Crystal structures of the human SUMO-2 protein at 1.6 A
˚
and 1.2 A
˚
resolution
Implication on the functional differences of SUMO proteins
Wen-Chen Huang
1,2
, Tzu-Ping Ko
1
, Steven S L Li
3
and Andrew H J. Wang
1
1
Institute of Biological Chemistry, Academia Sinica, Taipei, Taiwan;
2
Institute of Biomedical Sciences, National Sun Yat-Sen
University, Kaoshiung, Taiwan;
3
Department of Biotechnology, College of Life Sciences, Kaoshiung Medical University, Taiwan
The S UMO proteins are a class of small ubiquitin-like
modifiers. SUMO is attached to a s pecific lysine side chain
on the target protein via an isopeptide bond with its
C-terminal glycine. There are at least four SUMO proteins in
humans, wh ich are involved in protein trafficking and tar-
geting. A truncated human SUMO-2 protein that contains
residues 9–93 was expressed i n Escherichia c oli and crystal-
lized in two d ifferent unit cells, w ith dimensions of a ¼ b ¼
75.25 A
˚

packing a nalysis s uggests a possible trimeric assembly of
the SUMO-2 protein, of which the biological significance
remains t o be determined.
Keywords: homology m odeling; m olecular interactio ns;
protein mod ification; surface charge distributions; synchro-
tron radiations.
Control of protein expression and regulation of protein
activities are central to the cellular processes in an organism.
Many proteins are rather short lived, and are eventually
targeted to proteosomes f or degradation via conjugation
with ubiquitin [1]. However, the functions of various
proteins are not only a matter of time but also a matter of
place. T hus, n ewly synthesized proteins must be directed
toward specific subcellular compartments. SUMO is the
acronym for small ubiquitin-like modifier and named after
its three-dimensional structural similarity to ubiquitin. Both
SUMO and ubiquitin a re attached to target proteins by
forming an isopeptide bond between the C-terminal glycine
and a specific lysine side chain o n the target [2]. The extra
amino acids beyond the l ast g lycine–glycine m otif o f n ative
SUMO proteins are proteolytically removed in vivo.In
mammals, there are at least four different SUMO proteins,
SUMO-1, -2, -3 and -4. The h uman hSMT3 cDNA
encoding the SUMO-2 protein was first reported by
Mannen et al.[3].SUMO-2andSUMO-3share87%
sequence identity with each other, but they have only 47%
identity with SUMO-1 [4]. The novel SUMO-4 associated
with diabetes is also more similar in sequence to SUMO-2
than to SUMO-1 [5].
The first three-dimensional structure of SUMO-1 deter-

D
-galactoside.
(Received 2 1 May 2004, revised 14 July 2004,
accepted 31 August 2004)
Eur. J. Biochem. 271, 4114–4122 (2004) Ó FEBS 2004 doi:10.1111/j.1432-1033.2004.04349.x
functions in protein targeting. A rrangement of side chains
confers t he protein w ith unique surface properties. Thus,
comparison of SUMO-1, - 2 and -3 surface p roperties by
modelling provides an approach to understanding the
relationship b etween structure and function. To date, no
crystal structure of mammalian SUMO proteins has been
determined. In order to obtain more structural information,
especially about the protein side chains, we tried to
determine a three-dimensional structure of human SUMO
at high resolution by X-ray crystallography.
In this paper we present the crystal structure of a
truncated SUMO-2. To facilitate crystallization, our strat-
egy was to reduce the length of N-terminal arm while
preserving the sequence of Val10–Lys11–Thr12–Glu13, as
well as the C-terminal Gly92–Gly93 for conjugation via
an isopepti de bond. The VKTE s equence in SUMO-2 is
consistent with the S UMOylation consensus YKXE w here
Y represents a hydrophobic amino acid and X means a ny
amino acid in target proteins, and this consensus sequence is
functional for possible polymerization [11]. Furthermore,
the truncated SUMO-2 cDNA encoding sequence 9–93 was
fusedtoaHis
10
tag at the N terminus with a Factor Xa
cleavage site for efficient purificaion.

M
NaCl
(pH 8.0) with a French Press (Cell Disruption, Constant-
systems) at 206 843 kPa twice and centrifuged (18 592 g,
20 min ) for supernatant collection.
The S UMO-2 protein was purified using a column
packed with Ni–NTA HisBindÒ resin ( Novagen) in two
steps. In the first purification, major protein was eluted
using an i midazole gradient of 0–250 m
M
and the collec ted
fractions were analysed by SDS/PAGE. The SUMO-2
protein in peak fractions was pooled and dialysed three
times against 25 m
M
Tris-base, 150 m
M
NaCl (pH 8.0) and
incubated for 26 h at room temperature in the presence of
Factor Xa (Novagen). This step removes the His
10
tag to
generate the truncated SUMO-2 protein (9–93 amino acids).
The protein solution was then purified a second time,
in which the flow-through was collected using a wash buffer
that contained 20 m
M
inidazole, and dialysed t hree times i n
25 m
M

M
2-(cyclohexylamino)ethanesulfonic acid
(CHES) and 0.1
M
Tris/HCl pH 8.0, and diffracted to
1.6 A
˚
. The other one, of rectangular p olyhedron shape (type
II, Fig. 2B), grew in 40% (w/v) PEG-600, 0.1
M
CHES,
0.1
M
sodium HEPES pH 8.0, and diffracted well to a
resolution of 1.2 A
˚
.
Two data sets were collected using MSC R-AXIS
IV++ image plate detectors and processed using the
software package of
HKL
[12]. The first one was carried
out using the triangular plate crystal form (type I) at
Institute of Biological Chemistry, Academia Sinica, using
an MSC MicroMax 002 X-ray generator. The second data
set o f the polyhedral crystal form (type II) was c ollected
at t he National Synchrotron Radiation R esearch Center,
Hsinchu, Taiwan, using beam line 17B2 as an X-ray
source.
Crystallographic computing and modelling

representative electron density maps superimposed on the refined models of the two crystal forms I a nd II, respectively. B oth were contoured at
2.0 r levels using 2Fo–Fc maps phased by the refined mo dels. The side chain of Lys21 lacks well-defined density, presumably because i t is flexible.
4116 W C. Huang et al. (Eur. J. Biochem. 271) Ó FEBS 2004
SMT3. The models were then subjected to molecular
dynamics and energy minimization using
CNS
, while the
backbone atoms were restrained w ith t he original model
coordinates. For structural comparisons with ubiquitin,
yeast SMT3 and human SUMO-1, models directly from the
Protein Data Base (PDB) entries 1UBQ, 1 EUV (chain B)
and 1A5R (model 1), respectively, were used.
Figure 1 was produced using the program
ALSCRIPT
[16].
The r ibbon diagrams and the electron density m aps in
Figs 2, 3 and 5 were drawn using
MOLSCRIPT
[17],
BOBSCRIPT
[18] and
RASTER
3
D
[19]. The molecular surface properties
were examined using
GRASP
[20], w hich was a lso used t o
generate Fig. 4. Model geometry and crystal contacts were
analysed using the programs

˚
, with o ne exception
betweenAsp16andArg36,whichisseeninthe1.6A
˚
model. The amino acids are shaded in red, gree n and blue for acidic, neutral an d basic polar
residues, and in yellow for prolines and glycines. In (C) the polypeptide tracings of two SUMO-2 models from type I (12–89) and type II (17–88)
crystals, shown in green and red, are superimposed with th at of human ubiquitin (1–76), shown in blue. In (D ) the yeast SMT3 crystal structure
(20–98) and human SUMO-1 NMR structure ()2–101), coloured yellow and cyan, respectively, are compared with the SUMO-2 structure (type I
crystal), shown in red.
Ó FEBS 2004 Structure and function of human SUMO-2 (Eur. J. Biochem. 271) 4117
and 31.9% for the type I and type II crystal forms,
respectively.
The NMR model of human SUMO-1 (PDB code 1A5R)
contains full-length protein, whereas the N- and C-terminal
regions are fl exible. Molecular replacement search u sing the
NMR model did not yield a correct solution for the crystal
structure of SUMO-2, even with omission of the terminal
segments. Instead, i t was solved using yeast SMT3 (PDB
code 1EUV) as a search model. The initial R value for the
type I crystal was 0.465 after rigid-body refinement at 3.0 A
˚
resolution. The final m odel c ontains amino acid residues
12–89 and 67 water molecules, with R and R
free
values of
0.169 and 0.190, respectively. The R value for the type II
crystal based on the refined type I model was 0.409 at 1.5 A
˚
.
After refinement, the model contains amino acid residues

The protein models of SUMO-2 type I and type II
crystals superimpose with a n r .m.s.d. of 0.544 A
˚
for 288
backbone atoms and 1.201 A
˚
for all 584 atoms. Larger
deviations of Ca coordinates t han 1.0 A
˚
occur in the
residues 17, 26, 27, 5 6 a nd 88. A lthough type I I crystal
diffracts to higher resolution, its visible N terminus is
shorter than that o f type I crystal by five residues. As shown
in Fig. 3A, this segment extends away from t he protein core
and should be flexible because of exposure to the bulk
solvent. The smaller unit-cell dimension of type I crystal
allows the N terminus to be docked onto a neighbouring
molecule, specifically, near the region of Phe60–Thr70, and
thus stabilizes the extended conformation.
Also shown in F ig. 3C, the model of human ubiquitin
(PDB code 1UBQ) is superimposed on the S UMO-2
models of type I and II crystals, with an r.m.s.d.
of 0.952 A
˚
and 1.135 A
˚
for 55 and 65 Ca atoms,
Fig. 4. Surface properties of SUMO proteins.
The molecular surface of SUMO-2 (type I
crystal)isshownin(A)and(C);thatofthe

SUMO-2 and the equivalents of ubiquitin. Although the
sequences have only 18% identity, the protein folds of
SUMO-2 and ubiquitin are very similar, even without
insertion (Fig. 1). Yet th ese t wo classes of proteins h ave
very different functions, which m ay be explained by the
disparate surface charge distributions [6].
Significant difference between the yeast SMT3 crystal
structure and the human SUMO-1 NMR s tructure has been
observed by Mossessova and Lima [8]. In Fig. 3D the
SUMO-2 model is superimposed with those of SMT3
(1EUV) and S UMO-1 (1A5R). Based on a distance
criterion of 2 .0 A
˚
, the r.m.s.d. is 1.096 A
˚
between 43 pairs
of Ca atoms in S UMO-1 ( NMR) and SUMO-2 (type I
crystal). Under the same condition, the r msd is 0.918 A
˚
between 67 Ca pairs in SUMO-2 a nd SMT3, and it is
0.470 A
˚
for 40 matched pairs with a distance criterion of
1.0 A
˚
. Therefore, the crystal structure of human SUMO-2
is more similar to that of yeast SMT3 than to the NMR
structure of SUMO-1. The difference between SUMO-2
Table 1. X-ray data statistics for S UM O-2 crystals. Numbers in parentheses are for the highest resol ution shells.
Crystal form

-97
Total reflection used [F >0r(F)] 7868 (633) 20948 (1924)
R for 95% working data set 0.169 (0.266) 0.119 (0.217)
R
free
for 5% est data set 0.190 (0.273) 0.185 (0.239)
rmsd from ideal bond lengths (A
˚
) 0.017 0.013
rmsd from ideal bond angles (°) 1.8 2.3
rmsd from ideal dihedral angles (°)2726
rmsd from ideal improper angles (°) 1.3 1.8
Ramachandran plot: number of residues in most favored regions (%) 97.1 96.8
In additional allowed regions (%) 2.9 3.2
Average B-values/number of atoms for protein backbone (A
˚
2
) 17.7/312 18.4/288
For protein side chains (A
˚
2
) 22.8/322 27.6/297
For water molecules (A
˚
2
) 34.3/67 42.8/127
Table 2. Refinement procedures of the SUMO-2 crystals.
Description of steps Protein Water Resolution R/R
free
Type I crystal, yeast SMT3 model 13–98 (SMT3) 3.0 A

ularly evident in the regions of 28–43 and 71–83, that
correspond to the strand b2, the N terminus of the h elix a1,
the helix a2, and the connecting loop to the strand b5
(Fig. 3 A,D).
SuchalargedifferencebetweentheNMRandcrystal
structures may explain the fact t hat we were not able to
solve our crystal structure by the molecular replacement
method using SUMO-1 N MR structure as the starting
model. The high-resolution NMR structure of SUMO-1
determined later using heteronucle ar NOE also showed
difference from the s tructure of 1A5R [7 ]. Interestingly, this
new SUMO-1 NMR structure is similar to the SMT3 NMR
structure, whereas significant deviations betw een the crystal
structure and solution structure of SMT3 were also
observed [9]. Therefore, the deviations may be due to
different environments and different experimental tech-
niques used in the structure determinations.
Surface potential and functional difference
The mechanisms of protein ubiquitination and SUMOyla-
tion are similar, which involve the activating, conjugating,
and ligation enzymes E1, E2 and E3. A peptidase is also
required to remove the C-terminal peptide of a SUMO
protein to render the mature form, which has the C-terminal
Gly-Gly motif for conjugation with target proteins [4]. In
yeast, an E1-specific for SUMO has been identified as a
large heterodimeric Aos1/Uba2 of  11 0 kDa, and there i s
a heterodimeric homologue SAE1/SAE2 in man. The E2 in
both human and yeast is a highly conserved Ubc9 of
18 kDa, whereas the E3 proteins have a broader definition
and comprise s everal s ubclasses [ 4]. The enzymes Ulp1 a nd

Arg64–Arg71 (Fig. 1). These correspond to Arg59–Pro66 in
SUMO-2 and , with an adjacent Arg61 substituting Leu66 in
SMT3, the surface features in this region are also con served.
However, interactions between SUMO and other proteins,
including E3, may be established with other surface regions.
Although the sequences of human SUMO-2 and -3 are
87% i dentical, they a re located in different c ellular com-
partments: SUMO-2 was found in nuclear bodies but
SUMO-3 was located in the cytoplasm [10]. The s urface
charge distribution of SUMO-2/-3 is even more similar.
When these two protei n surfaces a re compared, t he only
visible difference corresponds to residue 77, which is a
negatively charged Glu in SUMO-2, but is a positively
charged A rg in SUMO-3. On the other hand, SUMO-1 is
47% identical to SUMO-2 in sequence, and has a longer
N-terminal arm. The r esulting difference in their surface
properties can be attributed to at least 10 residues. These
include Glu33, Lys48, Glu49, Gln53, Asn60, Leu6 5, Arg70,
Lys78, Gly81 and Glu93 in SUMO-1, whereas t he corres-
ponding surface residues in SUMO-2 are Val29, Met44,
Lys45, Glu49, Arg56, Arg61, Pro66, Ala74, Glu77 and
Gln89, respectively. The most prominent is a concave re gion
shown i n F ig. 4 C a nd D, which i s fl anked b y the helix a1
and the strands b3/b4 (Fig. 3A). This region is neutral in
SUMO-2 but positively charged in SUMO-1, probably
caused by the substitution of Met44 in SUMO-2 with Lys48
in SUMO-1, as shown in Fig. 4 E and F. In particular,
the concave surface is near the C terminus, and thus
may serve as a potential site for d iscrimination between
SUMO-1 and -2 i n humancells. The flexible N-terminal arms

, respectively.
The first and m ost conserved i nterface is between
molecules related by the crystallographic threefold axis.
The buried areas are 856 A
˚
2
and 821 A
˚
2
on each SUMO-2
monomer in the type I and type II crystals, respectively,
corresponding to about o ne-quarter and more than one-
third of the contact surfaces. T he interactions include two
hydrogen bonds between backbone atoms of Gly27(O)–
Lys33*(N) and Val29(N)–Gln31*(O), and a salt bridge
between the side chains o f Asp26 and Arg50*. (Amino acid
residues of t he symmetry-related molecules are denoted by
asterisks.) The latter is also hydrogen b onded to Tyr47(OH)
and Gln51(OE1). Such interactions, particularly those
between the strands b2, may stabilize a possible trimeric
assembly of SUMO-2 in solution, shown in Fig. 5. The
4120 W C. Huang et al. (Eur. J. Biochem. 271) Ó FEBS 2004
other four interfaces are not all conserved, whereas the
buried surface areas are much larger in type I crystal than in
type II. Because the c-axis is significantly shorter, more
lattice interactions were observed i n type I crystal. These
include docking of the flexible N-terminal segment onto a
neighbouring molecule.
Polymers of ubiquitin h ave been studied extensively since
they were discovered [28]. The site of self-conjugation is

We thank Drs Chia-Cheng Chou, Rey-Ting Guo and Cheng-Chung
Lee for their a ssistance in data collection. We also thank the National
Synchrotron Radiation Research Center for beam time allocation. This
work was supported by grants from National Science Council (NSC
92–3112-B-110 -001 and NSC 93-3112-B-110-001) to SSLL and from
Academia Sinica to A.H.J.W.
References
1. Pickart, C.M. (2004) Back to the future w ith ubiquitin. Cell 116,
181–190.
2. Mu
¨
ller,S.,Hoege,C.,Pyrowolakis,G.&Jentsch,S.(2001)
SUMO, ubiquitin’s mysterious cousin . Nat. Rev. Mol. Cell Biol. 2,
202–210.
3.Mannen,H.,Tseng,H M.,Cho,C L.&Li,S.S L.(1996)
Cloning and e xpression of human homolog H SMT3 to yeast
SMT3 suppressor of MIF2 mutations in a centromere protein
gene. Biochem. Biophys. Res. Commun. 222 , 178–180.
4. Melchior, F. (2000) S UMO-nonclassical ubiquitin. Annu. Rev.
Cell Dev. Biol. 16, 591–626.
5. Bohren, K.M., Nadkarni, V., Song, J.H., Gabbay, K.H. &
Owerbach, D. (2004) A M55V polymorphism in a novel SUMO
gene (SUMO-4) differentially activates h eat s hock transcription
factors and is asso ciated with su sceptibility to t ype I diabetes
mellitus. J. Biol. Chem. 279, 27233–27238.
6. Bayer, P., Arndt, A., Metzget, S., Mahajan, R . & Melchior, F.
(1998) St ructure d etermination of the small ubiquitin-related
modifier SUMO-1. J. M o l. Biol. 280, 275–286.
7. Jin,C.,Shiyanova,T.,Shen,Z.&Liao,X.(2001)Heteronuclear
nuclear magnetic reso nance assignments, s truc ture and d yn amics

392–400.
16. Barton, G.J. (1993) ALSCRIPT: a tool to format multiple
sequence alignments. Protein Eng. 6, 37–40.
Fig. 5. Trimer of SUMO-2. In both crystal forms the conserved
interactions between three molecules rela ted by the c rystallograph ic
threefold axis suggest the possible existence of a trimeric a ssembly in
solution. A no tew orthy feature is the association of t hree strands b2
around the triad axis, held together by several hydrogen bonds
between backbone atoms.
Ó FEBS 2004 Structure and function of human SUMO-2 (Eur. J. Biochem. 271) 4121
17. Kraulis, P.J. (1991) M OLSCRIPT: a program to produce both
detailed and schematic plots of protein structure. J. Appl. Crys-
tallogr. 24 , 946–950.
18. Esnouf, R.M. (1997) An extensively modified version of MolScri pt
that includes g rea tly enha nced coloring c apabilities. J. Mol. Graph.
15, 132–134.
19. Merrit, E.A. & Murphy, M.E.P. (1994) Raster3D, Version 2.0. A
program for p hotorealistic m olec ular grap hics. Acta Crys tallogr.
D 50, 869–873.
20. Nicholls, A., Sharp, K.A. & Honig, B. (1991) P rotein folding a nd
association: insights from the interfacial and thermodynamic
properties of hydrocarbons. Proteins 11, 281–296.
21. Collaborative Computational P roject Number, 4 (1994) The
CCP4 suite: p rograms for protein crystallography. Acta Crystal-
logr. D 50 , 760–763.
22. Matthews, B.W. (1968) Solvent content of protein crystals. J. Mol.
Biol. 33, 491–497.
23. Luzzati, P.V. (1952) Tra itement statistique des erreurs dans la
determination des structures cristallines. Acta Crystallogr. 5, 802–
810.