 100 solvents, without performing any experiment. which only the molecular structure of the target compound is
These methods have been developed in the field of pharmaceu- known whereas the content of the target compound in the nat-
ticals and polymers and are therefore of limited applicability ural product, the type and number of impurities, and data
in the field of natural product extraction; however, they show concerning extraction yield, extraction selectivity, and solubili-
reliable results and should only be used to get a rough approxi- ty of the target compound in different solvents are unknown.

[Link] © 2013 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chem. Eng. Technol. 2013, 36, No. 10, 1739–1748
 Solvent selection 1741

In Step 1 of the standard procedure, data like the molecular combined ranking position of 3.3 is assigned to the target
weight, the 1-octanol/water partition coefficient, the molar compound-ethylene glycol combination.
volume, the melting point and boiling point, and the enthalpy The values of the 1-octanol/water partition coefficient of the
of vaporization of the target molecule must be made available. solvents are implemented in the solvent database. The value of
The data are necessary to determine physically suitable sol- the 1-octanol/water partition coefficient of the target com-
vents and to estimate the solubility of the target compound in pound is calculated from its molecular structure as described
different solvents. These physical properties of the target com- in Step 1. The Hildebrand parameter and the three factors of
pound are estimated from the molecular structure by using the the Hansen solubility parameter (dispersion force, dipolar
freeware ChemSketch from Advanced Chemistry Develop- intermolecular force, hydrogen bonds) of each solvent are im-
ment, Canada [15]. In case of a known target compound, these plemented in the solvent database [21, 22]. The parameters for
values can be found with the web service SciFinder from the target compound are unknown. They are determined via a
Chemical Abstracts Service, USA [16]. group contribution method developed by van Krevelen [21],
In Step 2, the thermal operating window of the extraction which is implemented in the standard procedure. The user
process is fixed. Solvents with a higher boiling point than the only has to count the numbers of each functional group in the
temperature stability range of the target compound are ex- molecular structure and accordingly fill in a ready-made form.
cluded. To ensure a liquid phase, solvents with a higher melt- As an internal control, the numbers of carbon, oxygen and
ing point than room temperature are excluded as well. hydrogen atoms are determined from the input data and must
Furthermore, the operator can use this option to narrow the be identical with the chemical formula of the target com-
thermal operating window when, e.g., equipment for high pound. Therefore, prior knowledge of the group contribution
temperature is not available or should not be used. Solvents methods is not necessary.
with unsuitable thermal data can be excluded with the Excel In Step 5 of the standard procedure, short, easy and compar-
sorting function. The temperature stability of the target com- able extraction experiments are performed to determine the
pound must be determined in a preliminary experiment if extraction yield and extraction selectivity of the ten pre-
the corresponding data is not available. The boiling and melt- selected solvents. The extraction yield of the target compound
ing points of the solvents can be found in the solvent data- is related to the initial mass of dried plant material because the
base. total amount of the target compound is often unknown. The
In Step 3, the user must select the field of application for the selectivity of the separation is defined as the ratio between the
target compound to exclude solvents that are legally not amount of target compound and the amount of impurities in
allowed for this area. In case of no limitation, Step 3 can be the extract. The extraction yield and selectivity mainly depend
skipped. The database provides information about solvents on the extraction kinetics, the extraction temperature, the liq-
covering aspects of legislation for foodstuff, food supplements, uid-to-solid ratio, the swelling behavior, and the particle size
and green solvents [17–19]. Solvents with inappropriate legis- of the plant material. The strategy is to select special condi-
lation are excluded with the Excel sorting function. The data- tions in which the effects are equal or can be neglected, to
base can be easily extended to consumer requirements like sol- compare the extraction yield and selectivity of the ten pre-
vents allowed in the pharmaceutical or cosmetic industry. selected solvent.
In Step 4, ten suitable solvents with the highest solubility are The extraction kinetics is mainly limited by diffusion of the
selected for experimental investigation with the help of solubil- solvent through the porous plant material to the surface. The
ity estimation methods. A combined ranking position is calcu- target compound can be located in the pores of the plant ma-
lated for every target compound-solvent combination, using terial near the surface or, e.g., in the vacuole [23]. The rate of
three estimation methods for solubility: the 1-octanol/water diffusion can be described by Fick’s second law and increases
partition coefficient [20] and the Hildebrand and Hansen with larger concentration gradients, with larger diffusion coef-
parameters [20, 21]. ficients, and by reducing the particle diameter [24].
With the use of ranking positions, comparison of the differ- The diffusion coefficient depends on the temperature and
ent value ranges of the three solubility parameters is ensured. the dynamic viscosity of the solvent and increases with higher
In detail this means: First, the standard procedure separately extraction temperatures and lower viscosity [24]. Higher tem-
estimates the solubility using the 1-octanol/water partition peratures result in higher solubility, but attention must also be
coefficient and the Hildebrand and Hansen parameters paid to thermal decomposition.
[21, 22]. Thereafter, the solvents are ranked according to their Plant materials have different swelling behaviors depending
solubility. Each target compound-solvent combination then on the use of different solvents. The swelling behavior has an
has three provisional ranking positions between 1 (high solu- influence on the liquid-to-solid ratio needed and on the resid-
bility) and n (low solubility). The three provisional ranking ual moisture. A high liquid-to-solid ratio results in good mix-
positions of the target compound-solvent pair are summarized ing performance and higher concentration gradients, but a
as arithmetic mean value to determine a combined ranking dilution of the extract leads to much more effort during down-
position. Ten solvents with the highest combined ranking posi- stream processing. A low liquid-to-solid ratio results in bad
tion are assigned. mixing performance and lower yield of extraction but highly
An example: A target compound has the following provi- concentrated extracts. To compare solvents with different
sional solubility ranking positions: 4 by the 1-octanol/water swelling behavior, a liquid-to-solid ratio must be selected that
partition coefficient, 4 by the Hildebrand parameter, and 2 by ensures a liquid phase after solid-liquid extraction and a good
the Hansen parameter, using ethylene glycol as solvent. The mixing performance to transport enriched solvent from the

Chem. Eng. Technol. 2013, 36, No. 10, 1739–1748 © 2013 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim [Link]
1742 G. Schembecker et al.

boundary layer of the plant material to the bulk phase. With a

Si ˆ Si S min (2)
gravimetric liquid-to-solid ratio of 10:1, a good mixing perfor- S max S min
mance can be observed especially for strongly swelling plant
material. It should be noted at this point that, when using a Ci C min
i ˆ 1
C (3)
plant with low density, a higher amount of solvent (20:1 to C max C min
40:1) is needed.
The concentration gradient is the same for every solvent be- These relative values of a target compound-solvent combi-
cause the plant material is the same. To avoid different diffu- nation are per definition between zero and one. One indicates
sion path lengths, the extraction experiments are performed the solvent with the best performance and zero indicates the
with powdered plant material in which the plant cells are dis- solvent with the lowest performance. The solvents with the
rupted and the particle size is similar. highest single value of Yi , Si , and C
 i are displayed as a first
The influence of different diffusion coefficients can be ne- result of the standard procedure. Hence, the solvent with the
glected when using a long extraction time of 24 h. The extrac- highest extraction yield, the highest extraction selectivity, and
tion temperature is set to room temperature; this ensures ther- the lowest extraction costs are determined. The mobile phase
mal stability of the target compound because it is similar to performance index (MPIi) for each solvent from the ten pre-
the natural environment. selected solvents can be determined as final evaluation. The
In Step 6, the extraction yield, extraction selectivity, and the MPIi summarizes yield, selectivity, and costs to one value. The
expected solvent costs are the basis for comparison of the ten contribution of each part to the sum can be defined by the
pre-selected solvents. The amount of the target compound in user via the factors a, b, and c (Eq. 4). As default setting, the
the free liquid phase after phase separation is determined in performance values are equally distributed.
order to calculate the extraction yield. This allows for the dif-
ferent swelling behavior of each solvent and the residual mois- 1 1 1
MPIi ˆ Yi ‡ Si ‡ C with a ‡ b ‡ k ˆ 1 (4)
ture can be calculated. It should be considered that yield losses a b c i
due to residual moisture can be decreased with better phase
separation techniques. The MPIi results are displayed in a bar diagram called MPI
As a first approximation, the selectivity of extraction is cal- plot (see Fig. 1 and Sect. 4). This diagram displays the MPIi
culated from the peak area of the chromatogram of the target, values of the ten pre-selected solvents i in sorted order.
because it is too time consuming to quantify every single Furthermore, the single-performance values Yi , Si , and C i are
impurity in the extract. The peak area of the target compound included as line chart for better interpretation of the results.
is divided by the peak area of all chromatogram peaks, except After Step 6, the systematic selection is finished and the major
for the solvent peaks. Therefore, a combination of diode-array goal of the standard procedure, i.e. to find desirable solvents to
detector (DAD), evaporative light scattering detector (ELDS), extract a natural target compound from plant material with
and refractive index detector (RI detector) is used to detect as high yield, high selectivity, and low costs, is reached in a short
many impurities as possible. time.
Another option of the standard procedure is to take the sol-
vent costs of extraction with the different solvents into ac-
count. Under the assumption that the same equipment is used, 3 Material and Methods
re-buy, solvent losses, and evaporation for recycling with a
variable recovery rate are considered. The residual moisture in 3.1 Chemicals and Apparatus
the plant material after extraction plus 5 % of free liquid sol-
vent defines the minimum amount of new solvent. Powdered leaves of Stevia rebaudiana Bertoni were purchased
The energy demand for heating up the solvent to the boiling from MedHerbs, Germany. Rebaudioside A (97.6 wt %), stevio-
temperature and the energy demand for vaporization are taken side (96.8 wt %), rebaudioside B (96.7 wt %), rebaudioside C
into account for solvent recycling. The solvent price and the (98.2 wt %), rebaudioside D (90.3 wt %), rebaudioside F
thermodynamic data are implemented in the solvent database. (90.8 wt %), steviolbioside (98.9 wt %), dulcoside A (97.5 wt %),
To compare the solvents, the yield Yi1), the selectivity Si of and rubusoside (95.8 wt %) were purchased as certified stan-
extraction, and the costs Ci for each solvent i are scaled to the dards from LGC Standards, Germany. Powdered seeds of guar-
minimum values Ymin, Smin, Cmin and the maximum values aná were purchased from MedHerbs, Germany, and the caf-
Ymax, Smax, Cmax of all ten solvents, to calculate the relative feine standard (99.7 wt %) was purchased from Alfa Aesar,
yield Yi (Eq. 1), the relative selectivity Si (Eq. 2), and the rela- Germany. Cut leaves of Artemisia annua were purchased from
tive solvent costs C i (Eq. 3). CPQBA, Germany, and the artemisinin standard (98 wt %)
Y Y min was purchased from Xian Bo Sheng Biological Technology,
Yi ˆ i (1) China.
Y max Y min
Acetonitrile in ultra-gradient grade for high-performance
liquid chromatography (HPLC) analysis was obtained from
VWR, Germany. Water was obtained from a Milli-Q Synthesis
apparatus with 0.22 lm Millipak express filters from Millipore,
– USA. Solvents for extraction (methanol, propylene glycol, eth-
1) List of symbols at the end of the paper. ylene glycol, 1-propanol, ethanol, 2-butanol, isopropyl alcohol,

[Link] © 2013 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chem. Eng. Technol. 2013, 36, No. 10, 1739–1748
 Solvent selection 1743

nitromethane, acetone, ethyl acetate, tetrahydrofuran, cyclo- In Step 1 of the standard procedure, the following properties
hexanone, pyridine dichloromethane, butanone, chloroform, of the target compound rebaudioside A were determined using
methyl acetate, carbon disulfide) in gradient grade were pur- ChemSketch [15]: a molecular weight of 967.01 g mol–1, an
chased from VWR, Germany. 1-octanol/water partition coefficient of –1.12, a molar volume
Extraction experiments were performed using an overhead of 607.7 cm3mol–1, a melting point of 485 K, a boiling point of
shaker (PTR30; Grant Bio, UK) in a tempered oven (ED 240; 1375 K, and an enthalpy of vaporization of 183.24 kJ mol–1. In
Binder, Germany) to ensure a constant temperature. All ex- Step 2, room temperature (303 K) was defined as the reference
tracts were filtered through 0.2-lm PTFE syringe filters (Carl temperature. Steviol glycosides are stable at up to 473 K [26].
Roth, Germany) to remove particles before chromatography. The operating window of the process was defined between
283 K and 473 K, to take care of decomposition and liquid
physical state. As a result of Step 2, the list of suitable solvents
3.2 HPLC Analysis was reduced from 107 to 79 solvents. Excluded solvents were,
e.g., benzyl alcohol, dodecane, formamide, and triethylene gly-
All extracts were analyzed by HPLC. For Stevia rebaudiana col with too high boiling points, and dodecanol, cyclohexanol,
Bertoni, a detailed description of the analytical method can be and phenol with too high melting points.
found in Bergs et al. [25]. In Step 3, the area of application was defined. Rebaudiosi-
For guaraná, the analytical method was performed on a Pur- de A is used in food supplements; thus, the solvent database
ospher STAR 120-3 RP18 ec column (250 mm × 4.6 mm; was reduced to legally allowed solvents in this application area
Merck, Germany) at 323 K, with an eluent composition of [17, 27]. This reduced the list of suitable solvents to 43. Com-
water/acetonitrile of 90:10 vol %. For detection, a DAD was mon solvents in natural product extraction like chloroform,
used. For Artemisia annua, the analytical method was per- dichloromethane, n-heptane, n-hexane, and toluene were ex-
formed on a NUCLEODUR® 100-3 C18 ec column (250 mm cluded.
× 4.6 mm) from Macherey-Nagel, Germany, at 313 K under the In Step 4, the solubility of rebaudioside A in the remaining
gradient conditions described in Tab. 1. For detection, a DAD 43 solvents was estimated. The three parts of the Hildebrand
and an ELSD were used. The HPLC setup consisted of two and Hansen parameters were determined. After calculation of
Smartline pumps P1050, a Smartline Manager 5000 (NDG, the previous ranking positions of each target compound-sol-
degasser and interface), a spectrophotometer (Smartline PDA vent combination, the combined ranking positions were deter-
detector 2800 UV/FIS/NIR), and a Smartline column thermo- mined. The ten best solvents for rebaudioside A are summa-
stat from Knauer, Germany. The Triathlon autosampler from rized in Tab. 2, which took about 10 min in total to be
Spark Holland and the Sedex 85LT light scattering detector achieved. The standard procedure proposes the use of water,
from Sedere France were used. The system was controlled with mono- or bivalent alcohols, a nitrile, and a nitro compound.
Chromgate software (version 3.3.2) from Knauer, Germany. To show that the pre-selection is comprehensive the results are
compared to literature data. In the literature, steviol glycosides
from dried leaves of Stevia rebaudiana are extracted with water
Table 1. Eluent composition in artemisinin HPLC analysis.
or methanol [28–36]. These two solvents are in the pre-selec-
Time [min] Acetonitrile [vol %] Water [vol %] tion list. The other solvents determined by the standard proce-
dure are rather unusual solvents for the extraction of rebau-
0 68 32 dioside A.
5 68 32
7 100 0
Table 2. Solubility ranking of pre-selected solvents for rebaudio-
16 100 0
side A from Stevia rebaudiana.
18 68 32
Solubility ranking Pre-selected solvent
20 68 32
1 propylene glycol
2 ethylene glycol
3 ethanol
4 Results and Discussion
4 water
The standard procedure was tested for three natural products: 5 methanol
the polar steviol glycoside rebaudioside A from Stevia rebaudi-
ana Bertoni leaves, the semi-polar caffeine from guaraná seeds, 6 1-propanol
and the semi-polar artemisinin from Artemisia annua. The fol- 7 2-propanol
lowing section shows step by step the results of the test and the
8 2-butanol
validation for the target compound rebaudioside A from Stevia
rebaudiana Bertoni leaves. Additionally, the results are com- 9 acetonitrile
pared to literature data. For caffeine and artemisinin, only the
10 nitromethane
final results of the standard procedure are discussed.

Chem. Eng. Technol. 2013, 36, No. 10, 1739–1748 © 2013 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim [Link]
1744 G. Schembecker et al.

After the solubility estimation, extraction experiments were and costs of extraction, processes can be realized with different
operated under standard conditions as described in Sect. 2. operation targets.
HPLC analysis of the experiments was carried out and the data The comparison of solubility estimation and experimental
were evaluated. After input of the amount of extract after sol- yield of extraction is summarized in Tab. 3. Although the
id-liquid separation, the concentrations of the target com- swelling behavior, solvent viscosity, surface tension, and cellu-
pound in the extracts, the peak area of the target compound, lar localization of the target compound have an impact on the
and the total area of the chromatogram, the standard proce- extraction efficiency, the solubility estimation using a combi-
dure determined the relative yields, relative selectivities, and nation of 1-octanol/water partition coefficient and Hildebrand
relative costs. For calculation of the relative costs, the recovery and Hansen parameters to describe the extraction efficiency
rate of the solvent was defined to be 95 %, i.e. 95 % of the sol- shows only a small difference.
vent can be recycled and 5 % of free liquid solvent plus the re-
sidual moisture needs to be purchased. For the solvent water, Table 3. Comparison of experimental yield and solubility estima-
no recycling was selected. Water has the highest extraction se- tion for rebaudioside A from Stevia rebaudiana.
lectivity for rebaudioside A, the lowest costs, and the second
Ranking position
best extraction yield. The highest extraction yield is reached
with ethylene glycol. The MPIi was determined for each of the Experimental Solubility Difference
ten target compound-solvent combinations with equal weight- yield estimation
ing factors for yield, selectivity, and costs. The result of the Ethylene glycol 1 2 1
MPIi comparison is shown in the MPI plot in Fig. 2.
Water 2 4 2
Methanol 3 5 2
Propylene glycol 4 1 3
Nitromethane 5 10 5
Ethanol 6 3 3
1-Propanol 7 6 1
2-Propanol 8 7 1
2-Butanol 9 8 1
Acetonitrile 10 9 1

On average, the ranking positions between experimental in-

vestigation and solubility estimation differ by only two posi-
tions. Especially the best solvents ethylene glycol, water, and
methanol are among the top five in the solubility ranking. This
means that the experimental investigation of ten solvents cov-
ers the inaccuracies between solubility estimation and extrac-
tion efficiency. In order to prove that the methodology pro-
poses the most promising solvents, extraction experiments
were performed with solvents from ranking positions
Figure 2. MPI plot of selected solvents for the target compound 11, 12, 13, 16, 20, 23, and 30 of the solubility ranking. The rela-
rebaudioside A. tive extraction yield was used to compare these excluded sol-
vents with the relative yields of the MPI solvents (Tab. 4).
Water has the highest MPI. Therefore, water is the most de- In all cases, the relative yield of extraction is lower than for
sirable solvent for the extraction of rebaudioside A from dried solvent number 10, which is indicated by a negative value (cf.
leaves of Stevia rebaudiana, as stated in the literature. The sol- Eq. 1). Additionally, the yield of extraction is far away from the
vent was determined by the estimation of solubilities and with best solvents with the highest yield. In summary, the procedure
minimal experimental effort, i.e. with ten experiments per- was confirmed for the extraction of rebaudioside A from Stevia
formed in triplicate, needing only about 3 man hours. The rebaudiana Bertoni leaves.
standard procedure has an additional advantage: For users The results of the standard procedure for the semi-polar tar-
who are only interested in a high yield, the standard procedure get compound caffeine from guaraná seeds are depicted in the
provides the rather unconventional bivalent alcohol ethylene MPI plot in Fig. 3. No limitations were selected for the solvents
glycol as extraction solvent. Ethylene glycol has never been because the area of application of caffeine is diverse. This
used in the literature for the extraction of rebaudioside A, but means that all solvents implemented in the solvents database
it has a higher extraction yield than water and methanol. Thus, can be used for further investigation.
solvents that would probably not have been considered are The highest relative yield could be obtained using acetone,
presented and, by the use of a combination of yield, selectivity, the highest relative selectivity was reached with dichlorometh-

[Link] © 2013 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chem. Eng. Technol. 2013, 36, No. 10, 1739–1748
 Solvent selection 1745

Table 4. Comparison of excluded solvents for rebaudioside A Table 5. Comparison of experimental yield and solubility estima-
from Stevia rebaudiana to MPI solvents. tion for caffeine from guaraná seeds.

Solubility estimation Rebaudioside A Ranking position

Solvent
Ranking position Yi [–] Experimental Solubility Difference
yield estimation
Ethylene glycol 1 1
Acetone 1 3 2
Acetonitrile 10 0
Tetrahydrofuran 2 2 0
Isobutanol 11 –0.06
Ethyl acetate 3 7 4
n-Butanol 12 –0.01
Pyridine 4 9 5
Acetone 13 –0.02
Cyclohexanone 5 1 4
Tetrahydrofuran 16 –0.01
Methyl acetate 6 5 1
Ethyl acetate 20 –0.05
Dichloromethane 7 4 3
Octanol 23 –0.01
Butanone 8 6 2
Amyl acetate 30 –0.07
Chloroform 9 10 1
Carbon disulfide 10 8 2

The best solvents, acetone and tetrahydrofuran, are listed

among the top three of the solubility ranking. For ethyl acetate
and pyridine, the difference is a bit higher, but these two sol-
vents are still selected as top-ten solvents. Once again, the pro-
cedure was validated by evaluating extraction experiments
with typical solvents and lower-ranking positions (Tab. 6).

Table 6. Comparison of excluded solvents for caffeine from guar-

aná seeds to MPI solvents.

Solvent Solubility estimation Caffeine

Ranking position Yi [–]

Acetone 1 1
Carbon disulfide 10 0
2-Propanol 18 –0.27
Acetonitrile 22 0.03
Ethanol 34 0.17
Figure 3. MPI plot of selected solvents for the target compound
caffeine from guaraná seeds. Water 40 –0.03

ane, and the lowest relative costs were obtained using cyclo-
hexanone. For the equally divided weighting factors yield, Ethanol and acetonitrile have higher relative yields than car-
selectivity, and costs, the highest MPI was determined for ace- bon disulfide. In this case, two solvents with higher relative
tone. In the literature, caffeine from guaraná seed is usually yield than ranking position 10 are excluded. The relative yield
obtained using supercritical carbon dioxide [37], ethyl acetate, of ranking positions 1–9 is still higher, and the performance is
or dichloromethane. Supercritical carbon dioxide is not liquid far away from the best solvent for extraction. In summary, the
and is therefore not considered. Ethyl acetate is selected by the standard procedure seems to be successful. The validation of a
methodology and ranked at position 2. In comparison to ace- semi-polar component has shown that, in spite of larger differ-
tone, ethyl acetate has a lower yield, higher solvent costs, and a ences between solubility estimation and extraction experi-
negligibly higher selectivity. Dichloromethane is selected as ments, solvents are selected similar to those in the literature
well and is assigned to the MPI ranking position 6 because it and the best solvent has a better extraction behavior than the
has a higher selectivity and similar costs but a significantly solvents used in the literature.
lower yield. The comparison of the solubility estimation and A second semi-polar natural product was investigated be-
experimental yield of extraction for caffeine from guaraná cause of the larger differences between solubility estimation
seeds is summarized in Tab. 5. and extraction experiments in the previous validation. The

Chem. Eng. Technol. 2013, 36, No. 10, 1739–1748 © 2013 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim [Link]
1746 G. Schembecker et al.

results of the standard procedure for the semi-polar target Table 7. Comparison of experimental yield and solubility estima-
compound artemisinin from Artemisia annua are depicted in tion for artemisinin from Artemisia annua.
the MPI plot in Fig. 4. Here, a decomposition temperature of
Ranking position
423 K was considered [38]. The solvents were limited to green
solvents to improve the environmental performance of the Experimental Solubility Difference
process. yield estimation

Amyl acetate 1 1 0
Toluene 2 3 1
Tetrahydrofuran 3 5 2
Methyl acetate 4 8 4
Ethyl acetate 5 6 1
Butanone 6 7 1
2-Butanol 7 10 3
1-Pentanol 8 4 4
Cyclohexane 9 9 0
Butyl acetate 10 2 8

The best solvents, amyl acetate, toluene, and tetrahydrofu-

ran, are listed among the top five of the solubility ranking, and
for validation of the prediction method, all tested solvents with
ranking position lower than 10 have a lower relative yield in
comparison to the MPI solvents that were selected by the stan-
dard procedure (Tab. 8). Once again, the standard procedure is
tested successfully.
Figure 4. MPI plot for selected solvents of the target compound
Table 8. Comparison of excluded solvents for artemisinin from
artemisinin from Artemisia annua.
leaves of Artemisia annua to MPI solvents.

The highest relative yield could be obtained using amyl ace- Solvent Solubility estimation Caffeine
tate, the highest relative selectivity was reached with cyclohex- Ranking position Yi [–]
ane, and the lowest relative costs were obtained using buta-
none. For the equally divided weighting factors yield, Toluene 1 1
selectivity, and costs, the highest MPI was determined for tolu- Butyl acetate 10 0
ene, although this solvent did not perform best in selectivity
Acetone 11 –0.32
and costs, but all three factors showed relatively high values.
Toluene is therefore the most desirable solvent for the extrac- n-Butanol 12 –0.37
tion of artemisinin from dried leaves of Artemisia annua. In
n-Hexane 15 –1.04
the literature, toluene, cyclohexane, ethyl acetate, butanone,
and chloroform are usually used for the extraction of artemisi- 2-Propanol 19 –0.4
nin [38, 39]. Four of these solvents can be found in the pre- Ethanol 26 –0.48
selection list and have high MPI values. Chloroform is not in
the pre-selection list because it was excluded by limiting the
solvents to green solvents. In case of “no limitation” in the area The theoretical part of the standard procedure (Step 1 to
of application, chloroform would reach the highest MPI. Step 4) was further investigated with regard to five natural
Furthermore, the standard procedure provides the solvent products from plant material, i.e. capsaicin from chili pepper,
amyl acetate, a rather unconventional solvent for artemisinin furaneol from strawberries, nicotine from Nicotiana tabacum,
extraction, which has never been used in the literature but eucalyptol from Eucalyptus globulus, and limonene from or-
reached the highest extraction yield. The standard procedure ange. A comparison between the literature data of the best ex-
selects solvents for artemisinin extraction corresponding to traction solvents for these target compounds and the solubility
those given in the literature and provides an unconventional estimation shows a good approximation because the best sol-
solvent with better extraction behavior. vents for extraction are always placed among the ten estimated
The comparison of solubility estimation and experimental solvents using the combination of the three estimation meth-
yield of extraction of artemisinin from Artemisia annua is ods: 1-octanol/water partition coefficient and Hildebrand and
summarized in Tab. 7. Hansen parameters.

[Link] © 2013 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chem. Eng. Technol. 2013, 36, No. 10, 1739–1748
 Solvent selection 1747

5 Conclusion Symbols used

A standard procedure for the systematic selection of solvents C [u L–1] costs
for natural product extraction was developed and tested C [–] relative costs
experimentally. The standard procedure is implemented in i [–] control variable of solvents 1–10
Microsoft® Excel™ and is a hybrid approach of solubility esti- MPI [–] mobile phase performance index
mation and experimental investigation. Therefore, it is not S [–] selectivity (areatarger/areatotal)
meant to replace existing methods for modeling and estima- C [–] relative selectivity
tion of solubility, which are certainly more precise. The main Y [gtarget gleaves–1] yield
goal is to reduce the solvent screening time in the early stages Y [–] relative yield
of process development when solvent extraction is still one of
many alternatives. The standard procedure provides the neces- Greek symbols
sary information from a solvent database, creates test plans,
a, b, c [–] weighting factors
automatically performs calculations and helps the user by
structured descriptions of every single step. Ten solvents are Indices
selected using a combination of the three solubility estimation
methods, the 1-octanol/water partition coefficient and the Hil- max maximum value
debrand and Hansen parameters, which in sum has shown to min maximum value
yield a good approximation between estimated values and
experimental results. Special extraction conditions are defined
where effects on extraction, like kinetics, temperature, plant References
particle size, and swelling behavior, are equal or can be ne-
glected. The most desirable solvent is determined by compar- [1] Industrial Scale Natural Products Extraction (Eds: H.-J. Bart,
ing the relative values of yield, selectivity, and costs for the ten S. Pilz), Wiley-VCH, Weinheim 2011.
pre-selected solvents in a short time (about 3 man hours) and [2] A. P. Jarvis, E. D. Morgan, Phytochem. Anal. 1997, 8 (5), 217.
with low material consumption. The standard procedure was DOI: 10.1002/(SICI)1099-1565(199709/10)8:5<217:AID-
tested for the polar component rebaudioside A from leaves of PCA366>[Link];2-J
Stevia rebaudiana Bertoni, for the semi-polar component caf- [3] G. Brunner, J. Food Eng. 2005, 67 (1/2), 21. DOI: 10.1016/
feine from guaraná seeds, and for the semi-polar component [Link].2004.05.060
artemisinin from leaves of Artemisia annua. In all cases, solvents [4] E. Lack, B. Simándi, Ind. Chem. Libr. 2001, 9 (C), 537.
were found that were predicted in the literature. Additionally, [5] E. Rytting, K. A. Lentz, X.-Q. Chen, F. Qian, S. Venkatesh,
solvents were selected that show even better extraction behavior Pharm. Res. 2004, 21 (2), 237.
regarding yield, selectivity, and costs, but were out of radar, mak- [6] J. Huuskonen, D. J. Livingstone, D. T. Manallack, SAR QSAR
ing an optimization and customizing of processes possible. Environ. Res. 2008, 19 (3/4), 191.
In subsequent research work, the standard procedure should [7] G. Klopman, H. Zhu, J. Chem. Inf. Comput. Sci. 2001, 41 (2),
be used for a special application, namely the selection of the 439. DOI: 10.1021/ci000152d
mobile phase for an inverted chromatographic process design. [8] J. S. Delaney, J. Chem. Inf. Comput. Sci. 2004, 44 (3), 1000.
“In many cases the crude extract is evaporated and residue is DOI: 10.1021/ci034243x
dissolved in an appropriate solvent for further chromato- [9] M. Kassing, U. Jenelten, J. Schenk, J. Strube, Chem. Eng.
graphic purification” [1]. In inverted chromatographic process Technol. 2010, 33 (3), 377. DOI: 10.1002/ceat.200900480
design, the mobile phase is optimized first and, in a second [10] T. C. Frank, J. R. Downey, S. K. Gupta, Chem. Eng. Prog.
step, the adsorbent is selected. This means that the solvent se- 1999, 95 (12), 41.
lected by the standard procedure should be used as mobile [11] J. Shi, G. Mazza, M. Le Maguer, Functional Foods: Biochem-
phase for chromatography. Hence, the common solvent change ical and Processing Aspects, Technomic Publishing, Lancaster
between extraction and chromatography is avoided. This leads 1998–2002.
to a higher loading capacity of the chromatographic column [12] Natural Products Isolation (Eds: S. D. Sarker, Z. Latif, A. I.
and therefore to a higher time-volume yield. Gray), 2nd Edn., Humana Press, New York 2005.
[13] A. Pfennig, Wissensbasierte Designmethode zur Auslegung von
maßgeschneiderten Feststoffextraktoren auf der Basis von La-
Acknowledgment borversuchen, AiF-Schlussbericht (16146N), RWTH Aachen,
Aachen 2012.
Special thanks are due to my colleague Björn Dreisewerd for [14] M. Kassing, U. Jenelten, J. Schenk, R. Hänsch, J. Strube,
artemisinin HPLC method development. The research was Chem. Eng. Technol. 2012, 35 (1), 109. DOI: 10.1002/ceat.
financially supported by the German Federal Ministry of 201100268
Research and Education (Förderkennziffer: 0315663) and [15] ACDLabs, ChemSketch Freeware, Advanced Chemistry Devel-
CLIB2021 – Cluster Industrial Biotechnology. opment, Inc., Toronto 2010.
[16] CAS, SciFinder, [Link] Chemical Abstracts
The authors have declared no conflict of interest. Service 2012.

Chem. Eng. Technol. 2013, 36, No. 10, 1739–1748 © 2013 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim [Link]
1748 G. Schembecker et al.

[17] Richtlinie 2009/32/EG des europäischen Parlaments und des nung (EG) Nr. 1333/2008 des Europäischen Parlaments und
Rates zur Angleichung der Rechtsvorschriften der Mitglied- des Rates hinsichtlich Steviolglycosiden (Text von Bedeutung
staaten über Extraktionslösungsmittel, die bei der Herstel- für den EWR): L 295/205, Amtsblatt der Europäischen
lung von Lebensmitteln und Lebensmittelzutaten verwendet Union, European Union, Bruxelles 2011.
warden, 2009/32/EG, Amtsblatt der Europäischen Union, [28] J. Pól, E. Varadová Ostrá, P. Karásek, M. Roth, K. Benešová,
European Union, Bruxelles 2009. P. Kotlaríková, J. Cáslavský, Anal. Bioanal. Chem. 2007, 388
[18] Verordnung über die Verwendung von Extraktionslösungs- (8), 1847. DOI: 10.1007/s00216-007-1404-y
mitteln und anderen technischen Hilfsstoffen bei der Her- [29] S. de Blauth Slavutzky, J. Verbr. Lebensm. 2010, 5 (2), 213.
stellung von Lebensmitteln (Technische Hilfsstoff-Verord- [30] M. Carakastos, L. Curry, A. Boileau, D. Brusick, Food Chem.
nung – THV), Bundesministerium der Justiz, Bonn 1991. Toxicol. 2008, 46 (7), S1. DOI: 10.1016/[Link].2008.05.003
[19] C. Capello, U. Fischer, K. Hungerbühler, Green Chem. 2007, [31] A. S. Dacome, C. C. Da Silva, C. E. M. Da Costa, J. D. Fonta-
9 (9), 927. DOI: 10.1039/b617536h na, J. Adelmann, S. C. Da Costa, Process Biochem. 2005, 40
[20] W. M. Meylan, P. H. Howard, J. Pharm. Sci. 1995, 84 (1), 83. (11), 3587. DOI: 10.1016/[Link].2005.03.035
DOI: 10.1002/jps.2600840120 [32] W. Fuh, B. Chiang, J. Food Sci. 1990, 55 (5), 1454. DOI:
[21] D. W. van Krevelen, Properties of Polymers: Their Correlation 10.1111/j.1365-2621.1990.tb03956.x
with Chemical Structure, Their Numerical Estimation and Pre- [33] Y. Kitada, M. Sasaki, Y. Yamazoe, H. Nakazawa, J. Chroma-
diction from Additive Group Contributions, 3rd Edn., Elsevier, togr. 1989, 474 (2), 447.
Amsterdam 1992. [34] G. I. Kovylyaeva, G. A. Bakaleinik, I. Y. Strobykina, V. I.
[22] C. M. Hansen, Hansen Solubility Parameters: A User’s Hand- Gubskaya, R. R. Sharipova, V. A. Al’Fonsov, V. E. Kataev, A.
book, CRC Press, Boca Raton 2000. G. Tolstikov, Chem. Nat. Compd. 2007, 43 (1), 81. DOI:
[23] M. Bonakdar, Ph.D. Thesis, Universität Erlangen-Nürnberg 10.1007/s10600-007-0037-x
1996. [35] J. Liu, S. F. Y. Li, J. Liq. Chromatogr. 1995, 18 (9), 1703.
[24] H. D. Baehr, K. Stephan, Wärme- und Stoffübertragung, 5th [36] A. Esmat Abou-Arab, A. Azza Abou-Arab, M. Ferial Abu-
Edn., Springer-Verlag, Heidelberg 2006. Salem, Afr. J. Food Sci. 2010, 4 (5), 269.
[25] D. Bergs, B. Burghoff, M. Joehnck, G. Martin, G. Schem- [37] C. B. Mehr, R. N. Biswal, J. L. Collins, J. Supercrit. Fluids
becker, J. Verbr. Lebensm. 2012, 7 (2), 147. DOI: 10.1007/ 1996, 9 (3), 185. DOI: 10.1016/S0896-8446(96)90031-5
s00003-012-0760-5 [38] A. A. Lapkin, P. K. Plucinski, M. Cutler, J. Nat. Prod. 2006,
[26] G. Kroyer, J. Verbr. Lebensm. 2010, 5 (2), 225. DOI: 10.1007/ 69 (11), 1653. DOI: 10.1021/np060375j
s00003-010-0557-3 [39] J. Nti-Gyabaah, K. Gbewonyo, Y. C. Chiew, J. Chem. Eng.
[27] Verordnung (EU) Nr. 1131/2011 der Kommission vom Data 2010, 55 (9), 3356. DOI: 10.1021/je100125x
11. November 2011 zur Änderung von Anhang II der Verord-

[Link] © 2013 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chem. Eng. Technol. 2013, 36, No. 10, 1739–1748