Protein Engineering vol.12 no.6 pp.

439446, 1999

Crystal structure of human serum albumin at 2.5 resolution

S.Sugio1, A.Kashima, S.Mochizuki, M.Noda and

K.Kobayashi2
Osaka Laboratories, Yoshitomi Pharmaceutical Industries Ltd,
2-25-1, Shodai-Ohtani, Hirakata, Osaka 573-1153, Japan
1Present address: Mitsubishi Chemical Corporation, Yokohama 227-8502,
Japan
2To

whom all correspondence should be addressed

A new triclinic crystal form of human serum albumin

(HSA), derived either from pool plasma (pHSA) or from
a Pichia pastoris expression system (rHSA), was obtained
from polyethylene glycol 4000 solution. Three-dimensional
structures of pHSA and rHSA were determined at 2.5
resolution from the new triclinic crystal form by molecular
replacement, using atomic coordinates derived from a
multiple isomorphous replacement work with a known
tetragonal crystal form. The structures of pHSA and rHSA
are virtually identical, with an r.m.s. deviation of 0.24
for all C atoms. The two HSA molecules involved in the
asymmetric unit are related by a strict local twofold
symmetry such that the C atoms of the two molecules can
be superimposed with an r.m.s. deviation of 0.28 in
pHSA. Cys34 is the only cysteine with a free sulfhydryl
group which does not participate in a disulfide linkage
with any external ligand. Domains II and III both have a
pocket formed mostly of hydrophobic and positively
charged residues and in which a very wide range of
compounds may be accommodated. Three tentative binding
sites for long-chain fatty acids, each with different surroundings, are located at the surface of each domain.
Keywords: crystal structure/human serum albumin/recombinant protein

Introduction
Human serum albumin (HSA) is the most abundant protein
found in plasma and shows a typical blood concentration of
5 g/100 ml. Its physiological and pharmacological properties
have been extensively studied over several decades (Fehske
et al., 1981; Kragh-Hansen, 1981; Putnam, 1984; Peters, 1985).
Such studies have revealed that HSA has a high affinity to a
very wide range of materials, including metals such as Cu21
and Zn21, fatty acids, amino acids, metabolites such as
bilirubin and for many drug compounds. The most important
physiological role of the protein is therefore thought to be to
bring such solutes in the bloodstream to their target organs,
as well as to maintain the pH and osmotic pressure of
plasma. In addition to its ordinary clinical applications, such
as hypovolemic shock treatment, many investigators have
attempted to utilize HSA as a carrier to deliver various drugs
to their specific targets (Yapel, 1985; Fiume et al., 1988). The
primary sequence of HSA shows that the protein is a single
polypeptide with 585 residues containing 17 pairs of disulfide
bridges and one free cysteine (Dugiaczyk et al., 1982). Human
Oxford University Press

serum albumin, as well as serum albumin from other species,

has been found to consist of three homologous domains
probably derived through gene multiplication (Brown, 1976).
Despite many investigations using hydrodynamics, smallangle X-ray diffraction, electron microscopy and structural
prediction, its three-dimensional molecular structure has
remained largely unknown. Several crystal forms of HSA were
reported in the 1970s (McClure and Craven, 1974; Rao et al.,
1976), but they provided no structural information due to their
poor reproducibility. The first crystal structure of HSA at low
resolution was reported by Carter and co-workers in 1989
(Carter et al., 1989; Carter and He, 1990), and its refined
structure at 2.8 resolution was published by the same group
(He and Carter, 1992). The research group has also succeeded
through extensive efforts in producing half a dozen different
crystal forms of both HSA and non-human serum albumin,
improving the resolution of their crystal structures and enhancing the knowledge of the diverse chemistry of serum albumin
(Ho et al., 1993; Carter and Ho, 1994; Carter et al., 1994).
Recently, a couple of structures of HSA have been published
by another research group (Curry et al., 1998), in which five
myristates are accommodated in fatty acid sites. Not only did
these studies shed light on the structural features of the
protein, but knowledge of the three-dimensional structure also
contributed to the clarification of how the protein binds various
ligands. In this article, we describe a new crystal form of
defatted HSA and discuss various molecular aspects of the
protein as determined at the highest resolution so far reported.
Materials and methods
Crystallization of the tetragonal crystal
pHSA was purchased from Sigma Chemical Co. (catalog no.
A-3782), and was used for crystallization without further
purification. Crystallization took place via a conventional
hanging drop procedure. Plate-shaped yellowish crystals
(1.0 3 1.0 3 0.2 mm) were grown from a solution containing
150255 mg/ml protein, 50 mM potassium phosphate (pH 5.0
5.5), 3038% (v/v) polyethylene glycol 400 and 5 mM sodium
azide at 293 K over 23 weeks. Preliminary diffraction studies
revealed that the crystals belonged to the tetragonal space
group P4212 with cell dimensions a 5 187.1 , c 5 80.5 .
This crystal form is isomorphous to that previously reported
(Carter et al., 1989; Carter and He, 1990), containing one
HSA molecule in an asymmetric unit. Due to their high solvent
content (77%), diffraction of these crystals only reaches a
resolution of 3.0 .
Structure determination of the tetragonal crystal
Intensity data from a native crystal were collected with a
Rigaku R-AIXS IV area-detector on a Rigaku FR X-ray
generator ( 5 1.5418 ). Intensity data from derivative
crystals were collected with a Rigaku R-AXIS IIc area-detector
on a Rigaku RU-200H X-ray generator. Data to 3.0 resolution
for the native crystal and to 3.6 or 4.0 for derivatives were
439

S.Sugio et al.

Table I. Statistics in data collection and phasing for the tetragonal crystal
Crystal

Soaking
(mM)

Conditions
(h)

Resolution
()

No. of
refls.

Rmerge
(%)a

Completeness
(%)b

Figure of
merit

Phasing
powerc

Native
HgCl2-1
HgCl2-2
Thimerosal
Mersalyl-1
Mersalyl-2

0.53
0.50
1.04
0.51
0.50

29
24
24
24
72

3.00
3.56
3.55
3.56
4.00
3.55

26761
13534
12332
12498
9198
14118

9.9
12.5
13.4
11.1
11.1
11.3

92.0
78.2
70.9
71.3
75.4
80.9

0.461
0.464
0.427
0.415
0.452

1.665
1.708
1.232
1.265
1.552

merge 5 h (j|I(h)j I(h)|)/h jI(h).

bCompleteness 5 (No. of reflections observed)/(No.
cPhasing power 5 |F |/[(|F
2 1/2
H
PHobs| |FPHcalc|) ] .
aR

of reflections theoretically existing).

Table II. Statistics in data collection and refinement for the triclinic crystal
pHSA

rHSA

Intensity data statistics

No. of crystals used
Resolution range ()
No. of unique reflections (I.I)
Mean I/Ia
R mergeb (%)a
Completeness (%)a

2
93.62.38
40 737
9.4/2.4
9.7/27.2
81.4/46.9

5
93.52.37
36 149
4.4/1.5
13.4/42.8
71.3/36.1

Refinement statistics
Resolution range ()
No. of total reflections
R factorc,d
Free R factore (%)c
Completeness (%)c

50.02.5
37 461
21.8/26.0
28.2/34.7
86.2/64.8

50.02.5
33 040
20.7/29.2
29.2/42.0
77.1/48.1

No. of atoms
Protein
Solvent

460032
7

460032
4

R.m.s.d. from ideal values

Bond lengths ()
Bond angles ()
Dihedral angles ()
Improper angles ()

0.014
1.72
21.0
1.59

0.014
1.75
21.1
1.69

Average B factor (2)

48.5

40.7

aOverall/highest

resolution bin (pHSA, 2.502.38 , rHSA, 2.502.37).

bRmerge 5 |I I|/I .
i
i
cOverall/highest resolution

bin (2.592.50 ).

dR factor 5 |F
obs Fcalc|/|Fobs|.
eCalculated from 10% randomly selected

reflections which were excluded

from the refinement.

measured at 293 K. All diffraction data were processed with

PROCESS (Sato et al., 1992). Data collection statistics are
summarized in Table I. Isomorphous difference Patterson maps,
calculated at 4.0 resolution using the XtalView program
package (McRee, 1993), revealed that crystals soaked in HgCl2,
thimerosal or mersalyl could be used for phasing with multiple
isomorphous replacement. Heavy-atom sites for the HgCl2
derivative were located using the HASSP program (Terwilliger,
1994), and refined using a correlation search method implemented in XtalView. Subsequent cross difference Fourier
syntheses with phases derived solely from the HgCl2 derivative
clearly showed all the heavy-atom sites for other derivatives.
Initial phases at 3.6 resolution were obtained via the multiple
isomorphous replacement method with all derivatives. Phasing
statistics are listed in Table I. The initial phases were then
improved and extended to 3.0 through iterative cycles of
440

Fig. 1. Schematic drawing of the HSA molecule. Each subdomain is marked

with a different color (yellow for subdomain Ia; green, Ib; red, IIa; magenta,
IIb; blue, IIIa; and cyan, IIIb). N- and C-termini are marked as N and C,
respectively. Arg117, Lys351 and Lys475, which may be binding sites for
long-chain fatty acids, are colored white.

solvent flattening (Wang, 1985) implemented in the PHASES

program (Furey, 1994). In the first electron density map, a
number of helical features were clear and easy to interpret,
and a poly-alanine model containing a few cysteine and
proline residues was built using the molecular graphics program
QUANTA 4.1 (Molecular Simulation Inc.). The initial atomic
model consisted of 1806 non-hydrogen atoms from 359 residues, and was subjected to stereochemically restrained leastsquares refinement in X-PLOR (Brunger, 1992b) with the
parameter set proposed by Engh and Huber (1991). Sequence
assignment of the atomic model was carried out using the
amino acid sequence derived from the analysis of genomic
DNA (Minghetti et al., 1986). Further improvement of phases
were achieved using a density modification procedure in which
the partial model contribution was taken into account as well
as solvent flattening and histogram matching (Zhang and Main,
1990). The procedure was iterated using the CCP4 program
suite (Collaborative Computational Project Number 4, 1994)
and SQUASH (Zhang and Main, 1990), and no additional

Crystal structure of human serum albumin

Fig. 2. Stereographic illustration of C traces of three domains (A, domain I; B, domain II; C, domain III): the traces are superimposed by a least-squares
method with the corresponding C atoms, which are determined using a multiple sequence alignment of the HOMOLOGY program (Molecular Simulation,
Inc.), and are drawn in the same direction. All disulfide bands are drawn as thick lines. The side chains of Cys34 (in A), and residues forming site I (in C)
and site II (in C) are also shown as thick lines.

features emerged on the subsequent electron density map after

the fourth cycle of phase refinement. The last cycle of density
modification provided an atomic model containing 3794 out
of the 4652 possibly existing non-hydrogen atoms (81.6%) of
the HSA molecule, giving a crystallographic R factor of 31.4%
at 10.03.0 resolution. No further attempts were made to
refine this model with the data from the tetragonal crystal form.
Crystallization of the triclinic crystal
pHSA was purchased from Miles Inc. (catalog no. 82-301), and
rHSA was prepared using the procedure described previously
(Sumi et al., 1993; Ohi et al., 1994; Yokoyama and Ohmura,
1995). Prior to crystallization, fatty acids were removed from
the pHSA and rHSA solutions with powder-activated charcoal
at pH 3.0 with 1.0 N hydrochloric acid. After removal of the
charcoal, the solution was neutralized with 1.0 N sodium
hydroxide. An aliquot of the solution was then loaded on a
POROS 20-CM (PerSeptive Biosystems) which had been
equilibrated with 0.1 M sodium acetate (pH 4.5). Elution was
carried out with a linear salt gradient at NaCl concentrations
of 0.31.2 M at a pH of 4.55.0. Fractions corresponding to
the main peak were pooled, and the entire column work was

repeated until all the protein solution had been processed

by the Pharmacia FPLC system. The pooled fraction was
concentrated by ultrafiltration, and dialyzed against 10 mM
potassium phosphate (pH 7.0) containing 5 mM sodium azide.
The purified HSA was diluted to 120240 mg/ml with 5 mM
sodium azide. Crystallization was performed using a hanging
drop procedure where the reservoir solution contained 50 mM
potassium phosphate (pH 7.08.0), 2026% (w/v) polyethylene
glycol 4000 and 5 mM sodium azide. A few prismatic crystals
with an average size of 0.4 mm for each edge were grown at
288293 K within a month. Preliminary diffraction studies
revealed that crystals of both pHSA and rHSA obtained from
the above procedure diffract X-rays at up to 2.4 resolution,
belonged to the same space group, i.e. triclinic P1, and have
virtually identical unit cell parameters: a, 59.7/59.4 ; b,
97.0/96.9 ; c, 59.7/59.2 ; , 91.1/91.0; , 103.5/103.4; ,
75.1/74.9 (pHSA/rHSA). Two HSA molecules are involved
in the asymmetric unit with solvent content of 50/49%. The
unit cell volume of these crystals, 324600/319700 3, is
significantly different from that of any other crystal form of
serum albumin so far reported (McClure and Craven, 1974;
441

S.Sugio et al.

tightly restrained individual isotropic temperature factors were

incorporated. Although a local twofold symmetry was observed
between HSA molecules in the asymmetric unit, no restraints
based on this relation were applied during the refinement
process. Isolated peaks with significant height in Fo-Fc difference maps were carefully chosen as ordered water molecules
if they were located near the protein molecules. A free R factor
(Brunger, 1992a) for 10% of randomly selected reflections was
monitored for every cycle of the refinement to assess progress.
Crystallographic R factor and free R factor for the current
atomic model were 21.8 and 28.2%, respectively.
The triclinic structure of rHSA was refined in the same
way as described above, but with the refined structure of
pHSA as the starting model. Current R and free R were 20.7
and 29.2% at 502.5 resolution. The refinement statistics
for both pHSA and rHSA are shown in Table II. Atomic
coordinates and structure factors of triclinic pHSA and rHSA
have been deposited in the Brookhaven Protein Data Bank
with accession codes 1AO6 and 1BM0, respectively.
Results and discussion

Fig. 3. Schematic drawing of secondary structure elements and disulfide

bridges of HSA. Helices are represented by rectangles, and loops and turns
by thin lines. Disulfide bridges are drawn with thick lines. The sequence
nomenclature is derived from Minghetti et al. (1986).

Rao et al., 1976; Carter et al., 1989; Carter and He, 1990; He
and Carter, 1992; Ho et al., 1993; Carter and Ho, 1994).
Structural determination of the triclinic crystal
Diffraction data of the triclinic crystal form were measured at
288 or 293 K with a Rigaku R-AXIS IIc on a Rigaku RU200H X-ray generator. Data to a resolution of 2.38 for
pHSA and 2.37 for rHSA were collected from two and five
crystals, respectively, and were processed in the same way as
described above. Data collection statistics are summarized in
Table II. Structural determination of pHSA in the triclinic
crystal was performed using molecular replacement and crystallographic refinement facilities implemented in X-PLOR
(Brunger, 1992b). A real-space rotation function at a resolution
range of 10.04.0 followed by a PC-refinement with the
structure of the tetragonal crystal as a search model gave
two unique solutions (7.8 and 7.6) corresponding to two
molecules in an asymmetric unit. A subsequent translation
search showed unambiguously a solution from which the
relative positions of these molecules were derived. A rigidbody refinement followed by positional and overall temperature
factor refinements at a resolution range of 10.03.0 provided
a reasonable molecular packing in the triclinic crystal lattice
with a crystallographic R factor of 38.6%. The atomic model
obtained above was then subjected to a refinement procedure
where simulated annealing from 1000 to 300 K was followed
by conventional positional and overall temperature factor
refinements at 3.0 resolution. The resulting model was
manually revised using three-dimensional graphics. Hundreds
of side chain atoms which had not been observed in the
tetragonal crystal were identified and added to the model. All
reflections within a resolution range of 50.02.5 were then
included in the refinement, in which bulk solvent correction,
a resolution-dependent weighting scheme for reflections, and
442

Structural determination and quality

The structural determination of HSA was performed using two
different crystal forms which reveal complementary properties.
As the atomic coordinates of HSA had not been deposited in
the Brookhaven Protein Data Bank (Bernstein et al., 1977),
isomorphous replacement methods had to be employed to
solve the initial structure of the protein. Tetragonal crystals
were suitable for these experiments because of their high
tolerance of the changing conditions of the mother liquor
during heavy-atom derivative preparations. On the other hand,
all attempts to transfer triclinic crystals from the mother liquor
to any other solution failed. This was probably due to the
marked difference in packing density between the two crystal
forms. The solvent content of tetragonal crystals is very high
(77%) and HSA molecules are loosely packed in the crystal
lattice, while in triclinic crystals solvent content is rather low
(50%) and molecular packing is consequently tighter. After
the partial molecular model had been made, the subsequent
structure refinement proceeded using intensity data collected
from the triclinic crystals, which diffract better (up to 2.4
resolution) than the tetragonal crystals (below 3.0 ).
The current atomic coordinates of pHSA consist of two
HSA molecules, each of which contains residues 5582 (4600
non-hydrogen atoms), and seven water molecules, whereas
two copies of HSA (residues 5582) and four water molecules
are involved in the current coordinates of rHSA. An electron
density corresponding to residues 14 and residues 583585
of the HSA molecule is not clearly observed, probably due to
conformational flexibility at both termini. Geometrical analysis
of the atomic coordinates shown in Table II indicates that
bond distances, bond angles and dihedral angles were in good
agreement with the ideal values proposed by Engh and Huber
(1991). The peptide linkages at Pro96 adopt the cis-conformation. A Ramachandran plot (data not shown) reveals that
the backbone conformation for all residues falls into either
energetically favorable or allowed regions (Ramakrishnan
and Ramachandran, 1965; Laskowski et al., 1993). Average
coordinate error estimated from a Luzzati analysis (Luzzati,
1952) was about 0.3 on R (0.4 on free R) for both pHSA
and rHSA. The Fo-Fc difference map from the last refinement
cycle showed no significant features attributable to mistracing

Crystal structure of human serum albumin

Fig. 4. Stereographic representation of typical double disulfide bridges (Cys200Cys246 and Cys245Cys263) shown with electron density (contour level
1.5). Disulfide bridges are drawn with thick lines.

Fig. 5. Structure around Cys34 drawn in stereo with difference electron density (contour level 2.0): Gln33, Cys34, Pro35 and Thr83 face the proteinsolvent
interface. The sulfhydryl side chain of Cys34 is toward the interior of the protein, and is surrounded by residues Pro35, His39, Val77 and Tyr84.

or incorrect atomic coordinates. Average temperature factors

for pHSA and rHSA significantly differ residue by residue,
i.e. those for exposed residues tended to be 100 2, while
those for buried ones stayed at around 1520 2, the result of
which was the overall temperature factors, 48.5 2 for pHSA
and 40.7 2 for rHSA, were rather large.

Local symmetry in the triclinic lattice

Two HSA molecules are involved in the asymmetric unit of
the triclinic crystal, although HSA shows a mono-dispersive
state even in condensed solutions. The local twofold symmetry
is rather strict, although no assumption was made during the
refinement process regarding non-crystallographic symmetry
443

S.Sugio et al.

Fig. 6. Stereographic illustration of ligand binding site I (A) and site II (B) with charge distribution overlayed on the illustrations as dot surfaces (blue for
positive and red for negative charges, respectively). Charge distribution is derived from electrostatic potential calculations using the DelPhi program
(Molecular Simulation, Inc.). Helices h1, h2, h3, h4 and h6 in subdomains IIa and IIIa are colored magenta, yellow, green, orange and cyan, respectively.
Helix Ib-h3 (only shown in A) is colored grey.

restraints. The rotation angle to superimpose two crystallographically independent molecules in the asymmetric unit is
179.8, and translation vector length is less than 0.15 for
both pHSA and rHSA. These two independent molecules in
the pHSA structure have virtually identical structures, with
r.m.s. deviations of 0.28, 0.31 and 0.66 for equivalent
C atoms, backbone atoms and all non-hydrogen atoms,
respectively (0.36, 0.40 and 0.76 for rHSA). Further analysis
of the residue basis clearly showed that the r.m.s. deviations
for C atoms are fairly constant throughout the polypeptide
chain. All the regions in which structural deviations are
relatively distinct are found in loop conformations which
connect two -helices, and r.m.s. deviations in these regions
remain below 1.2 . No significant correlation was found
between the residual plots of r.m.s. deviations and those of
average temperature factors.
444

Overall structure
As can be expected from the isomorphism of crystals, the
refined structures of pHSA and rHSA are virtually identical,
with an r.m.s. deviation of 0.24 for all C atoms. Only the
structure of pHSA will therefore be described and discussed.
HSA is a helical protein with turns and extended loops, and
resembles a heart shape, with approximate dimensions of
80 3 80 3 30 , as illustrated in Figure 1. It consists of three
domains, I (residues 1195), II (196383) and III (384585),
which are not only topologically identical, as predicted from
amino acid sequence comparison, but also very similar in
three-dimensional structure (Figure 2). The overall r.m.s.
deviation of the corresponding C atoms is 3.78 for all three
domains. The r.m.s. deviation between domains I and II for
corresponding C atoms is 3.1 , while that for other pairs is
4.65.0 . Although the C trace suggests a resemblance

Crystal structure of human serum albumin

between domains I and II, the actual shapes of domains II and

III look more similar than those of I and II (Figure 2). Each
domain can be further divided into subdomains a and b, which
are further described in the next section.
Although all three domains of the HSA molecule have
similar three-dimensional structures, their assembly is highly
asymmetric (Figure 1). Domains I and II are almost perpendicular to each other to form a T-shaped assembly in which the
tail of subdomain IIa is attached to the interface region
between subdomains Ia and Ib by hydrophobic interactions
and hydrogen bonds. In contrast, domain III protrudes from
subdomain IIb at an angel of about 45 to form a Y-shaped
assembly for domains II and III. Domain III interacts only
with subdomain IIb. These features make the HSA molecule
heart-shaped. There are few contacts between domains I and
III due to a large channel formed by subdomains Ib, IIIa and
IIIb. The two helices at the C-terminus of domain III show
very high temperature factors (.75 2) due to their few
interactions with other parts of the molecule.
It has been said from physicochemical studies that HSA is
a flexible protein that easily changes its molecular shape
(Carter and Ho, 1994). A couple of interesting features are
revealed from a comparison of our defatted HSA with that in
complex with myristates (Curry et al., 1998). First of all, the
relative arrangement of these three domains are largely changed
when fatty acids bind to the protein. Secondly, each domain
retains its conformation during complex formation, except for
that of two C-terminal helices in domain III which move
towards the outside of the molecule. These features imply that
the molecular flexibility of HSA in different conditions is
mainly due to the relative motions of its domain structures.
Subdomains
Each domain can be further divided into subdomains a and b,
which are composed of six and four -helices, respectively
(Figure 2). Subdomain a is composed in each case of a cluster
of four -helices (a-h1 to a-h4) flanked by two short -helices
(a-h5 and a-h6), while subdomain b in each case has another
cluster of four -helices (b-h1 to b-h4). A long extended loop
traverses the two subdomains to link them together. The helix
clusters found in subdomains a and b are also similar in threedimensional structure, and are approximately related by a
pseudo-twofold axis situated between helices a-h2 and b-h2,
as shown in Figure 2. As the last helices (b-h4) of domains I
and II are fused with the first helices (a-h1) of the next
domains, the total number of helices in a HSA molecule is 28
rather than 30 (He and Carter, 1992).
Marked common structural features are observed between
the subdomains, not only in polypeptide chain folding but also
in their unique disulfide bond topology. There are 35 cysteine
residues in HSA, 34 of which form 17 disulfide bridges.
Sixteen of these participate in the formation of eight so-called
double disulfide bridges (Figures 3 and 4). The two adjoining
cysteine residues in the amino acid sequence on helices a-h3
and b-h3 lie at the center of this unique folding topology. The
first bridge connects helices h3 and h1, which are spatially
aligned nearly perpendicular to each other, while the second
links helices h3 and h4, which are nearly parallel. Consequently,
three helices (h2, h3 and h4) join to form a curved wall at
one side of the subdomain. Another series of disulfide bridges
is found around helices a-h4, a-h5 and a-h6, where two
consecutive cysteine residues are located on each a-h5 helix.
Residues having aliphatic and aromatic side chains are clustered

at the inner surface of the helix wall to form the hydrophobic

core of the subdomain, and are partly capped by helix h1.
This topological feature is seen in every subdomain except Iah3. Sequence comparison of serum albumins of various species
suggests that simultaneous mutation of two cysteine residues
may have occurred at the end of helix Ia-h3 in ancient proteins
(Carter and Ho, 1994).
Free sulfhydryl group at Cys34
Cys34, located in a loop between helices Ia-h2 and Ia-h3, is
the only cysteine residue that does not participate in any
disulfide bridges (Figure 2A). Its sulfhydryl group is oxidized
by cysteine or glutathione in 3040% of HSA molecules in
the bloodstream to form an intermolecular disulfide linkage
(Peters, 1985). In the current structure, however, none of the
features on the final difference electron density map is observed
around the sulfhydryl side chain of Cys34 (Figure 5). Although
Cys34 is located at the surface of the protein, its S atom is
toward the interior and is surrounded by side chains of Pro35,
His39, Val77 and Tyr84, which prevent the sulfhydryl group
from coupling with the external counterparts. Ion-spray mass
spectrometry revealed that the purified HSA used for crystallization contains much fewer oxidized species at Cys34 than
the commercially available substance (data not shown), and
that triclinic crystals grew only from solution of the former
and not of the latter. This observation suggests that the
molecular species mainly involved in the triclinic crystals is
one whose Cys34 is free of oxidation. When the protein is in
solution, the phenolic side chain of Tyr84 may flip over to
enable the external disulfide counterparts to bind to the S
atom of Cys34, or the backbone conformation may change
to bring the free sulfhydryl group toward the exterior of
the protein.
Binding sites for drugs and other compounds
In both subdomains IIa and IIIa, a part of the hydrophobic
core is further surrounded by the first few residues of the
extended loop together with helices a-h5 and a-h6 to form a
pocket (Figures 2B and C). The pocket of subdomain IIa
seems to correspond to the so-called site I, which is thought
to be a binding site for salicylates, sulfonamides and a number
of other drugs (Sudlow et al., 1975, 1976). The bilirubin
binding site overlaps with this locus (Brown and Shockley,
1982). The inside wall of the pocket is formed by hydrophobic
side chains, whereas the entrance of the pocket is surrounded
by positively charged residues, i.e., Arg257, Arg222, Lys199,
His242, Arg218 and Lys195 (Figure 6A). Site II, the putative
binding site for tryptophane, thyroxine, octanoate and other
drugs (Sudlow et al., 1975), would correspond to the pocket
of subdomain IIIa, which is almost the same size as site I.
The pocket is lined by hydrophobic side chains and the double
disulfide bridges of helix IIIa-h3. The side chain of Arg410 is
located at the mouth of the pocket while the hydroxyl of
Tyr411 faces toward the inside of the pocket (Figure 6B).
Unlike subdomains IIa and IIIa, subdomain Ia has no pocket
near its hydrophobic core (Figure 2A). Due to relaxation of
helix Ia-h4, the latter is no longer parallel to helix Ia-h3. The
side chains of the residues on helix Ia-h4 bury the putative
pocket, and subdomain IIa interacts with them to seal this
region.
Camerman et al. (1976) proposed that the backbone nitrogens
of the first three residues as well as the imidazole of His3
form a binding site of the Cu21 ion. No significant evidence
445

S.Sugio et al.

regarding metal binding can be drawn from our refined

coordinates due to disorder at the N-terminus.
It is well known that HSA is able to accommodate fatty
acids. Reed (1986) proposed that palmitic acid is bound to
Lys473, Lys349 and Lys116 in bovine serum albumin (equivalent to Lys475, Lys351 and Arg116 in HSA, respectively).
Lys475 and Lys351 are located on the molecular surface
(Figure 1). The Lys475 on helix IIIa-h5, however, projects
from the molecular surface. Two more clefts are found near
Lys351 on helix IIb-h3. One is at the interface between
subdomains IIa and IIb and the other between subdomains IIb
and IIIa. Arg117 is located in a crevice formed by the helix
cluster of subdomain Ib and the extended loop between
subdomains Ia and Ib.
Conclusion
We determined the X-ray structures of HSA, derived from
human pooled plasma or from a Pichia pastoris expression
system, at a resolution of 2.5 , and demonstrated that their
three-dimensional structures are identical within the margin of
experimental error. HSA consists of three similar helical
domains with eight pairs of double disulfide bridges. Each
domain is further divided into two subdomains. Cys34, the
only cysteine with a free sulfhydryl group, does not participate
in a disulfide linkage with any external ligand. Deep hydrophobic pockets with positively charged entrances are located
at similar positions in subdomains IIa and IIIa, and are thought
to correspond to the so-called site I and site II binding sites
for various compounds. Subdomain Ia does not have such a
pocket. Three possible binding sites for long-chain fatty acids
are located on the surface of the molecule, but their structural
environments are entirely different. Our proposed structure for
HSA, whose atomic coordinates are now available from
PDB, will serve as a guide for researchers investigating the
biochemical properties and medicinal applications of this
protein.
Acknowledgements
Some of the diffraction experiments were carried out using the Rigaku R-AXIS
IV area-detector and FR X-ray generator of the Nara Institute for Science and
Technology. We thank Dr Toshio Hakoshima for permitting the use of
this equipment.

References
Bernstein,F.C., Koetzle,T.F., Williams,G.J.B., Meyer,E.F.,Jr, Brice,M.D.,
Rodgers,J.R., Kennard,O., Shimanouchi,T. and Tasumi,M. (1977) J. Mol.
Biol., 112, 535542.
Brown,J.R. (1976) Fed. Proc., 35, 21412144.
Brown,J.R. and Shockley,P. (1982) In Jost,P. and Griffith,O.H. (eds), Lipid
Protein Interactions, Vol. 1. Wiley, New York, pp. 2568.
Brunger,A. (1992a) Nature, 355, 472474.
Brunger,A. (1992b) X-PLOR Version 3.1: A System for X-ray Crystallography
and NMR. Yale University Press, New Haven.
Camerman,N., Camerman,A. and Sarakar,B (1976) Can. J. Chem., 54,
13091316.
Carter,D.C. and He,X.M. (1990) Science, 249, 302303.
Carter,D.C. and Ho,J.X. (1994) Adv. Protein Chem., 45, 153203.
Carter,D.C., He,X.M., Munson,S.H., Twigg,P.D., Gernert,K.M., Broom,B. and
Miller,T.Y. (1989) Science, 244, 11951198.
Carter,D.C., Chang,B., Ho,J.X., Keeling,K. and Krishnasami,Z. (1994) Eur.
J. Biochem., 226, 10491052.
Collaborative Computational Project, Number 4 (1994) Acta Crystallogr.,
D50, 760776.
Curry,S., Mandelkow,H., Brick,P. and Franks,N. (1998) Nature Struct. Biol.,
5, 827835.
Dugiaczyk,A., Law,S.W. and Dennison,O.E. (1982) Proc. Natl Acad. Sci.
USA, 79, 7175.

446

Engh,R.A. and Huber,R. (1991) Acta Crystallogr., A47, 392400.

Fehske,K.J., Muller,W.E. and Wollert,U. (1981) Biochem. Pharmacol., 30,
687692.
Fiume,L., Busi,C., Mattioli,A. and Spinosa,G. (1988) CRC Crit. Rev. Ther.
Drug Carrier Syst., 4, 265284.
Furey,W. (1994) PHASES, Verona, PA, USA.
He,X.M. and Carter,D.C. (1992) Nature, 358, 209215.
Ho,J.X., Holowachuk,E.W., Norton,E.J., Twigg,P.D. and Carter,D.C. (1993)
Eur. J. Biochem., 215, 201212.
Kragh-Hansen,U. (1981) Pharmacol. Rev., 33, 1753.
Laskowski,R.A., MacArthur,M.W., Moss,D.S. and Thornton,J.M. (1993)
J. Appl. Crystallogr., 26, 283291.
Luzzati,V. (1952) Acta Crystallogr., 5, 802810.
McClure,R.J. and Craven,B.M. (1974) J. Mol. Biol., 83, 551555.
McRee,D.E. (1993) In Practical Protein Crystallography. Academic Press,
London.
Minghetti,P.P., Ruffner,D.E., Kuang,W.J., Dennison,O.E., Hawkins,J.W.,
Beattie,W.G. and Dugaiczyk,A. (1986) J. Biol. Chem., 261, 67476757.
Ohi,H., Miura,M., Hiramatsu,R. and Ohmura,T. (1994) Mol. Gen. Genet.,
243, 489499.
Peters,T.,Jr (1985) Adv. Protein Chem., 37, 161245.
Putnam,F.W. (1984) The Plasma Proteins, 2nd edn, Vol. 4. Academic Press,
London.
Ramakrishnan,C. and Ramachandran,G.N. (1965) Biophys. J., 5, 909933.
Rao,S.N., Basu,S.P., Sanny,C.G., Manely,R.V. and Harts,J.A. (1976) J. Biol.
Chem., 251, 31913193.
Reed,R.G. (1986) J. Biol. Chem., 261, 1561915624.
Sato,M., Yamamoto,M., Imada,K., Katsube,Y., Tanaka,N. and Higashi,T.
(1992) J. Appl. Crystallogr., 25, 348357.
Sudlow,G., Birkett,D.J. and Wade,D.N. (1975) Mol. Pharmacol., 11, 824832.
Sudlow,G., Birkett,D.J. and Wade,D.N. (1976) Mol. Pharmacol., 12, 1052
1061.
Sumi,A., Ohtani,W., Kobayashi,K., Ohmura,T., Yokoyama,K., Nishida,M. and
Suyama,T. (1993) Biotechnol. Blood Proteins, 227, 293298.
Terwilliger,T. (1994) Heavy Version 3. Los Alamos National Laboratory,
Los Alamos.
Wang,B.C. (1985) In Colowick,S.P. and Kaplan,N.O. (eds), Methods in
Enzymology, 115. Academic Press, London, 90112.
Yapel,A.F., Jr (1985) In Colowick,S.P. and Kaplan,N.O. (eds), Methods in
Enzymology, 112. Academic Press, London, 317.
Yokoyama,K. and Ohmua,T. (1995) Jpn. J. Apheresis, 14, 1920.
Zhang,K.Y.J. and Main,P. (190) Acta Crystallogr., A46, 377381.
Received November 22, 1998; revised January 27, 1999; accepted February
24, 1999