Structural basis for HNF-4alpha activation by ligand and coactivator binding.

In addition to suggesting that fatty acids are endogenous ligands, our recent crystal structure of HNF-4alpha showed an unusual degree of structural flexibility in the AF-2 domain (helix alpha12). Although every molecule contained a fatty acid within its ligand binding domain, one molecule in each homodimer was in an open inactive conformation with alpha12 fully extended and colinear with alpha10. By contrast, the second molecule in each homodimer was in a closed conformation with alpha12 folded against the body of the domain in what is widely considered to be the active state. This indicates that although ligand binding is necessary, it is not sufficient to induce an activating structural transition in HNF-4alpha as is commonly suggested to occur for nuclear receptors. To further assess potential mechanisms of activation, we have solved a structure of human HNF-4alpha bound to both fatty acid ligand and a coactivator sequence derived from SRC-1. The mode of coactivator binding is similar to that observed for other nuclear receptors, and in this case, all of the molecules adopt the closed active conformation. We conclude that for HNF-4alpha, coactivator rather than ligand binding locks the active conformation.

In addition to suggesting that fatty acids are endogenous ligands, our recent crystal structure of HNF-4␣ showed an unusual degree of structural flexibility in the AF-2 domain (helix ␣12). Although every molecule contained a fatty acid within its ligand binding domain, one molecule in each homodimer was in an open inactive conformation with ␣12 fully extended and colinear with ␣10. By contrast, the second molecule in each homodimer was in a closed conformation with ␣12 folded against the body of the domain in what is widely considered to be the active state. This indicates that although ligand binding is necessary, it is not sufficient to induce an activating structural transition in HNF-4␣ as is commonly suggested to occur for nuclear receptors. To further assess potential mechanisms of activation, we have solved a structure of human HNF-4␣ bound to both fatty acid ligand and a coactivator sequence derived from SRC-1. The mode of coactivator binding is similar to that observed for other nuclear receptors, and in this case, all of the molecules adopt the closed active conformation. We conclude that for HNF-4␣, coactivator rather than ligand binding locks the active conformation.
As a member of the nuclear receptor family of transcription factors, HNF-4␣ can be separated into DNA and ligand binding domains (LBD). 1 The HNF-4␣ LBD (residues 140 -368) contains distinctive features such as a hydrophobic ligand binding pocket, a dimerization interface, and an AF-2 transactivation domain (residues 360 -368) (1). HNF-4␣ expression is highest in the liver, kidney, and the small and large intestines and lower but detectable in pancreatic ␤ cells and the stomach (2). HNF-4␣ plays critical roles in nutrient transport and metabolism. Its target genes include PEPCK, a major enzyme in the gluconeogenic pathway (3), as well as Apo-AI, Apo-II, Apo-IV, Apo-B, Apo-CII, Apo-III, and Apo-E, which are involved in lipid transport and metabolism (4 -6). Targeted deletion studies indicate that HNF-4␣ is necessary for normal embryonic development (7,8).
Full activity is achieved through the interaction of HNF-4␣ homodimers with DNA and coactivators, including SRC-1, PGC-1, and GRIP-1. These coactivators contain nuclear receptor (NR) boxes comprised of LXXLL motifs. The sequence on either side of the NR boxes imparts specificity to binding that allows the same coactivator to have different affinities for different nuclear receptors. The ability of NRs to form various coactivator complexes provides a mechanism for activating discrete transcriptional processes (9 -11). HNF-4␣ has also been linked to nutrient metabolism through interactions with the coactivator PGC-1, potentially providing a driving force for gluconeogenesis (12).
Mutations in hnf-4␣ are associated with a rare monogenic form of diabetes referred to as MODY1 (maturity-onset diabetes of the young). MODY1 patients are characterized by an insulin secretion defect in the face of normal insulin sensitivity (13). Recent genetic studies have linked polymorphisms in or near the islet-specific P2 promoter of hnf-4␣ to type 2 diabetes (14,15), suggesting that reduced HNF-4␣ activity in pancreatic ␤ cells might also be associated with and possibly predispose to the development of this prevalent disorder.
Nuclear receptor LBDs are found in two distinct conformations. The difference between forms is in the position of helix ␣12, often referred to as the AF-2 domain. In the first conformation, ␣12 is extended colinearly with helices ␣10 and ␣11, leaving the ligand binding pocket accessible and the LBD in what is referred to as the "open" conformation. Because this structure does not allow the binding of coactivator molecules, it is also considered to be the inactive form. In the second conformation, helix ␣12 folds back onto the LBD to seal off the ligand binding pocket. Because this "closed" form interacts with coactivators, it is considered to be the active state of the NR. Structural comparisons between unliganded (apodomain) and ligand-bound and coactivator-bound forms of various NRs have led to the supposition that it is ligand binding that activates NRs by converting molecules in the open inactive conformation to an all closed active state (16,17).
Our previously reported crystal structure of HNF-4␣ challenges this hypothesis (1). One major finding from both the HNF-4␣ and HNF-4␥ structures was the serendipitous presence of fatty acids in their LBDs, suggesting that these might be endogenous ligands (1,18). However, only half of the fatty acid-bound molecules in the HNF-4␣ structure adopted the closed conformation expected for the activated state. The second molecule in each homodimer was in a conformation expected for the inactive state with helix ␣12 fully extended and colinear with helix ␣10. This finding suggested that fatty acid binding might be necessary but not sufficient for activating HNF-4␣. To better understand the structural mechanism for ligand-and coactivator-mediated activation of HNF-4␣, we have crystallized its LBD bound to both fatty acid ligand and a coactivator peptide sequence derived from SRC-1.

MATERIALS AND METHODS
Protein Production and Crystallization-The protein boundaries for expression of the human HNF-4␣ LBD were defined by our previous structure of the rat apodomain (1) (the LBDs of rat and human HNF-4␣ are 97% identical). DNA encoding the LBD of human HNF-4␣ (residues 140 -382) was subcloned into the pET 28a vector (Novagen) by PCR. Protein was expressed in Escherichia coli BL21(DE3) (Invitrogen) cells and isolated from lysates using Talon cobalt affinity resin (Clontech). The His 6 affinity tag was cleaved using bovine thrombin (10 units/ml), and the protein was further purified by ion-exchange chromatography (Mono Q fast protein liquid chromatography). A 3-fold molar excess of SRC-1 peptide (SSLTERHKILHRLLQEGSPS, residues 681-700) was incubated with the protein for 1 h at 4°C prior to concentration (15 mg/ml). Crystals were obtained at room temperature by the vapor diffusion method in 20-l drops containing equal volumes of protein/ peptide complex and crystallization buffer (0.1 M Hepes, pH 7.0, 0.7 M sodium/potassium tartrate and 0.01 M dithiothreitol). Crystals belonging to the space group P4 1 2 1 2 reached maximum dimensions of 0.3 ϫ 0.2 ϫ 0.2 mm within 5 days.
Data Collection and Structure Determination-Diffraction data were collected at beamline X12C of the National Synchrotron Light Source (Brookhaven, NY). Oscillation images (every 1°) were collected at 100 K, and the data were processed using the DENZO HKL software package (19). The structure was determined by molecular replacement using the program MOLREP (20) and our previous structure of HNF-4␣ in complex with fatty acid as a search model (1). Each asymmetric unit contains one ternary complex comprising the LBD of human HNF-4␣, one fatty acid, and one SRC-1 peptide. The initial R value was 0.48 with a correlation coefficient of 0.57. The subsequent -weighted 2F o ϪF c map after rigid body refinement clearly revealed density corresponding to the bound SRC-1 peptide that was not present in the search model. The model was refined using an amplitude-based maximum-likelihood target in the program CNS (21), applying bulk solvent corrections and individual B-factor corrections at the final stage. The data and refinement statistics are summarized in Table I. Graphics were generated using the program O (22).

RESULTS AND DISCUSSION
Domain Architecture-The LBD of human HNF-4␣ is dimeric in the crystals with one protein molecule in each asym-metric unit. It adopts the canonical nuclear receptor LBD fold containing nine ␣ helices and two ␤ strands (Fig. 1A). The helices were numbered according to conventional NR nomenclature as we had previously numbered the helices in the rodent protein (1). Helix ␣2, which is variably present in NR LBDs, is not present in HNF-4␣. ␣6 consists of only one helical turn, and because ␣10 and ␣11 are contiguous, this helix is referred to simply as ␣10. The nine ␣ helices are organized into three layers within the helical sandwich. Helices ␣4, ␣5, ␣8, and ␣9 form the central layer. Helices ␣1 and ␣3 comprise the outer layer at one side of the central layer, and helix ␣10 forms the opposite outer layer. Helix ␣12 is separate from ␣10, consistent with the HNF-4␣ LBD being in the active state and bound to coactivator. The dimerization interface, comprising residues from ␣9 and ␣10, lies along a plane of 2-fold crystallographic symmetry. At 2.1-Å resolution, the amino acid side chains are well defined in the electron density map with the exception of side chains from residues of the ␣1-␣3 loop.
There are two obvious differences between the structures of the binary complex of the FFA-bound rat domain (Fig. 1B) reported earlier (1) and the ternary complex of the human domain reported here (Fig. 1A). The first is the presence of the SRC-1 peptide in one structure and not the other, which provides the opportunity to determine the effects of coactivator binding both on global structure and more specifically on the orientation of ␣12 (the AF-2 domain). The second difference, linked to the first, addresses the issue of open and closed configurations and what they mean. The LBD adopts two distinct conformations in the binary complex, despite both molecules being bound to fatty acid ligand (Fig. 1, B and C) (1). The first molecule is in an open conformation in which ␣10 and ␣12 are contiguous and colinear. This long helix is fully extended. The second molecule of each homodimer adopts a closed conformation with ␣12 situated against the body of the LBD. In the closed state, a hydrophobic patch on ␣12 (Leu-360, Leu-361, Met-364, and Leu-365) affixes it via hydrophobic interactions to the body of the domain. Residues from ␣3 (Met-182, Ala-183, Leu-186, Leu-187, Leu-189, and Val-190) and ␣4 (Leu-211, Ala-215, Gly-216, and Leu-219) form a complementary patch on the body of the domain. Both molecules in the ternary LBD complex (HNF-4␣⅐FFA⅐SRC-1) adopt identical closed conformations (Fig. 1A). This structure is indistinguishable from that of the molecule in the closed conformation of the binary complex (Fig. 1C), yielding an root mean square deviation value of 0.57 Å when all of the C␣ atoms of the two structures are superimposed. We have concluded from these comparisons that ligand (FFA) binding is conformationally permissive, because it facilitates the formation of the activated closed state but does not lock it. Coactivator binding, on the other hand, is restrictive because it apparently locks FFA-bound HNF-4␣ into the activated closed state.
HNF-4␣/Fatty Acid Interactions-X-ray crystal structures of rat HNF-4␣ and human HNF-4␥ indicated that both of these proteins bind fatty acids as natural ligands (1,18). Because fatty acids incorporated spontaneously into the ligand binding pockets of the proteins during bacterial expression and because it was not possible to remove the fatty acids without at least partially denaturing the proteins, we had hypothesized that ligands, presumably fatty acids, might be critical to the stably folded domains (1,18). Our new structure shows that human HNF-4␣ similarly sequesters bacterial fatty acids during protein production. Fatty acids are once again present in all of the molecules of the structure, despite not having been added intentionally.
The ligand binding pocket forms a narrow channel that is lined almost exclusively with side chains of hydrophobic resi- dues (Fig. 2). The side chain of Arg-226 at the base of the binding pocket is the exception. Considerably smaller than the ligand binding pockets of other reported NR LBDs, HNF-4␣ has a cavity area of 370 Å 3 as calculated by the program Voidoo (23). This is the approximate molecular volume of a long chain fatty acid. The fatty acids bound to our structure are anchored in the ligand binding pocket via an interaction between the fatty acid headgroup and the side chain of Arg-226. Both of the oxygen molecules in the fatty acid headgroup interact with FIG. 1. Structures of HNF-4␣. A, ribbon diagram of the ternary complex of HNF-4␣ bound to fatty acid ligand and SRC-1 coactivator peptide. Helices ␣1-␣9 are colored turquoise, ␣10 is yellow, and ␣12 is in red. Bound fatty acid ligand is colored magenta, and the SRC-1 peptide is dark blue. B, ribbon diagram of the binary complex of HNF-4␣ bound to fatty acid ligand. Elements of secondary structure and the fatty acid ligand are colored as in A (1). C, structure so that the HNF-4␣ LBD homodimers are superimposed with the binary HNF-4␣⅐FFA structure colored turquoise and the ternary HNF-4␣⅐FFA⅐SRC-1 structure in magenta. the arginine guanidinium group, and one oxygen interacts as well with the backbone NH group of Gly-237. The ligand interaction is further stabilized by hydrophobic and Van der Waals interactions between the fatty acid carbon chain and residues lining the pocket. Residues that participate in these interactions include Ile-175, Val-178, Cys-179, Met-182, Leu-219, Leu-220, Arg-223, Leu-236, Met-252, Val-255, Ile-259, Met-342, Ile-346, Ile-349, and Ile-357 (Fig. 2).
Because our previous structure was of the binary FFA-bound LBD complex and our new structure is of the ternary HNF-4␣⅐FFA⅐SRC-1 complex, we are in a position to compare ligand binding pockets in the absence and presence of bound coactivator. Improved electron density for the ternary complex allowed visualization of 14 carbons of the fatty acid, as opposed to the 12 carbons seen in the binary complex (1). Improved electron density similarly facilitated the identification of additional water molecules in the ligand binding pocket of the ternary complex. Otherwise, there appear to be no significant differences (Fig. 2). As noted previously, backbone C␣ atoms are superimposable. Amino acid side chains that line the pockets are in similar positions in the two structures and maintain the same interactions, including the salt bridge between the side chain of Arg-226 and the fatty acid head. The trivial differences in fatty acid binding between the two structures can thus be accounted for by the enhanced resolution of the ternary structure, solved at a 2.1-Å resolution as opposed to a 2.8-Å resolution.
HNF-4␣/SRC-1 Interactions-As a recurring structural theme for nuclear receptors, bound coactivators form ␣-helices with hydrophobic side chains of the LXXLL motif directed inward to interact with the LBDs (24). The HNF-4␣⅐FFA⅐SRC-1 complex recapitulates this mode of interaction (Fig. 3). Of 20 residues in the crystallized SRC-1 peptide, 14 at the carboxyl terminus are well defined. The amino-terminal six residues are not visualized in the electron density map. Eight residues (ILHRLLQEG) encompassing the LXXLL motif form Ͼ2 full turns of an ␣-helix. The HNF-4␣/SRC-1 interaction is dominated by hydrophobic interactions involving the leucine residues of the LXXLL motif and the canonical "charge clamp" created by hydrogen bonds between backbone atoms and the side chains of invariant residues Glu-363 of ␣12 and Lys-194 of ␣3. The interaction with Gly-363 requires that ␣12 be folded into place in the ligand binding pocket and thus mandates that coactivator binding occur only with the closed LBD conformation.
The hydrophobic face of the LXXLL helix is packed into a hydrophobic pocket created by residues in ␣3 (Leu-187, , ␣3-␣4 loop (Phe-199), ␣4 (Leu-211), and ␣12 (Glu-363). The charge clamp orients the LXXLL motif into place, which explains why modes of interactions are well conserved between NRs and coactivators. Subtler interactions are thought to govern specificity and relative affinity of a given NR for different coactivator LXXLL motifs. These may be mediated by flanking residues at the carboxyl-terminal end of the LXXLL helix (25,26). In our structure, the amino-terminal flanking region of the SRC-1 peptide is disordered, whereas the carboxyl-terminal flanking region is well defined but lacks specific interactions with the LBD.
Helix ␣12 Structural Changes-Because ligand-bound HNF-4␣ adopts both closed (active) and open (inactive) conformations (1), whereas the SRC-1-bound LBD adopts only the closed conformation, ␣12 does not appear to be a ligand-dependent switch that upon ligand binding triggers a structural rearrangement, providing a surface for coactivator binding (27,28). Moreover, ␣12 does not participate in direct interactions with the ligand as is seen in the interactions of other NRs with many synthetic ligands. Therefore, it appears that both ligand and co-activator binding are required to lock HNF-4␣ in the closed and active state.
Comparisons between Unliganded NRs and Binary and Ternary Complexes-The first apodomain structure to be determined was that of RXR (29). Comparisons between the apodomain and structures of the binary ligand/RXR and ternary ligand/coactivator/RXR complexes formed the basis for developing a global theory regarding structural determinants in NR activation. The theory predicted that apodomains are open and that ligand binding induces a conformational transition that stabilizes a closed active state with ␣12 (AF-2) folded against the body of the domain to form a coactivator binding surface (16,17). The theory further predicted that although coactivator binding might further stabilize the closed conformation, it would not produce additional conformational changes. Because HNF-4␣ does not fit this scheme, we have re-examined the many structures that have now been solved. Of the four published apodomain structures (29 -32), RXR is the only one with ␣12 in the open conformation (Fig 4A). This indicates that, in the absence of ligand, ␣12 is conformationally flexible and can adopt both open and closed configurations. The predominance of the closed form seen in structural studies, in addition to data from thermal melt studies, suggests that the closed form of the apodomain may be energetically favored (33).
Three structures of binary complexes of LBDs bound to naturally occurring ligands are shown in Fig. 4B. RXR␣ is bound to 9-cis-retinoic acid (PDB 1FBY) (16), retinoid acid-related orphan receptor ␣ is bound to cholesterol (PDB 1N83) (34), and the androgen receptor is bound to dihydrotestosterone (PDB 1I37) (35). LBDs in previously reported structures of binary ligand-bound complexes adopt the active closed conformation, supporting the theory that ligand binding serves as the activating switch. However, the finding that binary HNF-4␣⅐FFA complexes adopt both open and closed conformations in our recently solved structure runs counter to the theory. In reexamining the structural basis for activation, found that many endogenous ligands in published structures do not interact with ␣12 directly (16,34,35). We conclude that although natural ligands undoubtedly stabilize the closed configuration, they may not lock the LBD into the active form. By contrast, ligands with larger appendages that do interact with ␣12 (36 -39) may well lock the LBD into an activated conformation.
Structures of ternary ligand/coactivator/LBD complexes are profiled in Fig. 4C . The LBDs in each of these structures and in the ternary HNF-4␣ structure presented in this paper all adopt the active closed conformation. The coactivator charge clamp located on either side of the LXXLL motif interacts directly with ␣12 in each of these structures. Thus, coactivator binding requires that the LBD be closed and, because the LBD must remain closed as long as the coactivator is bound, this essentially "locks" the LBD in an active conformation. We conclude for HNF-4␣ that although ligand binding and coactivator binding both stabilize the LBD, it is coactivator binding that locks it in the active state. As far as we can tell from available structures of other LBDs bound to endogenous ligands and coactivators, the same holds, i.e. the binding of natural ligands and coactivators stabilize the LBD but coactivator binding locks the active state. Synthetic ligands with bulky appendages that interact with ␣12 may have the added ability to lock LBDs in the active conformation. Structural Basis for HNF-4␣ Activation 23315