1006
SHORT COMMUNICATIONS
J. Appl. Cryst. (1999). 32, 1006±1009
An improved method for protein crystal density measurements
Adelaine K. W. Leung,a Michael M. V. Parkb and David W. Borhania* at aDepartment of Organic Chemistry, Southern
Research Institute, Birmingham, AL 35205, USA, and bJefferson County International Baccalaureate School, Birmingham, AL
35210, USA. E-mail: borhani@sri.org
(Received 23 December 1998; accepted 14 May 1999 )
Abstract
Determination of the density of protein crystals by ¯otation in
organic solvent density gradients using simple methods for the
preparation of the density gradients and the transfer of crystals
into these gradients is described. The method was tested with
crystals of hen egg white lysozyme. These methods are
especially suitable for use with fragile, high-solvent-content
protein crystals. These methods were applied to the measure-
ment of the density of human apolipoprotein A-I crystals.
1. Introduction
In a protein crystal structure determination, it is usually
necessary to determine the number of protein molecules in the
asymmetric unit of the crystal (Matthews, 1968, 1985). This can
be accomplished by direct measurement of the number of
protein molecules in a macroscopic crystal (by measurement of
the crystal volume and quantitative amino acid analysis of the
crystal contents) (Kiefersauer et al., 1996), or by the
measurement of the density of the crystal. As the ®rst method
requires complicated and specialized apparatus, most labora-
tories have focused on density measurements instead. There
are two main ways of measuring this density: ¯otation in a
Ficoll density gradient (Westbrook, 1976, 1985), and ¯otation
in an organic solvent density gradient (Low & Richards, 1952;
Colman & Matthews, 1971). Both methods have drawbacks.
With the Ficoll method, the extremely viscous density
gradients [formed from 30% and 60% (w/w) Ficoll] are very
dif®cult to prepare. Also, only one data point (the density of
the crystal in the Ficoll solution, which approximates that of
the crystal in which the interstitial solvent is replaced by pure
water) is obtained, thus rendering the measurement more
susceptible to systematic and random errors. In addition, some
crystals disintegrate partially or completely in Ficoll solutions,
making the density measurement dif®cult or impossible.
Finally, the transfer of fragile crystals into the gradient can be
dif®cult.
The organic solvent density-gradient method has the distinct
advantage that the densities of several crystals, previously
equilibrated in harvest buffers of varying density, can be
measured. These multiple data points reduce the in¯uence of
random errors, while simultaneously allowing the determina-
tion of the percentage of bound water and free solvent in the
crystal lattice (Colman & Matthews, 1971). Unfortunately,
these gradients can also be dif®cult to prepare, and the transfer
of crystals freed of excess mother liquor into the gradient is
tricky.
We have devised a modi®ed method for using organic
solvent (ethylbenzene and iodobenzene) density gradients that
# 1999 International Union of Crystallography
Printed in Great Britain ± all rights reserved
is an improvement over existing procedures. The modi®cations
address the preparation of the gradient, the removal of excess
mother liquor from the crystals, and the transfer of the crystals
into the gradient. We tested this modi®ed method by
measuring the density of hen egg white lysozyme crystals. We
then applied the method to measure the density of crystals of a
major fragment of human apolipoprotein A-I (apo A-I),
apo
(1±43)A-I (residues 44 to 243 of apo A-I). This crystal
density helped to de®ne the number of molecules of apo
(1±
43)A-I in the asymmetric unit, which was essential for the
correct interpretation of the electron density map at 4 A Ê
resolution (Borhani et al., 1997). Although we cannot state
whether all protein (or nucleic acid) crystals will be stable in
these gradients, two very different protein crystals were: hen
egg white lysozyme (pI 11, 34% exchangeable solvent,
crystals grown at 293 K, pH 4.7) and apo
(1±43)A-I (pI 6.5,
75% solvent, 277 K, pH 7.0). We think that this modi®ed
procedure will prove useful to other crystallographers working
with sensitive high-solvent-content crystals.
2. Methods
2.1. Density-gradient formation
A density gradient of water-saturated iodobenzene ( =
1.838 g mlÿ1) and water-saturated ethylbenzene ( =
0.866 g mlÿ1) was prepared using the apparatus shown in
Fig. 1. Two 20 ml disposable syringes (Becton-Dickinson
309661) equipped with 18 Ga needles were charged with 15 ml
of iodobenzene and 15 ml of air, and were clamped in place. A
25 ml Erlenmeyer ¯ask, charged with 15 ml of ethylbenzene
and a Te¯on-coated magnetic stir bar (13 mm), was sealed
with a septum (Aldrich Z10,074±9). The septum was carefully
pierced with two stainless-steel double-ended needles (18 Ga;
Aldrich Z10,242±3; cut to 50 mm length) connected by poly-
ethylene tubing (Clay Adams Intramedic PE-160; 18 Ga
needles ®t into this tubing) to the syringes. The septum was
pierced with a third double-ended needle (100 mm) connected
by polyethylene tubing to a fourth double-ended needle
(220 mm), which extended to the bottom of a 25 ml graduated
cylinder (Pyrex No. 3046, 0.2 ml graduations). The needles
were cut to the desired length by scoring with a triangular ®le
followed by snapping at the score mark. Care must be taken to
ensure that the various connections are tight. Air leaks, due to
poor septum/needle seals, lead to incorrect delivery of the
solvents to the graduated cylinder. The gradient in such a case
is, however, usually still usable.
While stirring vigorously, the plungers of both syringes were
slowly depressed at the same rate by pressing on the wooden
block (approximately two drops of iodobenzene per second).
Journal of Applied Crystallography
ISSN 0021-8898 # 1999
 SHORT COMMUNICATIONS
In this manner, a linear density gradient was formed; for each
volume of iodobenzene added to the ¯ask, twice this volume of
the ethylbenzene/iodobenzene mixture was delivered to the
cylinder. Gradient formation was complete when both syringes
were emptied. Care should be taken to stop before air is
pushed to the bottom of the gradient; air bubbles rising
through the gradient lead to undesirable mixing. At this point,
the double-ended needle was slowly withdrawn from the
graduated cylinder. If desired, a slightly larger amount of
iodobenzene and air (e.g. 16 ml) can be used, to provide a
`cushion' at the bottom of the gradient. Furthermore, the
organic solvents can be varied to obtain gradients that cover
different density ranges (e.g. appropriate mixtures of ethyl-
benzene and iodobenzene can be combined to form shallow
gradients for very accurate crystal density measurements).
Gradients prepared in this manner at both 277 K and room
temperature were stable for at least 24 h, allowing them to be
prepared in advance. Occasionally, the gradients became hazy
in the centre, apparently due to the lower solubility of water in
the ethylbenzene/iodobenzene mixture. This haziness
appeared not to affect the crystal density measurements, but
could be avoided by using dry iodobenzene and water-satu-
rated ethylbenzene.
This method of preparing the density gradient overcomes
two dif®culties: (i) standard (open to air) gradient makers
cannot be used, because of the great difference in density
between ethylbenzene and iodobenzene, which leads to irre-
gular and overly rapid entry of iodobenzene into the mixing
chamber of the gradient former; (ii) the apparatus is resistant
to organic solvents, unlike standard gradient makers
constructed of various plastics.
2.2. Density-gradient calibration
The gradient was calibrated with small droplets (diameter
0.5 mm) of H2O and aqueous salt solutions (e.g. CsCl) of
known density. Densities were determined gravimetrically in
5 ml `Class A' volumetric ¯asks; weights were corrected for
buoyancy. The ¯ask-®lling and weighing procedure was
Fig. 1. The density-gradient apparatus. A: Magnetic stirrer. B:
Erlenmeyer ¯ask charged with ethylbenzene and a stirring bar. C:
Septum. D: Double-ended needles connected with polyethylene
tubing. E: Graduated cylinder. F, G: Syringes charged with
iodobenzene and air. H: Wooden block or equivalent.
1007
checked by determining the density of deionized water ( =
0.9971 g mlÿ1 at 298 K). Droplets (most dense ®rst, least dense
last) were added to the top of the gradient (one side only, to
minimize the likelihood of a crystal hitting a droplet later)
using a ®ne ¯ame-drawn Pasteur pipette: the droplet is
partially squeezed out of the capillary under the surface of the
ethylbenzene and is then dislodged from the end of the
capillary by lifting the capillary out of the gradient. The
droplets fell to their equilibrium position within 30 s. This
position was read to the nearest 0.05 ml (avoid parallax
errors). It was important to read the equilibrium position
immediately, because an upward drift of some of the droplets
was noticeable after 5±10 min. This drift, which was progres-
sively larger at greater depths in the gradient, appeared to be
due to the absorption of water from the iodobenzene (the
droplets which were highest in the gradient, in the ethylben-
zene-rich phase, did not drift at all). Recalibration after several
hours revealed that the gradient itself was perfectly stable: the
original positions of all the droplets were reproducible to
within 0.0±0.2 ml.
2.3. Crystal preparation
The glass plate of a stereo microscope (Nikon SMZ-U) was
wrapped tightly with Dura-Seal (Diversi®ed Biotech, DS4-
500), a compliant ®lm of polyethylene. This ®lm is transparent
and solvent-resistant, and it can be stretched and rubbed to
remove wrinkles and air pockets. The crystal, which had been
previously equilibrated in the desired mother liquor, was
deposited onto the Dura-Seal using a 1 mm glass capillary
(Charles Supper Co.) attached to a 1 ml syringe. (Alternatively,
a cryocrystallography loop can be used.) While observing the
crystal through the microscope, excess mother liquor was
carefully removed from the crystal by using paper wicks and a
®nely drawn glass capillary. The dry crystal was immediately
¯ooded with several drops (100 ml) of water-saturated
ethylbenzene.
2.4. Crystal transfer to the density gradient
The crystal was picked up from the ethylbenzene pool by
applying gentle suction (pipette or 1 ml syringe) to a 200 ml
plastic pipette tip (with its end cut off to provide a 1 mm
opening). Also, it was helpful to bend the pipette tip in the
middle by about 90 . (A glass capillary was less useful.) The
crystal in ethylbenzene was placed into the top of the density
gradient. Release of suction and a gentle shake were usually
suf®cient to release the crystal. The crystal fell rapidly (usually
15±30 s) to its equilibrium point in the gradient, which was
read immediately to the nearest 0.05 ml. Even very small
crystals (0.2 mm diameter) could be easily observed as they fell
through the gradient.
3. Example 1: lysozyme
Tetragonal crystals of hen egg white lysozyme (Sigma L-6876)
were grown in hanging drops at 293 K by mixing 5 ml of lyso-
zyme (80 mg mlÿ1 in 40 mM NaOAc, pH 4.7) with 5 ml of well
solution [0.8±1.2 M NaCl, 40 mM NaOAc, pH 4.3±4.8
(Alderton & Fevold, 1946)]. All harvest buffers used below
contained 40 mM NaOAc, pH 4.7, and varying amounts of
NaCl and/or CsCl. Crystals (0.2±1.5 mm diameter) were
harvested (1 M NaCl) and then transferred, in a nine-well
depression plate, through solutions of increasing NaCl
1008 SHORT COMMUNICATIONS
concentration (0.25 M steps, 2±5 min each) to 2 M NaCl. Some
crystals were left to soak in either the 1 M NaCl or the 2 M
NaCl harvest buffers. The rest were transferred to 2 M CsCl, by
sequential passage through 0.5 M CsCl/1.5 M NaCl, 1 M CsCl/
1 M NaCl, and 1.5 M CsCl/0.5 M NaCl. Finally, the crystals
were transferred to various CsCl solutions (2.5 M to saturated,
in 0.5 M steps). Crystals were allowed to equilibrate for
between 30 min and several hours, either in the depression
wells (sealed with a cover slip) or in small ¯at-bottomed vials
containing 150 ml of the appropriate buffer. The densities of
the NaCl and CsCl harvest buffers were determined as
described above.
The equilibrated lysozyme crystals were dried and trans-
ferred to an ethylbenzene/iodobenzene density gradient
(calibrated with the NaCl and CsCl harvest buffers) as
described above. The positions of the standards in the gradient
(range 8.95±22.25 ml, which encompassed the observed crystal
positions) were linearly related to the densities of the stan-
dards (range 1.5907±1.1949 g mlÿ1; correlation coef®cient
0.991). Crystal density was determined from the position of the
crystal in the gradient by back-calculation using this linear
relationship. The crystals had densities ranging from 1.2354 to
1.5354 g mlÿ1 (Table 1).
Following the formalism of Colman & Matthews (1971), the
crystal density (Dc) is the sum of three component densities,
namely, the protein density (Dp), the density of the tightly
bound water (Dw), and that of the freely exchangeable solvent
(Ds), taken in proportion to the fractional volume of each
component. This relationship is expressed by
Dc ˆ Dp …Vp =V† ‡ Dw …Vw =V† ‡ Ds …Vs =V†; …1†
where V is the unit-cell volume. Therefore, as the solvent
(mother liquor) density Ds is varied, the apparent crystal
density Dc varies as a linear function of Ds with slope Vs=V,
assuming that the other parameters are unaffected by changes
in mother-liquor composition. Vs, Vp, Vw and n, the number of
protein molecules in the asymmetric unit of the crystal, can
then be derived from the slope and intercept values, Avoga-
dro's number, the unit-cell volume and symmetry, and the
protein molecular weight and partial speci®c volume (Colman
& Matthews, 1971). The lysozyme crystal density was linearly
related to the mother-liquor density (Table 1): Dc = 0.882 +
0.345Ds (correlation coef®cient 0.977). Combining this result
with the unit-cell volume and symmetry [space group P43212; a
Ê ; V = 237 133 A
= 79.1, c = 37.9 A Ê 3 (Blake et al., 1965)], protein
molecular weight (14 313 Da) and protein partial speci®c
volume [0.703 ml gÿ1 (Squire & Himmel, 1979)], yielded Vp=V
= 54%, Vw=V = 12%, Vs=V = 34% and n = 0.96. These
experimental values compare very favourably with those
calculated from the known crystal structure [101 bound water
molecules (Protein Data Bank entry 6lyz; Diamond, 1974):
Vp=V = 56%, Vw=V = 10%, Vs=V = 34% and n = 1].
4. Example 2: apo D(1±43)A-I
Large bipyramidal crystals of apo
(1±43)A-I were grown in
hanging drops at 277 K from 5 ml of apo
(1±43)A-I
(19.7 mg mlÿ1) mixed with 5 ml of well solution (1.2 M Na3-
citrate, 100 mM NaHEPES, pH 7.5). These crystals belong to
Ê . The
space group P212121; a = 97.47, b = 113.87, c = 196.19 A
crystals were harvested into an arti®cial mother liquor
consisting of 8 volumes of 1.8 M Na3citrate, 0.5 volumes of
Table 1. Density of hen egg white lysozyme crystals as a function
of mother-liquor density
Mother-liquor Mother-liquor Crystal density
composition² density (g mlÿ1)³ (g mlÿ1)§
Saturated CsCl 1.8985 1.5354
6.0 M CsCl 1.7812 1.4844
5.5 M CsCl 1.7194 1.4844
4.5 M CsCl 1.5907 1.4319
3.5 M CsCl 1.4623 1.3704
2.5 M CsCl 1.3297 1.3839
2.0 M NaCl 1.0721 1.2399
1.0 M NaCl 1.0369 1.2354
1.0 M NaCl 1.0369 1.2369
² Indicated molarity of CsCl or NaCl, plus 40 mM NaOAc, pH
4.7. ³ Determined gravimetrically. § Position in the density
gradient was converted to density by calculation from a linear
regression of gradient position versus density of seven standard
solutions of known (gravimetrically determined) density.
2.0 M NaHEPES, pH 7.5, and 1.5 volumes of H2O. Crystals
were then transferred to six harvest buffers of varying density
prepared from 3.5 M Cs2SO4, 1.8 M Na3citrate, 3.0 M
K3citrate, and H2O (Table 2). Crystals were allowed to equi-
librate in these harvest buffers at 277 K for 18 h.
The equilibrated apo
(1±43)A-I crystals were dried and
transferred to an ethylbenzene/iodobenzene density gradient
at 277 K, as described above. The positions of the standards in
the gradient (range 4.45±25.30 ml, which encompassed the
observed crystal positions) were linearly related to their
densities (range 1.7488±1.0000 g mlÿ1; correlation coef®cient
0.998). Crystal density was determined from the position of the
crystal in the gradient by back-calculation using this linear
relationship. The crystals had densities ranging from 1.2513 to
1.6816 g mlÿ1 (Table 2).
The apo
(1±43)A-I crystal density was linearly related to
the mother-liquor density (Table 2): Dc = 0.333 + 0.749Ds
(correlation coef®cient 0.941). Combining this result with the
unit-cell volume (2 177 495 A Ê 3) and symmetry (primitive
orthorhombic), protein molecular weight (23 264 Da), and
protein partial speci®c volume [0.725 ml gÿ1 (Brouillette,
1996)], yielded Vp=V = 22%, Vw=V = 3%, Vs=V = 75% and n =
4.24. The high exchangeable solvent volume fraction of 75%
turned out to be completely consistent with the apo
(1±
43)A-I crystal structure when it was solved (four molecules in
the asymmetric unit, n = 4), in which extremely large connected
solvent channels run throughout the crystal (Borhani et al.,
1997).
The self-rotation function (SRF) of the apo
(1±43)A-I
crystals exhibited a strong noncrystallographic dyad in the ab
plane, as well as two rather weak peaks that, when combined
with the dyad, were indicative of noncrystallographic 222
pseudosymmetry (Borhani et al., 1997). Although the SRF
result by itself was not suf®ciently clear to de®ne unambigu-
ously the contents of the asymmetric unit, the interpretation of
this result was bolstered by the crystal density results. Taken
together, these results suggested strongly that the apo

(1±43)A-I crystals had four molecules in the asymmetric
unit. This con®dent knowledge of the molecular contents of
the apo
(1±43)A-I crystals proved to be very helpful in the
correct interpretation of the experimental electron density
maps.
 SHORT COMMUNICATIONS
Table 2. Density of apo
(1±43)A-I crystals as a function of
mother-liquor density
Mother-liquor Mother-liquor Crystal density
composition² density (g mlÿ1)³ (g mlÿ1)§
Cs:Na, 2:1 1.7488 1.6816
Cs:Na, 1:1 1.6347 1.5327
Cs:Na, 1:2 1.5211 1.4419
Cs:Na, 1:5 1.4009 1.4220
K:H2O, 2:1 1.3630 1.3112
K:H2O, 1:2 1.1929 1.2513
² Stock solutions: Cs: 3.5 M Cs2SO4; Na: 1.8 M Na3citrate; K: 3.0 M
K3citrate. ³ Determined gravimetrically. § Position in the density
gradient was converted to density by calculation from a linear
regression of gradient position versus density of 17 standard solutions
of known (gravimetrically determined) density.
5. Sources of error
The procedure we have described is susceptible to several
possible sources of error, subject to the assumption that the
mother-liquor densities have been measured accurately. First,
the solvent content of a crystal can change (e.g. by absorption
of water) as it falls through the gradient, leading to systematic
errors in the measured densities. It is dif®cult to determine how
quickly solvent exchange occurs, though it is likely quicker and
more severe for smaller crystals. The problem can be mitigated
by using larger crystals, and in some cases by using dry rather
than water-saturated solvents. Similarly, crystals grown from
organic solvents (e.g. ethanol, MPD) may be particularly
sensitive to loss of these solvents into the gradient. Second,
application of equation (1) assumes that the parameters Dp, Vw
and, especially, Vp and V are independent of varying mother-
liquor composition. This may not be true for all crystals; it was
clearly not true for apo
(1±43)A-I. Although the crystals
were visually unaffected by soaking in various harvest buffers
(optically clear, sharp faces and edges), those crystals soaked
in the medium-density harvest buffers diffracted X-rays poorly,
and those soaked at higher densities did not diffract at all.
Thus, the assumption of a constant unit-cell volume V could
not be con®rmed with these crystals. This may explain the
1009
lower quality of these data compared to the lysozyme data.
Third, the crystal may not have been dried completely, or may
have adhering salt crystals. This problem was not encountered
with apo
(1±43)A-I, but it was noticed with lysozyme. We
found that ¯ooding the crystals with ethylbenzene immediately
upon mother-liquor removal was the best way to minimize this
problem. Saturated and near-saturated CsCl solutions were
used with lysozyme, because of its low solvent content, and
evaporation of these solutions was rapid at room temperature.
For these reasons, the lysozyme test crystals were actually
more dif®cult to handle than our apo
(1±43)A-I crystals!
Therefore, we recommend proceeding to a crystal of interest
after a test run with lysozyme if calibration of the gradient
indicates that a suitable linear gradient can be formed, even if
the lysozyme results prove less than satisfactory.
We thank Christie Brouillette and Jeffrey Engler for
supplying puri®ed apo
(1±43)A-I. DWB is an Established
Investigator of the American Heart Association (Grant No.
974008 N). This work was also supported by Southern
Research Institute (Project Nos. 1008 and 1026).
References
Alderton, G. & Fevold, H. L. (1946). J. Biol. Chem. 164, 1±5.
Blake, C. C. F., Koenig, D. F., Mair, G. A., North, A. C. T., Phillips, D. C.
& Sarma, V. R. (1965). Nature (London), 206, 757±761.
Borhani, D. W., Rogers, D. P., Engler, J. A. & Brouillette, C. G. (1997).
Proc. Natl Acad. Sci. USA, 94, 12291±12296.
Brouillette, C. G. (1996). Personal communication.
Colman, P. M. & Matthews, B. W. (1971). J. Mol. Biol. 60, 163±168.
Diamond, R. (1974). J. Mol. Biol. 82, 371±391.
Kiefersauer, R., Stetefeld, J., Gomis-RuÈth, F. X., RomaÄo, M. J.,
Lottspeich, F. & Huber, R. (1996). J. Appl. Cryst. 29, 311±317.
Low, B. W. & Richards, F. M. (1952). J. Am. Chem. Soc. 74, 1660±1666.
Matthews, B. W. (1968). J. Mol. Biol. 33, 491±497.
Matthews, B. W. (1985). Methods Enzymol. 114, 176±187.
Squire, P. G. & Himmel, M. E. (1979). Arch. Biochem. Biophys. 196,
165±177.
Westbrook, E. M. (1976). J. Mol. Biol. 103, 659±664.
Westbrook, E. M. (1985). Methods Enzymol. 114, 187±196.