Microfluidic approach to cocrystal screening of pharmaceutical parent compounds

Article abstract of DOI:10.1021/cg3011212

We describe a microfluidic approach to screen for the formation of cocrystalline solid forms of pharmaceutical parent compounds (PCs). Saturated solutions of PCs and of cocrystal formers dissolved in a variety of solvents are precisely metered in arrays o

Full text of DOI:10.1021/cg3011212

Article
pubs.acs.org/crystal
Microfluidic Approach to Cocrystal Screening of Pharmaceutical
Parent Compounds
Sachit Goyal,† Michael R. Thorson,† Geoff G. Z. Zhang,‡ Yuchuan Gong,*,‡ and Paul J. A. Kenis*,†
†Department of Chemical & Biomolecular Engineering, University of Illinois at Urbana−Champaign, 600 South Mathews Avenue,
Urbana, Illinois 61801, United States
‡Materials Science, Global Pharmaceutical R & D, Abbott Laboratories, 1401 Sheridan Road, NCR13-317B, North Chicago, Illinois
60064, United States
S
* Supporting Information
ABSTRACT: We describe a microfluidic approach to screen
for the formation of cocrystalline solid forms of pharmaceutical
parent compounds (PCs). Saturated solutions of PCs and of
cocrystal formers dissolved in a variety of solvents are precisely
metered in arrays of 48 wells to enable the combinatorial mixing
of all possible combinations. Key characteristics of the
microfluidic approach, including small quantities (∼240 μg/48
conditions), the ability to generate and screen 48 unique
conditions per chip, and the ability to identify solid forms on-
chip via Raman spectroscopy, enable solid form screening very
early in the drug development process. In contrast, current
approaches require on the order of ∼240 mg for 48 conditions,
thus delaying solid form screening to later stages of the drug development. Sequential screening experiments using caffeine as the
model compound were conducted to validate the on-chip approach reported here. Preliminary screens were executed to identify
conditions with the highest propensity for crystallization and to identify the cocrystal formers (CCFs) resulting in formation of
cocrystals via on-chip Raman spectroscopy. Next, the identified, promising conditions were replicated to confirm reproducibility
and consistency of the on-chip outcomes. Nine cocrystals of caffeine were identified in this way.
and/or the CCF. In such cases, crystallization of the PC by
itself may be thermodynamically more favorable, hence
preventing cocrystal formation.16 The use of nonstoichiometric
concentrations of PC and CCF, however, has been reported to
result in lower solubility of cocrystals, and hence an improved
chance for cocrystal formation.17,18 Other methods including
solid state grinding,19,20 neat and liquid assisted grinding,19,21
solvent drop grinding,22 slurry conversion (solution-mediated
phase transformation),16 melt crystallization, differential
scanning calorimetry,23 and hot stage microscopy24 have
avoided some of the problems associated with direct
crystallization approaches through thermodynamics. However,
these methods still require large amounts of PC for the same
number of experiments that one would conduct when using a
direct crystallization approach. In this study, we extensively
screen for cocrystals using much smaller amounts of PC, by
employing microfluidic platforms for direct crystallization of
saturated solutions of CCF and PC, often in nonstoichiometric
molar concentrations, in a variety of solvents.
1. INTRODUCTION
Cocrystals are multicomponent assemblies held together by
reversible, noncovalent interactions.1 Cocrystallization offers a
convenient way to alter the physical properties (e.g., dissolution
rate, melting point, solubility, hygroscopicity) of the solid form
of pharmaceutical parent compounds (PCs) without affecting
their chemical identity and hence their therapeutic effects.1−5
Cocrystal screening involves investigating a variety of cocrystal
formers (CCFs), typically compounds with one or more acidic,
basic, or nonionizable groups, under a variety of crystallization
conditions, including choice of solvent, pH, temperature, CCF
concentration, PC concentration, and PC-to-CCF ratio.5−9
Robotic systems have been implemented to automate and
thereby speed up these screening processes.10 However, these
systems require on the order of 0.5 g of PC to carry out a
screen using 96-well plates, thereby preventing solid form
screening in the early stages of drug development due to
limited availability of PC at that time.
Typically, cocrystal screening involves direct crystallization
(crystallization from clear solutions of PC and CCFs) utilizing
approaches such as reactive crystallization,9,11 temperature
gradients,12 solvent evaporation,13 antisolvent addition, spray
drying,14 and sonochemical crystallization.15 However, consid-
erable differences in solubility of the components of a cocrystal
hamper cocrystal formation due to too low solubility of the PC
Progress in microfluidics over the last few decades has
resulted in the development of large scale integrated micro-
Received: August 5, 2012
Revised: October 22, 2012
Published: October 23, 2012
© 2012 American Chemical Society
6023
dx.doi.org/10.1021/cg3011212 | Cryst. Growth Des. 2012, 12, 6023−6034

Crystal Growth & Design
Article
Figure 1. (a) View of cross section of four microfluidic crystallization wells (2 × 2 array) of 48 (4 × 12 array)-well multiplexed platform depicting
the layered assembly of the platform. The platform comprises the PDMS fluid and control layers sandwiched between layers of cyclic olefin
copolymer (COC) on top and bottom to minimize solvent loss, provide rigidity, and enable Raman compatibility. The numbers 1, 2, 3, and 4 refer to
different sets of control valves. (b) Perspective view of four microfluidic crystallization wells (2 × 2 arrays) depicts parent compound and cocrystal
former chambers in the fluid layer and pneumatic control lines and valves in the control layer. (c) Top view of the microfluidic platform depicts the
control layer aligned over the fluid layer illustrating the function of different valves in the filling and mixing of solutions on-chip.
fluidic platforms allowing precise control over mixing and on-
chip metering of solutions, on-chip detection and analysis of
chemical compounds, and combinatorial mixing of very small
sample sizes, thereby enabling screening of many conditions
with limited material for different applications.25−29 Specifically,
microfluidic platforms find application in diverse fields such as
biological studies (e.g., gene expression,30 single cell analysis,31
digital PCR,32 DNA sequencing33 and analysis34), point-of-care
and clinical diagnostics,35−37 and protein/pharmaceutical
crystallization.8,28,38,39 Free interface diffusion (FID) based
microfluidic platforms,25,28,39 droplet based microfluidics,40−43
evaporation-based platforms,44 and slip-chip approaches45,46
have been developed to crystallize proteins as well as salts and
polymorphs of PCs, and they have been used to study
crystallization kinetics. However, most of the microfluidic
crystallization tools reported to date have two major
limitations: (1) the materials the platforms are made of are
often incompatible with most of the common organic
solvents,47 and (2) they are not amenable to on-chip solid
form analysis. Typically, manual harvesting of crystals is needed
for analysis using techniques such as X-ray diffraction, Raman
spectroscopy, solid state NMR, and/or IR spectroscopy (FTIR,
near IR). Recently, we reported on a microfluidic platform for
pharmaceutical salt screening that is compatible with water,
alcohols, and dimethylsulfoxide (DMSO) and allows for on-
chip solid form analysis via Raman spectroscopy.28 Use of gold-
coated glass substrates was needed to suppress excessive
background in Raman spectra, resulting in acceptable signal-to-
noise ratios.
2. EXPERIMENTAL SECTION
Chip Assembly. The crystallization platform comprises thin
polydimethylsiloxane (PDMS, General Electric RTV 615, Part A/B)
fluid and control layers sandwiched between a cyclic olefin copolymer
(COC, 6013 grade, TOPAS Advanced Polymers) support layer and a
COC substrate layer. The fluid layer (FL) and the control layer (CL)
molds were created via patterning of SU-8 2050 photoresist
(MicroChem) onto silicon wafers (University wafer) to a thickness
of 50 μm, each using standard photolithography.26,48 The master
molds were evaporated with a monolayer of tridecafluoro-1,1,2,2-
tetrahydrooctyl)trichlorosilane (Gelest). CL (80 μm) and FL (70 μm)
were obtained by spin coating 5:1 A:B, PDMS and 15:1 A:B, PDMS at
1100 and 1200 rpm respectively. The CL and the FL were partially
cured on a digital hot plate/stirrer (Dataplate 730 series, Barnstead
Thermolyne) at 80 °C for approximately 5 and 15 min respectively. A
2 mil (50 μm) COC sheet was bonded irreversibly to the top of the
CL mold after treatment with an oxygen plasma in a plasma cleaner
(Harrick Plasma) for 1 min. The COC-CL assembly was then heated
at 80 °C for 10 min on a hot plate. The COC-CL assembly was
carefully lifted off the CL master. The embossed side of the PDMS CL
was covered with Scotch removable tape (3M) and through holes were
drilled (Dremel 300 series drill with a 750 μm McMaster-Carr drill bit)
to create inlets for vacuum actuation. The COC-CL assembly was
manually aligned and placed on the FL under an optical microscope
(Leica MZ6), followed by heating at 80 °C for 20−30 min on a hot
plate to bond the layers. After the three-layer assembly was lifted off
the FL master, the FL side was covered with Scotch removable tape
(3M). Through-holes were drilled at the locations of the inlets and
outlets in the FL. Subsequently, a 3−5 mm thick PDMS layer (5:1
A:B) was irreversibly bonded to the top of the three-layer (COC-CL-
FL) assembly covering the FL and the CL inlets, after treatment with
an oxygen plasma for 1 min. Through-holes were punched into the
thick PDMS layer with a 20-gauge needle (BD PrecisionGlide)
through the holes already drilled at the CL and the FL inlets. The
complete assembly, as schematically shown in Figure 1a,b, was then
reversibly bonded to a flat 2 mil COC substrate prior to setting up of
the pharmaceutical solid form screening experiment. Since negative
pressure was used for the operation of the valves incorporated in the
device, the FL did not need to be bonded irreversibly to the COC
substrate.27,28,49
In this study, we report a microfluidic approach that (1)
enables extensive and systematic cocrystal screening with
limited quantities of PCs (∼240 μg per 48 conditions) via
combinatorial mixing; (2) is compatible with solvents that are
often used in pharmaceutical crystallization, i.e., water, alcohols,
and acetonitrile; and (3) allows for on-chip solid form analysis
using Raman spectroscopy. We validate this microfluidic
approach through cocrystal screening of caffeine.
6024
dx.doi.org/10.1021/cg3011212 | Cryst. Growth Des. 2012, 12, 6023−6034

Crystal Growth & Design
Article
Figure 2. Tiled optical micrographs (taken 4−12 h after mixing) for an on-chip screen of cocrystal solid forms of the PC caffeine from the first phase
of screening. (a) The rows are filled with PC in acetonitrile (1), acetonitrile/water (v/v = 1/1) (2), methanol (3), and ethanol (4) respectively. The
columns are filled with the following CCFs in solvent: 2,4-dihydroxy benzoic acid in acetonitrile (A), acetonitrile/water (v/v = 1/1) (B), methanol
(C), and ethanol (D) respectively; 3,5-dihydroxy benzoic acid in acetonitrile (E), acetonitrile/water (v/v = 1/1) (F), methanol (G), and ethanol (H)
respectively; 2,5-dihydroxy benzoic acid in acetonitrile (I), acetonitrile/water (v/v = 1/1) (J), methanol (K), and ethanol (L) respectively. (b) The
rows are filled with caffeine in acetonitrile (1), acetonitrile/water (v/v = 1/1) (2), methanol (3), and ethanol (4), respectively. The columns are filled
with the following CCFs in solvent: 1-hydroxy 2-naphthotic acid in acetonitrile (A), acetonitrile/water (v/v = 1/1) (B), methanol (C), and ethanol
(D), respectively; 2-hydroxy 1-naphthotic acid in acetonitrile (E), acetonitrile/water (v/v = 1/1) (F), methanol (G), and ethanol (H), respectively;
3-hydroxy 2-naphthotic acid in acetonitrile (I), acetonitrile/water (v/v = 1/1) (J), methanol (K), and ethanol (L), respectively. Colored boxes
indicate the parent compound (green) and the cocrystal former (red) chambers and arrows indicate the different sets of control valves.
Preparation of Parent Compound and Cocrystal Former
Solutions. Caffeine and all CCFs were obtained from Sigma-Aldrich
and used as received. Five classes of CCFs were used: (1) carboxylic
acids: acetic acid, formic acid, and trifluoroacetic acid; (2) dicarboxylic
acids: oxalic acid, maleic acid, malonic acid, citric acid, glutaric acid,
and adipic acid; (3) hydroxy or dihydroxy benzoic acids: 2-hydroxy
benzoic acid, 4-hydroxy benzoic acid, 2,5-dihydroxy benzoic acid, 2,3-
dihydroxy benzoic acid, 2,4-dihydroxy benzoic acid, and 3,5-dihydroxy
benzoic acid; (4) amines/amides: indole, ethylenediamine, and
saccharin; and (5) hydroxy naphthotic acids: 1-hydroxy 2-naphthotic
acid, 2-hydroxy 1-naphthotic acid, and 3-hydroxy 2-naphthotic acid.
The following procedure was used for the preparation of the
saturated solutions of caffeine and CCFs. Fifty milligrams of caffeine or
50−200 mg of CCF was dispensed into a 7 mL glass vial (Kimble/
Chase). Solvents (acetonitrile, acetonitrile/water (v/v = 1/1), ethanol,
or methanol) were pipetted into the glass vial in increments of 200 μL
till the solid was completely dissolved. After each addition, the mixture
was vortexed (Maxi Mix II, Barnstead/Thermolyne), sonicated for 2
min (Branson 2510), and then allowed to sit for about 5 min. The
solubility of caffeine was estimated to be 10, 25, 7, and 8 mg/mL in
acetonitrile, acetonitrile/water (v/v = 1/1), ethanol, and methanol,
respectively. Extra solid was added in each of the vials at the end to
ensure the solution is saturated with caffeine or CCF. The suspension
was filtered, and the supernatant was used in the experiments.
On-Chip Crystallization. Fluidic routing and mixing was
controlled via an array of normally closed valves (four sets)
incorporated in the control layer (Figure 1b,c).27 Microcentrifuge
tubes (VWR International) were filled with PC or CCF solutions and
then connected to the inlets on the platform via tubing (30 AWG thin-
walled PTFE, Cole Parmer) (see Figure S1 in the Supporting
Information). The PC solutions are introduced horizontally by
actuating the valve sets 1 and 2 (Figure 1b), and applying vacuum
suction at the outlet of the row being filled. The valve sets 1 and 2 are
then closed such that the PC solutions in the PC and CCF chambers
adjacent to each other are isolated. The CCF chambers are then
purged with acetonitrile by actuating valve set 3 (Figure 1b) to flush
PC solution from the CCF chambers, followed by introduction of the
CCF solutions via application of vacuum suction at the outlet of the
column being filled. Subsequently, valve set 3 is closed to lock up the
CCF solutions in their chambers. Next, valve set 2 is opened for about
30 min (Figure 1b) to allow mixing of the combinatorial combinations
of PC and CCF solutions confined in adjacent chambers. The mixing
6025
dx.doi.org/10.1021/cg3011212 | Cryst. Growth Des. 2012, 12, 6023−6034

Crystal Growth & Design
Article
time can be reduced to about 10 min by repeated actuation of valves 4
(Figure 1b). After the solutions were fully mixed, all tubing for
pneumatic control and fluidic flow was disconnected and the chip was
sealed with Crystal Clear Tape (Hampton Research HR4−511) to
prevent solvent evaporation. The chip can then be moved to the solid
form analysis stations. Visualization of the whole procedure of on-chip
filling and mixing in 24-wells of a 48-well microfluidic platform is
available as a movie in the Supporting Information.
On-Chip Solid Form Analysis. Throughout the mixing and the
following 2−12 h, the wells were periodically monitored for solid
formation using an automated imaging setup comprised of an optical
microscope (Leica Z16 APO) equipped with an autozoom lens (Leica
10447176), a digital camera (Leica DFC280), and a motorized X-Y
stage (Semprex KL66) controlled by Image Pro Plus 7.1 software
(Media Cybernetics). Images of each well were acquired every 10 min
by moving the automated motorized stage in a sequential fashion from
well to well. Periodically birefringent images of the wells were taken
using crossed polarizers.
After the completion of the on-chip crystallization experiment, the
images of the crystalline solids were captured using a stereomicroscope
(Leica MZ12.5) equipped with a digital camera (Leica DFC295) and a
crossed polarizer. The identity of the crystalline solids was determined
by Raman spectroscopy (Renishaw mircoPL/Raman microscope).
The Raman spectrometer equipped with a 785 nm excitation source
(Renishaw NIR 100 mW diode laser) was connected to an upright
microscope (Leica DM2500M). The microfluidic chip was placed in
the sample holder and individual wells in the microfluidic chip were
centered in the bright field mode using low magnification (5×),
followed by use of higher magnifications (20× and/or 50×). The laser
is then switched on and set at 10% laser power and the Raman spectra
of individual crystals were collected in the range of 500−1700 cm−1 by
focusing to a spot size of ∼5 μm at 50× magnification with a long
working distance objective in the dark field mode. Data collection was
carried out at a spectral resolution of ∼0.5 cm−1 at 1800 gratings/mm,
with the exposure time set to 40 s, and each spectrum was averaged
over two accumulations.
Off-Chip Crystallization. The cocrystal screening experiments
were also conducted off-chip by mixing the same saturated solutions of
caffeine and CCFs at the same volumetric ratios (150 μL of CCF
solution and 300 μL of caffeine solution) in 1-mL glass vials (Kimble/
Chase). The mixture of solutions was vortexed, sonicated for 5 min,
and then sealed by capping the glass vials to avoid solvent evaporation.
The appearance of solids in the solutions was monitored over a period
of 12−72 h. The solids formed were collected by centrifugation
(MiniSpin plus with Rotor F-45-12-11, Eppendorf).
3. RESULTS AND DISCUSSION
3.1. Design and Operation of the Microfluidic
Platform. Chip Design. The platform used here is an
adaptation of the design of a microfluidic platform we reported
earlier.28 The microfluidic platform used here has 48 wells
(Figure 2), where each well is comprised of an isolated chamber
for a PC solution and an adjacent chamber for a CCF solution
(Figure 1b). Each well is isolated from the rest of the wells
using a series of normally closed valves27,49 and can host a
separate crystallization experiment (Figure 1). Here, we
conduct cocrystal screening of 48 unique conditions using a
single chip: four caffeine solutions with 12 CCF solutions. Each
PC chamber has a size of ∼90 nL (combined volume of the PC
and the CCF chamber). The dimensions of the chambers in the
chip were designed such that the solutions in the adjacent
chambers can be mixed on-chip by diffusion within 30 min as
estimated by Fick’s law, t = x2/(4D), where x is the combined
length along which the two solutions are mixing and D is the
diffusivity of the diffusing species.28 Therefore, each condition
screened only requires 0.90−4.5 μg of a PC with a solubility of
10−50 mg/mL in the crystallization solvent. This microfluidic
platform reduces the sample requirement by a factor of more
than 1000 compared to the traditional manual screening which
typically requires ∼5 mg of PC per condition. Therefore, the
platform developed here enables extensive solid form screening,
including cocrystal screening of a given PC, at the early stages
of drug development when only limited quantities of PC are
available.
PCs of interest usually have higher molecular weights and
lower solubilities than the commonly used CCFs. We were able
to deliver the PCs and CCFs with different solubilities on-chip
in approximately equivalent molar ratios by using PC chambers
that are twice as large as the CCF chambers. In addition, PCs
and CCFs are frequently delivered in organic solvent and water,
respectively, due to their adequate solubility in those solvents.
The smaller volume of the CCF chamber, to a certain extent,
prevents the precipitation of PCs dissolved in organic solvents
when brought in contact with aqueous solutions of CCFs due
to the antisolvent effect of water.
Chip Operation. To fill the microfluidic chips we reported
previously,28 a droplet (1−2 μL) of the PC or salt former
solution had to be pipetted on the inlet port of each fluid line,
and introduced into the individual chambers upon actuation of
the appropriate valve sets and gentle suction at the appropriate
fluid outlets. Evaporation of the solvent leading to solute
precipitation and even clogging of the inlets is a drawback of
this sample loading method. Here, we modified the platform to
allow for a more efficient solution loading procedure. The PC
and CCF solutions stored in individual off-chip containers
connected to the inlets on the platform via tubing were
introduced to the chambers on-chip upon actuation of the
appropriate set of valves while simultaneously applying gentle
suction at corresponding fluidic outlets (vide supra). This
closed system approach minimizes solvent loss.
Off-Chip Solid Form Analysis. PXRD patterns and Raman
spectra of the solids collected in the off-chip experiments were
collected and used as reference in the solid form identification.
Raman Spectroscopy. The crystals were placed on a gold-coated
microscope glass slide and centered in the bright field mode of the
microscope using low magnification (5×), followed by use of higher
magnifications (20×). The Raman spectra of the crystal were collected
using a 785 nm laser at 10% laser power in the range of 500−1700
cm−1 by focusing to a spot size of ∼20 μm at 20× magnification. Data
collection was carried out at a spectral resolution of ∼0.5 cm−1 at 1800
gratings/mm, with the exposure time set to 40 s, and each spectrum
was averaged over two accumulations. The gold coated glass slides
were prepared by evaporating a 20-nm layer of chromium followed by
a 200-nm layer of gold using E-beam evaporation system (Temescal
six pocket E-Beam Evaporation System).
Powder X-ray Diffraction. The solid forms crystallized off-chip was
analyzed using a D5000 diffractometer (Siemens/Bruker D-5000)
using Cu Kα radiation (λ = 1.54059 Å). The sample was loaded onto a
sample holder and leveled with a microscope glass slide. Data
collection was carried out in the 2θ range 5−35° with a step size of
0.02°. The copper anode tube (1.5 kW fine focus) voltage and
amperage were set at 40 kV and 40 mA, respectively. The instrument
was controlled by a computer with the Siemens DIFFRAC plus and
the data were analyzed using MDI Jade 9+.
The modified, closed sample loading method further reduced
the sample requirement. Previously droplets of PC solutions
prepared in volatile solvents such as methanol were pipetted
over the inlets. These droplets were exposed to the ambient
conditions for a short period of time before being pulled into
the chambers in the microfluidic platform. However, a small
droplet evaporates rapidly owing to a large surface exposed to
the ambient conditions. In practice, we frequently had to add
6026
dx.doi.org/10.1021/cg3011212 | Cryst. Growth Des. 2012, 12, 6023−6034

Crystal Growth & Design
Article
Table 1. Cocrystal Formation Data (On-Chip and off-Chip) for the Parent Compound (PC) Caffeine from the First Phase (“×”
Indicates Cocrystal Formation and “−” Indicates the Absence of Precipitation)
a
Cocrystal formers: 4- and 2-hydroxy benzoic acid: “4B” and “2B”; 2,5-, 3,5-, 2,3-, and 2,4-dihydroxy benzoic acids: “2,5B”, “3,5B”, “2,3B”, and
“2,4B”, respectively; saccharin “S”; acetic, formic, and trifluoroacetic acids: “AA”, “FA”, and “TFA”, respectively; 3-hydroxy 2-naphthotic, 2-hydroxy
1-naphthotic, and 1-hydroxy 2-naphthotic acids: “3,2N”, “2,1N”, and “1,2N”, respectively; adipic, citric, glutaric, maleic, malonic, and oxalic acids:
b
“Ad”, “C”, “G”, “M1”, “M2”, and “O”, respectively; ethylene diamine “ED”, and indole “I”. Solvents: acetonitrile “A”; acetonitrile/water (v/v = 1/1)
“AW”; methanol “M”; ethanol “E”.
large droplets of solutions to the inlets to have enough solution
to be filled into the wells despite evaporation. Therefore, the
sample required for a crystallization experiment using these
microfluidic platforms was determined by the volume of the
droplet required to fill the chambers instead of the size of the
chambers in the platform. The modified, closed sample delivery
method avoids solvent loss, thus removing the limitation on the
reduction of the sample requirement. In addition, this design
may allow us to further automate sample delivery.
Solvent Compatibility. Lee et al. studied the interaction of
PDMS with a wide range of solvents. They ranked the solvents
studied based on their solubility in PDMS and the degree to
which they swell PDMS.47 On the basis of this prior work, we
expect that the PDMS-based platforms are compatible with
solvents such as water, dimethylsulfoxide, dimethylformamide,
and various alcohols. However, the PDMS will absorb these
solvents and may swell by up to 9% in volume. Additionally, the
solvents may evaporate from the surface of PDMS-based
microfluidic chips, as PDMS is permeable to air and solvent
vapor.47 In crystallization chips, this solvent loss causes an
undesired increase of supersaturation levels, which reduces the
maximum length of experiments, typically to less than 2 h.
To minimize solvent absorption by bulk PDMS we reduced
the total thickness of the PDMS fluid and control layer to ∼150
μm. To minimize solvent loss due to the evaporation, we
sandwiched the PDMS layers between two 50 μm COC layers.
COC is less permeable to the solvents of interests, is chemically
more resistant to the solvents commonly used in crystallization,
and it increases by less than 3% by weight when in contact with
the solvents used in this study.50 After filling, the inlets and
outlets on the microfluidic platform were covered by Crystal
Clear tape to further minimize solvent loss. These improve-
ments allowed the chips to be used for much longer
crystallization screening experiments (4−12 h) using a broad
range of solvents, including volatile alcohols such as methanol,
ethanol, isopropanol, trifluoroethanol, as well as acetonitrile.
Strong nonpolar organic solvents such as acetone, tetrahy-
drofuran, ethyl acetate, and hexanes, cause extensive swelling of
PDMS, which results in significant changes in chamber volumes
and thus poorly controlled crystallization conditions. Micro-
6027
dx.doi.org/10.1021/cg3011212 | Cryst. Growth Des. 2012, 12, 6023−6034

Crystal Growth & Design
Article
fluidic platforms comprised of more solvent resistant materials
will be required to enable on-chip crystallization involving these
strong nonpolar organic solvents.
from this cocrystal screen of caffeine: Solid forms were
observed in 63 out of the 96 unique conditions screened in
these two chips. Table 1 summarizes the results from all on-
chip and off-chip experiments. Crystalline solids were observed
in 25% of the conditions in both on-chip and off-chip
experiments.
Raman Compatibility. The analytical techniques commonly
used in crystal form identification include powder X-ray
diffraction, single crystal X-ray diffraction, solid state NMR,
IR spectroscopy, and Raman spectroscopy.51−54 Raman
spectroscopy is preferred for on-chip solid form identification
because it is amenable to automated high throughput screening
and can achieve high spatial resolution.53 In our prior work
using PDMS-based chips, we were able to collect reasonable
Raman spectra from on-chip crystals, but only after integration
of a gold-coated bottom substrate to suppress excessive
background signal from the PDMS and fluorescence from the
glass substrate.28 Here, we significantly reduced the background
signal and enhanced the signal-to-noise ratio by reducing the
thickness of the PDMS layers to <150 μm and by incorporating
thin sheets of Raman transparent COC. The change to COC-
based chips also simplified chip fabrication and handling, and
made microscopic visualization of crystals easier. Raman spectra
of individual COC and PDMS layers, as well as an assembled
but empty chip, are provided in the Supporting Information
(Figure S2). In addition, the process of background correction
of on-chip Raman spectra is provided there (Figure S3).
In summary, the microfluidic platform reported here allows
us to conduct 48 crystallization experiments in a single chip
with minimal solvent loss and enhanced solvent and Raman
compatibility. These characteristics, as well as reduced sample
requirements, are critical to the solid form screening (especially
by cocrystallization) at the early stages of drug development
when only limited quantities of PC are available and knowledge
of the properties of the compounds is limited.
3.2. Cocrystal Screening of Caffeine. Caffeine has been
reported to form cocrystals with a wide range of
CCFs.6,16,45,55−57 These CCFs include carboxylic acids,6
dicarboxylic acids,16,57,58 hydroxy or dihydroxy benzoic
acids,45 nitro benzoic acids,55 hydroxy naphthotic acids,56,59
and anilines.55 We tested 21 CCFs representing dicarboxylic
acids, hydroxy or dihydroxy benzoic acids, hydroxy naphthotic
acids, carboxylic acids, and amines/amides. Saturated solutions
of caffeine and each of the 17 solid CCFs were prepared in four
solvents (acetonitrile, acetonitrile/water (v/v = 1/1), methanol,
and ethanol). Four CCFs are liquid at the ambient conditions
and were used directly.
Off-Chip Experiments (Glass Vials). In the off-chip experi-
ments, crystals appeared in the presence of 10 CCFs including
2,4-dihydroxy benzoic acid (2,4B), 3,5-dihydroxy benzoic acid
(3,5B), 2,5-dihydroxy benzoic acid (2,5B), 1-hydroxy 2-
naphthotic acid (1,2N), 2-hydroxy 1-naphthotic acid (2,1N),
3-hydroxy 2-naphthotic acid (3,2N), 2,3-dihydroxy benzoic acid
(2,3B), 2-hydroxy benzoic acid (2B), oxalic acid (O), and
ethylene diamine (ED). Mixing solutions of caffeine with those
of the other CCFs including 4-hydroxy benzoic (4B), maleic
(M1), malonic (M2), glutaric (G), citric (C), adipic (Ad), acetic
(AA), formic (FA), and trifluoroacetic (TFA) acids as well as
indole (I) and saccharin (S) did not result in precipitation
though most of the CCFs used in this study (19 of 21)6 have
been reported to form cocrystal with caffeine. All the glass vials
were sealed after solutions were mixed to prevent solvent
evaporation. Therefore, crystallization was only possible when
either the cocrystal or the individual components became
supersaturated in the initial caffeine/CCF solution mixture.
On-Chip Experiments. Use of the microfluidic chips led to
similar outcomes. Crystalline solids were observed in the
presence of the same group of CCFs in the on-chip
experiments as in the off-chip experiments. The fact that no
solid formation was observed for many conditions on-chip in
the 4−12 h post mixing time period suggests that solvent loss
indeed is minimal in the chips reported here. The levels of
supersaturation of either component (PC or CCF) and of the
intended cocrystal are not sufficient to induce crystallization,
and the lack of solvent loss prevents further increase of the
supersaturation after complete mixing has been achieved.
On the basis of the excellent agreement between on-chip and
off-chip cocrystal screening experiments, we expect the
microfluidic chips (∼90 nL wells, mixing via diffusion) to
deliver similar results when the same screen is performed off-
chip (in 1 mL vials, mixing via direct mixing) for a given
compound. Furthermore, the significant reduction in the
sample requirement gives the microfluidic approach a
significant advantage in solid form screening in the early stages
of drug development.
First Phase of Screening. Extensive screening was
conducted to identify promising crystallization conditions, i.e.,
those combinations of solvents and CCFs that exhibit the
highest propensity to form crystalline solids of a given PC, and
to identify those CCFs resulting in the formation of cocrystals
on-chip via Raman spectroscopy. The saturated solutions of
caffeine were introduced into the PC chambers in the
microfluidic chips horizontally, while those of the CCFs (or
liquid CCFs) were introduced into the CCF chambers
vertically. The combinatorial mixing of the solutions of each
caffeine/CCF pair allowed screening of 288 (4 caffeine
solutions × 72 CCF solutions/liquids) crystallizations using a
total of 6 microfluidic chips. On-chip crystallization experi-
ments were monitored for up to 12 h postmixing. For
comparison, the same crystallization experiments were repeated
off-chip, at a larger scale, in 1 mL vials by direct mixing of the
caffeine/CCF solutions.
Solid Form Analysis. Cocrystal screening requires scientists
not only to identify the conditions at which crystalline solids
are formed, but also to determine the nature of the solids. The
latter is important because the PC and/or CCFs may crystallize
by themselves if they have a lower solubility in the solvent
mixture after the solutions are mixed together, especially when
the PC and CCF initially were dissolved in solvents with very
different properties. Besides, either component may form a
solvate or a hydrate in the solvent mixture. For these reasons,
one cannot conclude that the total number of conditions that
produces a solid form within a screen equals the incidence of
cocrystals. So, to determine the number of cocrystals formed,
individual solid forms need to be analyzed. Here we first
located crystalline solids in the various chambers of the
microfluidic chips using bright field microscopy, and then we
further analyzed the solids using Raman spectroscopy.
The Raman spectra of caffeine, all CCFs, and all crystalline
solids isolated from the off-chip experiments were collected in
the range of 500−1700 cm−1 for comparison with the on-chip
Figure 2 shows the tiled optical micrographs of the two-
microfluidic crystallization chip with the highest success rate
6028
dx.doi.org/10.1021/cg3011212 | Cryst. Growth Des. 2012, 12, 6023−6034

Crystal Growth & Design
Article
results. In addition, the Raman spectra of the individual PDMS
and COC layers, as well as of an assembled, empty chip were
recorded (Figure S2) for background subtraction. An example
of background subtraction and subsequent comparison of the
Raman spectrum of an individual cocrystal with the spectra of
caffeine alone and the corresponding CCF is provided in the
Supporting Information (Figure S3). We determined the
identity of the crystalline solids corresponding to each CCF
by comparing their PXRD patterns to those reported in the
literature, since the Raman spectra of all targeted cocrystals are
not available. Specifically, the PXRD patterns for the cocrystals
grown in glass vials in acetonitrile or in the mixture of
acetonitrile/water (our off-chip experiments) were collected
and found to be in good agreement with the PXRD patterns for
the cocrystals reported in the literature suggesting that indeed
the intended cocrystals are formed in our off-chip experiments.
An example of such a comparison of PXRD patterns is provided
in Figure S4 in the Supporting Information.
We did not take PXRD data on every solid form obtained
from every PC and CCF solvent combination expected to yield
a specific PC−CCF cocrystal. Instead, we confirmed the
identity of those cocrystals by comparing their Raman spectra
to the Raman spectrum of the cocrystal whose identity we
already had confirmed using PXRD (see above). The observed
excellent agreement between all Raman spectra for a given PC
and CCF combination suggests that crystallization of a targeted
cocrystal is independent of the solvent mixture it was grown
from.
Figure 3. Raman spectroscopy data of the nine-caffeine cocrystals
formed off-chip (dashed lines) and in a 48-well diffusional mixing chip
(solid lines), empty microfluidic chip, and caffeine (off-chip).
For the off-chip experiments, all crystalline solids were
determined to be the targeted cocrystals of caffeine and the
corresponding CCF, except those obtained in the presence of
ethylene diamine (ED). Comparison of the PXRD pattern
collected for the solid forms crystallized in the presence of ED
with the PXRD for as-received caffeine powder reveals that only
crystals of the PC caffeine formed when ED was present.
Similarly, Raman spectra of the crystalline solids formed in the
on-chip experiments were in agreement with the reference
Raman spectra from the cocrystals obtained in the off-chip
experiments. Figure 3 shows representative examples of Raman
spectra of corresponding solid forms obtained on-chip and off-
chip. The unique peaks in the Raman spectra distinguished
different solid forms from each other and from caffeine and
confirmed that on-chip and off-chip data were in agreement.
These results confirm that nine cocrystals of caffeine were
successfully prepared via combinatorial mixing of PC and CCF
solutions using the microfluidic chips reported here.
Like in the off-chip experiments, the solid forms obtained
from on-chip conditions that contained ethylene diamine
turned out to be only crystals of caffeine itself, instead of the
intended cocrystal as confirmed by Raman Spectroscopy
(Figure S5). The precipitation of caffeine on-chip as well as
off-chip suggests that caffeine has low solubility in ethylene
diamine and ED behaves as an antisolvent when added to the
solution of caffeine.
crystallization outcomes: (1) Differences in the length of the
crystallization experiments. The off-chip experiments often
extended for 12−72 h, whereas most on-chip experiments
were limited to 12 h or less due to considerable solvent loss
encountered in the crystallization wells beyond 12 h. Beyond
12 h of experiments on-chip, we could not rely on the on-chip
crystallization outcomes because then crystallization of the
intended cocrystal or PC/CCF might result due to solvent
evaporation (not the scope of this study) as well as reactive
crystallization. The longer duration of the off-chip experiments
may have led to a few more conditions resulting in solid forms
resulting from slow nucleation kinetics of some of the
cocrystals. Here, this is a possible explanation for 3 of the 9
cases in which solid forms were observed off-chip, but no solids
were crystallized on-chip. (2) Differences in the method of mixing
of the PC and CCF solutions. In the off-chip experiments, the
solutions are brought in contact by pipetting one solution onto
the other, which can result in very high supersaturation levels
on contact. However, the rate of change of supersaturation level
is very rapid due to instantaneous mixing attained in off-chip
experiments resulting from vortexing and sonication of
solutions. This convective, direct mixing process led to
screening of a wide range of supersaturation levels rapidly
and thereby might result in instantaneous precipitation of
solids. In the on-chip experiments, the two solutions are
brought into contact with each other only at the interface of the
PC and CCF chamber by opening a valve, which is followed by
a slow diffusive mixing process (no convection). The
supersaturation level attained on-chip is defined by the relative
size of the chambers or the volumetric ratios in which the
solutions are mixed and the initial concentration of both the
solutions.39 On-chip diffusional mixing may lead to local
supersaturation levels that are lower than those achieved by
A closer examination of the results in Table 1 revealed that
on-chip crystallization did not perfectly replicate the outcomes
of the off-chip experiments for all conditions. Cocrystals were
formed in a total of 82/288 conditions out of which 64
conditions produced cocrystals on-chip as well as off-chip, while
9 conditions yielded cocrystals uniquely in the microfluidic
chips and the remaining 9 conditions resulted in cocrystals
uniquely in the off-chip experiments. A number of factors can
explain these differences in the on-chip and off-chip
6029
dx.doi.org/10.1021/cg3011212 | Cryst. Growth Des. 2012, 12, 6023−6034

Crystal Growth & Design
Article
Table 2. Cocrystal Formation Data (On-Chip and off-Chip) for the Parent Compound (PC) Caffeine from the Second Phase
(“×” Indicates Cocrystal Formation and “−” Indicates the Absence of Precipitation)
a
Cocrystal formers: 3-hydroxy 2-naphthotic acid, 2-hydroxy 1-naphthotic acid, and 1-hydroxy 2-naphthotic acids: “3,2N”, “2,1N”, and “1,2N”,
respectively; ethylene diamine “ED”; 2- and 4-hydroxy benzoic acids: “2B” and “4B”, respectively, 3,5-, 2,5-, 2,3-, and 2,4-dihydroxy benzoic acids:
b
“3,5B”, “2,5B”, “2,3B”, and “2,4B”, respectively; and oxalic acid “O”. Solvents: acetonitrile “A”, acetonitrile/water (v/v = 1/1) “AW”, and methanol
“M”.
direct mixing of solutions (off-chip) and thereby might not
result in crystallization on-chip when the supersaturation level
attained on-chip is low enough to lead to crystallization.
However, diffusional mixing on-chip allows better control over
the rate of change in supersaturation and might allow enough
time at each supersaturation level, which may be conducive to
crystal nucleation in some cases. Either mixing process can lead
to more solid forms. Determining which crystallization
outcome is due to differences in the on-chip and off-chip
mixing processes is hard unless, for example, instantaneous
precipitation was observed in the off-chip experiment, or in the
on-chip experiment nucleation and growth is observed close to
the valve location soon after the onset of mixing. (3) Insufficient
amount of one component (typically PC due to lower solubility
compared to the CCFs) to generate crystals of sufficient size
(depletion effect). Each PC chamber in the microfluidic chip has
a size of ∼90 nL. With the solubility of caffeine estimated as 10,
25, 7, and 8 mg/mL in acetonitrile, acetonitrile/water (v/v = 1/
1), ethanol, and methanol, respectively, about 0.63−2.25 μg of
caffeine is present in each of the microfluidic crystallization
wells. Specifically, when the caffeine was dissolved in methanol
or ethanol (only 0.63−0.72 μg caffeine present per well),
crystallization of its intended cocrystal might have been difficult
because of a low level of supersaturation reached after mixing.
This might explain at least 4 of the 9 cases in which
crystallization was observed off-chip but no solids formed on-
chip. Increasing the size of the chambers might eliminate
discrepancies in on-chip crystallization outcomes resulting due
to delivery of an insufficient amount of one component on-
chip. (4) Imperfect formulation of crystallization conditions on-
chip. Barring a pipetting error little can go wrong in an off-chip
experiment. In the on-chip experiments at times, bubbles can
get trapped in a chamber or a chamber may be deformed due to
a malfunctioning valve. Similarly, some PC solution may remain
behind once the CCF chambers are purged after the
introduction of the PC solutions preceding the introduction
of the CCF solutions. In all these cases, the metering of the PC
or CCF solution may be slightly off, resulting in the mixing of
PC and CCF solutions in volumetric ratios different from those
planned. The resulting change in solution composition and/or
supersaturation level might increase or reduce the chance of
crystal formation on-chip. In addition, mixing of the solutions
may be insufficient at times if a valve does not open completely,
or does not open at all, during the mixing process. By going
through multiple generations of chip designs, most of these
potential issues have been eliminated in the microfluidic
platforms used for the studies reported here. Still, these issues
may occur at times, causing occasional discrepancy between on-
and off-chip results. (5) The stochastic nature of crystallization
reduces the probability of crystal nucleation in smaller
volumes.8,60,61 This well-known phenomenon might explain 2
of the 9 cases in which solid forms were observed uniquely off-
chip, however, extensive replication of on-chip experiments
would be needed to say this with certainty. The discrepancies
between the on-chip and off-chip outcomes that might occur
due to the former three reasons are a result of characteristics of
the microfluidic chip. However, the discrepancies that might
occur due to the latter two reasons can be improved by
replicating the conditions on-chip, which is feasible thanks to
the limited sample requirements for on-chip experiments
compared to the needs of identical off-chip experiments.
Second Phase of Screening. Cocrystal screening of caffeine
was repeated in quadruplicate for several conditions to study
the reproducibility and the consistency of the on-chip
experiments. Caffeine crystallized with the CCFs with a higher
propensity when caffeine was dissolved in acetonitrile (35%
wells on-chip, 32% wells off-chip) or acetonitrile/water (33%,
32%), compared to when caffeine was dissolved in methanol
(19%, 19%) and ethanol (14%, 15%). Therefore, acetonitrile-
containing solvents were used for the caffeine solutions in the
second phase of the study. The on-chip crystallization
experiments were repeated with all 10 CCFs for which
crystallization was observed. In addition, 4-hydroxy benzoic
acid (4B) was used in the second phase because in the first
phase caffeine crystallized with five of the six (5/6) hydroxy or
dihydroxy benzoic acids screened on-chip. When caffeine was
delivered in acetonitrile-containing solvents in the first round of
screening, we observed that caffeine crystallized with 9/17
CCFs when the CCFs solutions were prepared in acetonitrile,
with 4/17 CCFs when the CCFs solutions were prepared in 1:1
(v/v) acetonitrile/water, with 7/17 CCFs when the CCFs
solutions were prepared in methanol, and with 6/17 CCFs
when the CCFs solutions were prepared in ethanol. Addition-
ally, caffeine crystallized with 1/4 liquid CCFs. Because of the
6030
dx.doi.org/10.1021/cg3011212 | Cryst. Growth Des. 2012, 12, 6023−6034

Crystal Growth & Design
Article
Figure 4. Tiled optical micrographs (taken 4−12 h after mixing) for an on-chip screen of cocrystal solid forms of the PC caffeine from the second
phase of screening, zooming in on suitable conditions identified in the first phase of screening. (a) The rows are filled with PC in acetonitrile (1 and
2), and caffeine in acetonitrile/water (v/v = 1/1) (3 and 4). The columns are filled with the following CCFs in acetonitrile: 2-hydroxy 1-naphthotic
acid (A), 3-hydroxy 2-naphthotic acid (B); 1-hydroxy 2-naphthotic acid (C); 2,3-dihydroxy benzoic acid (D) in methanol; oxalic acid (E); and
ethylene diamine (F), respectively. Brightfield images of 2-hydroxy 1-naphthotate (A1); 3-hydroxy 2-naphthotate (B2); 1-hydroxy 2-naphthotate
(C2); 2,3-dihydroxy benzoate (D3); oxalate (E2) cocrystals of caffeine, and the solid form of caffeine crystallized in presence of ethylene diamine
(B4) are visible in the enlarged views.
high propensity of crystallization when CCFs solutions were
prepared in acetonitrile and to avoid a change in solubility of
either of the cocrystal screening components during mixing, we
exclusively used acetonitrile as the solvent for the CCFs in the
second phase. Additionally, we repeated the cocrystal screening
of caffeine with 2,3-dihydroxy benzoic acid (2,3B) as the CCF
in methanol since opposite results were observed when it was
delivered in acetonitrile and methanol.
For each microfluidic chip in the second phase of screening,
caffeine solutions prepared in acetonitrile and acetonitrile/
water (v/v = 1/1) were each introduced in two adjacent rows
of a 4 × 12 array chip. The six CCFs dissolved in acetonitrile
were each introduced in two adjacent columns. So we screened
24 conditions in total: two 48-well chips, each with four
replicates of 12 conditions. The 24 corresponding off-chip
experiments were only performed once. All the solid forms
crystallized on-chip were analyzed using Raman spectroscopy.
The Raman spectra collected for each solid form crystallized
on-chip were compared with the representative Raman spectra
collected for the cocrystals obtained in the first phase of
screening to confirm the identity of the solid forms.
show typical examples of crystals formed on-chip for the
conditions replicated on-chip. Caffeine formed cocrystals with
all three naphthotic acids (1-hydroxy 2-naphthotic acid, 2-
hydroxy 1-naphthotic acid, and 3-hydroxy 2-naphthotic acid)
used in this study and oxalic acid in all crystallization wells.
Caffeine also crystallized with 2,5-dihydroxy benzoic acid, 2,4-
dihydroxy benzoic acid, and 3,5-dihydroxy benzoic acid, in
almost all wells. Cocrystal formation of caffeine with 2,3-
dihydroxy benzoic acid was observed to be dependent on the
solvent. When 2,3-dihydroxy benzoic acid was delivered in
acetonitrile, cocrystallization of caffeine and 2,3-dihydroxy
benzoic acid only occurred when caffeine was also dissolved
in acetonitrile, but not when it was dissolved in acetonitrile/
water. However, water needs to be present in the acetonitrile
solution of caffeine when 2,3-dihydroxy benzoic acid was
dissolved in methanol. The impact of water on the cocrystal
formation between caffeine and 2-hydroxy benzoic acid was
also reproduced in the second screening. The cocrystal only
formed when caffeine was delivered in acetonitrile, but not in
the presence of water. No crystalline solid appeared in the
presence of 4-hydroxy benzoic acid, in agreement with our
findings during the first phase of screening. Addition of
ethylene diamine to a solution of caffeine in the acetonitrile/
water mixture led to the precipitation of caffeine. The
agreement between the results in the first and the second
Table 2 summarizes the results of the on-chip cocrystal
screening experiments conducted during the second phase of
screening, which agreed well with those from the first phase of
screening (Table 1). The enlarged views in Figures 4 and 5
6031
dx.doi.org/10.1021/cg3011212 | Cryst. Growth Des. 2012, 12, 6023−6034

Crystal Growth & Design
Article
Figure 5. Tiled optical micrographs (taken 4−12 h after mixing) for an on-chip screen of cocrystal solid forms of the PC caffeine from the second
phase of screening, zooming in on suitable conditions identified in the first phase of screening. (a) The rows are filled with PC in acetonitrile (1 and
2), and caffeine in acetonitrile/water (v/v = 1/1) (3 and 4). The columns are filled with the following CCFs in acetonitrile: 2-hydroxy benzoic acid
(A); 4-hydroxy benzoic acid (B); 2,5-dihydroxy benzoic acid (C); 2,4-dihydroxy benzoic acid (D); 2,3-dihydroxy benzoic acid (E); and 3,5-
dihydroxy benzoic acid (F) respectively. Brightfield images of 2-hydroxy benzoate (A2); 2,5-dihydroxy benzoate (C1); 2,4-dihydroxy benzoate (D2);
2,3-dihydroxy benzoate (E1); and 3,5-dihydroxy benzoate (F2) cocrystals of caffeine are visible in the enlarged views.
screening suggests that the cocrystal screening using micro-
fluidic chips is reliable and reproducible.
chip configuration is compatible with on-chip Raman analysis.
The chip design, the fact that it is fully enclosed, and the way in
which solutions are introduced minimize the amount of PC/
CCF solution needed and avoid preconcentration of the PC/
CCF solutions due to evaporation. Only ∼90 nL of solution is
required for each condition screened, which represents a
reduction of at least 3 orders of magnitude in sample volume
compared to the conventional automated solid form screening
platforms (∼500 μL per condition). The small sample size
requirement allows solid form screening to be conducted at an
early stage in the drug development process, at a time when
only limited quantities of each PC are available. Ease of
operation, requiring only a vacuum source, enables immediate
application of these chips in laboratories for solid form
screening. Analysis of the resulting solid forms does require a
bright field microscope setup to determine size and
morphology, and a Raman Spectroscopy setup for chemical
identification. However, analysis can be performed separately
from the screening process due to the portability and enclosed
nature of the chips.
Using caffeine as the model compound, we validated the
capability of the microfluidic chip to screen and identify
multiple cocrystalline solid forms of PCs using only a limited
amount of material. We first conducted screens to identify
promising crystallization conditions, i.e., those combinations of
solvents and CCFs that exhibited the highest propensity for the
formation of crystalline solids of a given PC and identifying the
CCFs resulting in the formation of cocrystalline solid forms on-
chip via Raman spectroscopy. Subsequently, the identified
promising crystallization conditions were replicated on-chip, to
confirm reproducibility and consistency of the on-chip
Some inconsistencies in the cocrystallization outcomes were
also observed in the second phase of screening. Caffeine
crystallized with 2-hydroxy benzoic acid and 2,3-dihydroxy
benzoic acid in 2 out of the 4 and 1 out of 4 identical
experiments, respectively. Mixing caffeine dissolved in acetoni-
trile/water and 1-hydroxy-2-naphthotic acid dissolved in
acetonitrile did not lead to cocrystal formation in the first
phase of screening, but revealed a high propensity for cocrystal
formation in the second phase of screening. These differences
in the on-chip crystallization outcomes might be due to either
the stochastic nature of crystallization especially in small
volumes8,60,61 or imperfect formulation of solutions on-chip.
Overall, these outcomes of the second phase of screening still
suggest that reliable results can be obtained by replicating
crystallization conditions on-chip, which is feasible thanks to
the limited sample requirements for on-chip experiments
compared to the needs of identical off-chip experiments.
4. CONCLUSIONS
We developed and validated hybrid COC-PDMS based
microfluidic platforms for cocrystalline solid form screening
of PCs. The platform meters various PC and CCF solutions
and enables the creation of 48 (4 × 12 array) unique
combinatorial conditions on-chip upon mixing via free interface
diffusion. The use of thin PDMS layers sandwiched by COC
sheets rendered the screening chips compatible with mild
organic solvents typically used in pharmaceutical crystallization
such as methanol, ethanol, and acetonitrile for long duration
crystallization screening experiments (4−12 h). Similarly, this
6032
dx.doi.org/10.1021/cg3011212 | Cryst. Growth Des. 2012, 12, 6023−6034

Crystal Growth & Design
Article
crystallization outcomes. The identity of the solid forms grown
on-chip was confirmed by comparing their Raman spectra with
those of corresponding solid forms crystallized in off-chip
experiments. The COC/PDMS chips used here are also
compatible with X-ray analysis and subsequent structure
determination of solid forms, as we have recently shown for
on-chip protein crystallization.62
Micro- & Nanotechnology Laboratory as well as the Frederick
Seitz Materials Research Laboratory Central Facilities at
University of Illinois at Urbana−Champaign, which is partially
supported by the U.S. Department of Energy under Grants DE-
FG02-07ER46453 and DE-FG02-07ER46471. We thank Dr.
Amit V. Desai and Dr. Daria Khvostichenko for stimulating
discussions and Cassandra Schneider and Jose Gallegos-Lopez
for help in fabrication of microfluidic platforms.
In summary, the platform and protocol for its use presented
in this paper can be applied to broad screening of suitable
cocrystalline solid forms of PCs when only limited amounts of
each PC are available. One can also foresee using this platform
to study the effects of additives such as polymers, surfactants,
and antisolvents on solid form crystallization, crystal morphol-
ogy, and polymorphism of active pharmaceutical ingredients
(APIs) as well as on stabilization of the amorphous forms of
APIs. The solutions in these COC/PDMS-based crystallization-
screening platforms are fully enclosed to minimize solvent loss.
With further modifications to the current platform one could
also perform crystallization screens in which solid form
formation is driven by controlled evaporation of solvent. In
addition, by modifying the relative dimensions of the mixing
zone and the solution chambers multiple supersaturation
profiles can be screened with respect to suitability for solid
form crystallization. In fact, as we have shown previously for
salt and polymorph screening of PCs,28,36 the spatiotemporal
variations in local concentrations of the various chemical
species, and thus in local levels of supersaturation, can be
modeled and correlated with crystal nucleation and growth
events on-chip. The combined results of experiments and
modeling will enhance understanding of what parameters
determine desirable crystallization outcomes for a certain PC,
which in turn will help scale up efforts in later stages of drug
development, if the given PC is found to be promising for
further development.
REFERENCES
■
(1) Good, D. J.; Rodríguez-Hornedo, N. Cryst. Growth Des. 2009, 9,
2252−2264.
(2) Childs, S. L.; Chyall, L. J.; Dunlap, J. T.; Smolenskaya, V. N.;
Stahly, B. C.; Stahly, G. P. J. Am. Chem. Soc. 2004, 126, 13335−13342.
(3) Aakeroy, C. B.; Salmon, D. J. CrystEngComm 2005, 7, 439−448.
(4) Schultheiss, N.; Newman, A. Cryst. Growth Des. 2009, 9, 2950−
2967.
́ ́
(5) Fabian, L. s. Cryst. Growth Des. 2009, 9, 1436−1443.
(6) Musumeci, D.; Hunter, C. A.; Prohens, R.; Scuderi, S.; McCabe,
J. F. Chem. Sci. 2011, 2, 883−890.
(7) Remenar, J. F.; Morissette, S. L.; Peterson, M. L.; Moulton, B.;
MacPhee, J. M.; Guzman, H. R.; Almarsson, O. J. Am. Chem. Soc. 2003,
125, 8456−8457.
(8) Goh, L.; Chen, K.; Bhamidi, V.; He, G.; Kee, N. C. S.; Kenis, P. J.
A.; Zukoski, C. F.; Braatz, R. D. Cryst. Growth Des. 2010, 10, 2515−
2521.
(9) Childs, S. L.; Rodriguez-Hornedo, N.; Reddy, L. S.; Jayasankar,
A.; Maheshwari, C.; McCausland, L.; Shipplett, R.; Stahly, B. C.
CrystEngComm 2008, 10, 856−864.
(10) Desrosiers, P. J. Modern Drug Discovery 2004, 40−43.
(11) Gagniere, E.; Mangin, D.; Puel, F. o.; Bebon, C.; Klein, J.-P.;
Monnier, O.; Garcia, E. Cryst. Growth Des. 2009, 9, 3376−3383.
(12) ter Horst, J. H.; Deij, M. A.; Cains, P. W. Cryst. Growth Des.
2009, 9, 1531−1537.
(13) Weyna, D. R.; Shattock, T.; Vishweshwar, P.; Zaworotko, M. J.
Cryst. Growth Des. 2009, 9, 1106−1123.
(14) Alhalaweh, A.; Velaga, S. P. Cryst. Growth Des. 2010, 10, 3302−
ASSOCIATED CONTENT
* Supporting Information
3305.
■
(15) Friscic, T.; Childs, S. L.; Rizvi, S. A. A.; Jones, W.
CrystEngComm 2009, 11, 418−426.
S
Protocols for introducing solutions in the microfluidic platform,
Raman spectra for all materials that the microfluidic platform is
comprised of, and procedure for background correction of
Raman spectrum of on-chip grown crystals, comparison of
Raman spectra for off-chip and on-chip grown crystals,
comparison of PXRD patterns for solid forms crystallized
using our off-chip experiments to PXRD pattern for intended
cocrystals available in the literature, Raman spectra for caffeine
solid form crystallized in the presence of ethylene diamine, and
a movie demonstrating the combinatorial mixing capability of
the microfluidic platform is provided. This material is available
free of charge via the Internet at http://pubs.acs.org.
(16) Zhang, G. G. Z.; Henry, R. F.; Borchardt, T. B.; Lou, X. J.
Pharm. Sci. 2007, 96, 990−995.
(17) Nehm, S. J.; Rodríguez-Spong, B.; Rodríguez-Hornedo, N. Cryst.
Growth Des. 2005, 6, 592−600.
(18) Rodríguez-Hornedo, N.; Nehm, S. J.; Seefeldt, K. F.; Pagan-
Torres, Y.; Falkiewicz, C. J. Mol. Pharmaceutics 2006, 3, 362−367.
(19) Trask, A. V.; Jones, W. Crystal Engineering of Organic
Cocrystals by the Solid-State Grinding Approach. In Organic Solid
State Reactions; Toda, F., Ed.; Springer: Berlin/Heidelberg: 2005; Vol.
254, pp 41−70.
(20) Friscic, T.; Jones, W. Cryst. Growth Des. 2009, 9, 1621−1637.
(21) Karki, S.; Friscic, T.; Jones, W.; Motherwell, W. D. S. Mol.
Pharmaceutics 2007, 4, 347−354.
(22) Trask, A. V.; Motherwell, W. D. S.; Jones, W. Chem. Commun.
2004, 890−891.
AUTHOR INFORMATION
Corresponding Author
■
(23) Lu, E.; Rodriguez-Hornedo, N.; Suryanarayanan, R. CrystEng-
Comm 2008, 10, 665−668.
*(P.J.A.K.) Phone: (217) 265-0523. Fax: (217) 333-5052. E-
mail: kenis@illinois.edu. Web Address: http://www.scs.uiuc.
edu/∼pkgroup/ (Y.G.) Phone: (847) 938-6642. Fax: (847)
937-2417. E-mail: yuchuan.gong@abbott.com.
(24) Berry, D. J.; Seaton, C. C.; Clegg, W.; Harrington, R. W.; Coles,
S. J.; Horton, P. N.; Hursthouse, M. B.; Storey, R.; Jones, W.; Friscic,
T.; Blagden, N. Cryst. Growth Des. 2008, 8, 1697−1712.
(25) Hansen, C. L.; Skordalakes, E.; Berger, J. M.; Quake, S. R. Proc.
Natl. Acad. Sci. U. S. A. 2002, 99, 16531−16536.
(26) Thorsen, T.; Maerkl, S. J.; Quake, S. R. Science 2002, 298, 580−
584.
Notes
The authors declare no competing financial interest.
(27) Schudel, B. R.; Choi, C. J.; Cunningham, B. T.; Kenis, P. J. A.
Lab Chip 2009, 9, 1676−1680.
(28) Thorson, M. R.; Goyal, S.; Schudel, B. R.; Zukoski, C. F.; Zhang,
G. G. Z.; Gong, Y.; Kenis, P. J. A. Lab Chip 2011, 11, 3829−3837.
ACKNOWLEDGMENTS
■
We would like to thank Abbott Laboratories for financial
support. Part of this work made use of the facilities in the
6033
dx.doi.org/10.1021/cg3011212 | Cryst. Growth Des. 2012, 12, 6023−6034

Crystal Growth & Design
Article
(29) Hansen, C.; Quake, S. R. Curr. Opin. Struct. Biol. 2003, 13, 538−
544.
(30) Charvin, G.; Cross, F. R.; Siggia, E. D. PLoS ONE 2008, 3,
(62) Guha, S.; Perry, S. L.; Pawate, A. S.; Nair, S. K.; Kenis, P. J. A.
Proceedings of the 15th International Conference on Miniaturized Systems
for Chemistry and Life Sciences; Seattle, Washington, October 2−6,
2001; Landers, Ed.; Curran Associates, Inc.: Lakeville, MN, 2012; pp
876−878.
e1468.
(31) Balaban, N. Q.; Merrin, J.; Chait, R.; Kowalik, L.; Leibler, S.
Science 2004, 305, 1622−1625.
(32) Ottesen, E. A.; Hong, J. W.; Quake, S. R.; Leadbetter, J. R.
Science 2006, 314, 1464−1467.
(33) Blazej, R. G.; Kumaresan, P.; Mathies, R. A. Proc. Natl. Acad. Sci..
U. S. A. 2006, 103, 7240−7245.
(34) Grover, W. H.; Skelley, A. M.; Liu, C. N.; Lagally, E. T.;
Mathies, R. A. Sensors Actuators B-Chem. 2003, 89, 315−323.
(35) Wheeler, T. D.; Zeng, D.; Desai, A. V.; Onal, B.; Reichert, D. E.;
Kenis, P. J. A. Lab Chip 2010, 10, 3387−3396.
(36) Sista, R.; Hua, Z.; Thwar, P.; Sudarsan, A.; Srinivasan, V.;
Eckhardt, A.; Pollack, M.; Pamula, V. Lab Chip 2008, 8, 2091−2104.
(37) Meagher, R. J.; Hatch, A. V.; Renzi, R. F.; Singh, A. K. Lab Chip
2008, 8, 2046−2053.
(38) Perry, S. L.; Roberts, G. W.; Tice, J. D.; Gennis, R. B.; Kenis, P.
J. A. Cryst. Growth Des. 2009, 9, 2566−2569.
(39) Thorson, M. R.; Goyal, S.; Gong, Y.; Zhang, G. G. Z.; Kenis, P.
J. A. CrystEngComm 2012, 14, 2404−2412.
(40) Zheng, B.; Roach, L. S.; Ismagilov, R. F. J. Am. Chem. Soc. 2003,
125, 11170−11171.
(41) Zheng, B.; Gerdts, C. J.; Ismagilov, R. F. Curr. Opin. Struct. Biol.
2005, 15, 548−555.
(42) Laval, P.; Salmon, J. B.; Joanicot, M. J. Cryst. Growth 2007, 303,
622−628.
(43) Ildefonso, M.; Revalor, E.; Punniam, P.; Salmon, J. B.; Candoni,
N.; Veesler, S. J. Cryst. Growth 2012, 342, 9−12.
(44) Talreja, S.; Kim, D. Y.; Mirarefi, A. Y.; Zukoski, C. F.; Kenis, P. J.
A. J. Appl. Crystallogr. 2005, 38, 988−995.
̌
(45) Bucar, D.-K. i.; Henry, R. F.; Lou, X.; Duerst, R. W.;
MacGillivray, L. R.; Zhang, G. G. Z. Cryst. Growth Des. 2009, 9,
1932−1943.
(46) Li, L.; Du, W.; Ismagilov, R. F. J. Am. Chem. Soc. 2009, 132,
112−119.
(47) Lee, T.; Wang, P. Y. Cryst. Growth Des. 2010, 10, 1419−1434.
(48) Xia, Y. N.; Whitesides, G. M. Angew. Chem., Int. Ed. 1998, 37,
551−575.
(49) Mohan, R.; Schudel, B. R.; Desai, A. V.; Yearsley, J. D.; Apblett,
C. A.; Kenis, P. J. A. Sens. Actuators, B 2011, 160, 1216−1223.
(50) Nunes, P. S.; Ohlsson, P. D.; Ordeig, O.; Kutter, J. P. Microfluid.
Nanofluid. 2010, 9, 145−161.
(51) Allesø, M.; Velaga, S.; Alhalaweh, A.; Cornett, C.; Rasmussen,
M. A.; Berg, F. v. d.; Diego, H. L. d.; Rantanen, J. Anal. Chem. 2008,
80, 7755−7764.
(52) Vogt, F. G.; Clawson, J. S.; Strohmeier, M.; Edwards, A. J.;
Pham, T. N.; Watson, S. A. Cryst. Growth Des. 2008, 9, 921−937.
(53) Kojima, T.; Onoue, S.; Murase, N.; Katoh, F.; Mano, T.;
Matsuda, Y. Pharm. Res. 2006, 23, 806−812.
(54) Meng, Y.; Weidner, D. J.; Gwanmesia, G. D.; Liebermann, R. C.;
Vaughan, M. T.; Wang, Y.; Leinenweber, K.; Pacalo, R. E.; Yeganeh-
Haeri, A.; Zhao, Y. J. Geophys. Res. 1993, 98, 22199−22207.
(55) Ghosh, S.; Malla Reddy, C. CrystEngComm 2012, 14, 2444−
2453.
̌
(56) Bucar, D.-K.; Henry, R. F.; Lou, X.; Duerst, R. W.; Borchardt, T.
B.; MacGillivray, L. R.; Zhang, G. G. Z. Mol. Pharmaceutics 2007, 4,
339−346.
(57) Bucar, D.-K.; Henry, R. F.; Lou, X.; Borchardt, T. B.; Zhang, G.
G. Z. Chem. Commun. 2007, 525−527.
(58) Aitipamula, S.; Chow, P. S.; Tan, R. B. H. CrystEngComm 2012,
14, 2381−2385.
̌
(59) Bucar, D.-K.; Henry, R.; Duerst, R.; Lou, X.; MacGillivray, L.;
Zhang, G. J. Chem. Crystallogr. 2010, 40, 933−939.
(60) Newman, J.; Xu, J.; Willis, M. C. Acta Crystallogr., Sect. D 2007,
63, 826−832.
(61) Talreja, S.; Kenis, P. J. A.; Zukoski, C. F. Langmuir 2007, 23,
4516−4522.
6034
dx.doi.org/10.1021/cg3011212 | Cryst. Growth Des. 2012, 12, 6023−6034