Modeling the Solubility of Pharmaceuticals in Pure Solvents

and Solvent Mixtures for Drug Process Design

FEELLY RUETHER, GABRIELE SADOWSKI

Laboratory of Thermodynamics, Technische Universität Dortmund, Emil-Figge-Str. 70, D-44227 Dortmund, Germany

Received 16 October 2008; revised 6 January 2009; accepted 22 January 2009

Published online 12 March 2009 in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/jps.21725

ABSTRACT: The knowledge of the solubility of pharmaceuticals in pure solvents and

solvent mixtures is crucial for designing the crystallization process of drug substances.
The first step in finding optimal crystallization conditions is usually a solvent screening.
Since experiments are very time consuming, a model which allows for solubility pre-
dictions in pure solvents and solvent mixtures based only on a small amount of
experimental data is required. In this work, we investigated the applicability of the
thermodynamic model perturbed-chain statistical associating fluid theory (PC-SAFT) to
correlate and to predict the solubility of exemplary five typical drug substances
and intermediates (paracetamol, ibuprofen, sulfadiazine, p-hydroxyphenylacetic acid,
and p-aminophenylacetic acid) in pure solvents and solvent mixtures. ß 2009 Wiley-Liss,
Inc. and the American Pharmacists Association J Pharm Sci 98:4205–4215, 2009
Keywords: solubility; thermodynamics; crystallization; pharmaceuticals; PC-SAFT

INTRODUCTION as functions of different group contribu-

The solubility is probably the most important
property that is needed to be known for the
production and purification of pharmaceuticals,
for example, by crystallization. Most of the
approaches to estimate the solubility of drug
substances reported in literature can basically be
divided into two types:
tions,3 and molecular structures.4
(2) The solubility is related to the chemical
potentials of the drug substance in the solid
and liquid phase. The chemical potentials
are commonly expressed by the so-called
activity coefficients which can be calculated
via thermodynamic models (e.g., NRTL-
SAC5,6 and COSMO-RS7).

(1) Empirical/semi-empirical correlations for

the solubility, where the model coefficients
are considered either as functions of physi-
cal drug and solvent properties such as
polarity, for example, within the quantita-
tive structure property relationship (QSPR)
method, as reported among others by
Rytting et al.1 and Huuskonen et al.,2 or
Correspondence to: Gabriele Sadowski (Telephone: þ49-
231-755-2635; Fax: þ49-31-755-2572;
E-mail: gabriele.sadowski@bci.tu-dortmund.de)
Journal of Pharmaceutical Sciences, Vol. 98, 4205–4215 (2009)
ß 2009 Wiley-Liss, Inc. and the American Pharmacists Association
The first method requires rather a large number
of experimental solubility data to fit the respective
model coefficients. Hence, its reliability mostly
depends on the number of experimental data
available to correlate the coefficients. In contrast,
the second approach requires in general only a
much smaller set of experimental data to identify
the model parameters and to calculate solubilities
afterwards. The main reason is that the
solvents and solvent mixtures are characterized
by model parameters which were already identi-
fied independently, and only the solute–solvent
interactions need to be determined from solubility
data.

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 11, NOVEMBER 2009 4205
4206 RUETHER AND SADOWSKI
In this work, we apply the thermodynamic
model perturbed-chain statistical associating
fluid theory (PC-SAFT) equation of state8 to
correlate and predict the solubility of typical
pharmaceuticals and intermediates in different
solvents and solvent mixtures. ‘‘Correlation’’ in
this work means that a few numbers of experi-
mental data are used to identify the model
parameters and to give a quantitative representa-
tion of the solubility data, whereas the term
‘‘prediction’’ is applied when the solubility curve of
a pharmaceutical in a solvent or solvent mixture is
calculated without using any experimental data of
the considered system.
THEORY
Thermodynamic Solubility Calculation
Based on the thermodynamic phase equilibrium
condition for solid and liquid phases, that is,
equality of the chemical potentials in both solid
and liquid phases, the solubility can be calculated
as:
" !#
L 1 DhSL
0i T
xi ¼ L exp  1  SL (1)
gi RT T0i
Here, xLi and g Li represent the mole fraction and
the activity coefficient of drug substance i in the
liquid phase (L), which consists of the dissolved
drug compound and the solvent. DhSL SL
0i and T0i are
the enthalpy of fusion and the melting tempera-
ture of the pure drug substance i, respectively,
whereas R is the so-called gas constant and T is
the temperature.
From Eq. (1) it can be seen that the solubility of
a solute—besides its pure-component properties
(enthalpy of fusion and melting temperature)—
only depends on its activity coefficient in the
solution. Therefore, the activity coefficient is the
only quantity in the equation that influences the
varying solubility of one solute in different
solvents or solvent mixtures.
From the thermodynamic point of view, the
activity coefficient reflects the extent to what the
molecular structure between solute and solvent as
well as the interactions between the solute–solute
and solvent–solvent molecules are different from
each other. For unlike molecules the values of the
activity coefficients differ from one. In case of
pharmaceutical/solvent systems, the molecules
are usually very different from each other with
respect to volume, polarity, etc. Thus, the activity
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 11, NOVEMBER 2009
coefficients in the mixtures are in most cases not
unity and hence cannot be neglected in the
solubility calculations.
If the enthalpy of fusion and the melting
temperature of the solute are not available, it is
also possible to use the entropy of fusion instead,
which in some cases can be estimated from
the solute’s molecular structure.9 The melting
temperature can then be treated as adjustable
parameter and has to be adjusted to the solubility
data. In this case, Eq. (1) becomes:
" !#
SL
1 Ds T
xLi ¼ L exp  0i 1  SL (2)
gi R T0i
For solutes which decompose before or during
melting (e.g., amino acids), both the melting
temperature and the enthalpy of fusion can be
treated as adjustable parameters.10
When dealing with polymorphs, Eq. (1) can also
be applied by using different enthalpies of fusion
and melting temperatures for the different poly-
morphs.
Activity Coefficient Estimation
The activity coefficients of a drug substance i (g Li )
in the mixture with one or more solvents, are
estimated via the PC-SAFT model8 in this work.
Using this model, the residual Helmholtz energy
Aresidual of a system containing drug substance
and solvents can be considered as the sum of
different contributions resulting from repulsion
(hard chain), van der Waals attractions (disper-
sion) and hydrogen bonding (association) accord-
ing to:
Aresidual ¼ Ahard chain þ Adispersion þ Aassociation (3)
The residual Helmholtz energy of a system is a
function of temperature, pressure, concentration,
and density. The detailed expressions of the
various Helmholtz-energy contributions are given
elsewhere8 in detail. The activity coefficients are
related to the partial derivatives of the residual
Helmholtz energy with respect to the concentra-
tion (see Appendix).
To visualize the model, the underlying mole-
cular picture of paracetamol within the PC-SAFT
framework is exemplary depicted in Figure 1 as an
example. The molecules are assumed to be
composed of freely jointed spherical segments
which form a repulsive chain (hard chain). In
addition to van der Waals (dispersion) interac-
tions, the segments also exhibit association sites
DOI 10.1002/jps
 MODELING THE SOLUBILITY OF PHARMACEUTICALS
Figure 1. Molecular modeling of paracetamol within
the framework of PC-SAFT.
to account for strong short-ranged interactions
(hydrogen bonding).
Knowing the molecular structure of a drug
substance, one can choose the terms that have to
be included in the calculation of the Helmholtz
energy. Nonpolar or weakly polar components can
be satisfactorily described using only the hard-
chain and dispersion contributions. In this case,
three pure-component parameters are required
to characterize the drug: segment number m,
segment diameter s, and the dispersion-energy
parameter e. For the associating drug compounds
considered in this work such as paracetamol (see
Fig. 1), in addition two different types of associa-
tion sites are assumed (electron acceptor and
electron donor), each of them existing twice per
molecule. The two association types were assumed
to be of equal bonding strength and range. Thus,
only one value for the association energy ("Ai Bi )
and one value for the association volume (kAi Bi )
parameter need to be determined which are used
to calculate the Helmholtz-energy contribution
Aassociation.
In order to calculate thermodynamic properties
of mixtures, the conventional combining rules for
the hard-chain and dispersion term are applied.
Only one binary parameter kij is introduced
correcting the dispersion-energy parameter for
the mixture according to:
1 have less influence on the results of the solubility
s ij ¼ ðs i þ s j Þ (4)
2
pffiffiffiffiffiffiffiffi
"ij ¼ ð1  kij Þ "i "j (5)
For cross-associating systems, for example,
paracetamol and water, the strength of the
cross-associating interactions between the two
associating substances can be described by
applying simple combining rules as suggested
by Wolbach and Sandler11 without using any
DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 11, NOVEMBER 2009
4207
additional adjustable correction parameters:
1
"Ai Bj ¼ ð"Ai Bi þ "Aj Bj Þ (6)
2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi
si sj
3
kAi Bj ¼ kAi Bi kAj Bj (7)
ð1=2Þðs i þ s j Þ
RESULTS AND DISCUSSION
Estimation of PC-SAFT Parameters
Pure-component parameters for more than
130 solvents and a huge number of binary
parameters to describe the solvent–solvent inter-
actions were already determined in previous
articles.8,12 Hence, only the pure-component
parameters for the drug compound and the
drug–solvent binary interaction parameters
needed to be estimated here.
Generally, equation-of-state parameters are
best fitted to the liquid densities and vapor
pressures of the pure components. However, at
normal conditions most of the pharmaceuticals
exist only as solids and often no experimental
solution densities data are available. Hence, in
this work another approach to determine the
pure-component parameters for drug substances
is proposed and applied: the parameters are
determined by fitting them to a few number of
solubility data in one pure solvent.
Consequently, the required amount of experi-
mental solubility data depends on the number of
parameters (three or five pure-component para-
meters) which have to be adjusted. As binary
solubility data are used for determining the solute
parameters, the kij between drug and solvent has
also to be estimated simultaneously. Compounds
which exhibit OH-groups should be modeled as
associating compounds with at least two associa-
tion-site types, each having one or two sites.
The association volume (kAi Bi ) turned out to
calculations. Therefore, the parameter kAi Bi can be
excluded from the parameter estimation to reduce
the number of adjustable parameters. We set kAi Bi
to a value between 0.01 and 0.03, which are
common values for association volumes for most
organic substances.
Moreover, the enthalpy of fusion and melting
temperature need to be known as indicated in
Eq. (1). For paracetamol as an example, the
experimental enthalpy of fusion and melting
4208 RUETHER AND SADOWSKI

Estimation
Solvent for
Parameter
temperature are reported in literature13 and used

Acetone

Acetone
Water
Water

Water
in this work for solubility calculations.

—
—
—
—
—
—
—
—
As paracetamol is modeled as an associating
compound according to Figure 1, a total number of
five pure-component parameters have to be
fitted plus one binary parameter kij between

Volume, kAi Bj
Association
paracetamol and the solvent. Because para-

0.035
0.032

0.025
0.01
0.01
0.03
0.03
0.03

0.01

0.01
0.01

0.01
cetamol’s association volume was set to a constant

—
value of 0.01 (Tab. 1), consequently at least five
solubility data points of paracetamol at different
temperatures in one solvent are required to
identify the model parameters.

(Association), "Ai Bj (K)

If the solubility data in one solvent are not

Energy Parameter
accessible at several temperatures, one can also
use solubility data of the solute in several

1994.228

1954.408
1690.790

2500.671
2653.384

2544.560
879.415

2253.9
1000
solvents. In order to reduce the number of

—
Table 1. Pure Component PC-SAFT Parameters for Studied Drug Substances, Intermediates and Solvents

0
0
parameters to be fitted, we suggest dividing the
solvents according to their molecular structures
and physical properties into certain classes, for
example, nonpolar, weakly polar, strongly polar,
and associating components. The kij values
between one solute and all solvents belonging

Energy Parameter
(Dispersion), e (K)
to the same class can then be assumed to be
identical.

398.284

374.651
294.627
258.182

366.512
198.237
247.418
285.690
265.041
230.800
208.420
259.591
The summary of the so-obtained pure-

200
component parameters of five well-known drug
substances and intermediates is presented in
Table 1. The information which solvent was used
for the determination of the pure-component
parameters for each drug substance is also given
Diameter, s (Å)

in the table. For the sake of completeness, the

pure-component parameters for the solvents
Segment

3.508
2.914
4.432
3.635
3.544

3.001
3.177
3.228
3.717
3.292
3.308
3.209
3.614
are given in Table 1 as well. Table 2 lists the
determined kij parameter for each of the consid-
ered drug–solvent systems.
Number, m

Solubility of Paracetamol in Pure Solvents

Segment

14.029
7.524

2.522
3.321
2.234

1.066
2.383
2.891
2.815
3.212
3.537
3.093
2.752
Experimental values of the enthalpy of fusion
(27 kJ/mol) and melting temperature of 443.6 K
from Granberg and Rasmuson13 were used for
the solubility calculation of paracetamol in
pure solvents. The pure-component parameters
p-Hydroxyphenylacetic acid
p-Aminophenylacetic acid

of paracetamol—except for the association

volume which was set to 0.01—and the binary
parameter kij for paracetamol/water were fitted
to the experimental solubility data in water.13
Pharmaceuticals

Figure 2 shows the result of the modeling (line)

Ethyl acetate
Paracetamol
Sulfadiazine

1,4-Dioxane

compared to experimental data (symbols).

2-Propanol
1-Butanol
Ibuprofen
Compound

It turned out that by introducing a linear-

Toluene
Ethanol
Acetone
Solvents
Water

temperature dependency of the binary parameter

kij between one solute and one solvent, the
calculation results could be brought into a better

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 11, NOVEMBER 2009 DOI 10.1002/jps
 MODELING THE SOLUBILITY OF PHARMACEUTICALS 4209

Table 2. Binary Parameters for Pharmaceutical/Solvent and Solvent/Solvent

Interactions

System Binary Parameter, kij

Pharmaceutical/solvent
Paracetamol–water 0.00028276T (K)  0.10072067
Paracetamol–ethanol 0.00021306T (K)  0.13203682
Paracetamol–acetone 0.00026365T (K)  0.13297859
Paracetamol–toluene 0.00065331T (K)  0.24063338
Ibuprofen–acetone 0.000312T (K)  0.0665428
Ibuprofen–ethanol 0.00030433T (K) þ 0.14584507
Sulfadiazine–water 0.00003827T (K)  0.02844995
Sulfadiazine–1,4-dioxane 0.0383 (at 258C)
p-Hydroxyphenylacetic acid–water 0.00003679T (K)  0.10958812
p-Aminophenylacetic acid–acetone 0.00073743T (K)  0.42128367
Solvent/solvent
Water–acetone 0.055
Water–ethanol 0.02
Water–1,4-dioxane 0.01

agreement with the experimental data. The PC-
SAFT model is able to correlate the solubility of
paracetamol in water; the results show a very
good agreement (average absolute deviation for
the temperature range ¼ 1.63%) with experimen-
tal data over the temperature range from 0 to
308C.
Now that the pure-component parameters for
paracetamol are defined, one can principally
estimate the solubility of paracetamol in different
solvents. For this purpose, the only parameters
that are still unknown are the binary parameters
kij between paracetamol and the respective
solvent. If kij would be set to zero, then the
calculation would be pure prediction. But, for
sufficient quantitative good calculation results
the binary parameter kij between two unlike
molecules has to be fitted to at least one
experimental point in most cases.
As mentioned above, to reduce the number of
fitted parameters we suggest dividing the solvents
into different classes and using the same kij
between one solute and all solvents belonging to
the same class. Here, we chose three representa-
tive solvents: ethanol to represent the associating
solvents, acetone for weakly polar solvents, and
toluene for aromatic solvents. The binary para-
meters kij between paracetamol and each solvent
were subsequently fitted to only two experimental
solubility points (0 and 308C). The results of the
correlations are compared to experimental data
at other temperatures as shown in Figure 3.
The PC-SAFT model can again remarkably well
reproduce the temperature dependence of the
paracetamol solubility in these different solvents
for the considered temperature range.
The use of a linear temperature-dependent
binary parameter kij in our work can be justified
for two reasons:

(a) The thermodynamic relationship to calcu-

Figure 2. Solubility of paracetamol in water. Com-
parison of PC-SAFT correlation (solid line) and experi-
mental data13 (symbols).
late the solubility according to Eq. (1)
neglects the influence of the heat capacity
which is temperature dependent but experi-
mentally in most of the cases not available,
that is, the kij would be expected to be
temperature dependent to correct for this
assumption.

DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 11, NOVEMBER 2009
4210 RUETHER AND SADOWSKI
Figure 3. Solubility of paracetamol in different sol-
vents. Comparison of PC-SAFT correlations (solid lines)
and experimental data13 (symbols).
(b) The determination of the linear tempera-
ture function of kij requires only two data
points. The linear function can afterwards
be adopted for a safe extrapolation to other
temperatures which is demonstrated here
by comparing the modeling and experimen-
tal results, for example, in Figure 3.
We compared the correlation results of PC-
SAFT with those of NRTL-SAC,5 which is also a
thermodynamic model to calculate the activity
coefficients (Fig. 4) using the parameters as
reported in literature.5 It is worth mentioning,
Figure 4. Comparison of correlation results of PC-
SAFT (solid lines) and NRTL-SAC (dashed lines) for the
solubility of paracetamol in different solvents. Symbols
are experimental data.13
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 11, NOVEMBER 2009
that NRTL-SAC also requires some experimental
solubility data to identify the model parameters.
In the case of paracetamol, the same database
from literature13 for parameter fitting was used
for both models PC-SAFT and NRTL-SAC. For the
NRTL-SAC modeling, the experimental enthalpy
of fusion (26 kJ/mol) from Manzo and Ahumada14
was used, whereas the molecular parameters for
paracetamol as well as the melting temperature
were fitted to the solubility data of paracetamol in
eight solvents (water, ethanol, DMSO, acetone,
acetonitrile, THF, chloroform, and toluene) at
308C. The value of the fitted melting temperature
by NRTL-SAC was determined to be 401.2 K. It
can be seen, that NTRL-SAC model is not able to
quantitatively capture the temperature depen-
dence of the paracetamol solubility in different
solvents. In particular, the solubility of paraceta-
mol in ethanol correlated by NRTL-SAC is twice
as high as the experimental data, although
exactly this solubility value was also used for
fitting of paracetamol NRTL-SAC parameters.
To demonstrate the predictive capability of the
PC-SAFT model, the solubilities of paracetamol in
other pure alcohols (2-propanol and 1-butanol) are
subsequently predicted using the paracetamol
PC-SAFT parameters and the kij-value for para-
cetamol/ethanol. The latter means that all alco-
hols are considered here to belong to the same
solvent class, and the binary parameter kij
between paracetamol and different alcohols was
set to the kij-value for paracetamol/ethanol.
Hence, the solubility of paracetamol in 2-propanol
and 1-butanol calculated this way was purely
predicted, and the results are presented in
Figure 5. Moreover, the PC-SAFT results (solid
lines) are also compared to the results from NRTL-
SAC (dashed lines) and to experimental data
(symbols). It becomes obvious that (I) both models
can give qualitative predictions and (II) the
performance of the models is more or less of the
same quality. Although not in quantitative
agreement, these results are accurate enough
for solvent-screening purposes in the process
development. They clearly indicate that 2-propa-
nol is the better solvent for paracetamol when
compared to 1-butanol. Thus, by applying the PC-
SAFT model, the number of necessary screening
experiments can be significantly reduced.
Solubility of Other Pharmaceuticals in Pure Solvents
To point out the general ability of the PC-SAFT
model to calculate and to predict the solubility of
DOI 10.1002/jps
 MODELING THE SOLUBILITY OF PHARMACEUTICALS 4211

Figure 5. Prediction of paracetamol solubility in Figure 6. Solubility of ibuprofen in acetone and ethyl
2-propanol and 1-butanol (symbols: experimental acetate. Symbols are experimental data.15 Lines are
data;13 — PC-SAFT, - - - NRTL-SAC). correlation (ibuprofen–acetone) and prediction (ibupro-
fen–ethyl acetate) results using PC-SAFT.

drug substances, the results for other pharma- this, the kij between ibuprofen/ethanol was fitted
ceuticals and intermediates (ibuprofen, sulfadia- to two solubility points of ibuprofen in ethanol (at
zine, p-hydroxyphenylacetic acid, and p- 10 and 358C). The resulting kij-value was subse-
aminophenylacetic acid) are also presented in quently used to predict the solubility of ibuprofen
this chapter. in 2-propanol. Figure 7 shows the results for the
The PC-SAFT pure-component parameters for correlation (ibuprofen–ethanol) and the predic-
ibuprofen as well as the kij-value for ibuprofen/ tion (ibuprofen–2-propanol) compared to experi-
acetone were fitted to the solubility data of mental data. Again, using PC-SAFT a good
ibuprofen in acetone. The experimental enthalpy quality of the predicted solubility of ibuprofen in
of fusion and melting temperature were taken 2-propanol can be achieved.
from Gracin and Rasmuson.15 For sulfadiazine, p-hydroxyphenylacetic acid,
Like paracetamol, ibuprofen is modeled as an and p-aminophenylacetic acid the pure-component
associating component with two different asso-
ciation-site types, each having two associating
sites. Furthermore, the kij-value of ibuprofen/
acetone was subsequently used to predict the
solubility of ibuprofen in ethyl acetate, that is, (I)
acetone and ethyl acetate were considered to be in
the same solvent class (weakly polar) and (II) no
solubility data for ibuprofen in ethyl acetate was
used for this calculation. The results of the
correlation (ibuprofen–acetone) and the predic-
tion (ibuprofen–ethyl acetate) are presented in
Figure 6.
It can be seen, that the results are again in good
agreement with the experimental data. The
solubility of ibuprofen in acetone can be correlated
very precisely when compared to experimental
data. Without any experimental information, the
solubility of ibuprofen in ethyl acetate can be Figure 7. Solubility of ibuprofen in ethanol and 2-
purely predicted with sufficient accuracy. propanol. Symbols are experimental data.15 Lines are
This method can also be applied to other solvent correlation (ibuprofen–ethanol) and prediction (ibupro-
classes, for example, alcohols. To demonstrate fen–2-propanol) results using PC-SAFT.

DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 11, NOVEMBER 2009
4212 RUETHER AND SADOWSKI
parameters are fitted to the solubility data in water
(sulfadiazine16 and p-hydroxyphenylacetic acid15)
and acetone ( p-aminophenylacetic acid15), respec-
tively. Experimental enthalpies of fusion and
melting temperatures from Sunwoo and Eisen17
for sulfadiazine and from Gracin and Rasmuson15
for p-hydroxyphenylacetic acid as well as for
p-aminophenylacetic acid were used.
The results of the solubility correlations using
PC-SAFT are shown in Figure 8. Again, the
temperature dependence of the experimental data
can be reproduced by the PC-SAFT model in a
very good agreement with the experimental data.
The determined model parameters for the
respective drug compounds can also be applied
to other solvents in the same solvent class. The
performance of the prediction results is more or
less of the same quality as for paracetamol and
ibuprofen, respectively.
Prediction of the Solubility in Solvent Mixtures
To increase the solubility of the pharmaceuticals,
often solvent mixtures or cosolvents are used.
However, it is practically impossible to scan
all possible solvent mixtures and mixture compo-
sitions experimentally. Therefore, a thermody-
namic model is needed to give qualitative
information on the solubility in solvent mixtures
based only on the experimental solubility data in
the pure solvents only.
Figure 8. Solubility of sulfadiazine and p-hydroxy-
phenylacetic acid in water as well as the solubility of p-
aminophenylacetic acid in acetone. Symbols are experi-
mental data.15,16 Lines are correlation results using PC-
SAFT.
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 11, NOVEMBER 2009
Figure 9. Solubility of paracetamol in acetone–water
mixtures at 258C. Experimental data (symbols).18 Line
is the PC-SAFT prediction.
The effect of the cosolvent on the solubility can
show an unexpected behavior, as for example,
shown for the solubility of paracetamol in water–
acetone mixtures of various compositions (Fig. 9,
symbols). The x-axis characterizes the amount of
water in the water–acetone mixtures on a solute-
free basis. The y-axis represents the solubility of
paracetamol in the water–acetone mixtures of
varying composition ranging from pure acetone to
pure water. Although water is the worse solvent
compared to acetone, the solubility of paracetamol
in acetone can be almost quadrupled just by
adding a small amount of water to the solution.
One could have expected that water would act like
an anti-solvent for paracetamol if added to acetone
(decreasing solubility), but the opposite behavior
is observed experimentally.
Using only the pure-component as well as
binary parameters determined from the solubility
in the pure solvents in the previous section, the
solubility of pharmaceutical compounds in differ-
ent solvent mixtures is subsequently predicted
without fitting any additional parameters. This
means that no extra experimental data is required
to predict the solubility in the solvent mixtures in
this work. The kij-values between the different
solvents were mostly already determined inde-
pendently in the previous articles by fitting to
binary vapor–liquid or liquid–liquid equilibria of
the solvent mixtures, respectively.
Figure 9 also shows the predicted solubility of
paracetamol in acetone–water mixtures at 258C
(solid line). As it can be seen, the predicted curve
DOI 10.1002/jps

gives a good description of the nonlinear depen-
dency of the solubility on the solvent composition
based only on the knowledge of the solubility of
paracetamol in pure water and pure acetone,
respectively.
Similar results were obtained for paracetamol
in other mixed-solvent systems, for example, in
water and ethanol which are shown in Figure 10.
The predicted solubility is in fair agreement with
the experimental data without fitting any addi-
tional parameters.
Even the very low and again nonmonotonous
solubility of sulfadiazine in water–1,4-dioxane
mixtures at 258C can be predicted reliably well
using only the experimental data of sulfadiazine
solubility in pure water and pure 1,4-dioxane,
respectively, (Fig. 11).
All the prediction results of the solubility in
solvent mixtures show that the extrapolation from
the solubility in pure solvents to the solubility in
solvent mixtures always gives at least qualita-
tively reliable estimations. That means that
the experimental effort to find potential solvent
mixtures which can enhance the solubility
compared to that in the pure solvents can be
minimized tremendously by applying an appro-
priate thermodynamic model.
CONCLUSION
The solubility of several drugs and intermediates
in different pure solvents as well as in various
solvent mixtures was modeled and predicted
Figure 10. Solubility of paracetamol in various
water–ethanol mixtures at 258C. Experimental data
(symbols).19 Line is PC-SAFT prediction.
DOI 10.1002/jps
MODELING THE SOLUBILITY OF PHARMACEUTICALS 4213
Figure 11. Solubility of sulfadiazine in water–1,4-
dioxane mixtures at 258C. Experimental data (symbols)
by.20 Line is PC-SAFT prediction.
using the PC-SAFT model. In the calculations,
the independently determined experimental
melting temperature and enthalpy of fusion were
used. To identify the PC-SAFT parameters for
the pharmaceuticals, only a few solubility data in
one pure solvent are required.
For an adequate quantitative estimation of the
solute’s solubility in other solvents, it is recom-
mended to divide the solvents into certain classes
according to their molecular structures and
physical properties and to measure two solubility
points for one solvent in each solvent class.
It turned out, that the model allows for an
accurate description of the temperature depen-
dence of the solubility by applying a linear
temperature-dependent interaction parameter
for each binary system. The solubility in different
solvents in the same solvent class can be
estimated without any extra effort of experiments.
Based on the modeling results of the solubility
in pure solvents, the solubility in mixed solvents
can subsequently be reliably predicted without
any need of additional fitting of parameters, hence
without extra experiments. The prediction results
of solubility in solvent mixtures are in fair to very
good agreement with the experimental data for all
considered solvent mixtures and solvent–mixture
compositions.
NOMENCLATURE
A Helmholtz energy (J)
h enthalpy (J/mol)
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 11, NOVEMBER 2009
4214 RUETHER AND SADOWSKI

kij binary interaction parameter and temperature T:

L liquid phase pv
m segment number Z (A.3)
kNAV T
R gas constant (J/(mol K))
T temperature (8C, K) The residual chemical potential is the partial
x mole fraction derivation of the residual Helmholtz free energy
Ares divided with number of molecules N with
respect to the concentration xi at constant
temperature and volume and the concentrations
of other components in the system according to:
Greek Symbols
 
s segment diameter mres
i ðT; vÞ Ares @ðAres =NkTÞ
e energy parameter, dispersion (K) ¼ þZ1þ
kT NkT @xi T;v;xk6¼i
"Ai Bj energy parameter, association (K) "   #
kAi Bj association volume X @ðAres =NkTÞ
 xj
g activity coefficient j
@xj T;v;xk6¼j

(A.4)

With Eqs. (A.2) and (A.4), the fugacity coefficient

Subscript can finally be calculated as:
 
i, j substance i, j Ares @ðAres =NkTÞ
ln ’i ¼ þ
NkT @xi T;v;xk6¼i
"   #
X res
@ðA =NkTÞ (A.5)
 xj
j
@x j T;v;xk6¼j
Superscripts
þ Z  1  ln Z
L liquid phase
SL melting/phase change S ! L

REFERENCES
APPENDIX 1. Rytting E, Lentz KA, Chen XQ, Qian F, Venkatesh
S. 2004. A quantitative structure-property relation-
This appendix lists the equations to calculate the ship for predicting drug solubility in PEG 400/water
activity coefficient from the residual Helmholtz cosolvent systems. Pharm Res 21:237–244.
free energy: 2. Huuskonen J, Livingstone DJ, Manallack DT. 2008.
The activity coefficient is defined as a quotient Prediction of drug solubility from molecular struc-
of the so-called fugacity coefficient in the mixture ture using a drug-like training set. SAR QSAR
wi to the fugacity coefficient of pure component w0i: Environ Res 19:191–212.
3. Klopman G, Zhu H. 2001. Estimation of the aqueous
’i solubility of organic molecules by the group contri-
gi  (A.1)
’0i bution approach. J Chem Inf Comput Sci 41:439–
445.
The fugacity coefficient in the mixture is related 4. Delaney JS. 2004. ESOL: Estimating aqueous solu-
bility directly from molecular structure. J Chem Inf
to the residual chemical potential mres i , the Comput Sci 44:1000–1005.
Boltzmann’s constant k, temperature T, and to
5. Chen CC, Crafts PA. 2006. Correlation and predic-
the compressibility factor Z according to: tion of drug molecule solubility in mixed solvent
systems with the nonrandom two-liquid segment
mres
i ðT; nÞ activity coefficient (NRTL-SAC) model. Ind Eng
ln ’i ¼  ln Z (A.2)
kT Chem Res 45:4816–4824.
6. Tung HH, Tabora J, Variankaval N, Bakken D,
The compressibility factor Z is defined as a Chen CC. 2008. Prediction of pharmaceutical
function of system pressure p, molar volume v, solubility via NRTL-SAC and COSMO-SAC.
Boltzmann constant k, Avogadro number NAV, J Pharm Sci 97:1813–1820.

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 11, NOVEMBER 2009 DOI 10.1002/jps
 MODELING THE SOLUBILITY OF PHARMACEUTICALS
7. Klamt A, Eckert F, Hornig M, Beck ME, Burger T.
2002. Prediction of aqueous solubility of drugs and
pesticides with COSMO-RS. J Comput Chem 23:
275–281.
8. Gross J, Sadowski G. 2001. Perturbed-chain SAFT:
An equation of state based on a perturbation theory
for chain molecules. Ind Eng Chem Res 40:1244–
1260.
9. Dannenfelser RM, Yalkowsky SH. 1996. Estimation
of entropy of melting from molecular structure: A
non-group contribution method. Ind Eng Chem Res
35:1483–1486.
10. Fuchs D, Fischer J, Tumakaka F, Sadowski G.
2006. Solubility of amino acids: Influence of the
pH value and the addition of alcoholic cosolvents
on aqueous solubility. Ind Eng Chem Res 45:6578–
6584.
11. Wolbach JP, Sandler SI. 1998. Using molecular
orbital calculations to describe the phase behavior
of cross-associating mixtures. Ind Eng Chem Res
37:2917–2928.
12. Kleiner M, Sadowski G. 2007. Modeling of
polar systems using PCP-SAFT: An approach to
account for induced-association interactions.
J Phys Chem C 111:15544–15553.
13. Granberg RA, Rasmuson AC. 1999. Solubility of
paracetamol in pure solvents. J Chem Eng Data
44:1391–1395.
14. Manzo RH, Ahumada AA. 1990. Effects of solvent
medium on solubility. V. Enthalpic and entropic
DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 98, NO. 11, NOVEMBER 2009
4215
contributions to the free energy changes of di-sub-
stituted benzene derivatives in ethanol:water and
ethanol:cyclohexane mixtures. J Pharm Sci 79:
1109–1115.
15. Gracin S, Rasmuson AC. 2002. Solubility of pheny-
lacetic acid, p-hydroxyphenylacetic acid, p-amino-
phenylacetic acid, p-hydroxybenzoic acid, and
ibuprofen in pure solvents. J Chem Eng Data 47:
1379–1383.
16. Elworthy PH, Worthington HEC. 1968. Solubility of
sulfadiazine in water-dimethylformamide mix-
tures. J Pharm Pharmacol 20:830–835.
17. Sunwoo C, Eisen H. 1971. Solubility parameter of
selected sulfonamides. J Pharm Sci 60:238–
244.
18. Granberg RA, Rasmuson AC. 2000. Solubility of
paracetamol in binary and ternary mixtures of
water þ acetone þ toluene. J Chem Eng Data 45:
478–483.
19. Romero S, Reillo A, Escalera B, Bustamante P.
1996. The behavior of paracetamol in mixtures of
amphiprotic and amphiprotic-aprotic solvents.
Relationship of solubility curves to specific and
nonspecific interactions. Chem Pharm Bull 44:
1061–1064.
20. Bustamante P, Escalera B, Martin A, Selles E.
1993. A modification of the extended hildebrand
approach to predict the solubility of structurally
related drugs in solvent mixtures. J Pharm Phar-
macol 45:253–257.