Tonicity-responsive enhancer binding protein (TonEBP), also known as NFAT5, is a unique member of the NFAT family of transcription factors that regulates gene expression induced by osmotic stress in mammalian cells. Unlike monomeric members of the NFAT family, TonEBP exists as a homodimer and binds asymmetric TonE DNA sites; furthermore, the affinity of TonEBP for DNA is much lower than that of other NFAT proteins. How TonEBP recognizes the TonE site and regulates the activation of hypertonicity response genes has not been clear. Here we show that TonEBP adopts a NF-B-like structure upon binding to DNA, providing a direct structural link between the NFAT and NF-B family of transcription factors. We also show that TonEBP completely encircles its DNA target and present biochemical evidence that the DNA encirclement may lead to increased kinetic stability of the TonEBP−DNA complex. Thus, the list of proteins that bind DNA by encirclement is now expanded to include sequence-specific transcription factors.
The NFAT and NF-B families of transcription factors play essential roles in diverse biological functions such as immune response and development. Proteins from these two families are thought to be distantly related to each other because their DNA binding domains share a low but significant sequence identity1. However, previous structural studies indicate that NFAT and NF-B have distinct DNA binding modes2,
3,
4. While members of the NF-B family bind DNA as a cooperative dimer, those from the NFAT family bind DNA as a monomer or in complex with other transcription factors.
The NFAT family has recently been expanded by the discovery of a new member, NFAT5, or TonEBP. TonEBP is the only known mammalian transcription factor that regulates gene expression in response to hypertonicity5,
6. In addition to regulation of osmotic stress, TonEBP also plays important roles in other cellular processes such as lymphocyte activation and brain development7,
8. TonEBP is classified as a member of the NFAT family because its DNA binding domain shares a high sequence identity (up to 43%) with that of NFAT proteins. Furthermore, the consensus DNA binding site of TonEBP is TGGAAANNYNY (where N represents any nucleotide, and Y represents any pyrimidine), which closely resembles that of the NFAT family of transcription factors1,
9,
10. However, unlike the 'classical' NFAT members, NFAT1−4, that are monomeric in solution and often bind DNA cooperatively with Fos−Jun4, TonEBP exists as a preformed dimer and binds asymmetric DNA sites. The DNA affinity of TonEBP is much lower than that of NFAT1−4. These observations raise an intriguing question as to how TonEBP might recognize the TonE site and regulate the activation of hypertonicity genes. Here we show that TonEBP binds DNA as a dimer in a conformation similar to the classical 'butterfly' mode of the NF-B−DNA complexes2,
3. A surprising feature of the TonEBP−DNA complex is that the TonEBP dimer forms a complete circle around DNA. We further present biochemical evidence that the TonEBP−DNA complex has an unusually high kinetic stability, as shown by the slow dissociation rate, that probably arises from the DNA encirclement by TonEBP.
Overall structure of the TonEBP−DNA complex The crystal structure of the human TonEBP DNA binding region (amino acids 170−470) in complex with a DNA closely related to the TonEC1 site from the promoter of the human sodium/myo-inositol cotransporter (SMIT) gene was determined by MIRAS to 2.86 Å (Table 1)11. TonEBP binds DNA as a homodimer. Each monomer has two immunoglobulin (Ig)-like domains and resembles structurally the DNA binding region, termed Rel Homology Region (RHR), of the NFAT and NF-B family of transcription factors (Fig. 1a)2,
3,
4.
a, Overall structure. The following scheme is used for all illustrations except Fig. 3c: the N-terminal (RHR-N) and C-terminal (RHR-C) domains of TonEBP are colored yellow and green, respectively. Strands and helices are labeled; the lettering corresponds to that used for NFAT1 (ref. 4). DNA is shown in the stick model. The sequence of the DNA used in the crystals is shown below the figure. b, Structure-based sequence alignment of human TonEBP (T), NFAT1(N), and NF-B p52 (P) in the DNA-binding region. The numbering for TonEBP is used. The secondary structure assignments for TonEBP are shown as colored bars (-helices) and arrows (-strands) above the aligned sequences; those for NF-B p52 are shown below the sequence. The colored blocks show residues that participate in contacts to DNA (magenta), RHR-C dimerization (green) and E'F loop dimerization (yellow).
Structural comparison by backbone superposition indicates that TonEBP is more closely related to NFAT than to NF-B, which is consistent with the higher sequence identity between TonEBP and NFAT1 (43%) than between TonEBP and NF-B p52 (25%), the NF-B family member most closely related to TonEBP (Fig. 1b)12. However, the relative orientation between the N-terminal (RHR-N) and C-terminal (RHR-C) Ig domains in TonEBP differs drastically from that in NFAT1 (ref. 4). In the NFAT1−Fos−Jun−DNA complex involved in cytokine expression4, the RHR-C of NFAT1 is adjacent to its RHR-N and makes a small contact to Fos−Jun. In the TonEBP−DNA complex, on the other hand, the RHR-C of TonEBP extends away from its RHR-N and makes a large dimer interface with the RHR-C of the other monomer, much like the NF-B dimer−DNA complexes2,
3,
12,
13,
14.
Unlike the NF-B−DNA complexes, however, the TonEBP−DNA complex has a second dimer interface mediated by the RHR-N domains of the two monomers. This N-terminal dimer interface is on the opposite side of the DNA with respect to the dimer interface of the RHR-C domains; as a result, the TonEBP dimer forms a complete circle around the DNA (Fig. 1a).
DNA binding by TonEBP The encircled DNA, in a straight B form conformation, is bound by the TonEBP dimer where the surface potential is relatively positive (Fig. 2a). Because the inner diameter of the protein ring is larger than the outer diameter of DNA, the surface of the DNA is not fully contacted by protein around the circle. Instead the DNA is bound to one side of the circle through specific interactions between the RHR-N of one monomer and the TGGAAA half of the TonE site, referred to as the consensus half site (Fig. 1a).
a, The TonEBP dimer (surface model) encircles DNA; the region of the circle that interacts with DNA has a relatively positive surface potential (blue). The orientation of this figure is the same as that in Fig. 1a. b, Detailed view of consensus half-site, showing the roles of Arg 217, Glu 223, Arg 226, Gln 364, and Tyr 220 in specifying base pairs. Arg 226 is supported by salt links to Glu 223, which also accepts hydrogen bonds from the main chain NH of Tyr 220 and from N4 of C2'.
Sequence-specific DNA binding by TonEBP is mediated mostly by the AB loop of its RHR-N, which is also the major DNA binding element in NFAT and NF-B. Arg 217, Arg 226, Glu 223, and Tyr 220 from the AB loop and Gln 364 from the linker between RHR-N and RHR-C interact extensively with DNA in the major groove to specify the base sequence GGAAA (Fig. 2b). Consistent with our crystal structure, mutation of Glu 223 (ref. 15), Arg 217, or Arg 226 (ref. 16) in TonEBP abolished its DNA binding ability. TonEBP shares a conserved Arg (Arg 219) with the NF-B proteins (Arg 54 in NF-B p52) (Fig. 1b). However, unlike its NF-B counterpart12, Arg 219 of TonEBP does not bind DNA directly in our crystal structure. Instead, Arg 219 of TonEBP adopts a conformation similar to that of His 423 in NFAT1 which does not bind DNA in the NFAT1−Fos−Jun−DNA complex (Figs 1b, 2)4. Thus, the DNA binding mode of the AB loop of TonEBP more closely resembles that of NFAT than that of NF-B. However, there is a major difference between TonEBP and NFAT1 in DNA binding. In NFAT1, the linker connecting RHR-N and RHR-C forms an -helix and contacts DNA extensively in the major groove and along the backbone; the corresponding linker in TonEBP is shorter and extended, making no DNA contacts other than Gln 364 (Figs 1b, 2).
The RHR-N of the other monomer in the TonEBP−DNA complex does not have the consensus sequence in its DNA site (TGGAAAAATAG, site of binding is underlined). As a result, the RHR-N binds the nonconsensus sequence mostly through contacts to the DNA backbone. Our structure suggests that the orientation and DNA binding interactions of the second RHR-N on the nonconsensus half-site can vary to accommodate the variety of sequences found in different tonicity responsive enhancers9,
10. The TonE sites with a second half more closely resembling the consensus sequence (GGAAA) have higher affinities for the TonEBP dimer (J.C.S. and L.C., unpublished results), which may explain the partial sequence conservation in the nonconsensus half of the TonE sites.
The C-terminal dimer interface The major dimer interface in the TonEBP−DNA complex is formed between the RHR-C domains of the two monomers (Fig. 3a). The dyad axis of the RHR-C dimer is roughly perpendicular to the DNA, similar to the situation in the NF-B−DNA complexes2,
3,
12,
13,
14. Unlike the NF-B−DNA complexes, in which the RHR-C dimer makes extensive DNA contacts, the RHR-C dimer in TonEBP 'hangs above' the DNA with few DNA contacts (Fig. 1a). The RHR-C dimer interface in TonEBP consists of a hydrophobic center surrounded by extensive polar interactions, burying 1,200 Å2 surface area (Fig. 3a). The large dimer interface explains why TonEBP exists as a stable homodimer in solution, as demonstrated by gel filtration and chemical crosslinking (J.C.S. and L.C., unpublished data). Double mutation of the interface residues, Phe 388 and Ile 390 to Ala, or His 424 and Ile 429 to Ala, completely abolishes or significantly reduces TonEBP dimerization15.
a, The C-terminal dimer interface viewed in the same orientation as Fig. 1a. Hydrophobic residues Leu 372, Ile 390, Leu 422, Ile 429, and Phe 388 from each TonEBP monomer form the center of the interface. Polar residues Asn 426, His 424, His 427, Lys 373, Glu 386, and Ser 375 from each monomer form the peripheral of the interface through networks of hydrogen bonding and electrostatic interactions. b, The N-terminal dimer interface viewed from underneath the DNA with respect to Fig. 1a. The -helix of the E'F loop from each monomer supplies residues for dimerization, which include Arg 315, Ala 317, Asp 318, and Glu 320. The conformation of the -helix is stabilized by the DNA backbone (N-terminal capping) and by the hydrophobic interactions with the main body of the protein through residues Val 321, Leu 312, and Phe 267. c, Simulated-annealing omit map showing the electron density of the E'F dimer interface residues (Arg 315, Asn 316, Ala 317, Asp 318, Val 319, and Glu 320) in stereo. The 2.86 Å A-weighted Fo - Fc map is contoured at 2.0 level.
Although the overall RHR sequence identity between TonEBP and NF-B p52 is low (25%), the RHR-C domains of TonEBP and NF-B p52 share 34% sequence identity. Remarkably, many dimerization residues in TonEBP, such as Lys 373, Glu 386, Phe 388, Ile 390, His 424, and Ile 429, have either identical or similar counterparts in NF-B p52 and p65 (Fig. 1b), raising the possibility that TonEBP and NF-B proteins might form mixed dimers.
Despite the fact that TonEBP and NFAT1 share 35% sequence identity in their RHR-C domains, many RHR-C dimerization residues in TonEBP differ from the corresponding residues in NFAT1. For example, Glu 386, Phe 388, and Ile 390 of TonEBP are changed to Gln 595, Ile 597, and Thr 599 respectively in NFAT1 (Fig. 1b). These substitutions most likely destabilize the dimer interface in NFAT1. This observation is consistent with the fact that other members of the NFAT family (NFAT1−4) do not form stable dimers in solution17,
18.
The N-terminal dimer interface A second dimer interface in the TonEBP−DNA complex is formed by the E'F loop of the RHR-N domain of each monomer (Fig. 3b,c). The overall conformation of the E'F loop in TonEBP is very similar to that of NFAT1, with a short -helix followed by an extended loop. However, many residues of the E'F loop in NFAT1 that bind Fos−Jun in the NFAT−Fos−Jun−DNA complex, such as Lys 530, Thr 533, and Ile 535, are changed to Ile 323, Ala 326, and Ser 328 respectively in TonEBP (Fig. 1b), explaining why TonEBP does not bind DNA cooperatively with Fos−Jun6. Instead, the side chains of the E'F loop -helix of each monomer interact with each other directly in the TonEBP−DNA complex. In the middle of the helix dimer interface, the Ala 317 residues of each monomer make van der Waals contacts with each other and with the side chain of Arg 315 of the opposing -helix. On each side of the dimer interface, Arg 315 from one -helix, whose side chain conformation is held by Asp 318 within the same helix, forms salt bridges with Glu 320 of the opposing helix. Although the E'F loop dimer interface has a much smaller buried surface area (238 Å2) than the RHR-C dimer interface (1200 Å2), its extensive van der Waals and electrostatic interaction networks most likely make a significant contribution to the stability of the TonEBP−TonE complex. Interestingly, all residues involved in the E'F loop-mediated dimerization in TonEBP are also conserved in NFAT1−4 (Fig. 1b), raising the possibility that other NFAT members may use similar dimerization mechanisms in DNA binding19.
Kinetic stability of the TonEBP−DNA complex The affinities of TonEBP for most TonE sites are low with Kds 50 nM (data not shown), consistent with the limited protein−DNA contacts seen in the TonEBP−DNA complex. Dimerization of TonEBP may improve its DNA affinity by the cooperative DNA binding of both RHR-N domains. However, even on the TonE site containing two consensus half sites9, the affinity of the TonEBP dimer for DNA (30 nM) is still much lower than that of NFAT1 when it binds one half-site as a monomer (Kd7.6 nM). Based on the studies of toroidal proteins that encircle DNA20, it is likely that a major mechanism for TonEBP binding to its DNA target may be through increased kinetic stability. To test this, we analyzed the dissociation of the TonEBP−DNA and the NFAT1−DNA complexes. In the presence of excess amount of unlabeled DNA, the NFAT1−DNA complex dissociated almost completely in <2 min (Fig. 4a). In contrast, the TonEBP−DNA complex, despite having a much weaker affinity, had a much slower dissociation rate. Quantitative analysis revealed that the dissociation of the TonEBP−DNA complex is biphasic (Fig. 4b). While a large fraction (60%) of the complex dissociated rapidly (within 1min), which may represent nonspecific TonEBP−DNA complex (see Methods), a significant amount of the TonEBP−DNA complex dissociated much more slowly (half life 39 6 min). For comparison, we also measured the dissociation half life of the NF-B p50−p65 complex bound to Ig-B DNA under similar conditions and found it to be 3 min, significantly shorter than that of the TonEBP−DNA complex (data not shown). The dissociation experiments were carried out with a short DNA fragment. We expect that the dissociation rate of the TonEBP−DNA complex will be even slower from a longer or circular DNA substrate. Mutations at the RHR-C dimer interface (double mutants F388A/I390A and H424A/I429A) that break up the DNA encirclement abolished the kinetic stability of the TonEBP−DNA complex even though these mutations are far from the DNA binding surfaces (data not shown)15.
Figure 4. Biochemical analysis of the TonEBP−DNA complex.
a, Electrophoresis mobility shift assay of the NFAT1−DNA and the TonEBP−DNA complexes. The radiolabeled NFAT1−DNA and TonEBP−DNA complexes were formed as described in Methods. An aliquot of the NFAT1−DNA complex (lane 1) and the TonEBP−DNA complex (lane 5) was run on the gels as time zero. 1000-fold excess of unlabeled DNA probe was added to each complex and an aliquot of the complex was analyzed at different time points (NFAT1−DNA complex: lanes 2−4; TonEBP−DNA complex: lanes 6−11). Lanes 1−4 and 5−11 are from two separate gels that were run under the same conditions. The sequence of the DNA probe is shown below the gels. b, The fraction of the TonEBP−DNA complex left at time t (Dt) relative to the total complex at time zero (D0) was plotted versus time. Data were taken from five independent experiments and fit to a biphasic kinetic model (see Methods).
Discussion TonEBP shares many structural features with both the NFAT and NF-B families of transcription factors. On one hand, the C-terminal dimer interface of TonEBP is remarkably similar to that of the NF-B proteins, raising the possibility that TonEBP and members of the NF-B family may form mixed dimers inside cells and that the TonEBP and NF-B pathways may crossregulate gene expression in certain cellular contexts. On the other hand, the N-terminal dimer interface of TonEBP in the TonEBP−DNA complex is formed by a protein surface (the E'F loop) that is used by NFAT1 to contact Fos−Jun in the NFAT1−Fos−Jun−DNA complex4. Thus, although TonEBP and NFAT1 share an almost identical protein fold, the protein surface formed by the E'F loop functions differently in the assembly of these two distinct transcription complexes. The crystal structure of the TonEBP−DNA complex reveals that the NFAT and NF-B families of transcription factors are much more closely related than their sequence suggests. The fact that NFAT and NF-B share similar DNA binding mechanisms also suggests that transcription factors from these two families may regulate gene expression through shared enhancer elements in response to different cellular signals.
Overall, TonEBP binds its TonE target site through base-specific contacts as well as DNA encirclement, giving rise the increased kinetic stability of the TonEBP−DNA complex. Binding via encirclement has been proposed for a number of proteins involved in DNA modifications20,
21. Our studies extend this mode of DNA binding to a sequence-specific transcription factor.
DNA encirclement by proteins may occur widely in transcription factor complexes that bind DNA cooperatively. For example, many members of the NF-B family of transcription factors, such as p50 and p52, have a helix-rich motif corresponding to the E'F loop of TonEBP. Although the helix-rich motifs in p52 and p50 do not interact with each other directly in their respective protein−DNA complexes, these motifs are poised to bind additional accessory factors to encircle DNA. The high mobility group protein HMG I(Y), which enhances the stability of NF-B−DNA complexes and NF-B dependent transcription, may be one such accessory factor22. HMG I(Y) has been shown to interact with NF-B and bind DNA in the minor groove at the position equivalent to the E'F loop dimer interface in our TonEBP−TonE complex23. Thus, a complex of NF-B dimer and HMG I(Y) may form a ring-like structure on DNA similar to that seen here3.
Methods Sample preparation and crystallization. The DNA binding domain of human TonEBP (amino acids 170−470) was expressed and purified as described6. The TonEBP−DNA complex was prepared by mixing equal molar amounts of TonEBP and DNA at a concentration of 0.2−0.4 mM in a storage buffer containing 10 mM HEPES, pH 7.5, 1 mM dithiothreitol (DTT) and 100 mM NaCl. Crystals of the TonEBP−DNA complex were grown at room temperature by the hanging drop method using a reservoir buffer of 50 mM Bis-Tris-Propane (BTP) HCl, pH 6.6, 100 mM NaCl, 3−6% (w/v) PEG 3000 and 10 mM MgCl2. Typically, crystals grow to 800 m 200 m 50 m in one or two days. Crystals belong to the space group P212121, with cell dimensions a = 59.57 Å, b = 95.37 Å, and c = 158.43 Å.
Data collection, structure determination and analysis. Crystals were stabilized in the harvest/cryoprotectant buffer: 10 mM BTP HCl pH 6.6, 100 mM NaCl, 10−12% (w/v) PEG 3000 and 25% (w/v) glycerol. In addition to the SeMet labeled TonEBP, isomorphous heavy atom derivatives were also obtained by soaking crystals in methyl mercury nitrate (1−2 mM). All native and derivative crystals were flash frozen in liquid nitrogen for storage and data collection under cryogenic conditions (100 K). The native and SeMet derivative data were collected at the Advanced Photon Source (APS, Argonne National Laboratory) beamline (14-BM-C) and (14-BM-D) (wavelength 0.9795 Å) respectively.
Data were reduced using DENZO and SCALEPACK24. The positions of selenium atoms were determined by inspecting isomorphous and anomalous difference Patterson maps calculated using programs from the CCP4 suite25. Refinement of heavy atom parameters and phase calculation was performed using MLPHARE. The phases were further improved by DM with a two-fold ncs averaging. The overall figure of merit after DM was 0.78 at 30−3.1 Å resolution for the initial map. DNA and all protein residues except 170−187 and 469−470, which were disordered in the crystals, were built into the density using O26. The refinement was carried out using CNS27. Final models have very good geometry as examined by PROCHECK. All residues have / angles in the 'allowed' region of a Ramachandran plot, with 80% in the 'most favored region'. The statistics of crystallographic analysis are presented in Table 1. Figures of structure illustration were prepared using MOLSCRIPT28, Ribbons29, and GRASP30. Structural comparison between TonEBP, NFAT and NF-B was carried out in O26.
Electrophoresis mobility shift assay. 32P labeled DNA (0.02 nM) with a TonE site was mixed with various concentrations of TonEBP or NFAT1 (1 nM−500 nM) in 20 l of binding buffer (10 mM HEPES, pH 7.4, 100 mM NaCl, 1 mM DTT, 5% glycerol) and incubated at room temperature for 3 h. The equilibrated complexes were analyzed on a 6% nondenaturing polyacrylamide gel. The free and bound DNA were quantified by phosphor image analysis, and the Kds were determined by standard methods.
To analyze the dissociation rate, the TonEBP−DNA and the NFAT1−DNA complexes were formed as above, followed by addition of 1000-fold excess of unlabeled DNA. At various time points, an aliquot was loaded on a 6% nondenaturing polyacrylamide gel. The bound and unbound DNA probes were separated by electrophoresis and quantified by phosphor imaging. Quantitative analysis indicated that the dissociation of the TonEBP−DNA complex is biphasic. The following equation was used to fit the dissociation curve:
where Dt is the amount of complex (normalized by the total input of DNA probe) left at time t and D0 is the amount of the complex at time zero. The terms kf and ks are the rate constants governing the fast and the slow phases, respectively, af and as are the corresponding amplitudes. The half life of the slow phase (39 6 min) is calculated from ks.
Coordinates. Coordinates and diffraction data have been deposited in the Protein Data Bank under accession code 1IMH.
Acknowledgments The authors thank Z. Ren and H. Tong from APS beamline 14-BM; M. Giffin, G.A. Murphy and D. Theobald for help in data collection; T.R. Cech, O.C. Uhlenbeck and J.A Goodrich for critical reading of the manuscript. This research was supported by a scholar award from the Damon Runyon-Walter Winchell Foundation (L.C.), a grant from the W. M. Keck foundation (L.C.), and an NIH training grant (J.C.S.). C. L.R. is a recipient of a career development Special Fellowship of the Leukemia and Lymphoma Society.
B-like structure upon binding to DNA, providing a direct structural link between the NFAT and NF-
B families of transcription factors play essential roles in diverse biological functions such as immune response and development. Proteins from these two families are thought to be distantly related to each other because their DNA binding domains share a low but significant sequence identity
-helices) and arrows (
-strands) above the aligned sequences; those for NF-
1,200 Å2 surface area (
A-weighted Fo - Fc map is contoured at 2.0 
6 min). For comparison, we also measured the dissociation half life of the NF-
m 
200 
/
angles in the 'allowed' region of a Ramachandran plot, with 80% in the 'most favored region'. The statistics of crystallographic analysis are presented in 
