Structure of Human Cyclophilin A in Complex with the Novel Immunosuppressant Sanglifehrin A at 1.6 Å Resolution*

Sanglifehrin A (SFA) is a novel immunosuppressant isolated from Streptomyces sp. that binds strongly to the human immunophilin cyclophilin A (CypA). SFA exerts its immunosuppressive activity through a mode of action different from that of all other known immunophi-lin-binding substances, namely cyclosporine A (CsA), FK506, and rapamycin. We have determined the crystal structure of human CypA in complex with SFA at 1.6 Å resolution. The high resolution of the structure revealed the absolute configuration at all 17 chiral centers of SFA as well as the details of the CypA/SFA interactions. In particular, it was shown that the 22-membered macrocycle of SFA is deeply embedded in the same binding site as CsA and forms six direct hydrogen bonds with CypA. The effector domain of SFA, on the other hand, has a chemical and three-dimensional structure very different from CsA, already strongly suggesting different immunosuppressive mechanisms. Furthermore, two CypA (cid:1) SFA complexes form a dimer in the crystal as well as in solution as shown by light scattering and size exclusion chromatography experiments. This observation raises the possibility that the dimer of CypA (cid:1) SFA complexes is the molecular species mediating the immunosuppressive effect.

More than 2 decades ago, the discovery of cyclosporine A (CsA) 1 allowed spectacular progress in the field of organ transplantation (1). Since then the search for novel immunosuppressants that interfere with T cell activation/proliferation has resulted in the isolation of several new molecules from natural sources (2). The most prominent examples are FK506 and rapamycin. Besides their important therapeutic use, these drugs have also proven to be powerful tools for dissecting signal transduction pathways at the molecular level (3). It has been shown that the biological activity of CsA, FK506, and rapamycin is mediated by intracellular binding proteins called immunophilins. CsA binds to cyclophilins (e.g. cyclophilin A (CypA)), whereas FK506 and rapamycin bind to FK506-binding proteins (e.g. FK506-binding protein-12). Immunophilin binding is required but not sufficient for the immunosuppressive activity of these drugs. Each complex interacts with a third partner, the effector protein, to achieve the full biological response. Thus, the CypA⅐CsA and FK506-binding protein-12-FK506 complexes inhibit the serine/threonine phosphatase activity of calcineurin, thereby blocking the production of cytokines, including interleukin-2 (4). The FK506-binding protein-12-rapamycin complex inhibits the kinase FK506-binding protein-12/rapamycin-associated protein kinase (also known as RAFT or mTOR (5)) that is involved in interleukin-2 receptor-mediated T cell proliferation (6).
We have screened microbial broth extracts to identify CypA ligands with an immunosuppressive mechanism different from CsA (i.e. searching for the "rapamycin counterpart," which binds to CypA). This search resulted in the isolation from Streptomyces sp. A92-308110 of a new class of compounds named sanglifehrins (7,8). Among the 20 different sanglifehrins isolated so far, sanglifehrin A (SFA) is the most abundant component (Fig. 1A). The affinity of SFA for CypA in a cell-free assay is remarkably high (IC 50 ϭ 6.9 nM) (i.e. approximately 60-fold higher than that of CsA) (IC 50 ϭ 420 nM) (9). Furthermore, SFA displays potent immunosuppressive activity in the murine mixed lymphocyte reaction (IC 50 ϭ 170 nM), an in vitro immune response assay (9). The details of the mechanism by which SFA exerts its immunosuppressive activity at the molecular level are still under intense investigation. The first results describing the effects of SFA on T cells (10 -12) and dendritic cells (13,14) were published in the last few years. These data clearly indicate that SFA acts via a mode of action that is different from that of all other known immunophilinbinding compounds, namely CsA, FK506, and rapamycin.
In addition to its interesting biological profile, SFA has a unique chemical structure. The compound consists of a 22membered macrocycle, bearing in position 23 a nine-carbon tether terminated by a highly substituted spirobicyclic moiety. The macrolide contains an E,E-diene, a short polypropionate fragment, and a tripeptide unit composed of valine and two unusual amino acids, piperazic acid and meta-tyrosine. It is the ␤-nitrogen of piperazic acid that is involved in the amide bond formation, which stands in contrast to all other piperazic acid containing natural products isolated so far. The unique chemical structure of sanglifehrin combined with its immunosuppressive activity generated broad interest. Chemical efforts resulted in the synthesis of several fragments and finally culminated with the total syntheses by Nicolaou et al. (15) and Paquette et al. (16). Finally, degradation and synthetic work on the SFA macrocycle, together with x-ray analysis of the 22membered macrocycle alone bound to CypA, has been published recently (17).
Here we present the crystal structure of the complex between human CypA and SFA. The high resolution of the x-ray data allowed the determination of the absolute configuration of all chiral centers of SFA and revealed the details of the CypA/SFA interactions. Furthermore, the structure of the CypA⅐SFA complex was compared with the structure of the CypA⅐CsA complex, since both immunosuppressants bind to the active site of CypA. Also, the finding that the CypA⅐SFA complex is present as a dimer in the crystal as well as in solution raises the possibility that the immunosuppressive effect of SFA might be mediated through a dimer of CypA⅐SFA complexes.

Protein Preparation, Crystallization, and Data Collection-Human
CypA was purified to homogeneity and SFA was obtained as described (7). The complex CypA⅐SFA was prepared by adding the ligand (100 mM solution in ethanol) in a 3-fold molar excess to the protein solution (12 mg/ml cyclophilin A, 20 mM NaCl, 20 mM Hepes, 0.02% NaN 3 , pH 7.0). Cocrystallization was performed using a standard vapor diffusion hanging drop set-up, with VDX crystallization plates and siliconized microscope coverslips from Hampton Research. Crystallization droplets were made by mixing on the coverslips 1.0 l of the protein solution with 1.0 l of reservoir solution and equilibrated against 700 l of reservoir solution at 20°C. Commercially available screening kits from Hampton Research were used to find preliminary crystallization conditions. In the refined conditions, crystals grow within 3 weeks at 20°C to a size of 0.1 ϫ 0.1 ϫ 0.5 mm 3 with a reservoir of 80 mM magnesium acetate, 50 mM sodium cacodylate, pH 6.5, and 30% (w/v) polyethylene glycol 4000. A crystal was mounted in a glass capillary for data collection at 20°C. X-ray diffraction data (347 frames of 1°rotation) were collected with a mar345 detector at the Swiss-Norwegian beamline of the European Synchrotron Radiation Facility, using a wavelength of ϭ 0.873 Å. Intensity data were processed with DENZO/SCALEPACK (18) and converted to structure factors with the CCP4 suite (19).
Structure Determination, Model Building, and Refinement-The molecular replacement solution was found with X-PLOR, version 3.1 (20), using the protein coordinates of CypA (21) and diffraction data from 8 to 3 Å. For the subsequent refinement steps, all data from 15 to 1.6 Å were used. The two complexes were refined without noncrystallographic symmetry restraints. Alternate cycles of model building with the graphics program O (22) and refinement yielded a model with 256 added water molecules, but no ligands yet, with an R-factor of 23.2%.
The resulting 2F o Ϫ F c and F o Ϫ F c maps were of excellent quality for the missing ligands and allowed the clear determination of the absolute configurations at all 17 chiral centers of SFA. Further refinement with REFMAC version 5.0 (23), water insertion with ARP/wARP (24), and rebuilding steps yielded the final model, consisting of two complexes CypA⅐SFA (including residues 2-165 of CypA) and a total of 361 water molecules, which has an R-factor of 16.3% and R free of 18.5% (no cut-off, 15 to 1.60 Å, working set of 39,452 unique reflections, test set of 2096 reflections).
The quality of the model was assessed with PROCHECK version 3.3 (25) and REFMAC. 86.5% of the amino acids are in the most favored region, and 13.5% are in the additionally allowed region of the Ramachandran plot (no amino acids are in disallowed regions). The overall G-factor is 0.16, with r.m.s. bond lengths of 0.007 Å and r.m.s. bond angles of 1.18°. Buried solvent-accessible areas were calculated with the program AREAIMOL as implemented in the CCP4 suite (19), using a probe radius of 1.4 Å.
Dynamic Light Scattering and Size Exclusion Chromatography-Dynamic light scattering experiments were performed with a DynaPro-801 molecular sizing instrument in conjunction with the software package "Dynamics (version 3.30)," both from Protein Solutions Ltd. The complexes were prepared as 1:1 mixtures of CypA and ligand (both at 56 M), in 100 mM NaCl, 20 mM HEPES, pH 7.4, and measurements were performed at 20°C.
Size exclusion chromatography to detect monomer/dimer formation was done by injecting samples of 40-l total volume containing 29 M CypA and 126 M ligand (or no ligand, as a reference) at pH 7.0 in 20 mM ammonium bicarbonate buffer. The 5 ϫ 100-mm column was packed with BioGel P30 (45-90 m) from Bio-Rad, and the flow rate was 0.06 ml/min. The collected fractions with CypA (detected at 210, 254, and 280 nm) were then subjected to liquid chromatography/mass spectroscopy analysis for verification of complex formation and molecular weight of ligand.

RESULTS
The SFA-binding Site-We have determined the three-dimensional structure of CypA bound to SFA by x-ray crystallography. The 1.6 Å resolution data were phased by molecular replacement using CypA (21) as a search model, yielding a dimer of CypA⅐SFA complexes as solution. The atomic structure includes residues 2-165 of two CypA molecules, two SFA ligands and a total of 361 water molecules. The results of the crystallographic refinement are summarized in Table I. The high resolution of the x-ray data allowed the determination of the absolute configuration of all 17 asymmetric centers of the macrocyclic and spirobicyclic fragments of SFA (Fig. 1, A  and B).
CypA is an enzyme that catalyzes the cis-trans isomerization of prolyl peptide bonds (1). Like CsA, SFA binds to the active site of CypA (Fig. 2). In particular, the piperazic acid (Pip) moiety of SFA occupies the same hydrophobic pocket of CypA that is used by the proline ring of the model substrate Ala-Ala-Pro-Ala (26), by the proline ring of dipeptides (27) and by the side chain of Me-Val 11 for CsA (1, 28) (Fig. 3, A and B). This hydrophobic pocket is formed by the amino acids Phe 60 , Met 61 , Phe 113 , and Leu 122 of CypA. The tripeptide moiety Pip-meta-Tyr-Val of SFA forms six direct hydrogen bond interactions (cut-off 3.2 Å) with four amino acids of the protein (Fig. 3A). Three of these amino acids (Arg 55 , Gln 63 , and His 126 ) make intermolecular hydrogen bonds through their side chains, whereas one amino acid (Asn 102 ) forms an antiparallel ␤-sheet interaction with the ligand through its main chain. In detail, these hydrogen bonds are located between NE2-His 126 and the meta-tyrosine hydroxy group (2.71 Å/2.73 Å), between the main-chain nitrogen of Asn 102 and the main-chain carbonyl oxygen of meta-tyrosine (2.99 Å/2.96 Å), between the mainchain carbonyl oxygen of Asn 102 and the main-chain nitrogen of meta-tyrosine (2.80 Å/2.85 Å), between NE2-Gln 63 and the main-chain carbonyl oxygen of valine (3.03 Å/3.03 Å), between OE1-Gln 63 and the ␣-nitrogen of the piperazic acid (3.06 Å/3.10 Å), and between NH 2 -Arg 55 and the main-chain carbonyl oxygen of valine (3.21 Å/3.19 Å). In the latter list, distances are given between heteroatoms in the two independently refined complexes of the asymmetric unit (the averaged values are displayed in Fig. 3A). In addition, SFA makes two water-mediated hydrogen bonds to the protein, namely between the main-chain carbonyl oxygen of valine and NH1-Arg 55 , NE2-His 54 , and CO-Gly 72 and between the C31 hydroxyl and N-Trp 121 . The latter interaction and/or the side-chain reorientation of Trp 121 enforced by the binding of SFA (see below) could account for the change in Trp 121 fluorescence described previously (29). The 3-oxo-butyl side chain attached at C14 extends on the surface of the complex and does not form important interactions with the protein. The C15 hydroxy group points toward a pocket in the cyclophilin structure occupied by water molecules, whereas the methyl group at C16 is directed toward the interior surface of the macrolide, and the C17 hydroxy group is exposed to the solvent. The E,E-diene region C18 -C22 of the macrocycle is not involved in direct contacts with the protein but forms vdW contacts with the meta-tyrosine of SFA in the dyad-related monomer. The linker region C24-C32 between the macrocycle and the spirobicyclic moiety makes vdW contacts with side-chain atoms from amino acids Ile 57 , Phe 60 , Thr 119 , and Trp 121 . The presence of the linker region of SFA imposes a side-chain reorientation on Trp 121 , compared with unliganded CypA (30). Finally, within the spirobicycle, only the methyl group C45 makes vdW contacts with side-chain atoms from Ile 57 and Phe 60 . Upon complex formation of the monomer, a total of 1232/1225 Å 2 of the solvent-accessible surface from CypA and SFA is buried (547/547 Å 2 from the protein molecules and 685/678 Å 2 from the SFA ligands). In the latter list, values are given for the two independently refined complexes of the asymmetric unit.
The following amino acids have a nonhydrogen atom closer than 4 Å to the ligand SFA in both independently refined monomers: Arg 55 , Ile 57 , Gly 59 , Phe 60 , Met 61 , Gln 63 , Gly 72 , Thr 73 , Ala 101 , Asn 102 , Ala 103 , Gln 111 , Phe 113 , Thr 119 , and His 126 . The average B-factor for the ligand (11.1 Å 2 ) is lower than the average B-factor for the protein (15.4 Å 2 ), consistent with the fact that excellent electron density for all nonhydrogen atoms of SFA is visible. The most flexible regions of SFA, with B-factors Ͼ 15 Å 2 , are the 3-oxo-butyl side chain and the ethyl group attached to the spirobicycle at C40.
Overall, the x-ray crystal structure of SFA bound to CypA confirms that the affinity of the ligand for CypA is primarily mediated by the 22-membered macrocycle, whereas the spirobicyclic unit remains essential for the immunosuppressive activity.

Comparison of the CypA⅐SFA and CypA⅐CsA Complexes: A Case of Induced Fit and Binding with Opposite Peptide Backbone Directions-A least squares superposition of CypA⅐SFA
and CypA⅐CsA shows that the protein backbones are similar in both complexes. The r.m.s. deviation of C␣ positions for CypA in complex with SFA (monomer A in the present structure) and CsA (accession code 1CWA), using residues 2-165, is 0.30 Å. The largest difference in the position of backbone atoms occurs for the 68 -73 loop, with maximal deviations of 1.7 and 1.4 Å for the C␣s of Asn 71 and Gly 72 , respectively. In addition, there are two important side-chain differences in the ligand binding pockets that occur for Arg 55 (Fig. 3, A and B) and Trp 121 . These latter differences are essential for binding the two respective ligands. The side chain of Arg 55 , mediating a hydrogen bond with the carbonyl oxygen of valine (SFA), rotates into a new position in the CypA⅐CsA complex in order to form a hydrogen bond with the carbonyl oxygen of MeLeu 12 (CsA) and to avoid a steric clash with Val 5 (CsA). In return, the latter position of Arg 55 from the Cyp⅐CsA complex would lead to a steric clash with the ester and diene groups of SFA. The side chain of Trp 121 rotates into a new position in the CypA⅐CsA complex in order to form a hydrogen bond with the carbonyl oxygen of MeLeu 9 (CsA). In return, the latter position of Trp 121 from the CypA⅐CsA complex would lead to a steric clash with the linker that extends between the macrocycle and spirobicycle of SFA. These two fundamental side-chain adaptations can be regarded as examples of induced fit in order to bind two chemically very different ligands.
Interestingly, the parts of the backbone moieties of the tripeptides Pip-meta-Tyr-Val (SFA) and MeVal 11 -MeBmt 1 -Abu 2 (CsA) that undergo key interactions with CypA superimpose closely in space, although the peptide directionalities are opposite (Fig. 4). In particular, the carbonyl groups of valine (SFA) and MeBmt 1 (CsA), both mediating the hydrogen bond with NE2-Gln 63 , superimpose. The situation is very similar for the carbonyl groups of meta-Tyr (SFA) and Me-Val 11 (CsA), both mediating the hydrogen bond (weaker in the case of CsA) with the backbone nitrogen N-Asn 102 . Also, the backbone nitrogens of meta-Tyr (SFA) and Abu 2 (CsA) are positioned in such a way that they both can mediate a hydrogen bond with CO-Asn 102 .
Whereas both immunosuppressants SFA and CsA, despite their completely different chemical structures, fit nicely into the active site of CypA, their effector domains are clearly very different, already strongly suggesting that the respective immunosuppressive mechanisms are different. Indeed, a superposition with the ternary complex CypA⅐CsA⅐calcineurin (31,32) shows that CypA⅐SFA cannot interact with calcineurin.

Structure of the Local Dimer of CypA⅐SFA Complexes: The
Dimer Hypothesis-The two CypA⅐SFA complexes are related by an approximate local dyad axis (rotation angle ϭ 176°, calculated from the least squares superposition of protomers A and B using residues 2-165), which is located between the two SFA ligands and oriented almost parallel to the longest dimension of SFA (Fig. 5, A and B). The slight disturbance of the C2 symmetry can be explained by crystal packing (see below). With the exception of the spirobicyclic and ␣-ketobutyrate moieties (which extend toward the outside of the dimeric complex), all of the remaining parts of the SFA ligands are deeply buried in the dimer. Upon dimer formation, a total of 1504 Å 2 of the solvent-accessible surface from the two monomers is buried (721 Å 2 from the protein molecules and 783 Å 2 from the SFA ligands). Interestingly, these values are even larger than the ones for monomer complex formation itself (see above). The two SFA ligands make extensive vdW contacts with each other along the local dyad. In particular, the meta-tyrosine side chain is in close proximity to the E,E-diene fragment of the neighboring SFA molecule, and the C26 -C28 fragments form vdW contacts with each other. In addition, a number of water-mediated hydrogen bonds are formed between the two SFA ligands (e.g. between the C41 carbonyl oxygen and the C31 hydroxy group), between a SFA ligand and the neighboring protein molecule of the dimer (e.g. between the C17 hydroxy group and the Asn 102 side chain), and between the two protein molecules. Finally, a direct hydrogen bond links NE1-TrpA 121 and CO-ArgB 148 (distance 3.1 Å). Wall-eyed stereo view of the superposition between the tripeptide moiety of SFA (cyan, carbon; red, oxygen; blue, nitrogen) and the corresponding region of CsA (magenta, carbon; red, oxygen; blue, nitrogen). The peptide directionality is from left to right for CsA and from right to left for SFA. The main-chain carbonyl oxygens of meta-Tyr and Val for SFA superimpose with the carbonyl oxygens of MeVal 11 and MeBmt 1 for CsA, respectively (cf. black circles). The main-chain nitrogens of meta-Tyr (SFA) and Abu 2 (CsA), although not coincident (cf. black arrows), are both located such that they can accept a hydrogen bond from O-Asn 102 (Fig. 3). The piperazine ring and the valine side chain of SFA superimpose with the CsA side chains of MeVal 11 and Abu 1 , respectively.
The two CypA⅐SFA complexes in the asymmetric unit were refined without noncrystallographic symmetry restraints. They show an r.m.s deviation of 0.26 Å for the C␣ positions of residues 2-165 after least squares superposition. The main side-chain differences in proximity of the SFA binding site occur for Trp 121 , Lys 125 , and Arg 148 . These differences can be explained by differences in crystal contacts: for protomer A, ArgA 148 is involved in a salt bridge with GluBЈ 134 of a neighboring protomer BЈ in the lattice and thus adopts a different side-chain conformation than ArgB 148 . As a consequence, TrpB 121 , which stacks on ArgA 148 , adopts a different side-chain conformation than TrpA 121 . Finally, LysA 125 , which stacks on TrpA 121 , adopts a different side-chain conformation than LysB 125 . Since the side-chain conformations of TrpA 121 and LysA 125 are not influenced by crystal packing, they probably represent the state for the dimer in solution.
The fact that the crystallographically observed CypA⅐SFA dimer is stabilized by numerous interactions raised the possibility that this dimer also exists in solution. Toward this end, monomer/dimer formation was followed by size exclusion chromatography as well as dynamic light scattering. Using the latter method, CypA alone has an apparent molecular mass of 14 kDa, the actual mass being 18 kDa. In contrast, the apparent molecular mass of the CypA⅐SFA complex was 31 kDa, clearly confirming the existence of a dimer in solution. In a further experiment, a SFA derivative was tested that retains a high affinity for CypA but for which dimer formation is predicted to be impaired. The apparent molecular mass of (26S,27S)-dihydroxysanglifehrin A (33) in complex with CypA is 17 kDa, confirming the prediction, based on the crystal structure, that the two additional hydroxy groups would lead to steric clashes with themselves and thus block dimerization (in the intact dimer, the atoms C26 and C27 are at distances of 3.8 and 4.0 Å with their corresponding partners) (Fig. 5B). Also, size exclusion chromatography (followed by liquid chromatography/mass spectroscopy for verification of complex formation and molecular mass of ligand) showed that CypA complexed with the latter derivative displayed a retention time (6.5-7 min), which is very similar to unliganded CypA and CypA⅐CsA and thus corresponds to a monomer. CypA bound to the 22membered SFA macrocycle (17) alone showed the same result, again indicating a monomer. The complex CypA⅐SFA, on the other hand, had a clearly shorter retention time of 5 min, which indicates probable dimer formation.
The spirobicycle linker, which is missing in the case of the 22-membered SFA macrocycle alone, mediates important vdW contacts within the dimer and also induces a side-chain conformation of Trp 121 (Fig. 2), which enables interactions with the other monomer. The latter influence on Trp 121 could be denoted as the "Trp switch," which yields the intersubunit hydrogen bond between NE1-TrpA 121 and O-ArgB 148 . Importantly, both derivatives of SFA (dihydroxy-SFA and the macrocycle) lack immunosuppressive activity in the murine mixed lymphocyte reaction, although their affinity for CypA is unaffected (Table  II). 42-N-methyl-SFA, on the other hand, is an example of a derivative retaining immunosuppressive activity (Table II) and still forming a dimer in solution. The apparent molecular mass of the CypA⅐42-N-methyl SFA complex using the dynamic light scattering method was 26 kDa. Taken together, these results suggest that the ability of SFA derivatives to form a dimer similar to the one observed by crystallography might be related to their immunosuppressive activity as measured in the murine mixed lymphocyte reaction assay. DISCUSSION The present report describes the x-ray structure of the novel immunosuppressive natural product SFA bound to human CypA at 1.6 Å. The x-ray structure not only allowed the determination of the absolute configuration at all chiral centers of SFA but also revealed the details of the binding mode of SFA to CypA. In particular, it showed that the macrocycle of SFA forms six direct hydrogen bonds with CypA. The x-ray structure also revealed that the chemical structure of SFA is nicely suited to enable strong interactions with the active site of CypA (or possibly of CypA homologs possessing similar binding sites). The hydrogen bond to NE2-His 126 , for example, can only be formed because the hydroxyl group on the tyrosine side chain is not in the usual para-position but rather in the meta-position of the aromatic ring. The removal of this hydroxyl group leads to a significant reduction in the affinity for CypA as measured in a cell-free assay, as well as to substantial loss of the immunosuppressive activity (34). Similarly, the hydrogen bond with the side-chain atom OE1-Gln 63 is only possible because the ␤-nitrogen of the piperazic acid, instead of the usual ␣-nitrogen, is involved in the amide bond formation. This feature is unique among all piperazic acid-containing natural products. The observation that two CypA⅐SFA complexes form an intimate dimer in the crystal and in solution has led to the hypothesis that this dimer formation is related to the immunosuppressive effect. Again, many aspects of the chemical structure of SFA are nicely tailored toward promoting dimer formation. For instance, the meta-tyrosine ring forms vdW interactions with the E,E-diene of the other monomer, and the linker to the spirobicycle induces an important modification of the side chain conformation of Trp 121 . The dimer hypothesis states that SFA derivatives that induce a dimeric structure similar to the CypA⅐SFA complex should retain immunosuppressive activity, whereas SFA derivatives that block dimer formation (or form a different dimer) are not immunosuppressive. In summary, this x-ray structure assists in clarifying the mechanism of action of SFA and may help to identify the effector protein mediating the biological effects of SFA. a CypA binding and the MLR were performed as described previously (7). IC 50 values represent mean Ϯ S.D. of 3-5 independent experiments.