5858 J. Med. Chem.

2007, 50, 5858-5862

Poorly Soluble Marketed Drugs Display Solvation Limited Solubility

Christel A. S. Bergström,* Carola M. Wassvik, Kajsa Johansson, and Ina Hubatsch

Pharmaceutical Screening and Informatics, Department of Pharmacy, Uppsala UniVersity, Uppsala Biomedical Centre,
Post Office Box 580, SE-751 23 Uppsala, Sweden

ReceiVed June 5, 2007

We determined the intrinsic aqueous solubility of 15 poorly soluble drugs with solubilities ranging from 2.9
nM to 1.1 µM. We then analyzed the data from a physicochemical perspective, using experimentally
determined solid-state properties and easily interpretable two-dimensional molecular descriptors, to better
understand the factors underlying poor solubility. The analysis shows that poorly soluble drugs that have
reached the market are solubility limited by solvation rather than by their solid state.

Introduction the dataset. The solubility and solid-state characteristics of astemi-

Downloaded via INSTITUTO TECNOLOGICO DE COSTA RICA on July 23, 2024 at 20:01:17 (UTC).

zole, glyburide, and indomethacin had been determined previously

See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

The methods applied in the drug discovery process in use in our laboratory,7 and these data were included in the present study.
nowadays often result in molecules of poor solubility. Low Solid-State Characterization by Differential Scanning Cal-
solubility can cause low bioavailability or give rise to large orimetry (DSC). Thermograms were recorded with a Seiko
fluctuations in the fraction absorbed in humans that can often DSC220C analysis module with an automatic cooling controller
not be compensated by a high permeability. Furthermore, low (Seiko Instruments, Inc., Japan). Triplicate samples of 1-3 mg were
solubility may be associated with stability problems and weighed in sealed and pierced aluminum pans (TA Instruments,
difficulties in developing an acceptable formulation.1 Delaware). Only one sample was run for felodipine and troglitazone,
Early in the drug discovery process, compounds are often owing to the small quantities available. Samples of each compound
available only as virtual products or synthesized in limited were heated from room temperature to approximately 50 °C above
quantities. As it is not always possible to measure the solubility their melting point at a rate of 10 °C/min and purged with nitrogen
gas at a flow rate of 80 mL/min. The melting temperature, (Tm),
directly, good predictive computational models are desirable.2 entropy of melting (∆Sm), and enthalpy of melting (∆Hm), were
Several computational approaches for predicting solubility have determined for each compound. Danazol decomposed directly after
been developed, but many of these used training sets consisting melting when measured at 10 °C/min and was, therefore, determined
primarily of nondrug-like molecules such as alcohols, pesticides, at 20 °C/min to allow the melting to finish completely before
and herbicides,3-5 as a result of which they may not be very decomposition.
accurate in predicting the solubility of drugs.1,6 Additionally, Solubility Determinations by Shake-Flask Method. Each drug
there is a distinctive lack of quantitative experimental data for was added in excess to 1.5 mL Eppendorf tubes containing 1 mL
poorly soluble drugs and drug candidates with a solubility of of Milli-Q water. The pH of the suspensions were adjusted to at
less than 1 µM. Because the drug space of poorly soluble least 2 pH units below pKa (acids) or 2 pH units above pKa (bases)
compounds is essentially uncovered, difficulties in correctly with 0.5 M HCl or 0.5 M NaOH. This kept the drugs in their
uncharged states in accordance with the Henderson-Hasselbalch
predicting poorly soluble drug-like compounds prevail.
equation and allowed the intrinsic solubility to be determined.
In this study, we have experimentally determined the solubil- Neutrals (bases with a pKa below 2 and acids with a pKa above 12)
ity of a series of poorly soluble drugs, and solubility data ranging and zwitterions were not pH adjusted. The tubes were placed onto
from 2.9 nM to 1.1 µM are presented. This unique dataset has a plate shaker at 300 rpm at room temperature. The experiments
been used to analyze the following in detail: (i) which (using 3 to 5 replicate samples) were terminated after 24, 72, and
experimental setup is preferable for measuring poorly soluble 192 h or later if equilibrium solubility had not been obtained after
compounds and (ii) which physicochemical properties are 192 h. The excess solid was separated from the solution by
fundamental for the solubility of such compounds. Finally, using centrifugation in an Eppendorf Centrifuge 5403 (Eppendorf AG,
the data obtained, we discuss which poorly soluble compounds Hamburg, Germany) at 23 000 g for 15 min. After the centrifuga-
are most likely to successfully complete the drug development tion, approximately 0.25 mL of the supernatant was sampled with
Pasteur glass pipettes and dispensed into glass insert vials. These
process. vials were placed in Eppendorf tubes and centrifuged for 15 min a
second time to obtain complete separation of the solution and the
Experimental Section
remaining solid. After the centrifugation, the supernatant was
Selection of Dataset. A series of 15 poorly soluble compounds withdrawn with Pasteur pipettes and dispensed into glass HPLC
was selected for this study (Figure 1). The choice was made from vials. Glass was used throughout this procedure to minimize the
a literature search using PubMed and the search string “poor* AND risk of underestimating the solubility owing to adsorption of the
solubility AND drug* AND formulation” to extract information drugs to plastics.
on poorly soluble drug-like compounds. Compounds that were HPLC-MS/MS Analysis. Directly after the termination of the
judged to (i) be more soluble than 1 µM from the publications, (ii) solubility experiment, the concentration of the samples was
be unstable, (iii) only exist as salt forms, (iv) not be commercially determined with a ThermoFinnigan TSQ Quantum Discovery triple-
available, or (v) be very expensive were excluded. Principal quadrupole mass spectrometer using electrospray ionization (ESI),
component analysis (PCA) was conducted using Simca-P version coupled to a ThermoFinnigan Surveyor autosampler and Surveyor
10 (Umetrics, Umeå, Sweden) to verify the structural diversity of HPLC-MS pump (Thermo Electron Corp., Waltham, U.S.A.). For
separation, an XTerra MSC18 column (3.5 µm, 2.1 × 20 mm;
* To whom correspondence should be addressed. Phone: +46184714645. Waters, Milford, U.S.A.) and a flow rate of 200 µL/min were used.
Fax: +46184714223. E-mail: [email protected]. The samples (5 µL) were injected and run with a gradient using
10.1021/jm0706416 CCC: $37.00 © 2007 American Chemical Society
Published on Web 10/11/2007
Brief Articles Journal of Medicinal Chemistry, 2007, Vol. 50, No. 23 5859

Figure 1. Chemical structures of the compounds studied. The compounds had the following physicochemical profile: MW 261.1-705.7 g/mol,
ClogP 3.5-6.8, and PSA 2.4-137.2 Å2.
water and acetonitrile with or without formic acid in either the Extended time studies were performed, but a long agitation
negative or positive polarity mode (see Supporting Information). time was not needed for all compounds to reach their solubility
For each compound, a standard curve consisting of seven concen- equilibrium (see the Supporting Information). The shaking time
trations was established, with a separate quality control being made ranged from 24 to 1104 h, and for eight compounds, there was
of three concentrations.
no statistically significant difference between the means of the
Statistics. The solubility values are presented as means ( SD.
ANOVA was used to test whether the differences between means first and last time point at the 95% CI. However, it was not
were statistically significant (p < 0.05). The coefficient of deter- possible to know beforehand which compounds would need a
mination (R2) was used to assess the goodness-of-fit of standard longer time. Danazol and tamoxifen had the largest difference
curves of measured concentrations. Partial least-squares projection between the first and last concentrations. It was not surprising
to latent structures (PLS; Simca-P v.10) was used to further analyze that danazol required a long time to reach equilibrium, as
the importance of different physicochemical properties, as described previous results from our laboratory have shown that steroids
previously.8 Calculations of two-dimensional molecular descriptors may require a long time to attain equilibrium. This we have
reflecting among other the size, polarizability, hydrophobicity/ found for hydrocortisone6 and corticosterone (unpublished
hydrophilicity and electron distribution were made with the program results). The solubility of danazol increased 3.2-fold with time,
Selma.7,8 Only nonskewed descriptors (n ) 75) were included in
the PLS analysis, and a variable selection was performed to exclude and tamoxifen decreased 5-fold with time. Such variations in
noise and increase the transparency of the model. the solubility indicate that, to roughly estimate the equilibrium
solubility of poorly soluble compounds, a 24 h equilibrium time
Results and Discussion period is long enough. However, if the solubility is supposed
Experimental Results. The solubility and solid-state char- to be used as an input in computational solubility models, it is
acteristics are presented in Table 1. The solubility of the 15 preferable to extend the time scale for the experiment to ensure
drugs ranged from 2.9 nM to 1.1 µM. The limit of detection of that the equilibrium solubility is obtained. By adopting this
troglitazone was 15.9 nM, but even though the experiment was approach, solubility data of higher quality are used in the model
extended to 1104 h (46 days), all shake-flask samples were development, thereby increasing the chances of producing an
below this limit. It is unlikely that this was because of accurate in silico model for the solubility and minimizing the
degradation of the sample, because an LC-MS scan of the m/z risk of modeling noise.
200-500 of the samples did not reveal any other significant Physicochemical Properties and their Relation to Solubil-
peaks. However, as an m/z scan of cinnarizine samples shaken ity. The relation between physicochemical properties and
for longer than 72 h indicated degradation, the solubility value solubility was investigated (Table 2 and Supporting Information
from 72 h is reported. Table S3). First, the relationship to the widely used molecular
5860 Journal of Medicinal Chemistry, 2007, Vol. 50, No. 23 Brief Articles

Table 1. Experimental Solubility Values and Solid-State Characteristics

timea Sb Tmc ∆Hmd ∆Sme
(h) (nM) (°C) (kJ/mol) (J/mol/K)
albendazole 216 983.0 ( 112.2 178.1 ( 4.3 98.6 ( 16.1 218.7 ( 37.8
astemizolef 24 66.8 ( 21.1 174.4 ( 0.1 51.1 ( 0.8 114.1 ( 1.7
carvedilol 744 713.0 ( 56.0 114.1 ( 0.5 57.6 ( 1.0 148. 7 ( 2.6
cinnarizine 72 18.6 ( 1.7 120.2 ( 1.0 45.7 ( 2.3 116.1 ( 5.6
danazol 1032 36.0 ( 28.3 228.6 ( 0.5 35.5 ( 1.3 70.8 ( 2.6
felodipine 504 276.0 ( 17.7 139.1h 34.8h 84.3h
glimepiride 213 12.7 ( 2.9 212.5 ( 0.6 53.3 ( 4.5 109.9 ( 9.2
glyburidef 24 89.1 ( 40.6 173.6 ( 0.1 46.3 ( 0.1 103.7 ( 0.2
indomethacinf 24 1133.3 ( 251.7 159.8 ( 0.0 37.9 ( 0.2 87.6 ( 0.4
itraconazol 336 3.3 ( 4.6 165.4 ( 0.9 69.9 ( 3. 5 159.5 ( 7. 8
rimonabant 380 96.5 ( 12.5 154.7 ( 0.4 36.1 ( 1.7 84.4 ( 4.2
tamoxifen 888 2.9 ( 1.0 97.8 ( 0.3 34.0 ( 0.4 91.6 ( 1.1
terfenadine 864 11.6 ( 7.2 149.6 ( 0.3 58.1 ( 2.0 137.5 ( 4.8
tolfenamic acid 192 13.6 ( 1.5 212.1 ( 0.3 41.2 ( 1.3 84.9 ( 2.7
troglitazone 1104 <15.9g 139.2h 48.8h 118.3h
mean 246.9 161.3 49.9 115.3
min 2.9 97.8 34.0 70.8
max 1133.3 228.6 98.6 218.7
a Time (h) shows the equilibrium time used in the solubility study. b S (nM) is the solubility value presented in nanomolar concentration. c T denotes the
m
melting point in Celsius. None of the compounds displayed polymorphism. d ∆Hm denotes the enthalpy of melting. e ∆Sm denotes the entropy of melting.
f Data taken from Wassvik et al., 2006, ref 7 in this paper. g Troglitazone could not be detected within the sample even after an equilibrium time of

1104 h. The presented value is the limit of detection for troglitazone using the applied LC-MS/MS settings. The sample showed no indication of degradation
of troglitazone, in spite of the long time used for the study. h Only one sample was measured for solid-state characteristics as only a small amount of the
substance was available.

Table 2. Physicochemical Characteristics of the Compounds Studied

MW PSAa NPSAa acid/
(g/mol) (Å2) (Å2) ClogPb pKab baseb
albendazole 265.3 65.6 231.9 3.5 11.4; 5.6 a; b
astemizole 458.6 35.4 488.5 6.1 9.0 b
carvedilol 406.3 78.4 381.4 4.0 8 b
cinnarizine 368.5 2.4 452.4 6.1 7.5 b
danazol 337.5 45.9 356.1 3.9 neutral
felodipine 384.3 68.7 346.9 5.6 2.7 b
glimepiride 490.6 137.2 439.3 4.2 5.0 a
glyburide 494.0 126.9 415.0 4.2 5.0 a
indomethacin 357.6 68.8 306.8 4.2 4.0 a
itraconazol 705.7 84.7 654.2 6.5 4.0 b
rimonabant 463.8 50.7 418.2 6.6 3.6 b
tamoxifen 371.5 10.3 464.2 6.8 8.5 b
terfenadine 471. 7 46.3 561.1 6.1 9.5 b
tolfenamic 261.1 54.8 227.1 5.7 3.7 a
acid
troglitazone 441.6 92.2 400.6 5.6 6.1 a
mean 418.5 64.6 409.6 5.3
min 261.1 2.4 227.1 3.5
max 705.7 137.2 654.2 6.8
a Selma descriptors obtained from Dr. Ulf Norinder, AstraZeneca R&D,

Södertälje. b Calculated from SciFinder Scholar.

descriptor ClogP, that is, the calculated partition coefficient

between octanol and water was studied. We have previously
shown that this relationship for 270 compounds spanning more
than 9 log units in solubility, has an R value of 0.74.7 For the
15 poorly soluble compounds studied in this dataset, the R was Figure 2. Graphs showing the correlation between the physicochemical
properties and the solubility. (a) Correlation between the solubility and
0.71 (Figure 2a). However, the relationship obtained with the the calculated lipophilicity (R of -0.71). (b) Correlation between the
270 compounds suggests that ClogP needs to be 6.6 or larger solubility and the melting point (R of 0.13). (c) The correlation between
to enable a solubility of less than 1 µM to be predicted, as a the solubility and the nonpolar surface area. R went from -0.48 to
result of which most of the compounds investigated in the -0.85 after exclusion of the outlier tolfenamic acid.
present work would not be predicted as poorly soluble. Instead,
data for such compounds.7 Therefore, poorly soluble compounds
it was noted that all compounds with a ClogP value larger than did not have the power to influence the general equation to any
6 showed an intrinsic solubility less than 100 nM, that is, a great extent.
solubility value more than 10-fold lower than that predicted by Several calculated and experimentally determined properties
the general equation. One reason that the role played by were investigated to obtain a deeper knowledge of what other
lipophilicity in restricting solubility is underestimated could be physicochemical properties influence the solubility of poorly
that only a limited number of poorly soluble compounds were soluble compounds. First, we performed linear regression
included in our previous study because of the lack of solubility analysis in which we correlated those descriptors, one at the
Brief Articles Journal of Medicinal Chemistry, 2007, Vol. 50, No. 23 5861

Figure 3. The poorly soluble compounds superimposed on the two first principal components (PC) describing 52% of the structural diversity of
the oral drug space. The ellipse shows the 95% CI of the structural diversity of orally administered drugs registered in Sweden (n ) 527, Selma
descriptors used as input). PC1 mainly reflects the size (the higher the value, the larger the molecule) and PC2 reflects the lipophilicity (the lower
the value, the higher the lipophilicity). The poorly soluble compounds are all clustered into one-quarter of the plot. The compounds are colored
according to their solubility from the lowest solubility (red) to the highest (blue). A trend emerges whereby the closer a compound is to the ellipse
in the lower right quadrant, the poorer the solubility. The arrow is drawn to indicate in which direction the solubility increases. This trend was
confirmed by superimposing solubility data previously published by our laboratory,8 resulting in that no soluble compounds were found in the
lower right quadrant. The compounds are abbreviated to the four first letters of their names; for the full names, see Tables 1 or 2.

time, that have previously been identified as being important was used as a tool to further investigate the properties of
for solubility with the solubility values obtained (see Table S3 importance for poor solubility. The PLS analysis showed that
in the Supporting Information). No correlation was observed descriptors related to lipophilicity, size, and polarizability
between the solubility and the solid-state properties investigated (reflected in the molecular refractivity) were important for
(Tm, ∆Sm, and ∆Hm), all of which resulted in R less than 0.13 restricting solubility (Figure S2 in Supporting Information),
(Figure 2b and Table S3, Supporting Information). Hence, the confirming the results from the correlation matrix. Additionally,
poor solubility of the compounds studied was not dependent this analysis also showed that the larger the difference between
on the stability of the crystal, although the compounds displayed the highest occupied and lowest unoccupied molecular orbital
a wide range of Tm (97.8-228.6 °C, Table 1). Furthermore, (HOMO and LUMO, respectively), the lower the solubility. The
none of the hydrogen bond descriptors investigated (the number energy levels of these molecular orbitals have also previously
of acceptors and donors and the sum of these) proved to be been reported to influence the solubility.10
important for this dataset. This is in agreement with the findings How can we interpret the results concerning the properties
for the solid state, because increased polarity and the number that underlie poor solubility? The results suggest that increased
of hydrogen bonds are important for making stable crystals.9 molecular size, increased polarizability, increased lipophilicity,
The molecular size proved to be important for decreasing and an increased energy gap between the HOMO and LUMO
the solubility for all compounds except for tolfenamic acid. will decrease the solubility. First, the larger the size of the
Molecular descriptors such as molecular weight, polarizability, molecule, the larger the cavity in the water needs to be, which
and nonpolar surface area, all of which are highly correlated to implies a greater amount of energy is required to break the tight
the size of the molecule, displayed R values of 0.61, 0.79, and structure of water. For this dataset, the molecular size was highly
0.85 when correlated to the solubility after the exclusion of correlated with the polarizability, resulting in an R of 0.93.
tolfenamic acid (Figure 2c and Table S3 in Supporting Informa- Hence, we believe that the negative influence of polarizability
tion). The reason for tolfenamic acid being an outlier is probably on solubility found in this dataset also is partly a reflection of
the combination of its high lipophilicity, high melting point, the energy penalty involved in the cavity formation process.
and small size in comparison to the other compounds of the As the cavity is formed, the molecule is incorporated and needs
dataset. Hence, for this specific compound, both the lipophilicity to make bonds with the water molecules to remain in solution.
and the stability of the crystal can be reasons for its poor The higher the lipophilicity of a compound, the less favorable
solubility. However, it is unlikely that size itself is the limiting the hydration, and, subsequently, the lower the solubility.
descriptor for tolfenamic acid. Finally, the multivariate data analysis identified a descriptor
It is well-known that molecular size and lipophilicity are related to the energy gap between the HOMO and LUMO as
highly correlated and that larger molecules generally display a being important for poorly soluble compounds. Indeed, it was
higher ClogP value than smaller ones. However, the correlation found that the larger the energy difference between these two
(R) between ClogP and MW for the 15 compounds in this orbitals, the poorer the solubility. It is known that the energy
dataset was only 0.36, indicating that the size per se is also gap is related to the degree of conjugation and that a higher
restricting the solubility of these molecules (see Table S3, degree of conjugation results in smaller energy gaps. Hence,
Supporting Information). the analysis revealed that for this dataset the solubility will
In a second step, we analyzed the influence of the different improve for compounds that are more conjugated than others.
descriptors on solubility using multivariate data analysis. The One explanation for this could be that conjugated systems are
model obtained was not regarded as a model for the prediction more rigid and often more compact than nonconjugated ones,
of poor solubility because the dataset was too small. Instead, it and therefore, the cavity formation in water will not be as
5862 Journal of Medicinal Chemistry, 2007, Vol. 50, No. 23 Brief Articles

extensive. Hence, this finding further indicated that a majority of these compounds. Finally, the statistical analysis showed that
of the compounds investigated in this study are solvation limited poorly soluble compounds are located in a specific volume of
in their solubility rather than limited by their solid state. If the the oral drug space. This indicates that this simple tool can give
compounds were to be solubility limited by their solid state, it direct feedback on solvation limited solubility even before drug
is more likely that a decrease in conjugation would result in an synthesis.
increase in solubility because flat and rigid compounds tend to
form more stable crystals than flexible ones.11 Acknowledgment. We would like to acknowledge Dr. Ulf
All of the compounds examined have been developed as oral Norinder at AstraZeneca R&D Södertälje for providing us with
dosage forms (Table S4 and Table S5 in Supporting Informa- the molecular descriptors. Financial support from the Knut and
tion). We therefore superimposed the poorly soluble compounds Alice Wallenberg Foundation, the Swedish Foundation for
onto the chemical space covered by oral drugs, as identified by Strategic Research, and The Swedish Fund for Research without
a PCA based on all orally registered compounds in Sweden (n Animal Experiments is greatly appreciated.
) 527), and found the poorly soluble compounds clustered in
a dedicated chemical volume (Figure 3). Hence, by using Supporting Information Available: LC-MS/MS settings for
physicochemical descriptors and the structural diversity of the analysis of the compounds studied; time versus solubility profiles
oral drug space, an estimation of the likeliness of poor solubility of the compounds studied; a correlation matrix of physicochemical
could be obtained. Such a tool is convenient and easily applied properties, molecular descriptors, and solubility; formulation ap-
in the early drug discovery process and can be used to guide proaches to increase solubility for the compounds studied; dose
the medicinal chemist away from poor solubility issues. regimen of the compounds; and PLS model statistics and loading
Poorly soluble compounds are often referred to as “brick dust” plots for the final model based on poorly soluble compounds. This
and “grease balls”. Brick dust represents a stable crystal in which information is available free of charge via the Internet at http://
the strong intermolecular bonds within the crystal restrict the pubs.acs.org
solubility of the compound in water, whereas grease balls
represent highly lipophilic compounds that are unable to form References
bonds with the water molecules. The poorly soluble compounds (1) Lipinski, C. A.; Lombardo, F.; Dominy, B. W.; Feeny, P. J.
investigated in this study have all passed through the develop- Experimental and computational approaches to estimate solubility
ment process and, by adopting strategic formulation procedures, and permeability in drug discovery and development settings. AdV.
Drug DeliVery ReV. 1997, 23, 3-25.
they have reached the market in spite of their limited solubility (2) Delaney, J. S. Predicting aqueous solubility from structure. Drug
(Table S4, Supporting Information). The thorough analysis of DiscoVery Today 2005, 10 (4), 289-295.
the physicochemical properties important for lowering solubility (3) Cheng, A.; Merz, K. M. Prediction of aqueous solubility of a diverse
presented above did not support solid-state limited solubility set of compounds using quantitative structure-property relationships.
J. Med. Chem. 2003, 46, 3572-3580.
for the dataset as a whole. Rather, the successful development (4) Sun, H. A universal molecular descriptor system for prediction of
of the investigated compounds is probably related to them acting logP, logS, logBB, and absorption. J. Chem. Inf. Comput. Sci. 2004,
as grease balls. For such compounds, the solubility can be 44 (2), 748-757.
improved largely by incorporating excipients, such as disinte- (5) Huuskonen, J. Estimation of aqueous solubility for a diverse set of
organic compounds based on molecular topology. J. Chem. Inf.
grants, solubility enhancers/wetting agents, cyclodextrins, and Comput. Sci. 2000, 40 (3), 773-777.
lipids, in the formulation.12 Furthermore, bile salts in the (6) Bergström, C. A. S.; Norinder, U.; Luthman, K.; Artursson, P.
intestinal fluid will markedly improve the intestinal solubility Experimental and computational screening models for prediction of
of highly lipophilic compounds.13 For highly lipophilic com- aqueous drug solubility. Pharm. Res. 2002, 19 (2), 182-188.
(7) Wassvik, C. M.; Holmen, A. G.; Bergström, C. A. S.; Zamora, I.;
pounds with a ClogP value of three or more, we therefore
Artursson, P. Contribution of solid-state properties to the aqueous
recommend the solubility assay to be performed in the presence solubility of drugs. Eur. J. Pharm. Sci. 2006, 29 (3-4), 294-305.
of an emulsifying agent or in a simulated intestinal fluid14 to (8) Bergström, C. A. S.; Wassvik, C. M.; Norinder, U.; Luthman, K.;
avoid underestimating the intestinal solubility. Moreover, the Artursson, P. Global and local computational models for prediction
pH gradient in the intestine will affect the solubility of several of aqueous solubility of drug-like molecules. J. Chem. Inf. Comput.
Sci. 2004, 44 (4), 1477-1488.
of the compounds investigated, and their apparent solubility will (9) Bergström, C. A. S.; Norinder, U.; Luthman, K.; Artursson, P.
be higher at a physiological pH than the intrinsic solubility. Molecular descriptors influencing melting point and their role in
However, for most of the compounds investigated in this work, classification of solid drugs. J. Chem. Inf. Comput. Sci. 2003, 43
the solubility will still be in the lower micromolar range at (4), 1177-1185.
(10) Wang, Z.-W.; Feng, J.-L.; Wang, Z.-N.; Li, G.-Z.; Lou, A.-J.
physiological pH, confirming that solubilizing agents are of the Quantitative structure-property relationship on prediction of the
utmost importance for further increasing the solubility and interaction parameters δ2t of organic compounds. J. Dispersion Sci.
obtaining a therapeutic effect with these compounds. Technol. 2006, 27, 11-14.
(11) Wassvik, C. M. Computational analysis of aqueous drug solubility-
influence of the solid state. Thesis from the Faculty of Pharmacy,
Conclusion Uppsala University, Uppsala, 2006.
We have presented a unique dataset including experimental (12) Avdeef, A.; Bendels, S.; Tsinman, O.; Tsinman, K.; Kansy, M.
Solubility-excipient classification gradient maps. Pharm. Res. 2007,
solubility values and solid-state characteristics for poorly soluble 24 (3), 530-545
drugs. The thorough analysis of the factors underlying the poor (13) Persson, E. M.; Gustafsson, A. S.; Carlsson, A. S.; Nilsson, R. G.;
solubility of drugs showed that, for this dataset, the solubility Knutson, L.; Forsell, P.; Hanisch, G.; Lennernas, H.; Abrahamsson,
is restricted by solvation rather than the solid state. The most B. The effects of food on the dissolution of poorly soluble drugs in
human and in model small intestinal fluids. Pharm. Res. 2005, 22
important properties causing poor solubility were a high (12), 2141-2151.
lipophilicity, a large molecular size, and a poorly conjugated (14) Galia, E.; Nicolaides, E.; Horter, D.; Lobenberg, R.; Reppas, C.;
electron system. The study showed that the poor solubility Dressman, J. B. Evaluation of various dissolution media for predicting
caused by these properties can be treated by optimizing the in vivo performance of class I and II drugs. Pharm. Res. 1998, 15
(5), 698-705.
formulations, but components of the intestinal fluid are also
likely to play an important role in the successful development JM0706416