LCP-Tm: An Assay to Measure and Understand Stability of Membrane
Proteins in a Membrane Environment
Wei Liu, Michael A. Hanson, Raymond C. Stevens, and Vadim Cherezov*
Department of Molecular Biology, The Scripps Research Institute, La Jolla, California
membrane environment. The development of novel methods to efficiently stabilize membrane proteins immediately after purifi-
cation is important for biophysical studies, and is likely to be critical for studying the more challenging human targets. Lipidic cubic
phase (LCP) provides a suitable stabilizing matrix for studying membrane proteins by spectroscopic and other biophysical
techniques, including obtaining highly ordered membrane protein crystals for structural studies. We have developed a robust
and accurate assay, LCP-Tm, for measuring the thermal stability of membrane proteins embedded in an LCP matrix. In its
two implementations, protein denaturation is followed either by a change in the intrinsic protein fluorescence on ligand release,
or by an increase in the fluorescence of a thiol-binding reporter dye that measures exposure of cysteines buried in the native
structure. Application of the LCP-Tmassay to an engineered human b2-adrenergic receptor and bacteriorhodopsin revealed
a number of factors that increased protein stability in LCP. This assay has the potential to guide protein engineering efforts
and identify stabilizing conditions that may improve the chances of obtaining high-resolution structures of intrinsically unstable
Structural and functional studies of membrane proteins are limited by their poor stability outside the native
Membrane proteins are essential cellular components that
perform a variety of functions including transport of ions
and nutrients, transformation of energy and transduction of
signals across the cell membrane. The involvement of
membrane proteins in many crucial cellular and physiolog-
ical processes and their strategic location at the cell surface
makes them important pharmaceutical drug targets (~60%
of approved drugs on the market act on membrane proteins)
(1). Rational design of new drugs with improved efficacy and
selectivity is facilitated by the knowledge of the 3D structure
of the target protein at atomic resolution, yet acquiring this
information in the case of membrane proteins remains a
formidable task. There are only ~200 nonredundant mem-
brane protein structures out of ~20,000 total nonredundant
structures in the Protein Data Bank, and this disparity is
even greater when only human proteins are considered
(>2000 nonredundant human protein structures were solved
at <3 A˚resolution, but only five of these are membrane
proteins). The structures of two such membrane proteins,
members of the G protein-coupled receptor family (2–4),
were obtained using crystals grown in a lipidic cubic phase
The success of LCP for growing highly ordered crystals of
challenging membrane proteins can be attributed to at least
two factors. First, LCP allows the membrane proteins to
remain in a more native-like membrane environment
throughout crystallization instead of being transiently
exposed to a number of physically and chemically distinct
environments that are encountered when a protein is solubi-
lized with detergent. Second, crystals grown in LCP have
type I packing (see Deisenhofer and Michel (6) for a descrip-
tion of packing in membrane protein crystals) with protein
molecules making contacts not only through hydrophilic
but also through hydrophobic moieties resulting in lower
solvent content and better crystal ordering. By the end of
2009, the LCP crystallization technique yielded high-
resolution structures of 12 different membrane proteins con-
tributing 56 total entries to the Protein Data Bank (7) (Table
S1 in the Supporting Material).
Crystallization efforts of eukaryotic membrane proteins
are plagued by challenges associated with the heterologous
expression of functional proteins and by their low intrinsic
stability in detergent micelles. Both of these problems likely
relate to the more complex nature of biological membranes
in higher organisms. The relatively large number of chemi-
cally distinct components that make up eukaryotic mem-
branes and their heterogenous distribution in the membrane
allows tight control of function and selective stabilization
of embedded proteins. For example, cholesterol, an essential
component of eukaryotic membranes, was shown to directly
modulate the activity of several membrane proteins, such as
the oxytocin G protein-coupled receptor (8) and nicotinic
acetylcholine receptor (9).
Although it is natural to assume that the lipid bilayer of
LCP provides a more stabilizing environment for membrane
proteins than detergent micelles, this has neither been
convincingly demonstrated nor quantified. A number of
techniques have been used to quantify the thermal stability
of soluble proteins as well as membrane proteins in detergent
solutions, the most successful of which are based on
spectroscopic approaches (10–12). Because LCP material
Submitted September 21, 2009, and accepted for publication December 11,
Editor: Lukas K. Tamm.
? 2010 by the Biophysical Society
Biophysical Journal Volume 98 April 2010 1539–15481539
is optically transparent when formed under proper condi-
tions, proteins reconstituted in LCP are amenable to spectro-
scopic measurements. Recently, Lunde et al. (13) showed
that the intrinsic fluorescence of bacteriorhodopsin (bR)
embedded in LCP increases more than two times after heat-
ing the protein to 90?C and cooling it down to room temper-
ature. This increase was attributed to protein denaturation on
heating. We have extended these results and developed a
robust protocol, LCP-Tm, for comparing the thermal denatur-
ation temperatures of membrane proteins in LCP. Herein we
describe the details of the LCP-Tmprotocol and report on the
stability characteristics of two constructs of an engineered
human b2-adrenergic receptor (b2AR) and of the wild-type
bR in a variety of conditions.
MATERIALS AND METHODS
Expression and purification of b2AR constructs
Two engineered human b2-adrenergic receptor constructs were used in this
study. The first construct, b2AR, contained a stabilizing mutation E122W
(14), C-terminal truncation at residue 348, mutated out glycosylation site
N187E (3,15), and deleted residues 245–259 in the third intracellular
loop. In addition to all these initial modifications, the second construct,
b2AR-T4L, included a replacement of the third intracellular loop residues
231–262 between the transmembrane helices 5 and 6 with cysteine-free
T4 lysozyme (C54T, C97A) residues 2–161 (15). Both constructs were
expressed in baculovirus infected insect cells (sf9) and purified following
published protocols (3). The final protein was in 0.05% w/v n-dodecyl-
b-D-maltopyranoside (DDM, Anatrace, Maumee, OH), 0.01% w/v choles-
teryl hemisuccinate (CHS; Sigma, St. Louis, MO), 20 mM Hepes pH 7.5,
150 mM NaCl, 0.5 mM ligand. Additional details on expression and purifi-
cation are provided in the Supporting Material.
Expression and purification of bR
Cultivation of Halobacterium salinarum (strain S9) was carried out as
described previously (16,17). Wild-type bR was solubilized with 1.2%
w/v n-octyl-b-D-glucoside (OG, Anatrace) using purple membranes isolated
from H. salinarum following established protocols (16,18). The concentra-
tion of bR was determined by absorbance at 550 nm using the extinction
coefficient 3550~5.8 ? 104M?1cm?1(5). Purified bR was stored at
?80?C for up to 6 months until used.
Mixing monoolein with lipid additives
1,2-dioleoyl-sn-glycero-3-phosphatidylethanolamine (DOPE), 1,2-dioleoyl-
phosphatidylserine (DOPS), and cholesterol (all lipid additives were from
Avanti Polar Lipids, Alabaster, AL), were co-dissolved with monoolein
(Nu Chek Prep, Elysian, MN) in chloroform at appropriate molar ratios.
The bulk of the solvent was evaporated using a gentle stream of nitrogen
gas. The remaining traces of chloroform were removed under a vacuum,
150 mTorr, at room temperature (RT, 21–23?C) for at least 12 h. The dried
lipid mixtures were stored at ?20?C until used.
additives, 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC),
Preparation of LCP samples
LCP host lipids, monoolein (MO), monopalmitolein (MP), monovaccenin
(MV), and monoeicosenoin (ME) (all lipids were from Nu Chek Prep),
were taken from a ?20?C freezer, and melted by incubating them for
10 min at RT (MP), 40?C (MV, MO, and mixtures of additive lipids with
MO), or 60?C (ME). Molten lipid was cooled down to RT and immediately
mixed with a protein solution to form a homogeneous LCP using a syringe
mixer (19), as described in Caffrey and Cherezov (20). Final concentrations
for all target proteins in the LCP samples were ~0.5 mg/mL. Protein-free
buffer (washing buffer at the last purification step) was used to make control
samples for the background correction. To obtain optically clear samples for
spectroscopic measurements, sample compositions were selected based on
published phase diagrams: 40% w/w hydration for MO (21) and ME (22),
50% w/w hydration for MP (23) and MV (24). Deviations in the actual
hydration from the nominal values were within 2% w/w. If any of the addi-
tives significantly shifted the hydration boundary (>2% w/w), the sample
composition was adjusted accordingly. To estimate the hydration boundary
in the presence of additives, the aqueous content was varied by 2% w/w
steps in the vicinity of the hydration boundary of the additive-free control
until the boundary between the optically clear and hazy samples was deter-
mined. For example, the LCP samples containing bR and 1 M Na/K phos-
phate pH 5.6 were prepared at 35% w/w hydration to account for a reduced
LCP hydration capacity induced by the high concentration of salt.
After achieving a homogeneous and transparent LCP sample in the
syringe mixer, ~50 mL of the sample was transferred into a 3-mm quartz
cuvette (Hellma USA, Plainview, NY). The top of the cuvette was sealed
with a Teflon tape to prevent dehydration at high temperatures. The cuvette
was centrifuged at 5600 ? g and 20?C for 10 min (Allegra 25 R refrigerated
centrifuge with TS-5.1-500 swinging bucket rotor; Beckman Coulter, Full-
erton, CA) to remove any defects and air pockets that were introduced
during the sample loading.
Ultraviolet-visible absorption spectroscopy
Absorption spectra were recorded with a DU800 spectrophotometer (Beck-
man Coulter). Data were collected in the 750–250 nm range with 1-nm steps
at 1200 nm/min using air as the reference. The absorption spectrum of a
protein-free LCP or buffer sample recorded against air was subtracted
from the protein-containing sample spectra as appropriate.
Emission spectra were recorded using a Cary Eclipse fluorescence spectro-
photometer (Varian, Walnut Creek, CA). For measuring intrinsic protein
fluorescence the excitation wavelength was 280 nm and the emission was
scanned between 500 and 300 nm with 1 nm steps at 100 nm/min.
Appropriate inner filter, as well as background corrections were applied to
all relevant data following established procedures (25), as described (26,27).
Typical maximum absorbance values for the samples in this study were in
the 0.5–1 range corresponding to the inner filter correction factors ranging
between 1.9 and 3.5.
To overcome the problem of clouding LCP samples during temperature
ramps, a step-wise heating/cooling protocol was designed. The protocol
included 13 data points, starting at RT and ending at 80?C with 5?C incre-
ments between points (Fig. 1). LCP samples were prepared and loaded in
3-mm quartz cuvettes as described above. The initial absorbance and
intrinsic protein fluorescence spectra were taken immediately after making
the samples. At each step samples were incubated at the desired temperature
for 7 min in a VWR (West Chester, PA) Digital Heatblock. Temperature
increase induced shrinkage of LCP lattice and shedding of water, making
samples turbid (Fig. 2). Samples were reset back into a transparent state
by centrifugation at 5600 ? g and 20?C for 10–15 min (Fig. 2). Absorbance
and fluorescence spectra were collected from transparent samples at RT after
each incubation/centrifugation cycle, and the process was repeated with the
same samples incrementing the incubation temperature by 5?C on each step
until the whole data set was collected. Protein-free samples for background
correction were prepared at each different condition used in this study.It was
verified that both ultraviolet absorbance and fluorescence spectra of these
Biophysical Journal 98(8) 1539–1548
1540Liu et al.
background samples did not change when they were subjected to the same
heating/cooling/centrifugation treatment as the protein-containing samples.
Therefore, in all subsequent measurements, background data were collected
only once at RT and used for correction of data recorded from protein-
containing samples at all temperature treatment points.
Using a CPM probe in the LCP-Tmassay
The LCP-Tmassay can also be carried out using a thiol-reactive fluorescent
probe 7-diethylamino-3-(40-maleimidylphenyl)-4-methylcoumarin (CPM)
(Invitrogen, Carlsbad, CA), if the protein contains free buried cysteine resi-
dues. The CPM probe has very low fluorescence in solution (excitation
387 nm, emission 463 nm) that strongly increases when the probe binds
to cysteine. Use of the CPM probe reduces the amount of protein sample
required for measurement, typically by an order of magnitude. For this
assay, the typical protein concentration was ~0.015 mg/mL. CPM dye was
dissolved in dimethyl formamide at 4 mg/mL and added to the protein
solution at the final concentration of 0.007 mg/mL. Introducing a probe
into a protein sample, however, is an invasive technique, potentially desta-
bilizing the protein (28). Certain precautions should be taken to reduce the
interference between the CPM probe and other components in the sample.
Imidazole was identified as one of the most common components that in-
teracted with CPM resulting in high background signal. The backgrounddue
to imidazole binding could be reduced by decreasing the concentration of
imidazole in the protein samples eluted from the Ni-charged IMAC by de-
salting (PD-10, GE HealthCare, Piscataway, NJ). Additionally, we observed
that the fluorescence signal from the CPM dye in the protein-free control
sample increased after each heating/cooling step of the LCP-Tmassay,
possibly due to interactions of the dye with the lipid bilayer. Thus, it was
essential to process the control sample with exactly the same thermal treat-
ment as the protein-containing samples and acquire data from the control
samples lacking protein at all temperature points.
For liquid samples of proteins in detergent solutions, thermal denaturation
data were collected in 1 cm path-length cuvettes by a Cary Eclipse spectro-
fluorometer temperature ramping function (20–100?C, 1?C/min) as previ-
ously described (12).
Isothermal stability assay
An isothermal stability assay was carried out by incubating samples at RT in
the dark for 30 days. Fluorescence and absorption spectra were taken daily.
Protein stability in LCP was assessed via the intrinsic protein fluorescence
signal (F) as described in the LCP-Tmassay protocol. Fraction of the native
protein (Rnat) was estimated by assuming that the initial signal (FRT) corre-
sponded to 100% folded protein and the signal after heating the sample to
80?C (F80) corresponded to a completely unfolded protein using the
Rnat ¼ 1 ? ðF ? FRTÞ=ðF80? FRTÞ:
5,600 x g
5,600 x g
5,600 x g
0 1020 3040 50290 300
sequence of temperature treatment steps. After each heating/cooling step
samples are centrifuged at 20?C and 5600 ? g for 10 min to return the
LCP sample to a transparent state. Absorbance and fluorescence spectra
are taken for each sample at RT. The whole protocol contains 13 heating/
cooling treatment steps, takes ~300 min and allows for processing of up
to six to eight samples simultaneously (A, absorbance; F, fluorescence).
Schematic diagram of the LCP-Tm protocol showing a
Photographs of a cuvette filled with LCP: (i) after loading; (ii) after heating
to 80?C and cooling to RT; and (iii) after 10 min centrifugation at 5,600 ? g
and 20?C. The first layer of zoom windows shows the domain-like meso-
scopic structure of LCP on a scale of ~100 mm. The second layer of zoom
windows depicts the LCP microstructure at a 100 A˚scale, showing a cubic
lattice formed by a single lipid bilayer. In the initial sample (i) the LCP
domains are tightly packed resulting in a homogeneous and transparent
appearance. Heating (ii) induces shrinkage of the cubic lattice shedding
water into the interdomain space. The water droplets of a few microns in
size scatter light impeding spectroscopic measurements (see Fig. S1). Cool-
ing alone does not restore transparency due to a prominent hysteresis in the
LCP swelling behavior. A mechanical force such as centrifugation is
required to achieve a transparent sample (iii). (B) MO/water temperature-
composition phase diagram (re-drawn from Briggs et al. (21)). The phase
diagram is metastable at temperatures below 17?C (36). Ia3d and Pn3m
represent two bicontinuous cubic phases with different symmetries. Initial
samples are preparedat RT and40% w/w hydrationinthe cubic-Pn3m phase
(yellow dot). Heating brings the samples along the yellow dashed line into
a region where the cubic-Pn3m phase and water coexist. The higher the
temperature, the more water separates from the LCP.
Heat-induced water shedding from the LCP sample. (A)
Biophysical Journal 98(8) 1539–1548
Thermal Stability of Membrane Proteins1541
For bR, absorbance at 542 nm was used as an additional indicator of protein
stability. The amount of correctly folded bR was assumed to be proportional
to the absorbance at 542 nm. Fractions of folded bR estimated from the
intrinsic fluorescence and from the absorbance at 542 nm agreed within
the experimental errors.
Protein samples in detergent solutions were centrifuged in cuvettes at
5600 ? g for 15 min before each measurement. We observed some protein
aggregation and precipitation over time, correlating with a decrease in the
absorption signal at 280 nm. Visual pellet was observed after a few days
in the case of b2AR-T4L. To compensate for the loss of protein, the intrinsic
protein fluorescence signal was corrected by normalizing it to the absor-
bance at 280 nm, which is proportional to the concentration of protein in
solution. The fraction of folded protein remaining in solution in respect to
the total protein was estimated by the increase in the intrinsic fluorescence
in a fashion similar to the analysis used in LCP.
Ligand release and binding in LCP
Ligand release experiments were carried out by layering 80 mL buffer
(20 mM Hepes pH 7.5, 150 mM NaCl) on top of ~50 mL LCP sample
made of MO and b2AR-T4L/timolol in a 3-mm quartz cuvette. Absorbance
and fluorescence spectra of the protein in LCP were taken every hour along
with replacement of the top layer of the solution with fresh buffer.
After the fluorescence signal reached equilibrium, corresponding to
complete ligand dissociation, ligand binding was initiated by replacing the
ligand-free buffer with 80 mL of ligand-containing buffer (20 mM Hepes
pH 7.5, 150 mM NaCl, 812.5 mM timolol). The timolol concentration
used in the ligand binding experiment was chosen to bring the final concen-
tration of timolol after equilibration back to the original value of 500 mM.
Absorbance and fluorescence spectra were collected every hour after ligand
Samples for all experiments were prepared and analyzed at least in triplicate.
The protein thermal denaturation data were fit by a Boltzmann sigmoidal
function using a GraphPad Prism Software (La Jolla, CA):
FðTÞ ¼ Fminþ ðFmax? FminÞ=ð1þ
where F is the normalized fluorescence, T is the temperature, Fminand Fmax
are fitting parameters describing fluorescence before and after transition, Tm
is the transition temperature, and S is the slope factor. Fluorescence signal
is expressed in arbitrary units, and both temperature and slope are expressed
Development of the LCP-Tm protocol
A central issue to the development of spectroscopic methods
for measuring the thermal stability of membrane proteins in
LCP is shrinkage of LCP and shedding of water induces by
heating (Fig. 2 A). This phenomenon results in excess light
scattering that hampers the use of spectroscopy under these
conditions (Fig. 2 A and Fig. S1). To overcome this problem
we designed a step-wise heating/cooling/centrifugation pro-
tocol (Fig. 1). At each step the sample is heated to a desired
temperature, incubated for 7 min, cooled to 20?C and centri-
fuged for 10–15 min at 5600 ? g. Initially transparent LCP
becomes opaque on heating and resets back into a transparent
state after cooling and centrifugation (Fig. 2 A), allowing
absorbance and fluorescence measurements to be carried out.
Starting from 25?C the incubation temperature is increased
by 5?C at each step. The process is repeated until a final
temperature of 80?C is reached.
The phenomenon of shedding water can be understood by
inspecting the MO/water temperature-composition phase
diagram (Fig. 2 B) (21). Initially, at RT and 40% w/w hydra-
tion, the sample is in a pure cubic-Pn3m phase. On
increasing temperature, (Fig. 2 B, dashed yellow line), the
sample enters a region where water and the cubic-Pn3m
phase coexist, and the sample becomes cloudy. Cooling
the sample back to the original temperature alone typically
does not restore the transparent state due to a pronounced
hysteresis in the LCP swelling behavior; rather a physical
force, such as centrifugation, is required to accelerate the
equilibration. To ensure that an LCP sample returns back
to a transparent state after cooling and centrifugation, it
was important to prepare the sample at slightly below the
full hydration limit. Lipid or soluble additives can shift
the full hydration boundary, making it necessary to adjust
the sample composition to maintain the transparency of the
To reduce consumption of lipid and, more importantly,
of protein, we used quartz microcuvettes with 3-mm
pathlength. These cuvettes hold ~50 mL of LCP sample, cor-
responding to ~10 mg of protein. To prevent sample dehydra-
tion at high temperatures the cuvettes were sealed with a
Teflon tape. Special holders were made for centrifugation
and for mounting these cuvettes into a spectrophotometer
and a spectrofluorimeter (20).
Initial experiments were carried out with b2AR-T4L
bound to a partial inverse agonist, timolol, in an MO cubic
phase at 40% w/w hydration. At each step, intrinsic protein
fluorescence was measured using excitation at 280 nm and
emission scanned from 500 nm to 300 nm. Absorption spec-
trum in the 250–750 nm range was also taken to verify that
the samples did not scatter light after heating, and to carry
out necessary inner filter effect corrections (Fig. S2).
Acquiring a complete denaturation curve by the LCP-Tm
protocol takes ~300 min (Fig. 1). Several samples can be
processed in parallel, limited mainly by the number of avail-
able cuvettes and slots in the centrifuge, typically six or
eight. After the inner filter corrections and background
subtraction, measured data points follow a Boltzmann
sigmoidal curve, describing a transition between two states
(Fig. 3). Curve fitting provides the transition temperature,
Tm, and the slope factor, S. This protocol was extensively
optimized (see Supporting Material) resulting in typical
accuracy and reproducibility of the transition temperatures
to be within a few tenths of a degree (see Supporting Mate-
rial). We have shown that the apparent increase in intrinsic
protein fluorescence detected by LCP-Tmis due to irrevers-
ible ligand dissociation (see Supporting Material), meaning
that the use of intrinsic fluorescence for measuring the
thermal stability of proteins in LCP is limited to certain
Biophysical Journal 98(8) 1539–1548
1542Liu et al.
classes of ligand-bound proteins containing tryptophan
residues in the vicinity of the bound ligand.
LCP-Tmwith CPM probe
To extend the LCP-Tmmethod to a broader range of targets,
we used a thiol-sensitive fluorophore, CPM. The CPM probe
has been used previously for monitoring unfolding of
membrane proteins in detergent solutions (12). It has a low
fluorescence that increases dramatically when the probe
binds to cysteine. Protein unfolding can be detected when
cysteines that are buried in the native protein structure are
exposed to solvent on a protein conformational change or
on unfolding. We compared thermal denaturation curves
for b2AR-T4L/timolol in LCP collected using intrinsic
protein fluorescence and the CPM probe (Fig. 3 A, Fig. S3,
and Fig. S4). The b2AR-T4L construct contains 11 cysteines,
four of which are engaged in forming two disulfide bonds,
one cysteine is palmitoylated, one cysteine is capped with
iodoacetamide, and the remaining five cysteines are buried
in the protein interior (3). The Tm measured using the
CPM probe (Tm¼ 47.3 5 0.7?C) was within experimental
error of the Tm obtained by using intrinsic fluorescence
(Tm¼ 46.4 5 0.3?C), confirming that the receptor is unfold-
ing in parallel with ligand dissociation.
Effects of environment and protein engineering
on stability of b2AR-T4L in LCP
After establishing the parameters of the base condition for
b2AR-T4L/timolol in an MO cubic phase, we evaluated
the effects of ligands, host LCP lipids, lipid additives, and
pH on stability, and correlated these results with the out-
comes of crystallization trials. Most of the experiments
were carried out using intrinsic fluorescence, unless noted
Effect of ligands
The effects of four different b2AR ligands—carazolol
(partial inverse agonist), timolol (partial inverse agonist), al-
prenolol (partial inverse agonist / antagonist) and clenbuterol
(partial agonist)—on b2AR-T4L thermal stability were
tested by the LCP-Tmassay. The same set of ligands was
also used in extensive crystallization trials. Crystals of
b2AR-T4L bound to all four ligands were grown in LCP
and analyzed for diffraction quality using a 10-mm minibeam
at GM/CA CAT at the Advance Photon Source (Argonne,
IL) (29). A remarkable correlation was observed between
the ligand-induced stability of the receptor, as measured by
the LCP-Tmassay (Fig. 3 B, Fig. 4 A, and Table S3), and the
diffraction quality of crystals (Table S4). The ligands were
ranked by their ability to both improve the thermal stability
and increase the crystal resolution limit in the following
order: carazolol (Tm ¼ 50.8 5 0.2?C; resolution ¼
2.4 A˚(2)), timolol (Tm¼ 46.4 5 0.3?C; resolution ¼ 2.8 A˚
(3)), alprenolol (Tm¼ 44.5 5 0.2?C; resolution ~3.5 A˚
(Fig. S8 A)) and clenbuterol (Tm¼ 43.0 5 0.2?C; resolution
~7 A˚(Fig. S8 B)). Thermal stability of the apo-receptor was
measured using the CPM probe (Tm¼ 37.2 5 0.4?C) and
compared to stability of the timolol-bound receptor (Tm¼
47.3 5 0.7?C) obtained under identical conditions in side-
by-side experiments. The apo-receptor did not yield crystals
under combined optimization screening conditions selected
from the conditions used for the successful crystallization
trials of the ligand-bound receptor.
Effect of lipids
There are several known monoacylglycerols (MAGs) that
support formation of LCP at full hydration (30,31). Micro-
structural parameters of the LCP, such as lipid bilayer thick-
ness and water channel diameter, strongly depend on the
identity of the host MAG, and, thus, can affect the stability
of the reconstituted proteins. We selected four different
2030 4050 6070 80
20 3040 50607080
LCP-Tmprotocol.(A) Comparison between the intrinsicprotein fluorescence
and the CPM probe fluorescence for monitoring the unfoldingof b2AR-T4L/
timolol in the MO cubic phase. (B) Effect of ligands on denaturation of
b2AR-T4L in the MO cubic phase. Individual points represent averaged
data obtained from at least three samples. Continuous curves represent fits
by the Boltzmann sigmoidal function.
Thermal denaturation curves for b2AR-T4L obtained by the
Biophysical Journal 98(8) 1539–1548
Thermal Stability of Membrane Proteins1543
MAGs for our studies: monopalmitolein (MP, C16:1c9),
monoolein (MO, C18:1c9), monovaccenin (MV, C18:1c11)
and monoeicosenoin (ME, C20:1c11). Protein stability,
when plotted against the lipid identities arranged in the order
of increasing bilayer thickness, followed a bell-shaped curve
centered on MO (Fig. 4 B and Table S5), suggesting that the
bilayer thickness of MO provides a better match to the
hydrophobic thickness of b2AR-T4L.
It has been shown that the MO cubic phase can tolerate
supplementation with a variety of native membrane lipids,
such as phospholipids, cholesterol, etc. (32). Lipid additives
can affect protein stability either through a change in the
structural properties of the LCP or through direct lipid/
protein interactions. To minimize the effect of added lipids
on the structural parameters of LCP, most of the additive
lipids were used at a concentration of 5 mol %, except for
cholesterol, which was used at 10 mol %. The effects of
cholesterol, two zwitterionic phospholipids (DOPC and
DOPE) and two anionic phospholipids (DOPS and DOPG)
on the thermal stability of b2AR-T4L/timolol in LCP are
summarized in Fig. 4 C, Fig. S9 B, and Table S3. Cholesterol
was the only lipid that stabilized b2AR-T4L in LCP,
increasing Tmby 2.3?C. The most destabilizing lipid was
DOPC, decreasing Tmby 5.6?C. The rest of the lipid addi-
tives were mildly destabilizing, lowering Tm by 2–3?C.
Cholesterol was also the best additive lipid for crystalliza-
tion, substantially improving the crystal size and shape
(2,3). The stabilizing effect of cholesterol in LCP is likely
due to direct interactions with the protein, which is consistent
with the identification of two bound cholesterol molecules in
the b2AR-T4L crystal structure (3). The addition of DOPE
also increased the b2AR-T4L crystal size, despite a slightly
destabilizing effect observed by LCP-Tm. The increase in
crystal size, however, was mostly in one dimension and
did not improve the diffraction quality. The addition of
DOPC almost completely abolished crystal growth, in agree-
ment with the destabilizing effect of this lipid on b2AR-T4L
in LCP. The last two lipids (DOPS and DOPG) were not
used in crystallization trials.
Effect of pH
The effect of pH was rather straightforward, showing dimin-
ished stability of b2AR-T4L/timolol in a MO-based LCP at
lower pH levels (Fig. 4 D, Fig. S9 C, and Table S3). This
result is consistent with a decrease in the ligand binding
activity of b2AR at pH values below 7.0 (33). Crystals of
b2AR-T4L were obtained in a pH range from 6.5 to 8.0.
The best diffraction quality crystals were grown at pH 6.5–
7.0. Outside of this range crystals were small or irregularly
shaped and were not suitable for x-ray diffraction experi-
ments. At pH values >8.0 the LCP becomes unstable due
to the increased rate of MO transesterification and hydrolysis
(34); therefore, stability and crystallization experiments were
not carried out at higher pH values.
Effect of T4 lysozyme fusion
Lysozyme from T4 phage (T4L) was initially fused in place
of the third intracellular loop of b2AR to improve the proteo-
lytic stability of the protein and to facilitate crystallization
(15). We evaluated the effect of the T4L fusion on the
thermal stability of b2AR in LCP via the LCP-Tmassay using
two protein constructs that differed only by the presence or
absence of the T4L fusion. The stability of both constructs
was measured in the timolol and carazolol bound states
(Fig. 4 A), and, for timolol-bound proteins, in the presence
pH 4.0pH 6.0pH 7.5 pH 8.0
with the LCP-Tmassay using intrinsic protein fluorescence.
(A) Effect of ligands. (B) Effect of LCP host lipids. (C)
Effect of lipid additives. (D) Effect of pH. The solid bars
represent data for b2AR-T4L and the open bars for b2AR
samples. Experiments in B, C, and D were carried out
with receptors bound to timolol. The Tmdata represent
averaged values obtained by curve fitting fromat least three
samples. The error bars show standard deviation for Tm
Apparent melting temperatures obtained
Biophysical Journal 98(8) 1539–1548
1544Liu et al.
and absence of cholesterol (Fig. 4 C). In all of these cases,
the T4L fusion increased the Tmby 5–6?C, confirming a
substantial stabilizing effect of this fusion partner on the
receptor. The stabilizing effect of the T4L domain is likely
due to reduced conformational flexibility of the intracellular
interface between the transmembrane helices 5 and 6 (2,3).
Effect of mesophase
During the course of in meso crystallization, LCP can trans-
form to or transiently pass through different lipidic meso-
phases (35). We have tried to mimic such situations and to
estimate the impact of passing through various mesophases
on protein stability.
Below 17?C the MO-based LCP is metastable and can flip
like packingof thelipidmolecules (36). AnLCP samplewith
b2AR-T4L/timolol was prepared at RT, cooled to 4?C, and
incubated for 10 min. Transition to the Lc phase was
teristic birefringency detected using polarized light micros-
copy (20). The sample was warmed up to RT and incubated
for 20 min to return it to the LCP. LCP-Tmmeasurements
showed that the total increase in fluorescence dropped to
~35% (compared to ~75% in a control sample) and Tm
decreased to 42.8 5 0.4?C (Fig. S10 and Table S3). Because
the total increase in fluorescence is proportional to the
amount of folded receptor bound to the ligand, we concluded
that transient passage through the Lc phase damaged more
than half of the protein and destabilized the rest.
Transient formation of an inverse hexagonal HIIphase was
induced by mixing LCP with 2 M sodium sulfate at RT
(37,38). Formation of a birefringent phase was confirmed
by polarized light microscopy. After a 10 min incubation,
the salt concentration was decreased to <0.2 M using two
washes in a syringe lipid mixer (19) with more than three
times excess buffer solution. This treatment destroyed
~40% of protein and decreased Tmof the remaining protein
to 44.4 5 0.4?C (Fig. S10 and Table S3). In contrast, similar
treatment with 1 M sodium sulfate, which kept the LCP
intact, did not affect protein stability.
Finally, a lamellar liquid crystalline phase (La) was
achieved by either preparing samples at low hydration
(15% w/w, Fig. 2 B), incubating for 10 min and increasing
hydration to 40% w/w to convert the sample to the LCP,
or by using 40% hydration with high concentration of
DDM (4.1% w/v), incubating for 10 min and adding 1 M
sodium sulfate to convert the sample back into the LCP
(39). All manipulations were carried out using a syringe lipid
mixer (19) as described in Caffrey and Cherezov (20).
Although the amount of folded receptor decreased by
~20% after the low hydration treatment, the transient forma-
tion of the La phase induced by detergent did not appre-
ciably affect either the amount of folded receptor or its
stability (Fig. S10 and Table S3).
Therefore, we conclude that formation of the Lc or HII
phase during crystallization trials is highly detrimental to
the protein and should be avoided as much as possible.
However, a transient passage through the La phase can
likely be tolerated. These results are consistent with previous
observations from in meso crystallization trials with bR (39).
An excess of detergent may originally induce formation of
the La phase, which converts into a cubic phase on addition
of a precipitant. Such phase transitions typically do not
prevent subsequent crystal growth.
LCP provides a better stabilizing environment for
membrane proteins than detergent micelles at RT
Whereas the LCP-Tm protocol is a fast and quantitative
method for measuring protein stability in LCP that allows
comparison of different protein constructs and conditions,
it does not give clear evidence on long-term protein stability
and it does not allow comparison of protein stability in
different environments, such as detergent micelles and
LCP. To obtain this information we carried out isothermal
denaturation experiments, comparing stability of b2AR-
T4L/timolol and bR in both MO-based LCP and detergent
solutions. Samples were incubated at RT, and the state of
the proteins was assessed by intrinsic fluorescence and
UV-Vis absorbance daily for 30 days (Figs. S11–S18 in
the Supporting Material).
As expected, the b2AR-T4L/timolol sample in DDM/CHS
solution was the least stable, losing ~50% of the active
protein within a week and completely denaturing after
20 days (Fig. 5). Reconstitution of b2AR-T4L/timolol in
the MO-based LCP greatly improved its stability, retaining
Fraction of folded protein
bR in LCP+1M salt
bR in LCP
β2AR-T4L in LCP
bR in detergent
β2AR-T4L in detergent
detergent (open symbols) and LCP (solid symbols) environments at RT.
Fractions of folded proteins were estimated using absorbance and fluores-
cence as described in the Materials and Methods. The original raw data
are shown in Figs. S11–S18 in the Supporting Material. bR was prepared
in 1.2% w/v OG, 25 mM Na/K phosphate pH 5.6, and b2AR-T4L was
prepared in 0.05% w/v DDM, 0.01% w/v CHS, 20 mM Hepes pH 7.5,
150 mM NaCl, 0.5 mM timolol.
Comparisons of stability of bR and b2AR-T4L/timolol in
Biophysical Journal 98(8) 1539–1548
Thermal Stability of Membrane Proteins1545
~70% of the active protein after 30 days. Consistently, bR
samples in OG solution were moderately unstable, whereas
reconstitution in LCP increased the stability to about the
same level as b2AR-T4L/timolol. The increase in bR
stability in LCP was, however, not as dramatic as for
b2AR-T4L, which is consistent with observations of dis-
torted retinal spectrum for bR in MO LCP reported by Lunde
et al. (13). On the other hand, it has been known that under
crystallization conditions bR is stable in LCP for months.
Indeed, the addition of 1 M Na/K phosphate pH 5.6 (bR crys-
tallizes at >2 M Na/K phosphate pH 5.6) substantially
increased bR stability bringing the level of properly folded
protein to 80% after 30 days. Assuming a linear dependence
of protein stability on the salt concentration, we would
expect >90% viability of bR in LCP at crystallization condi-
tions after 30 days. This pronounced effect of salt on bR
stability in the isothermal experiments translated well into
the LCP-Tmmeasurements. Addition of 1 M Na/K phosphate
pH 5.6 increased the Tmof bR in LCP from 43.8 5 0.6?C
to 50.7 5 0.4?C (Fig. S9 F and Table S3).
Measuring the thermodynamic stability of membrane pro-
teins in a membrane environment is not a straightforward
task due to the difficulties associated with establishing equi-
librium, reversible transition conditions, and the concomitant
effects of denaturants on the properties of the lipid meso-
phases. For example, the thickness of lipid membranes in
a physiologically relevant lamellar liquid crystalline phase
shrinks with increasing temperature (40). This shrinkage
can denature the protein through a hydrophobic mismatch
(41). Properties of detergent micelles also depend on temper-
ature, but in a different way (42). In the case of LCP, heating
induces shrinkage of the water channel diameter along with
thinning of the lipid bilayer, increasing the membrane curva-
ture, which can also affect the protein stability (21). These
effects preclude the direct comparisons of denaturation tem-
peratures obtained in different environments. For example,
Tmmeasured by LCP-Tmfor b2AR-T4L/timolol is 46.4 5
0.3?C, whereas measured by the CPM probe in DDM/CHS
solution is 66.1 5 1.2?C (Fig. S19). On the other hand, using
the isothermal measurements we have shown that the same
protein was substantially more stable in LCP than in deter-
gent solution at RT (Fig. 5). Thus, meaningful comparisons
of denaturation temperatures for membrane proteins can
only be obtained when the proteins are analyzed in similar
We believe the LCP-Tmprotocol introduced in this work
provides a robust and accurate method for comparing the
stability of membrane proteins in LCP at a variety of condi-
tions. In one of its implementations, the method uses an
increase in the intrinsic protein fluorescence on heating to
assess denaturation of the protein. We have shown that this
increase is associated with an irreversible release of the
ligand coupled with the protein unfolding. Therefore, one
of the requirements for applying the intrinsic protein fluores-
cence for detection of protein denaturation in LCP is the
presence of both a bound ligand and tryptophans, preferably
near the ligand binding site. A large group of integral
membrane proteins that satisfies these criteria is the super-
family of G protein-coupled receptors (GPCRs). Most
GPCRs contain a highly conserved tryptophan residue, often
referred to as a toggle switch (43), near where most of the
small molecule ligands are believed to bind. Therefore, for
most GPCRs bound to ligands, the intrinsic fluorescence is
expected to change on temperature-induced protein unfold-
ing. It is possible that not all combinations of ligands and
tryptophan-containing proteins will work, therefore for any
new protein and/or ligand it is imperative to verify the
intrinsic fluorescence response by heating to 80?C before
starting to use the LCP-Tm protocol. Alternatively, for
proteins lacking ligands or tryptophans but containing buried
cysteines, the CPM probe can be used to monitor protein un-
folding. A simple analysis of all unique membrane proteins
of known structure in the Protein Data Bank indicates that
>60% of proteins contain buried cysteine residues (12).
For proteins that do not fit in either of these two categories,
a tryptophan or cysteine residue could be engineered, if
stability data are crucial for the success of the project.
One has to be careful with the interpretation of LCP-Tm
data. Similar to the temperature effect, most salts and other
kosmotropes, as well as some nonlamellar phase forming
lipids, can decrease the lattice parameter of LCP, increase
membrane curvature and change the membrane thickness
(32,38). In contrast, the addition of chaotropes, certain poly-
mers such as PEG 400, detergents, or lamellar phase forming
lipids can induce swelling of the cubic phase and decrease
the membrane curvature (32,37,44). Some of these additives
can transform LCP into another phase or change the phase
transition temperature. All of these effects should be taken
into account when the experiments are planned and con-
ducted. If carried out under controlled conditions, however,
LCP-Tmmeasurements are very accurate and sensitive to
factors affecting protein stability. They can be extremely
helpful in guiding protein engineering efforts and selecting
the most stabilizing host lipids and lipid additives, as well
as for studying specific lipid-protein interactions.
We observed a remarkable correlation between the stabi-
lizing effect of ligands on b2AR-T4L and the quality of ob-
tained crystals. Additionally, the LCP host lipid, MO, and
the lipid additive, cholesterol, used in obtaining high-resolu-
tion crystals of b2AR-T4L provided the most stabilizing lipid
matrix for this protein. In the case of bR, we found that salt
used to induce crystallization substantially stabilized the
protein in LCP. There are debates in the literature regarding
the existence of a correlation between protein stability and
its ability to crystallize. Positive correlations were found
between the stabilizing effect of additives and their effect
on promoting crystallization in several studies (11,45). On
Biophysical Journal 98(8) 1539–1548
1546Liu et al.
the other hand, a recent comprehensive analysis of 117
proteins suggested no correlation between denaturation
temperatures and the propensities of proteins to crystallize
(46). We should note that most if not all of such studies
were carried out with soluble proteins. Our experiments
were carried out on a relatively small scale reflecting a scar-
city of available crystal forms of selected membrane protein
targets. This analysis will be expanded to better understand
the relationship between the stability of proteins in LCP
and their crystallization propensities. We anticipate that for
the human membrane proteins, and especially such intrinsi-
cally flexible molecules as GPCRs, increasing stability is
the key toward obtaining high-resolution structures.
Supplemental methods, supplemental results, 19 figures, and five tables are
available at http://www.biophysj.org/biophysj/supplemental/S0006-3495
The authors acknowledge E. Chien, K. Allin, and T. Trinh for the help with
protein expression, and A. Walker and A. Pia Abola for assistance with the
This work was supported by the National Institutes of Health Roadmap
Initiative (P50 GM073197).
1. Yildirim, M. A., K. I. Goh, ., M. Vidal. 2007. Drug-target network.
Nat. Biotechnol. 25:1119–1126.
2. Cherezov, V., D. M. Rosenbaum, ., R. C. Stevens. 2007. High-
resolution crystal structure of an engineered human b2-adrenergic G
protein-coupled receptor. Science. 318:1258–1265.
3. Hanson, M. A., V. Cherezov, ., R. C. Stevens. 2008. A specific
cholesterol binding site is established by the 2.8 A structure of the
human b2-adrenergic receptor. Structure. 16:897–905.
4. Jaakola, V. P., M. T. Griffith, ., R. C. Stevens. 2008. The 2.6 angstrom
crystal structure of a human A2A adenosine receptor bound to an antag-
onist. Science. 322:1211–1217.
5. Landau, E. M., and J. P. Rosenbusch. 1996. Lipidic cubic phases:
a novel concept for the crystallization of membrane proteins. Proc.
Natl. Acad. Sci. USA. 93:14532–14535.
6. Deisenhofer, J., and H. Michel. 1989. Nobel lecture. The photosynthetic
reaction centre from the purple bacterium Rhodopseudomonas viridis.
EMBO J. 8:2149–2170.
7. Raman, P., V. Cherezov, and M. Caffrey. 2005. The Membrane Protein
Data Bank. Cell. Mol. Life Sci. 63:36–51.
8. Gimpl, G., K. Burger, and F. Fahrenholz. 1997. Cholesterol as modu-
lator of receptor function. Biochemistry. 36:10959–10974.
9. Addona, G. H., H. Sandermann, Jr., ., K. W. Miller. 1998. Where does
cholesterol act during activation of the nicotinic acetylcholine receptor?
Biochim. Biophys. Acta. 1370:299–309.
10. Pantoliano, M. W., E. C. Petrella, ., F. R. Salemme. 2001. High-
density miniaturized thermal shift assays as a general strategy for
drug discovery. J. Biomol. Screen. 6:429–440.
11. Ericsson, U. B., B. M. Hallberg, ., P. Nordlund. 2006. Thermofluor-
based high-throughput stability optimization of proteins for structural
studies. Anal. Biochem. 357:289–298.
12. Alexandrov, A. I., M. Mileni, ., R. C. Stevens. 2008. Microscale
fluorescent thermal stability assay for membrane proteins. Structure.
13. Lunde, C. S., S. Rouhani, ., R. M. Glaeser. 2006. Membrane-protein
stability in a phospholipid-based crystallization medium. J. Struct. Biol.
14. Roth, C. B., M. A. Hanson, and R. C. Stevens. 2008. Stabilization of
the human b2-adrenergic receptor TM4-TM3-TM5 helix interface by
mutagenesis of Glu122(3.41), a critical residue in GPCR structure.
J. Mol. Biol. 376:1305–1319.
15. Rosenbaum, D. M., V. Cherezov, ., B. K. Kobilka. 2007. GPCR
engineering yields high-resolution structural insights into b2-adrenergic
receptor function. Science. 318:1266–1273.
16. Gordeliy, V. I., R. Schlesinger, ., J. Heberle. 2003. Crystallization
in lipidic cubic phases: a case study with bacteriorhodopsin. Methods
Mol. Biol. 228:305–316.
17. Oesterhelt, D., and W. Stoeckenius. 1974. Isolation of the cell
membrane of Halobacterium halobium and its fractionation into red
and purple membrane. Methods Enzymol. 31(Pt A):667–678.
18. Nollert, P. 2004. Lipidic cubic phases as matrices for membrane protein
crystallization. Methods. 34:348–353.
19. Cheng, A., B. Hummel, ., M. Caffrey. 1998. A simple mechanical
mixer for small viscous lipid-containing samples. Chem. Phys. Lipids.
20. Caffrey, M., and V. Cherezov. 2009. Crystallizing membrane proteins
using lipidic mesophases. Nat. Protoc. 4:706–731.
21. Briggs, J., H. Chung, and M. Caffrey. 1996. The temperature-composi-
tion phase diagram and mesophase structure characterization of the
monoolein/water system. J. Phys. II. 6:723–751.
22. Qiu, H. 1998. The mesophase behavior of the monoacylglycerol/water
system: application in drug delivery. PhD thesis. The Ohio State
University, Columbus, OH.
23. Briggs, J. 1994. The phase behavior of hydrated monoacylglycerols and
the design of an x-ray compatible scanning calorimeter. PhD thesis. The
Ohio State University, Columbus, OH.
24. Qiu, H., and M. Caffrey. 1998. Lyotropic and thermotropic phase
behavior of hydrated monoacylglycerols: structure characterization of
monovaccenin. J. Phys. Chem. B. 102:4819–4829.
25. Lakowicz, J. R. 1983. Instrumentation of Fluorescence Spectroscopy.
Plenum Press, New York, NY.
26. Liu, W., and M. Caffrey. 2005. Gramicidin structure and disposition in
highly curved membranes. J. Struct. Biol. 150:23–40.
27. Liu, W., and M. Caffrey. 2006. Interactions of tryptophan, tryptophan
peptides, and tryptophan alkyl esters at curved membrane interfaces.
28. Smoot, A. L., M. Panda, B. T. Brazil, A. M. Buckle, A. R. Fersht.,
2001. The binding of bis-ANS to the isolated GroEL apical domain
fragment induces the formation of a folding intermediate with increased
hydrophobic surface not observed in tetradecameric GroEL. Biochem-
29. Cherezov, V., M. A. Hanson, ., R. C. Stevens. 2009. Rastering
strategy for screening and centering of microcrystal samples of human
membrane proteins with a sub-10 microm size x-ray synchrotron beam.
J. R. Soc. Interface. 6(Suppl 5):S587–S597.
30. Misquitta, Y., V. Cherezov, ., M. Caffrey. 2004. Rational design
of lipid for membrane protein crystallization. J. Struct. Biol. 148:
31. Caffrey, M. 2009. Crystallizing membrane proteins for structure deter-
mination: use of lipidic mesophases. Annu. Rev. Biophys. 38:29–51.
32. Cherezov, V., J. Clogston, ., M. Caffrey. 2002. Membrane protein
crystallization in meso: lipid type-tailoring of the cubic phase. Biophys.
33. Ghanouni, P., H. Schambye, ., B. K. Kobilka. 2000. The effect of pH
on b(2) adrenoceptor function. Evidence for protonation-dependent
activation. J. Biol. Chem. 275:3121–3127.
34. Murgia, S., F. Caboi, ., T. Nylander. 2002. Acyl migration and
hydrolysis in monoolein-based systems. Progr. Colloid Polym. Sci.
Biophysical Journal 98(8) 1539–1548
Thermal Stability of Membrane Proteins1547
35. Caffrey, M. 2008. On the mechanism of membrane protein crystalliza-
tion in lipidic mesophases. Cryst. Growth Des. 8:4244–4254.
36. Qiu, H., and M. Caffrey. 2000. The phase diagram of the monoolein/
water system: metastability and equilibrium aspects. Biomaterials.
37. Cherezov, V., H. Fersi, and M. Caffrey. 2001. Crystallization screens:
compatibility with the lipidic cubic phase for in meso crystallization
of membrane proteins. Biophys. J. 81:225–242.
38. Vargas, R., L. Mateu, and A. Romero. 2004. The effect of increasing
concentrations of precipitating salts used to crystallize proteins on the
structure of the lipidic Q224 cubic phase. Chem. Phys. Lipids.
39. Misquitta, Y., and M. Caffrey. 2003. Detergents destabilize the cubic
phase of monoolein: implications for membrane protein crystallization.
Biophys. J. 85:3084–3096.
40. Gordeliy, V. I., V. Cherezov, and J. Teixeira. 2005. Strength of thermal
undulations of phospholipid membranes. Phys. Rev. E Stat. Nonlin. Soft
Matter Phys. 72:061913.
41. Killian, J. A. 1998. Hydrophobic mismatch between proteins and lipids
in membranes. Biochim. Biophys. Acta. 1376:401–415.
42. Balmbra, R. R., J. S. Clunie, ., J. F. Goodman. 1962. Effect of temper-
ature on the micelle size of a homogeneous non-ionic detergent. Trans.
Faraday Soc. 58:1661–1667.
43. Schwartz, T. W., T. M. Frimurer, ., C. E. Elling. 2006. Molecular
mechanism of 7TM receptor activation—a global toggle switch model.
Annu. Rev. Pharmacol. Toxicol. 46:481–519.
44. Cherezov, V., J. Clogston, ., M. Caffrey. 2006. Room to move: crys-
tallizing membrane proteins in swollen lipidic mesophases. J. Mol. Biol.
45. Vedadi, M., F. H. Niesen, ., A. M. Edwards. 2006. Chemical
screening methods to identify ligands that promote protein stability,
protein crystallization, and structure determination. Proc. Natl. Acad.
Sci. USA. 103:15835–15840.
46. Price 2nd, W. N., Y. Chen, ., J. F. Hunt. 2009. Understanding the
physical properties that control protein crystallization by analysis of
large-scale experimental data. Nat. Biotechnol. 27:51–57.
Biophysical Journal 98(8) 1539–1548
1548Liu et al.