The EMBO Journal Vol.19 No.5 pp.1045–1054, 2000
Structure of the RXR–RAR DNA-binding complex on
the retinoic acid response element DR1
Fraydoon Rastinejad1,2,3, Trixie Wagner1,
Qiang Zhao1and Sepideh Khorasanizadeh2,3
1Department of Pharmacology, X-ray Crystallography Laboratory and
2Department of Biochemistry and Molecular Genetics, School of
Medicine, University of Virginia, Charlottesville, VA 22908, USA
e-mail: email@example.com or firstname.lastname@example.org
The 9-cis retinoic acid receptor (retinoid X receptor,
RXR) forms heterodimers with the all-trans retinoic
acid receptor (RAR) and other nuclear receptors on
DNA regulatory sites composed of tandem binding
elements. We describe the 1.70 Å resolution structure
of the ternary complex of RXR and RAR DNA-binding
regions in complex with the retinoic acid response
element DR1. The receptors recognize identical half-
sites through extensive base-specific contacts; however,
RXR binds exclusively to the 3? site to form an
asymmetric complex with the reverse polarity of other
RXR heterodimers. The subunits associate in a strictly
DNA-dependent manner using the T-box of RXR and
the Zn-II region of RAR, both of which are reshaped
in forming the complex. The protein–DNA contacts,
the dimerization interface and the DNA curvature in
the RXR–RAR complex are distinct from those of the
RXR homodimer, which also binds DR1. Together,
these structures illustrate how the nuclear receptor
superfamily exploits conformational flexibility and
locally induced structures to generate combinatorial
Keywords: nuclear receptor/RAR/RXR/structure/
The 9-cis retinoic acid receptor (retinoid X receptor, RXR)
is an important member of the nuclear receptor family
because it forms heterodimers with many other receptors,
including the all-trans retinoic acid receptor (RAR), vit-
amin D3receptor (VDR), thyroid hormone receptor (TR),
the peroxisome proliferator activated receptor (PPAR) and
the nerve growth factor induced-B (NGFI-B) receptor
(Leblanc and Stunnenberg, 1995; Mangelsdorf and Evans,
1995; Mangelsdorf et al., 1995). RAR and RXR, each of
which have α, β and γ isotypes, transduce the retinoid
signal into a variety of genetic responses, and their
functions underlie the essential role played by retinoids
in the development and homeostasis of vertebrates
(Chambon, 1996). A consensus sequence 5?-AGGTCA-3?
is the core recognition element for the entire family of
nearly 150 non-steroid nuclear receptors to which RXR
and RAR belong (Mangelsdorf and Evans, 1995). How-
ever, RXR and RAR alone display negligible binding to
© European Molecular Biology Organization
this consensus site as monomers (Leid et al., 1992).
Heterodimerization increases their joint affinity and select-
ivity for retinoic acid response elements (RAREs), which
are composed of both binding sites arranged in tandem.
The RXR–RAR heterodimer binds more efficiently to
RAREs than either of the homodimeric forms of RXR or
RAR (Leid et al., 1992).
The core sequence of the DNA-binding domain (DBD)
is highly conserved in the nuclear receptor superfamily,
with ?40% amino acid identity over a 67-residue region.
This sequence is comprised of two zinc-nucleated modules
and two α-helices that fold into a single globular domain.
The use of a highly conserved receptor DBD together
with a single major core binding sequence raises an
important question as to how diverse signaling pathways
are created in this transcription factor superfamily
(Gronemeyer and Moras, 1995). Formation of combinator-
ial transcription factors involving RXR is one important
mechanism by which complex DNA sites can be employed
with greater selectivity (Yu et al., 1991; Bugge et al.,
1992; Kliewer et al., 1992b; Marks et al., 1992). The
functional consequence of forming receptor heterodimers
is that in some cases two distinct ligands can co-regulate
transcription from a single gene-regulatory element
(Forman et al., 1995).
Importantly, the DBDs of the nuclear receptors can
generate the same pattern of DNA selectivity and dimeriz-
ation as the full-length receptors (Mader et al., 1993;
Perlmann et al., 1993; Zechel et al., 1994). A second
dimerization function in the ligand-binding domains of
some receptors has no selective role for response element
recognition, but can further stabilize some dimeric assem-
blies (Perlmann et al., 1996). All RXR heterodimers
preferentially bind response elements composed of two
AGGTCA sites arranged in a direct repeat configuration
with characteristic inter-half-site spacings of 1–5 bp (these
are known as DR1–DR5, respectively; Umesono et al.,
1991; Mangelsdorf and Evans, 1995). Interestingly, the
sequence of the spacer is less critical than its size for
discrimination. The selective recognition of DRs based on
inter-half-site spacings has been formalized in a 1–5 rule
that defines the highest affinity binding sites for each
RXR heterodimer (Umesono et al., 1991; Mangelsdorf
and Evans, 1995). However, the binding repertoire for
other members of the nuclear superfamily extends beyond
the usage of direct repeat geometry, as some factors bind
efficiently to inverted and everted repeats of core hexamer
(Mangelsdorf and Evans, 1995). The multiple arrange-
ments with which the same consensus AGGTCA sequence
is utilized in this family suggest that combinatorial nuclear
receptors possess a remarkable readout mechanism that
allows them to decipher the geometry of a binding site in
addition to its sequence.
An associated feature of any asymmetric response
F.Rastinejad et al.
Fig. 1. The protein and DNA constructs used in structure determination. (A) The DNA-binding regions of RXR (residues 129–212) and RAR
(residues 82–167) used for co-crystallization. The polypeptides contain two zinc-nucleated modules whose coordinating cysteines are indicated in
yellow. Each protein contains two α-helices (α1, α2) in its core domain. A third α-helix, α3, in the T-box of RXR (underlined) is seen in the
absence of DNA, but is disrupted upon DNA binding (Holmbeck et al., 1998a,b). Disordered residues not found in the electron density maps are
indicated by the small letters. The numbering of the amino acids is referenced to the first Zn-coordinating cysteine (1) for simplicity. (B) The 15 bp
DR1 sequence with the consensus recognition half-sites and spacer sequence shown in white and black boxes, respectively. The 5? overhangs,
indicated by the small letters, are not seen in the electron density maps. (C) 2Fo– Fcmap revealing the identity of the RAR subunit. RAR is
distinguished by clear densities for sites such as Ser16 and Pro(–1). (D) 2Fo– Fcmap showing how RXR can be distinguished by unambiguous
density at sites such as Tyr16, His(–2) and Ile(–1)
element is the polarity adopted by the subunits of the
bound receptor heterodimer. The non-symmetric RXR–
RAR complexes allow the position of each receptor to be
distinguished as upstream or downstream, even though
the two hexameric binding sites may have an identical
sequence. The RXR–RAR complex, therefore, can adopt
different polarities when bound to its various response
elements, and these alternative arrangements have been
shown to confer different functions on the heterodimer in
terms of hormone and co-repressor binding (Kurokawa
et al., 1994, 1995). A previous structural analysis of the
RXR–TR DBD complex bound to a thyroid response
element DR4 allowed the direct visualization of the
polarity associated with that complex (Rastinejad et al.,
1995). The asymmetric assembly was established through
the cooperation between DBD subunits, which occurs only
when the inter-half-site spacing is 4 bp in length and RXR
is positioned upstream of TR (Rastinejad et al., 1995).
The response element repertoire of the RXR–RAR
complex is considerably less selective than that of the
RXR–TR complex, which was restricted to a single direct
repeat element (DR4). In fact, the RXR–RAR heterodimer
mediates transcriptional control through DR1, DR2 and
DR5 sequences found in naturally occurring RAREs
(Mangelsdorf et al., 1991; Predki et al., 1994; Chambon,
1996). The multiple high-affinity DNA-binding targets
suggest a large degree of versatility on the part of their
DBDs in forming favorable interaction surfaces, each of
which has a distinct effect on target gene regulation
(Kurokawa et al., 1995; Chambon, 1996). To explore how
Structure of the RXR–RAR–DR1 complex
the joint assembly of RXR and RAR on DNA is mediated,
we determined the 1.7 Å crystal structure of the ternary
complex containing RXR and RAR DBDs bound to the
DR1 element, the sequence first identified in the promoter
of thecytosolic retinol-binding
(Mangelsdorf et al., 1991). In vivo, RXR–RAR hetero-
dimers on DR1 repress transcription in both the absence
and presence of their hormones (Kurokawa et al., 1995).
In examining the structure of the complex, we asked how
the DR1 specifies the polar assembly of the RXR and
RAR DBDs, how cooperativity and target selectivity are
achieved, and how DNA-induced changes in the secondary
and tertiary structures of RXR and RAR promote the
assembly of their complex.
Overview of the structure
The protein and DNA constructs used in co-crystallization
are shown in Figure 1A and B. The polypeptides were
designed to support the ability of RXR and RAR to form
cooperative interactions on DR1. Using these constructs,
the heterodimer of RXR and RAR DBDs gives half-
maximal DR1 binding when each protein is at a
concentration of ~300–350 nM (data not shown). In
contrast, the DBD of RXR alone shows a ?10-fold
reduction in binding to DNA sites containing a single
AGGTCA sequence (Leid et al., 1992; Lee et al., 1993;
Zhao et al., 2000). The T-box sequence of RXR, which has
the sequence KREAVQEER at the C-terminal extension of
the DBD, is required for the formation of the RXR
homodimer DR1 complex, as well as the formation of
other heterodimeric complexes when RXR occupies the
site downstream of DR1 (Wilson et al., 1992; Predki
et al., 1994; Ijpenberg et al., 1997).
There is a good structural overlap between the backbone
atoms of RXR and RAR DBDs in their homologous
67-residue core sequences, and this fold is shared with
receptor (ER), glucocorticoid receptor (GR), NGFI-B and
by extension the entire nuclear receptor family (Luisi
et al., 1991; Schwabe et al., 1993; Rastinejad, et al., 1995;
Zhao et al., 1998; Meinke and Sigler, 1999). To identify
the polarity associated with the complex, specific and
density maps were required. The polypeptides differ in
34 amino acid positions within the region –2 to 75, but
most of these residues are conservative substitutions or
solvent-exposed positions that are not fully ordered. The
high-resolution X-ray data proved helpful in this regard.
Figure 1C and D shows that the electron densities
allowing the protein electron density containing the former
to be assigned to RAR and the latter to RXR. Similarly,
the identification of non-conserved residues that precede
the core DBD, such as His(–2), Ile(–1) and Pro(–1), also
with the binding of RAR and RXR to the upstream and
downstream half-sites, respectively.
Figure 2 shows the overall structure of the complex.
The observed mode of polarity agrees with a previous
biochemical study that examined DR1 binding by full-
length RAR and RXR polypeptides (Kurokawa et al.,
Fig. 2. Overall structure of the ternary complex. (A) Top view looking
down the 5? end of the DNA. The DNA phosphates and the zinc
atoms are shown in red and white, respectively. The DNA recognition
helices (α1) of RXR and RAR are partly invisible in this orientation
as they insert into the major grooves of DR1. The C-terminally
positioned T-box of RXR points towards the RAR Zn-II module to
mediate the subunit interactions. (B) Side view showing the
complementary surfaces formed between RAR and RXR and the
polarity associated with the complex. Figures are made with the
program GRASP (Nicholls et al., 1991).
1994). The binding of RXR at the downstream site is
uniquely associated with DR1 elements, since most RXR
heterodimers studied to date, including the two other
RXR–RAR heterodimers on DR2 and DR5, the RXR–
VDR heterodimer on DR3, the RXR–TR and RXR–Liver
X receptor (LXR) heterodimers on DR4, and the RXR–
NGFI-B interactions on DR5, use the opposite polarity,
in which RXR binds exclusively at the upstream position.
However, the DR1 sequence also supports the cooperative
binding of RXR–PPAR with 3? RXR binding as observed
in the current structure (Ijpenberg et al., 1997). The unique
asymmetry is a consequence of the subunit’s ability to
and not the other.
The characteristic spacing of the DR1 response element
sets the rotation and translation geometry by which the
F.Rastinejad et al.
Fig. 3. Stereo diagrams showing the contacts between the protein subunits. (A) The RXR–RAR interface involves the DNA minor groove and well
ordered water molecules (red spheres). Dotted lines indicate hydrogen-bonding between atoms. The yellow spheres indicate the positions of the Zn-II
atoms. Arg75 and Asn51 form complementary van der Waals interactions. (B) The corresponding region of the RXR–DBD homodimer interface on
DR1 (Zhao et al., 2000). These and the following molecular display graphics were made using the program Ribbons (Carson and Bugg, 1986).
RXR and RAR subunits must interact while maintaining
a favorable registration with respect to the hexameric half-
site. The minor groove of the DR1 reaches its narrowest
point (5.6 Å) at the spacer base pair 8, precisely where
the two proteins meet. The DNA helical axis deviates a
total of 6° from that of standard B-DNA. The two
AGGTCA sequences are contacted in the major groove
by the recognition helix of each DBD (α1 in Figure 1A).
These contacts account for all of the major groove inter-
actions of the RXR–RAR complex. In their complex,
~3500 Å2of water-accessible surface area is buried,
1900Å2of whichderivesfromthe RXR–DNAinteractions
and 1600 Å2of which is from the RAR–DNA interface.
These values are in line with other protein–DNA inter-
actions that involve similar sized recognition modules and
DNA-binding sites (Nadassy et al., 1999).
The heterodimerization interface and its DNA
RXR and RAR DBDs do not form homo- or heterodimeric
associations in the absence of specific response elements.
Figure 3A shows that the contacts between RXR and RAR
in their complex take advantage of the polar functional
groups in the minor grooves of base pairs 8 and 9, as well
as the well ordered hydration spine associated with DNA
structure (Shui et al., 1998). The intimate role of the DNA
Structure of the RXR–RAR–DR1 complex
Fig. 4. Topological comparison of the RXR–RAR–DR1 complex with the (RXR)2–DR1 complex. (A) The superposition of the two complexes shows
that the DNA structure and the subunit position at the 5? (top) half-sites are different. (B) Superposition of the DNA molecules shows a divergence
in the helical axes associated with DR1 in its complexes with the RXR homodimer and the RXR–RAR heterodimer. The divergence is most severe
at the 5? half-sites, which are occupied by different proteins in these complexes. (C) For comparison, the superposition of two crystallographically
distinct RXR DBD homodimer–DR1 complexes, showing very close overlap throughout, despite different crystal packing environments (Zhao et al.,
in aligning the subunits explains why stable interactions
between RXR and RAR DNA-binding regions are not
possible off DNA. The minor groove compression at the
protein–protein junction (base pair 8) is electrostatically
favorable, bringing the negatively charged DNA phos-
phates into closer proximity with the guanidino group of
Arg75. This compression, as well as hydrogen-bonding
with Gln72 and the ordered water molecules in the minor
groove, reduces the conformational flexibility of Arg75,
precisely orienting its side-chain to make van der Waals
contact with Asn51 of RAR. The orientation of Asn51
with respect to the dimer interface is also restricted through
hydrogen-bonding with DNA. An interaction between
Glu73 of RXR, Val48 of RAR and bridging water gives
rise to a secondary stabilizing interface between subunits.
In all, the subunit interactions lead to a total loss of
less than ~400 Å2of solvent-accessible surface from both
proteins. This suggests that the observed dimer interface
is inferior to bona fide protein–protein interfaces, as
these typically bury ?700 Å2of solvent-exposed area
(Janin et al., 1998). However, the extent of the solvent-
excluded surface is comparable to and in some cases even
favorable relative to other transcription factor interfaces
that form exclusively on DNA (Klemm and Pabo, 1996;
Zhao et al., 1998, 2000; Fuji et al., 1999; Zheng et al.,
1999). The interactions in the RXR–RAR complex are
also distinct from those of the homo-cooperative complex
of RXR, which forms on the same DR1 response element
(see Figure 3B; Zhao et al., 2000). But in both cases, the
subunits utilize the DNA infrastructure to stabilize their
interactions. This observation suggests that these and other
combinatorial RXR complexes do not pre-assemble their
DNA-recognition surfaces in solution, but rather assemble
only when an appropriate gene-regulatory site is available.
Relation to other nuclear receptor assemblies
In both the homo- and heterodimeric complexes shown in
Figure 3, the downstream RXR uses exclusively its T-box
to mediate protein–protein interactions with the upstream
Zn-II region. However, there are differences in terms of
the induced T-box conformation, the protein–protein and
the protein–DNA interactions in these assemblies. For
example, the subunit interactions in the homodimer rely
to some extent on hydrogen-bonding interactions between
Glu74 from the T-box of the downstream subunit, and
Gln49 and Arg52 from the Zn-II region of the upstream
subunit (see Figure 3B). This is not possible in the RXR–
RAR interaction, since RAR has a T49 in place of
Q49, disabling the salt-bridge at the dimerization surface
The DR1 configuration is also used by the RXR–PPAR
complex (Kliewer et al., 1992a; Gearing et al., 1993;
Palmer et al., 1995; Ijpenberg et al., 1997). Interestingly,
PPARα, PPARβ and PPARγ all contain the same Asn51
and Glu54 residues that provide the upstream contact
surfaces shown in Figure 3A. In addition, an A·T base
pair is also highly preferred in the DR1 spacer of PPAR
response elements, suggesting the possibility of a similar
role for the DNA minor groove in aligning subunit
interactions (Ijpenberg et al., 1997). However, the binding
site for RXR–PPAR is extended relative to that of the
RXR–RXR and RXR–RAR, since PPAR also recognizes
the minor grooves of 5? flanking sequences using a
F.Rastinejad et al.
Fig. 5. DNA sequence recognition by RXR and RAR. (A) A view along the DNA-recognition helix (α1) of RAR showing residues Tyr13, Arg26,
Lys22, Glu19 and Arg27 and their direct and water-mediated base contacts. Hydrogen-bonds and water molecules are shown as dotted blue lines and
red spheres, respectively. The DNA sequence is numbered as in Figure 1B. (B) The corresponding view of the RXR interface. (C) Summary of the
protein–DNA contacts of RAR. Bridging water molecules are shown as black circles. The base pairs in blue form the consensus AGGTCA
recognition element. The gray circles indicate the DNA phosphates. (D) Summary of RXR–DNA interactions. The base pairs in orange form the
C-terminal extension of its core DBD known as the GRIP-
box (Ijpenberg et al., 1997; Zhao et al., 1998). These
distinctions suggest that a graded affinity is associated
with each of these RXR complexes, depending on the
sequence of the spacer, the induced curvature in the DNA
and the unique sequences flanking the DR1.
Factors responsible for the cooperative assembly
Figure 3A suggests that protein occupancy at one site can
lead to facilitated binding at the second site via protein–
protein interactions. An extended DNA-binding surface
in a considerable gain in DNA-binding cooperativity.
Structure of the RXR–RAR–DR1 complex
Fig. 6. Protein structural transitions induced in the RXR–RAR–DR1 complex. (A) The superposition of the free RAR DBD structure derived from
NMR studies with its structure in the RXR–RAR–DR1 complex. The dotted circle indicates that a major structural rearrangement induced on
assembly occurs along the Zn-II loop. (B) The superposition of the free RXR DBD structure from NMR studies with its structure in the RXR–RAR–
DR1 complex. The dotted circle indicates that a major structural reorganization occurs along the α3 helix of the T-box, which is disrupted in
forming the complex.
However, given the limited extent of the subunit contacts,
it is likely that additional indirect effects must also
contribute to cooperation between RXR and RAR. One
such mechanism is adjacent DNA site stabilization, when
binding of one subunit reduces the conformational flexibil-
ity at the adjacent site, thus pre-paying some of the
entropic costs associated with its tight binding. It has been
shown that the POU domains of Oct-1 use partially
overlapping DNA contacts at the center of their octamer
site to achieve cooperative binding, even though there is
a total absence of protein–protein interactions (Klemm
and Pabo, 1996). Similarly, Ubx–Exd subunits cooperate
to a significant extent through their partially overlapping
DNA contacts at the center of their binding site, which is
enough to produce a cooperative enhancement even if
their primary protein–protein interactions are removed
(Chan et al., 1994; Passner et al., 1999; Piper et al.,
1999). Tandem site stabilization may also enhance the
assembly of RXR and RAR on DR1, given that the half-
sites are separated by only 1 bp and that the subunits form
overlapping contacts at the spacer (see section on DNA
sequence recognition and Figure 5).
The RXR and RAR proteins also cooperate through
DNA- and dimerization-induced structural transitions in
RXR that overcome an inhibition in its DNA binding. We
have shown previously that the RXR T-box, in its α-helical
structure off DNA (Holmbeck et al., 1998a), poses
unfavorable contacts with the DNA via Glu74 and Glu75
and also blocks the DNA-binding surface in α1. A
reconfiguration of the T-box structure overcomes both
inhibitory effects in the RXR homodimer (Holmbeck
et al., 1998b; Zhao et al., 2000). Interestingly, either of
the two extended structures seen in downstream subunits
(Figure 3) leads to an effective dimerization surface (one
in the homodimer, the other in the heterodimer), and
relieves the inhibitory effects on DNA binding. A similar
mechanism by which the DNA element and a heterodim-
eric partner can help overcome an inhibitory DNA function
associated with a transcription factor has also been
described for the Hox–Exd complex (Chan et al., 1996).
The role of DNA curvature
Further cooperation in a ternary transcriptional complex
is achieved when protein occupancy at the one DNA
site induces structural distortions at the second site to
accommodate protein binding better. The importance of
even small DNA deformations, especially bends in the
DNA helix, for fine-tuning protein–DNA interactions has
been pointed out previously (Gewirth and Sigler, 1995;
Dickerson, 1998; Dickerson and Chiu, 1998). The bending
of the DNA has also been suggested to be responsible for
(IRF-1)–DNA complex, which does not use any contacts
between DBDs in its cooperative assembly (Fuji et al.,
1999). The utilization of this mechanism in achieving
cooperativity in the RXR–RAR–DR1 is illustrated in
Figure 4A and B, which shows that the DNA helical axis
bends, especially at the 5? binding site. This distortion
appears to facilitate RAR and RXR interactions with their
half-sites while being engaged as a heterodimer.
The comparison in Figure 4B demonstrates how struc-
tural versatility in DNA structure is exploited in accom-
modating two distinct types of receptor interactions with
RXR. The RXR homodimer induces a 12° kink in DR1
compared with the RXR–RAR complex, which bends the
DNA by 6°. These induced curvatures are a direct result
of the distinct subunit associations on DR1, since the
DNA sequences in the two complexes are otherwise
identical along 14 out of 15 bp. The kinks in both cases
are mainly localized to the central spacer, which adopts a
different combination of tilt and roll angles to give rise
to the 12 and 6° bends, respectively. Because the spacer
F.Rastinejad et al.
Table I. Summary of data collection and refinement
last shell (1.76–1.70 Å)
Resolution range (Å)
Total number of non-hydrogen atoms [?B? (Å2)]
R.m.s. deviation from ideal values
bond lengths (Å)
is also the convergence point for the protein–protein
interactions in each complex, the different helical distor-
tions are a direct consequence of the dimerization arrange-
ments. Figure 4C shows that two RXR homodimer–DR1
complexes from two different crystallographic environ-
ments (these being two distinct complexes in the asymmet-
ric unit of the RXR–RXR–DNA crystals) are nearly
superimposable with respect to their DNA curvatures
(Zhao et al., 2000). This observation suggests that the
distinct distortions observed in the helical axes of the
RXR–RAR and RXR–RXR binding sites are not caused
merely by their different crystallographic packing envir-
Recognition of DNA sequence
The DNA contacts for both proteins are shown in Figure 5.
The base-specific contacts of RXR and RAR occur along
6 and 5 bp, respectively. In the RXR–DNA interface, two
of the six base contacts occur along the minor grooves of
base pairs 8 and 9. The pattern of major groove contacts
for these proteins is similar, as shown in Figure 5A and
B. The recognition helix of each receptor forms direct
and water-mediated base contacts using Arg27, Lys22,
Glu19 and Lys26/Arg26. RXR and RAR both also use a
number of highly ordered water molecules to bridge their
base-specific and phosphate interactions. In addition, RXR
and RAR each form extensive backbone contacts along
7 bp of DNA (Figure 5C and D).
We compared the DNA contacts with those seen in the
RXR homodimer–DR1 complex (Zhao et al., 2000). In
terms of the number of specific protein–DNA interactions,
the RXR–RAR complex has a clear advantage over
the homodimeric RXR complex. In the homodimer, the
subunits together form specific contacts with only six
positions along the entire DR1, compared with 11 contacts
formed by the heterodimer. Moreover, the three base-
specific contacts made by each RXR in the homodimer
vary in each independent crystallographic representation
of RXR (Zhao et al., 2000). The weak, and by comparison
relaxed, binding of the homodimer to DR1 has also been
demonstrated by biochemical studies (Dowhan et al.,
1994; Yang et al., 1995; Castelein et al., 1996).
Protein structural transitions in the complex
The structures of the isolated RXR and RAR DBD
polypeptides have previously been studied by NMR
(Knegtel et al., 1993; Holmbeck et al., 1998a; van Tilborg
et al., 1999). In comparing these protein structures with
their counterparts bound to DR1, we asked what secondary
and tertiary protein structural changes are associated with
the formation of the complex. Figure 6 indicates that
the two regions undergoing the most dramatic structural
rearrangement in each subunit are the T-box and the Zn-II
module, both of which interact to stabilize the RXR–
RAR–DR1 complex and can adopt other conformations
in different contexts (Holmbeck et al., 1998b; van Tilborg
et al., 1999). The ability of the highly dynamic Zn-II
region of RXR to facilitate the assembly of the RXR–
RAR–DR5 complex was suggested in a recent dynamics
study of these proteins (van Tilborg et al., 1999). Further-
more, the Zn-II regions of nuclear receptors can in some
cases undergo transitions between helix and loop in
binding and dissociating from DNA (Schwabe et al., 1993;
Holmbeck et al., 1998a,b). Similarly, the T-box sequence
of RXR has an inherently large degree of structural
freedom, and this property is suggestive of its ability to
form cooperative protein–DNA complexes (Lee et al.,
1993; Holmbeck, 1998a,b).
The current structure illustrates the ability of a DNA
regulatory element to assume the role of a classic allosteric
ligand, inducing new conformations and/or interactions
that ultimately enhance its own binding. Conformational
changes that occur with DNA binding have not often been
studied in detail, as relatively few proteins have their
high-resolution, three-dimensional structures known for
both DNA-bound and free states. Our analysis suggests
that two distinct types of allosteric changes could be
conferred on nuclear receptors by their DNA response
element. First, there are conformational changes that can
occur within a DBD, such as the deformation of the T-box
α-helix, which leads to the enhanced binding of RXR to
DNA. A second type of allosteric influence involves
reshaping DNA structures to facilitate protein–protein
contacts. Favorable interactions between RXR and RAR
give rise to a substantial enhancement in their DNA
affinity. In the context of the full-length nuclear receptors,
the response elements are also known to have important
effects on the interactions of these factors with their
ligands, co-repressors, coactivators, AP-1 and other factors
that can influence gene expression (Kurokawa et al., 1993,
1994, 1995; Glass et al., 1997; Lefstin and Yamamoto,
The current study also provides important lessons about
the usefulness of flexible protein and DNA surfaces that
can be precisely altered on a DNA site. The inter-half-
site spacing of the retinoid response elements specifies a
fixed geometry through which a pair of nuclear receptors
must interact, since a change of one nucleotide in the
spacer causes RXR and its partners to rotate ~35° around
Structure of the RXR–RAR–DR1 complex
the double helix and be displaced from each other by
3.4 Å. Therefore, RXR DBD must either possess a number
of fixed and distinct surfaces to accommodate its many
combinatorial interactions, or more efficiently make use
of a few adaptable protein elements that can adjust to the
rotations, displacements and polarity of these binding sites
with one or more receptor partners. The essential role
played by the DNA in conferring the protein structures
required for dimerization also means that the various
combinatorial complexes of RXR need not be preformed
until they are required at particular control sites. In this
way, there is substantial economy gained given the large
number of pairwise interactions that RXR can form with
Materials and methods
The DBDs were expressed in Escherichia coli as fusions with
glutathione S–Sepharose, using the pGEX-4T vector (Pharmacia). The
proteins and the DNA oligonucleotides were purified as described
previously (Rastinejad et al., 1995; Zhao et al., 1998). The crystals used
for data collection grew at 8°C in hanging drops made up of 2 µl of
protein/DNA solution in 25 mM Tris buffer pH 7.5 and 2 µl of the
reservoir solution that contained 18–23% PEG3350, 25 mM Tris pH 7.5,
5 mM MgCl2and 0.4 M NH4Cl. The final protein and DNA concentra-
tions were 0.90 and 0.45 mM, respectively.
X-ray data were collected at –160°C on beamline X25 at the Brookhaven
National Laboratory on an MAR image plate detector (λ ? 0.98 Å). The
data were processed and scaled to 1.70 Å using HKL (Otwinowski and
Minor, 1997). The crystals belong to space group P212121with cell
parameters a ? 80.66, b ? 30.90 and c ? 101.86 Å, and one complex
per asymmetric unit. The detailed data parameters are summarized in
Table I. The Rmergevalue was calculated as in Zhao et al. (2000).
Structure solution and refinement
The structure was solved by molecular replacement using the program
AMORE (Navaza, 1994). The search model was based on a polyalanine
‘mutant’ made from the refined coordinates of the RXR homodimer on
its response element (Zhao et al., 2000). All zinc ions and terminal
residues (preceding residue 1 or following residue 67; see Figure 1A)
were omitted from the search. The rotation and translation function
searches each gave distinct solutions. Rigid body refinement resulted in
an R-factor of 47.7%, which dropped to 43.8% after the first cycle of
simulated annealing with X-PLOR (Bru ¨nger, 1993). At this point, metal
atoms, conserved side-chains and C-terminal residues were built into
their Fo– Fcand 2Fo– Fcelectron density maps with the program O
(Jones et al., 1991). Only after almost all the identical RXR and RAR
residues had been modeled were the resulting difference maps inspected
for electron density that could be assigned uniquely to RXR or RAR.
Many RXR- and RAR-specific side-chains were clearly identifiable in
all subsequent difference maps. Iterative steps of map inspection, model
building and simulated annealing cycles resulted in an R-factor of 25.4%
for 2σ reflections between 6.0 and 1.7 Å (Rfreefor 5% of the data was
29.9%). The refinement procedure was then changed to the conjugate
gradient least-squares method employed in SHELXL-97. Restraints for
the DNA base pairs were included according to geometric parameters
described by Parkinson et al. (1996), with bond angles addressed by
their 1,3-distances. Purine and pyrimidine atoms were restrained to the
plane of the aromatic ring system. We did not impose any torsion
restraints on the nucleotides. The stereochemistry at the chiral carbon
atoms, however, was fixed by restraining their chiral volume. Zinc and
sulfur atoms were refined anisotropically to account for disorder visible
in the final difference maps. The final model included 342 defined water
molecules, refined to the values given in Table I. The R and Rfreewere
calculated as in Zhao et al. (2000).
The authors wish to thank Scott A.Chasse and Srikripa Devarakonda for
assistance with biochemical studies, Christine S.Wright and Michael
L.Sierk for computational advice, and the staff of beamline X25 at
Brookhaven National Laboratory for assistance with data collection. The
coordinates can be obtained from the corresponding authors, or from
the Protein Data Bank. T.W. was supported in part by the Alexander
von Humboldt Foundation. This study was made possible by grant
support from the Leukemia Society of America (S.K.) and the National
Institutes of Health GM55217 (F.R.).
Bru ¨nger,A.T. (1993) X-PLOR Version 3.1: A System for X-ray
Crystallography and NMR. Yale University Press, New Haven, CT.
Bugge,T.H., Pohl,J., Lonnoy,O. and Stunnenberg,H.G. (1992) RXR α, a
promiscuous partner of retinoic acid and thyroid hormone receptors.
EMBO J., 11, 1409–1418.
Carson,M. and Bugg,C.E. (1986) Algorithm for ribbon models of
proteins. J. Mol. Graph., 4, 121–122.
Castelein,H., Janssen,A., Declercq,P.E. and Baes,M. (1996) Sequence
requirements for high affinity retinoid X receptor-α homodimer
binding. Mol. Cell. Endocrinol., 119, 11–20.
Chambon,P. (1996) A decade of molecular biology of retinoic acid.
FASEB J., 10, 940–954.
Chan,S.-K., Jaffe,L., Copovilla,M., Botas,J. and Mann,R.S. (1994) The
DNA binding specificity of Ultrabithorax is modulated by cooperative
interactions with extradenticle, another homeodomain. Cell, 78,
Chan,S.-K., Po ¨pperl,H., Krumlauf,R. and Mann,R.S. (1996) An
extradenticle-induced conformational change in a HOX protein
overcomes an inhibitory function of the conserved hexapeptide motif.
EMBO J., 15, 2476–2487.
Dickerson,R.E. (1998) DNA bending–the prevalence of kinkiness and
the virtues of normality. Nucleic Acids Res., 26, 1906–1926.
Dickerson,R.E. and Chiu,T.K. (1998) Helix bending as a factor in
protein/DNA recogntion. Biopolymers, 44, 361–403.
Identification of deoxyribonucleic acid sequences that bind retinoid-
X receptor-γ with high affinity. Endocrinology, 135, 2595–2607.
Forman,B.M., Umesono,K., Chen,J. and Evans,R.M. (1995) Unique
response pathways are established by allosteric interactions among
nuclear hormone receptors. Cell, 81, 541–550.
Fuji,Y., Shimizu,T., Kusumoto,M., Kyogoku,Y., Tadatsugu,T. and
Hakoshima,T. (1999) Crystal structure of an IRF–DNA complex
reveals novel DNA recognition and cooperative binding to a tandem
repeat of core sequences. EMBO J., 18, 5028–5041.
Gearing,K.L., Gottlicher,M., Teboul,M., Widmark,E. and Gustafsson,J.A.
(1993) Interaction of the peroxisome-proliferator-activated receptor
and retinoid X receptor. Proc. Natl Acad. Sci. USA, 90, 1440–1444.
Gewirth,D.T. and Sigler,P.B. (1995) The basis for half-site specificity
explored through a non-cognate steroid receptor–DNA complex.
Nature Struct. Biol., 2, 386–394.
Glass,C.K., Rose,D.W. and Rosenfeld,M.G. (1997) Nuclear receptor
coactivators. Curr. Opin. Cell Biol., 9, 222–232.
Gronemeyer,H. and Moras,D. (1995) How to finger DNA. Nature, 375,
Holmbeck,S.M.A., Foster,M.P., Casimoro,D.R., Sem,D.S., Dyson,H.J.
and Wright,P.E. (1998a) High-resolution solution structure of the
retinoid X receptor DNA-binding domain. J. Mol. Biol., 281, 271–284.
Holmbeck,S.M.A., Dyson,H.J. and Wright,P.E. (1998b) DNA-induced
conformational changes are the basis for cooperative dimerization by
the DNA binding domain of the retionoid X receptor. J. Mol. Biol.,
Ijpenberg,A., Jeannin,E., Wahli,W. and Desvergne,B. (1997) Polarity and
specific sequence requirements of peroxisome proliferator-activated
receptor (PPAR)/retinoid X receptor heterodimer binding to DNA.
J. Biol. Chem., 272, 20108–20117.
Janin,J., Miller,S. and Chothia,C. (1998) Surface, subunit interfaces and
interior of oligomeric proteins. J. Mol. Biol., 204, 155–164.
Jones,T.A., Zou,J.Y., Cowan,S.W. and Kjeldgaard,M. (1991) Improved
methods for building protein models in electron density maps and the
location of errors in these models. Acta Crystallogr. A, 47, 110–119.
Klemm,J.D. andPabo,C.O. (1996)
interactions: cooperative binding of isolated subdomains and effects
of covalent linkage. Genes Dev., 10, 27–36.
Kliewer,S.A., Umesono,K., Noonan,D.J., Heyman,R.A. and Evans,R.M.
(1992a) Convergence of 9-cis retinoic acid and peroxisome proliferator
signalling pathways through heterodimer formation of their receptors.
Nature, 358, 771–774.
F.Rastinejad et al.
Kliewer,S.A., Umesono,K., Mangelsdorf,D.J. and Evans,R.M. (1992b)
Retinoid X receptor interacts with nuclear receptors in retinoic acid,
thyroid hormone and vitamin D3signalling. Nature, 355, 446–449.
Knegtel,R.M.A., Katahira,M., Schilthuis,J.G., Bonvin,A.M., Boelens,R.,
Eib,D., van der Saag,P.T. and Kaptein,R. (1993) The solution structure
of the human retinoic acid receptor-β DNA-binding domain. J. Biomol.
NMR, 3, 1–17.
Kurokawa,R., Yu,V.C., Na ¨a ¨r,A., Kyakumoto,S., Han,Z., Silverman,S.,
Rosenfeld,M.G. and Glass,C.K. (1993) Differential orientations of the
DNA-binding domain and carboxy-terminal dimerization interface
regulate binding site selection by nuclear receptor heterodimers. Genes
Dev., 7, 1423–1435.
Rosenfeld,M.G., Heyman,R.A. and Glass,C.K. (1994) Regulation of
retinoid signalling by receptor polarity and allosteric control of ligand
binding. Nature, 371, 528–531.
Kurokawa,R., Soderstrom,M., Horlein,A., Halachmi,S., Brown,M.,
Rosenfeld,M.G. and Glass,C.K. (1995) Polarity-specific activities of
retinoic acid receptors determined by a co-repressor. Nature, 377,
Leblanc,B.P. and Stunnenberg,H.G. (1995) 9-cis retinoic acid signaling:
changing partners causes some excitement. Genes Dev., 9, 1811–1816.
Lee,M.S., Kliewer,S.A., Provencal,J., Wright,P.E. and Evans,R.M. (1993)
Structure of the retinoid X receptor α DNA binding domain: a helix
required for homodimeric DNA binding. Science, 260, 1117–1121.
Lefstin,J.A. and Yamamoto,K.R. (1998) Allosteric effects of DNA on
transcriptional regulators. Nature, 392, 885–888.
Leid,M. et al. (1992) Purification, cloning and RXR identity of the HeLa
cell facor with which RAR or TR heterodimerizes to bind target
sequences efficiently. Cell, 68, 377–395.
Luisi,B.F., Xu,W.X., Otwinowski,Z.,Freedman,L.P., Yamamoto,K.R. and
Sigler,P.B. (1991) Crystallographic analysis of the interaction of the
glucocorticoid receptor with DNA. Nature, 352, 497–505.
Mader,S., Chen,J.Y., Chen,Z., White,J., Chambon,P. and Gronemeyer,H.
(1993) The patterns of binding of RAR, RXR and TR homo- and
heterodimers to direct repeats are dictated by the binding specificites
of the DNA binding domains. EMBO J., 12, 5029–5041.
Mangelsdorf,D.J. and Evans,R.M. (1995) The RXR heterodimers and
orphan receptors. Cell, 83, 841–850.
Mangelsdorf,D.J., Umesono,K., Kliewer,S.A., Borgmeyer,U., Ong,E.S.
and Evans,R.M. (1991) A direct repeat in the cellular retinol-binding
protein type II gene confers differential regulation by RXR and RAR.
Cell, 66, 555–561.
Mangelsdorf,D.J. et al. (1995) The nuclear receptor superfamily: the
second decade. Cell, 83, 835–839.
Nikodem,V.M. and Ozato,K. (1992) H-2RIIBP (RXR β) hetero-
dimerization provides a mechanism for combinatorial diversity in the
regulation of retinoic acid and thyroid hormone responsive genes.
EMBO J., 11, 1419–1435.
Meinke,G. and Sigler,P.B. (1999) DNA-binding mechanism of the
monomeric orphan receptor NGFI-B. Nature Struct. Biol., 6, 471–477.
nucleic acid recognition sites. Biochemistry, 38, 1999–2017.
Navaza,J. (1994) AMoRe: an automated package for molecular
replacement. Acta Crystallogr. A, 50, 157–163.
Nicholls,A., Sharp,K.A. and Honig,B. (1991) Protein folding and
association: insights from the interfacial and thermodynamic properties
of hydrocarbons. Proteins, 11, 281–296.
Otwinowski,Z. and Minor,W. (1997) Processing of X-ray diffraction
data collected in oscillation mode. Methods Enzymol., 276, 307–326.
Palmer,C.N.A., Hsu,M.-H., Griffin,K.J. and Johnson,E.F. (1995) Novel
sequence determinants in peroxisome proliferator signaling. J. Biol.
Chem., 270, 16114–16121.
Berman,H.M. (1996) New parameters for the refinement of nucleic
acid containing structures. Acta Crystallogr. D, 52, 57–64.
Passner,J.M., Ryoo,H.D., Shen,L., Mann,R.S. and Aggarwal,A.K. (1999)
Structure of a DNA-bound Ultrabithorax–Extradenticle homeodomain
complex. Nature, 397, 714–719.
Perlmann,T., Rangarajan,P.N., Umesono,K. and Evans,R.M. (1993)
Determinants for selective RAR and TR recognition of direct repeat
HREs. Genes Dev., 7, 1411–1422.
Evans,R.M. (1996) Two distinct dimerization interfaces differentially
modulate target gene specificity of nuclear hormone receptors. Mol.
Endocrinol., 10, 958–966.
Piper,D.E., Batchelor,A.H., Chang,C.-P., Cleary,M.L. and Wolberger,C.
(1999) Structure of a HoxB1–Pbx1 heterodimer bound to DNA: role
of the hexapeptide and a fourth homeodomain helix in complex
formation. Cell, 96, 587–597.
Predki,P.F., Zamble,D., Sarkar,B. and Giguere,V. (1994) Ordered binding
of retinoic acid and retinoid-X receptor to asymmetric response
elements involves determinants adjacent to the DNA-binding domain.
Mol. Endocrinol., 8, 31–39.
Rastinejad,F., Perlmann,T., Evans,R.M. and Sigler,P.B. (1995) Structural
determinants of nuclear receptor assembly on DNA direct repeats.
Nature, 375, 203–211.
Schwabe,J.W., Chapman,L., Finch,J.T. and Rhodes,D. (1993) The crystal
structure of the estrogen receptor DNA-binding domain bound to
DNA: how receptors discriminate between their response elements.
Cell, 75, 567–578.
dodecamer at high resolution reveals a spine of water on sodium.
Biochemistry, 37, 8341–8355.
Umesono,K., Murakami,K.K., Thompson,C.C. and Evans,R.M. (1991)
Direct repeats as selective response elements for the thyroid hormone,
retinoic acid and vitamin D3receptors. Cell, 65, 1255–1266.
van Tilborg,P.J.A. et al. (1999) Millisecond to microsecond time scale
dynamics of the retinoid X and retinoic acid receptor DNA-binding
domains and dimeric complex formation. Biochemistry, 38, 1951–
Participation of non-zinc finger residues in DNA binding by two
nuclear orphan receptors. Science, 256, 107–110.
Yang,Y.Z., Subauste,J.S. and Koenig,R.J. (1995) Retinoid X receptor α
binds with the highest affinity to an imperfect direct repeat response
element. Endocrinology, 136, 2896–2903.
Yu,V.C. et al. (1991) RXR β: a coregulator that enhances binding of
retinoic acid, thyroid hormone and vitamin D receptors to their cognate
response elements. Cell, 67, 1251–1266.
Gronemeyer,H. (1994) The dimerization interfaces formed between
the DNA binding domains of RXR, RAR and TR determine the
binding specificity and polarity of the full-length receptors to direct
repeats. EMBO J., 13, 1425–1433.
Zhao,Q., Khorasanizadeh,S., Miyoshi,Y., Lazar,M.A. and Rastinejad,F.
(1998) Structural elements of an orphan nuclear receptor–DNA
complex. Mol. Cell, 1, 849–861.
Zhao,Q., Chasse,S.A., Devarakonda,S., Sierk,M.L., Ahvazi,B. and
Rastinejad,F. (2000) Structural basis of RXR–DNA interactions. J.
Mol. Biol., in press.
Zheng,N., Fraenkel,E., Pabo,C.O. and Pavletich,N.P. (1999) Structural
basis of DNA recognition by the heterodimeric cell cycle transcription
factor E2F-DP. Genes Dev., 13, 666–674.
Received December 8, 1999; accepted January 7, 2000