Kozikowski

ARTICLE
10.1177/1087057103252618
Effect of Freeze/Thaw
et al. Cycles on DMSO

The Effect of Freeze/Thaw Cycles on

the Stability of Compounds in DMSO

BARBARA A. KOZIKOWSKI, THOMAS M. BURT, DEBRA A. TIREY, LISA E. WILLIAMS,

BARBARA R. KUZMAK, DAVID T. STANTON, KENNETH L. MORAND, and SANDRA L. NELSON

A diverse set of 320 compounds from the Procter & Gamble Pharmaceuticals organic compound repository was prepared as
20-mM DMSO solutions and stored at 4 °C under argon in pressurized canisters to simulate a low-humidity environment. The
plates were subjected to 25 freeze/thaw cycles while being exposed to ambient atmospheric conditions after each thaw to sim-
ulate the time and manner by which compound plates are exposed to the atmosphere during typical liquid-handling and high-
throughput screening processes. High-performance liquid chromatography–mass spectrometry with evaporative light-
scattering detection was used to quantitate the amount of compound remaining after every 5th freeze/thaw cycle. Control
plates were stored either at room temperature under argon or at 4 °C under argon without freeze/thaw cycling and were evalu-
ated at the midpoint and the endpoint of the study. The study was conducted over a short time period (i.e., 7 weeks) to minimize
the effect of compound degradation over time due to the exposure of the compounds to DMSO. The results from this study will
be used to determine the maximum number of freeze/thaw cycles that can be achieved while maintaining acceptable com-
pound integrity. (Journal of Biomolecular Screening 2003:210-215)

Key words: mass spectrometry (MS); high-performance liquid chromatography–mass spectrometry (HPLC-MS); pharma-
ceutical drug discovery; compound repository; compound stability; high-throughput screening (HTS)

INTRODUCTION found effects on the integrity of a compound in solution because

sample degradation and/or precipitation may occur.

T HE PROLONGED STORAGE OF ORGANIC COMPOUNDS in solu-

tion can lead to significant compound degradation and to a
subsequent increase in the number of false positives for high-
throughput biological assays. As a consequence, many pharma-
ceutical organizations store their high-throughput screening (HTS)
collections as frozen DMSO solutions. The hygroscopic nature of
DMSO1-3 can complicate sample storage because water absorption
by DMSO can affect compound solutions by causing changes in
compound concentration, solubility, and/or degradation. As a re-
sult, soluble screening collections are generally stored at 4 °C and
low relative humidity. As a consequence, compound solutions are
thawed to access the samples for HTS testing and are frozen again
for long-term storage. This freeze/thaw cycling may have pro-
Procter & Gamble Pharmaceuticals, Mason, OH.
Received Aug 6, 2002, and in revised form Dec 31, 2002. Accepted for publica-
tion Jan 3, 2003.
Journal of Biomolecular Screening 8(2); 2003
DOI: 10.1177/1087057103252618
Published by Sage Publications in association with The Society for Biomolecular
Screening
The quality of an HTS screening collection can be maintained if
expiration limits for plate retention are used to trigger the creation
of fresh compound plates. Unfortunately, there is a paucity of in-
formation on this topic, as most information on the effects of
freeze/thaw cycling and prolonged storage on compound stability
has only been presented in conference proceedings.4-9 Accord-
ingly, we designed 2 studies to monitor the effects of storage and
subsequent compound handling on the integrity of a diverse set of
pharmaceutically relevant compounds.8,9 The first study examined
the fate of compounds stored up to 1 year under ambient conditions
and is presented in another manuscript.9 The study described in this
paper was undertaken to measure the effects of multiple freeze/
thaw cycles on the stability of DMSO solutions of 320 compounds.
The thawed solutions were exposed to ambient conditions after
each thaw to simulate the time that compound plates are exposed to
the atmosphere during liquid handling. The study was conducted
over a short period of time (7 weeks) to separate the effects of
freeze/thaw cycling from sample degradation due to other pro-
cesses. The goal of the study was to identify the maximum number
of freeze/thaw cycles that are allowable to maintain compound in-
tegrity at a predefined level of acceptability.

210 www.sbsonline.org © 2003 The Society for Biomolecular Screening

 Effect of Freeze/Thaw Cycles on DMSO
MATERIALS AND METHODS
Diversity analysis
A total of 320 organic compounds from various commercial
vendors were obtained from the Procter & Gamble Pharmaceutical
(P&GP) compound repository and were used in this study. The
analysis of the diversity of the subset was done using the computer
program DiverseSolutions.10 The subset was compared to the
whole P&GP repository collection to ensure that it was generally
representative of the entire collection. The comparison was made
on the basis of a “universal” chemistry-space definition compris-
ing 6 BCUT descriptors.11 The chemistry-space definition was de-
signed based on the examination of approximately 1 million drug-
like structures and captures important molecular features such as
hydrogen-bond donating and accepting ability, polarizability, charge
distribution, and solvent-accessible surface area. Principal compo-
nents analysis (PCA)12 was used to compress the 6-dimensional
chemistry space coordinates into a 3-dimensional data set for the
purpose of visualization. The computer program Minitab for
Windows13 was used for this analysis.
Materials
From the P&GP compound repository, 5 to 7 mg of each sample
was received as a dry powder and was dissolved on a custom-
designed automated compound dissolution system. All com-
pounds were dissolved to 20 mM in 99.8% anhydrous DMSO
(Aldrich, Milwaukee, WI).
The dissolved compounds were transferred to four 96-well
deep-well mother plates (Beckman Coulter, Fullerton, CA).
Daughter plates (Costar, Corning, NY) containing 100 µL of solu-
tion per well were prepared from each mother plate and were used
in the study. All daughter plates were sealed with microtiter cap
mats (Costar, Corning, NY) and stored within modified pressure
canisters (Speedaire, Chicago). A positive pressure of argon was
maintained in each canister to simulate a low-humidity
environment.
Freeze/thaw study design
Each sample was submitted for high-pressure liquid chroma-
tography–mass spectrometry (HPLC-MS) analysis with evapora-
tive light-scattering detection (ELSD) at the beginning of the study.
Four sets of daughter plates were stored for 7 weeks under the fol-
lowing conditions and were submitted for HPLC-MS/ELSD anal-
ysis as indicated:
1. Freeze/thaw plates: One set of 4 plates was stored under argon in a
4 °C cold room. The plates were thawed daily under argon and
were then exposed to ambient conditions for ~2 h to simulate the
time that plates are exposed to the atmosphere during liquid han-
dling. The canisters were then purged with argon and returned to
the cold room. A sample was sent for HPLC-MS/ELSD analysis
after every 5th freeze/thaw cycle.
Journal of Biomolecular Screening 8(2); 2003
2. Room-temperature controls: One set of 4 plates was stored under
argon at room temperature. The canisters were opened daily, and
the plates were exposed to ambient conditions for ~2 h to simulate
the time that plates are exposed to the atmosphere during liquid
handling. The canisters were then purged with argon. A sample
was sent for HPLC-MS/ELSD analysis at 2 weeks and at the end
of the study.
3. Frozen, never thawed controls: Two sets of 4 plates were stored in
separate canisters under argon in a 4 °C cold room. One set of
plates was thawed at 2 weeks and was submitted for HPLC-MS/
ELSD analysis. The other set of plates was thawed at the end of the
study and was submitted for HPLC-MS/ELSD analysis.
Analytical procedure
First, 10-µL aliquots were transferred with mixing from daugh-
ter plates into 96-well deep-well plates (Beckman Coulter,
Fullerton, CA) containing 200 µL of spectrophotometric-grade
DMSO (EM Science, Gibbstown, NJ). The diluted solutions were
then transferred to a vial fitted with a microinsert (Agilent, Palo
Alto, CA) and were placed in an Agilent 1050 autosampler. Then,
5 µL of the diluted sample solution was injected and analyzed via
HPLC-MS/ELSD. Both positive and negative atmospheric pres-
sure chemical ionization (APCI) were used for mass spectral de-
tection. ELSD was the method of choice for purity assessment and
quantitation in this application due to its more universal nature
when compared with UV detection. Mass spectrometry was the
method of choice for molecular weight confirmation. Experimen-
tal conditions for the analytical analyses are summarized in Table 1.
Ciprofloxacin (Seriologicals, Inc., Kankakee, IL) and (±)-4
[α-[2-Chloro-5-[4-(2.4-di-tert-pentylphenoxy)butyramido]-
phenylcarbamoyl]-α-pivaloyl-methoxy]benzoic acid (Aldrich,
Milwaukee, WI) were used as surrogate compounds to calibrate
the ELSD response because individual calibration of each analyte
in the study was not feasible. These 2 compounds were chosen as
“universal” surrogates (1) to encompass the MS range for most of
the study compounds, (2) to display many of the drug-like moieties
present in the study compounds, and (3) to elute within the ex-
pected retention time for most targets.
Fresh calibration standards and quality control (QC) solutions
were prepared from stock solutions and analyzed with each batch
run of the study. The calibration stock solution was serially diluted
to provide calibration standards ranging in concentrations from
0.02 to 1.5 mg/mL. Separately weighed QC stock solutions (from
0.4 to 0.8 mg/mL) were prepared from the surrogate compounds.
QC solutions and DMSO blanks were interspersed with the study
samples.
Samples, standards, and QC solutions were quantitated against
an averaged-value ELSD calibration curve constructed using data
from the surrogate calibration. A new calibration curve was con-
structed with each set of samples for every time point in the study
and was used to determine an absolute concentration in mmol/L
for each sample at each time point. The relative percentage of each
compound after each time point was determined by comparing the
www.sbsonline.org 211
 Kozikowski et al.

Table 1. Instrumental Conditions for the HPLC-MS Analyses A repeated-measures regression model was used to describe the
Performed in This Study relationship between the number of thaws and the percentage of
compound remaining. The fixed effect was the number of thaws,
APCI-MS experimental and the random effects were plates and compounds within the
conditions for HPLC/
plates. The maximum number of freeze/thaw cycles can be deter-
ELSD/APCI-MS
Instrumentation Micromass Platform-II with APCI ionization mined whereby the lower one-sided confidence interval intersects
Ionization/data type APCI + and –, centroid data a predetermined percent remaining. The analysis was performed in
Drying gas N2, 300 L/h SAS 8.0.14
APCI sheath gas N2, 100 L/h
Mass range 100-1000 (+, –)
Scan rate 0.12 interscan delay, 1.2 sec/scan RESULTS AND DISCUSSION
Cone voltage 8V
Source temperature 150 °C A total of 320 compounds from commercial vendors were in-
Probe temperature 450 °C
HPLC experimental cluded in the freeze/thaw study. However, 88 of the 320 com-
conditions for HPLC/ pounds were not detected by LC-MS at t = 0 and therefore could
ELSD/APCI-MS not be followed over the course of the study or included in the sta-
Instrumentation HP 1050 HPLC with vial autosampler tistical analysis of the data. A compound could have not been de-
Column Waters Symmetry Shield Rx C8, 4.6 × 50 mm
Mobile phase A: 95% ACN/5% H20/.005% TFA; B: 3% ACN/
tected due to rapid compound degradation upon dissolution, com-
97% H20/.005% TFA; C: 3% ACN/97% H20/ pound incompatibility with the HPLC and/or ionization methods
.005% TFA; D: 95% ACN/5% H20/.005% TFA used, or compound misidentification upon receipt. All of the com-
Gradient: 50/50 B/C to 50/50 A/D over 3 min, pounds not detected were obtained from commercial vendors. Er-
hold 1.5 min rors in the identity of compounds purchased from vendors are not
Equilibrate at 50/50 B/C for 1.5 min
(total runtime = 6.0 min) uncommon and may be due to errors in data transfer, sample label-
LC flow rate 3.0 mL/min ing, or original identification. The average purity of the remaining
Amount injected 5 µL 232 compounds at the start of the study was 96%.
Split ratio 2:3:1 (UV:ELSD:MS) The three 2-dimensional plots in Figure 1 show the relationship
ELSD Sedere Sedex 75C, Gain = 9, 2.8 bar, 41 °C
UV wavelength 254 nm
of the compounds included in the stability study to the whole P&
GP repository. The x, y, and z axes correspond to the first, second,
HPLC-MS, high-performance liquid chromatography–mass spectrometry; APCI-MS, at-
mospheric pressure chemical ionization–mass spectrometry; ELSD, evaporative light- and third principal components (PCs), respectively, from the PCA
scattering detection.
analysis. A total of 68.1% of variance in the original 6-dimensional
space is accounted for by the first 3 PCs. The compounds are
chemically diverse as a set and are representative of the diversity of
absolute concentration value for a given time point (tm) with its as- the chemical repository at P&GP.
sociated value at the beginning of the study (t0) via the following
A summary of the statistical analysis for comparing the 3 differ-
simple equation:
ent storage methods is shown in Table 2 and plotted in Figure 2.
Percentage of compound remaining = The treatment, time, and treatment-by-time interaction terms in the
(Concentration at tm冫Concentration at t0) • (100). statistical model were highly statistically significant (p = 0.0027, <
0.0001, and < 0.0001, respectively). At 10 freeze/thaw cycles, the
frozen control was statistically different from the room-tempera-
Statistical analysis
ture control and the freeze/thaw samples by the t-test but not by
A mixed model was used to model the effect of the 3 different Tukey’s test. Note that the > 100% midpoint reading for the frozen
storage conditions on the percentage of compound remaining. The control can be attributed to the use of external standards, a nonlin-
fixed effects were the treatment and time (or the number of thaws) ear detector, and sampling errors during plating and subsequent
and treatment-by-time interaction, with time being modeled as a liquid handling. At the end of 25 freeze/thaw cycles, all 3 storage
repeated measure. The random effects were plates, compounds methods were statistically different by the t-test, and both the
within the plates, and the plate-by-treatment interaction. For each frozen and room-temperature storage methods were statistically
treatment and time combination, least square means and standard different from the freeze/thaw storage method by Tukey’s test. The
errors were generated. Multiple treatment comparisons were made percentage of compound remaining at the end of the study was
using t-test and Tukey’s test. The t-test controls for comparison- greatest for the frozen and never-thawed controls and was least for
wise error rates, whereas Tukey’s test controls for experiment-wise the compounds exposed to the multiple freeze/thaw cycles (83.1%
error rates. Comparison-wise error rates are more liberal than vs. 55.8%, respectively), as seen in Figure 2. As expected, the per-
experiment-wise error rates. Statistical significance was α = 0.05, centage of compound remaining decreased for all 3 storage
and all tests were two-sided. methods.

212 www.sbsonline.org Journal of Biomolecular Screening 8(2); 2003

 Effect of Freeze/Thaw Cycles on DMSO

FIG. 1. A comparison of the diversity of the compounds included in the freeze/thaw stability study (red) to those in the Procter & Gamble Pharmaceu-
ticals repository (black) is shown. Principal components analysis was used to convert the 6-dimensional chemistry space into three 2-dimensional views.

Table 2. Summary of the Statistical Analysis for Comparing Compound loss was probably not due to degradation because
the 3 Different Storage Conditions additional peaks were not observed in any of the HPLC
chromatograms. Solid precipitate was observed in many of the so-
Least Square Multiple Comparisons
lutions at the end of the study, indicating that the apparent sample
Means
Treatment Time Standard Error t-Test Tukey’s Test
losses could be due to compound precipitation rather than com-
pound degradation.
Frozen control Midpoint 106.7 ± 3.2 A A The relationship between the percentage of compound remain-
Room-temperature
ing and the number of thaws can be described by the following re-
control Midpoint 96.7 ± 3.2 B A
Freeze/thaw 10 freeze/thaws 96.3 ± 3.2 B A gression equation, which is plotted in Figure 3:
Frozen control Endpoint 83.1 ± 3.2 A A
Room-temperature Percentage of compound remaining =
control Endpoint 72.4 ± 3.2 B A 104.1% – 1.67 • (Number of thaws).
Freeze/thaw 25 freeze/thaws 55.8 ± 3.2 C B
Column means followed by the same letter are not statistically significantly different at α = Both the slope and intercept terms in the regression model were
0.05 within a given time. highly statistically significant (p < 0.0001 and < 0.0001, respec-

Journal of Biomolecular Screening 8(2); 2003 www.sbsonline.org 213

 Kozikowski et al.
120
110
Percentage of Compound Remaining

Observed mean

Percentage of Compound Remaining

110 Predicted mean
100 Lower 95% confidence interval
Example of an acceptable
limit for compound loss
100
90
90

80
80

70 70

Frozen, never thawed observed means

60 Room temperature observed means 60
Freeze/thaw observed means

50
50
0 5 10 15 20 25 30
0 5 10 15 20 25 30
Time, marked by the number of freeze/thaw cycles Number of Freeze/thaws

FIG. 2. A plot of the statistical results for comparing the 3 different FIG. 3. The relationship between the percentage of compound re-
storage conditions. maining and the number of freeze/thaw cycles.

tively). The maximum number of freeze/thaw cycles can be deter-

mined from where the lower one-sided confidence interval inter-
sects a predetermined percentage of compound remaining. For
example, a maximum of 15 freeze/thaw cycles are allowed to
maintain compound losses of ~25%, as shown in Figure 3. The
amount of sample loss that can be tolerated is determined by con-
sidering the impact of less than pristine solutions on HTS data with
the costs of replacing compound solutions. Although there is no in-
dustry standard currently available to make this assessment, the
type of data presented in Figure 3 can be used to set retention limits
for compound storage.
CONCLUSION
This study provides the first published results to evaluate the ef-
fects of freeze/thaw cycling on solutions that were thawed in a low-
humidity environment and then exposed to atmospheric conditions
that mimic usage in the HTS lab environment. The lack of addi-
tional peaks in the HPLC chromatograms indicates that significant
compound degradation did not occur. An analysis of the data for
the percentage of compound remaining relative to the number of
freeze/thaw cycles can be used to determine retention limits for
compound storage to maintain a desired level of compound
integrity.
ACKNOWLEDGMENTS
We thank Barbara Hynd for suggesting the use of the pressure
canisters and D. Lee Garbutt and the machine shop at the Miami
Valley Laboratory for modifying the storage canisters.
REFERENCES
1. Nie D, Hilton A, Gosnell P: DMSO in HTS: the effect of water absorption and
evaporation. Paper presented at the 4th annual conference of the Society for
Biomolecular Screening, Baltimore, September 1998.
2. Heaton Z, Connell M, Holland S, Hughes R, Lifely R: The Importance of
Karl Fischer in quality control for the support of automated liquid stores. Pa-
per presented at the 7th annual conference of the Society for Biomolecular
Screening, Baltimore, September 2001.
3. Heaton Z, Ramirez-Molina C, Holland S, Hughes R, Lifely R: The effect of
sealing versus lidding plates on degradation and water uptake of compounds
held in DMSO at 4 °C. Paper presented at the 7th annual conference of the So-
ciety for Biomolecular Screening, Baltimore, September 2001.
4. Heaton Z, Holland S, Hughes R, Lifely R, Robb L: The effect of freeze thaw-
ing and storage time on degradation of compound at 4 °C under humidity con-
trol using experimental design. Paper presented at the 7th annual conference
of the Society for Biomolecular Screening, Baltimore, September 2001.
5. Cockburn-Price S, Adamson V: Industrializing HTS. Drug Discov Devel No-
vember/December 2001:81.
6. Darvas F, Karancsi R, Slegel P, Dorman G: Estimating stability for HTS li-
brary compounds. Genet Eng News 2000;20:30-31.
7. Aperghis PN, Battle C, Boulton M, Holland S, Lindsell T, O’Neill M, et al:
Automated liquid sample management for lead discovery. Paper presented at
Lab Automation ’99, San Diego, January-February 1999.
8. Kozikowski BA, Williams LE, Tirey D, Kuzmak B, Morand KL: The effect of
freeze/thaw cycles on the stability of compounds in DMSO. Paper presented
at the 7th annual conference of the Society for Biomolecular Screening, Balti-
more, September 2001.
9. Williams LE, Kozikowski BA, Burt T, Kuzmak B, Morand KL: High
throughput screening degradation study: determination of storage options
and retention limits. Paper presented at the 7th annual conference of the Soci-
ety for Biomolecular Screening, Baltimore, September 2001.

214 www.sbsonline.org Journal of Biomolecular Screening 8(2); 2003

 Effect of Freeze/Thaw Cycles on DMSO

10. DiverseSolutions, Version 4.1. Austin: University of Texas at Austin. Address reprint requests to:
11. Pearlman RS, Smith KM: Novel software tools for chemical similarity. In Sandra L. Nelson
Kubini H, Folkers G, Martin YC (eds): 3D-QSAR and Drug Design: Vol. 2. High Throughput Screening Laboratory
Ligand-Protein Interactions and Molecular Similarity. Dordrecht, the Neth- Procter & Gamble Pharmaceuticals
erlands: Kluwer, 1997:339-353. 8700 Mason Montgomery Road
12. Johnson RA, Wichern DW: Applied Multivariate Statistical Analysis. 2nd ed. Mason, OH 45040-8006
Englewood Cliffs, NJ: Prentice Hall, 1998.
13. Minitab for Windows, Release 12.23. State College, PA: Minitab, Inc. E-mail: nelson.sl@pg.com
14. SAS, Release 8.0. Cary, NC: SAS Institute.

Journal of Biomolecular Screening 8(2); 2003 www.sbsonline.org 215