Implications for Isoform-selective Inhibitor Design Derived from the Binding Mode of Bulky Isothioureas to the Heme Domain of Endothelial Nitric-oxide Synthase*

Nitric oxide produced by nitric-oxide synthase (NOS) is not only involved in a wide range of physiological functions but also in a variety of pathological conditions. Isoform-selective NOS inhibitors are highly desirable to regulate the NO production of one isoform ben-eficial to normal physiological functions from the uncontrolled NO production of another isoform that ac-companies certain pathological states. Crystal structures of the heme domain of the three NOS isoforms have revealed a very high degree of similarity in the immediate vicinity of the heme active site illustrating the challenge of isoform-selective inhibitor design. Isothioureas are potent NOS inhibitors, and the structures of the endothelial NOS heme domain complexed with isothioureas bearing small S- alkyl substituents have been determined (Li, C.S., Poulos, Bio-chem. In the present communication, the binding mode of larger bisisothioureas complexed to the the of the omit 2 o F c density of all four complex struc- tures running simu-lated in at an initial of 1000 with the of during the

using L-arginine (L-Arg) as the substrate. First, L-Arg is converted to N G -hydroxy-L-arginine, and then in a second step N G -hydroxy-L-arginine is oxidized to L-citrulline and NO (1). The three NOS isoforms (nNOS, iNOS, and eNOS) identified in mammals share more than 50% sequence identity and have identical overall architecture. Each has a heme-and H 4 Bcontaining N-terminal oxygenase domain, a FAD/FMN-containing C-terminal reductase domain, and a CaM-binding motif linking the two functional domains (2). The heme domain provides the site for L-Arg oxidation, while the FAD and FMN of the reductase domain shuttle electrons from NADPH to the heme (3). The cofactor, H 4 B, is necessary for NO production and apparently plays an active redox role during the catalytic cycle (4 -6).
NO contributes to a broad range of physiological functions in neurotransmission, control of platelet adhesion, vascular homeostasis, and cytotoxicity in the immune response (7). Accordingly, the different NOS isoforms are expressed in different tissues and are highly regulated transcriptionally or posttranscriptionally. nNOS and eNOS are expressed constitutively in neurons and endothelial cells among other cell types, but NO production is completely dependent on Ca 2ϩ /CaM binding (8,9). However, iNOS activity is controlled at the transcriptional level and, once expressed, that isoform will produce NO at a high rate. In addition, iNOS is not regulated by CaM, but instead CaM is bound with high affinity and functions as a permanent subunit (10). Unregulated NO production is associated with various pathological conditions (11), e.g. ischemiareperfusion injury in stroke, septic shock, and inflammatory disorder including arthritis. Pathology is due to nNOS and/or iNOS. Under such conditions, it is essential to block nNOS and/or iNOS but not eNOS because this last isoform is critical for maintaining proper vascular tone.
Because of the double-edged nature of NO in both basic physiological functions and various pathological conditions, the development of isoform-selective NOS inhibitors is a highly desirable goal. Considerable effort has been directed toward developing NOS inhibitors (12,13), and a vast majority of the existing NOS inhibitors were developed before any of the NOS crystal structures were available. Most of these inhibitors are L-arginine analogues and compounds bearing amidino or ureido functional groups that can simulate the guanidino group of L-Arg. Some of these inhibitors, indeed, show significant isoform selectivity (14).
The development of NOS inhibitors has entered a new era now that crystal structures of the catalytic heme domain of three NOS isoforms (5,(15)(16)(17) (nNOS) 2 are available. These structures reveal a striking similarity in dimeric quaternary structure as well as in the heme and pterin binding sites. Given such close similarity, structure-based isoform-selective inhibitor design presents an especially challenging problem. A first step toward this goal requires a correlation between what is known about the inhibition of NOS by various inhibitors and the structure of these inhibitors complexed with the various NOS isoforms. We anticipate that a large number of such structures will be required to provide the structural data base required for isoform-selective inhibitor design. We previously reported the structures of the eNOS heme domain complexed with isothioureas containing small alkyl side chains, and those studies provided a structural rationale for inhibition by this class of compounds (5,18). In this communication we report the structures of much larger bisisothioureas bound to the eNOS heme domain.

EXPERIMENTAL PROCEDURES
Materials-1,3-PBITU, 1,4-PBITU and SENPITU were purchased from Alexis. 1,14-BITU was synthesized by reaction of 1,14-diaminotetradecane with thiophosgene followed by anhydrous ammonia. The resulting bis-thiourea was then reacted with methyl iodide to give 1,14-BITU. 3 Crystal Growth, X-ray Data Collection and Processing-The bovine endothelial NOS heme domain protein sample was generated by trypsinolysis from the holo-eNOS expressed in Escherichia coli (5,19). The eNOS-inhibitor complex crystals were grown at 280 K using the sitting drop vapor diffusion method developed for the substrate complex so that the only difference here is to replace L-Arg with mM amounts of isothiourea ligands in the crystallization mixture (5).
The x-ray diffraction data were collected at the Stanford Synchrotron Radiation Laboratory BL7-1 at 100 K where a MarResearch Mar345 imaging plate was equipped. The raw data frames were integrated and scaled with HKL Suite (20). To locate the bound ligands, difference Fourier electron density maps were calculated with CNS (21). Once the model of ligands were built into the structures using TOM/FRODO (22), further structure refinements were carried out with CNS using the maximum likelihood target function. The data collection and refinement statistics are summarized in Table I. To illustrate the ligand binding, omit 2F o ϪF c electron density maps of all four complex structures are shown in Fig. 1. The maps were generated by running simulated annealing protocol in CNS at an initial temperature of 1000 K with the ligand of interest omitted during the calculation.

RESULTS AND DISCUSSION
Binding Mode of Bisisothioureas-Crystal structures of the eNOS heme domain complexed with S-alkyl-isothioureas bearing a simple alkyl group (5,18) show that the thiourea group occupies the same position as the guanidino group of L-arginine with the aulfur occupying the site normally taken by the reactive guanidinium nitrogen of L-Arg (Fig. 2). The high potency of this class of isothiourea inhibitors is primarily attributed to the extensive nonpolar contacts between the small S-alkyl group and an apolar protein pocket adjacent to the substrate binding site (Fig. 2) (18). In a comprehensive structure-activity-relationship (SAR) study on isothiourea inhibitors, Garvey et al. (23) measured competitive binding against substrate using a class of bulky bisisothioureas. Among them 1,3-PBITU and 1,4-PBITU (Fig. 3) showed good potency comparable with the small S-alkyl-isothioureas. To understand how these larger inhibitors bind, we determined crystal structures of endothelial NOS heme domain complexed with 1,3-PBITU and 1,4-PBITU, respectively.
The initial difference electron density maps of both structures revealed the binding of inhibitor at the distal side of the heme plane. However, compared with 1,4-PBITU, the electron density of 1,3-PBITU is weaker indicating lower occupancy or higher thermal motion (Fig. 1). Given that similar concentration of inhibitors were used for crystallization setups of both complexes, the lower occupancy of 1,3-PBITU observed in the structure is consistent with its weaker binding affinity (K i ϭ 9 M) to eNOS than 1,4-PBITU (360 nM) (23). Alternatively, the weaker electron density also could be attributed to higher thermal motion.
As expected from the known binding mode of a single iso- thiourea, the first ureido group of both bisisothioureas still fits into the guanidino site, donating hydrogen bonds to both the Glu-363 carboxylate side chain and the Trp-358 carbonyl oxygen (Fig. 4). However, the remainder of the bisisothiourea molecule does not adopt a favored extended conformation but instead makes an unexpected twist into an "S" shape. The energetic incentive for adopting the twisted conformation appears to be the positioning of the second ureido group close to the heme propionate of pyrrole ring D for H-bonding interactions (Fig. 4).
A comparison of both structures reveals differences in inhibitor conformation as well as inhibitor-protein contacts that provide a structural basis for understanding why 1,4-PBITU exhibits a 20-fold higher inhibition for all three NOS isoforms compared with 1,3-PBITU (23). For example, in 1,4-PBITU the alkyl group of the second ureido group is able to adopt a more extended conformation than in 1,3-PBITU enabling the ureido group to form an H-bond with the heme propionate. The alkyl group in 1,3-PBITU must kink to H-bond with the heme propionate. The second ureido plane is, therefore, flipped almost 180°from 1,4-PBITU to 1,3-PBITU (Fig. 4). The phenyl rings in each inhibitor also are oriented slightly differently. The phenyl ring in 1,4-PBITU complex is approximately perpendicular to the plane of the first ureido group, whereas that in 1,3-PBITU is tilted. The angles between the phenyl ring and first ureido plane in 1,3-PBITU are 64°and 48°for two subunits compared with 77°and 89°in 1,4-PBITU. As a consequence, all six carbon atoms of the phenyl moiety in the 1,4-PBITU complex make good van der Waals contacts with the side chain of Val-338, but in 1,3-PBITU only C-1 and C-2 of the phenyl moiety are close enough to contact Val-338 (Fig. 4).
The observed binding mode of the PBITU inhibitors provides an excellent structural basis for interpreting the SAR studies with other bisisothiourea analogues (23). A good bisisothiourea inhibitor requires that charge-charge interactions be satisfied for both ureido groups as well as concomitant, optimal nonpolar contacts between the bridging functional group of bisisothiourea and Val-338. Therefore, either shorter or longer methylene chains inserted between the ureido and phenyl groups will weaken interactions between the second ureido group and heme propionate. These changes will also shift the phenyl ring away from the proximity of Val-338. The even poorer binding with substituents at the 1,2-positions (23) would result from the crowding of two neighboring substituents and disposition of the phenyl ring relative to Val-338.
Structure of the S-Ethyl-N-phenyl-isothiourea Complex-Although small S-alkyl-isothioureas are potent NOS inhibitors, they exhibit poor isoform selectivity (23). To improve isoform selectivity, Shearer et al. (24) added N-substituents to S-ethylisothiourea, and SENPITU was found to have submicromolar binding affinity for all three isoforms, a good scaffold for SAR studies. To provide a structural basis for potential isoformselective inhibitors built upon this scaffold, we determined the crystal structure at 1.93 Å of eNOS heme domain with SEN-PITU bound.
SENPITU, as with S-ethyl-isothiourea, binds at the active site using both the phenyl-substituted nitrogen and the unsubstituted nitrogen of its thioureido function to H-bond with Glu-363 while making nonpolar contacts to the protein surroundings with its S-ethyl and N-phenyl groups (Fig. 2). Methylation of either of the two ureido nitrogen atoms abolishes the NOS inhibition (24), implying that the H-bond-donating ability of the ureido group is important for binding. The crystal structure reveals that the pocket bordered by the carboxylate of Glu-363 and the carbonyl of Trp-358 is too narrow to accommodate even a monomethyl substitution on the ureido nitrogens if the bifurcate hydrogen bonds between enzyme and ligand are kept in place. This is consistent with the observations based on binding affinity assays of a series of L-Arg analogues. (25) In those cases, the tightly bound L-Arg analogues always place an unsubstituted -NH 2 or ϭNH 2 ϩ in the distal guanidinium nitrogen pocket. This pocket favors the H-bond donating nitrogen over ϭS or ϭO group and exhibits very little affinity for -CH 3 .
The N-phenyl group of SENPITU in the active site is oriented such that the aromatic ring adopts an angle of 153°r elative the ureido plane (Fig. 2). Similar to the situation observed in the PBITU structures, this phenyl ring makes close van der Waals contacts (3.4 Å) with the Val-338 side chain. These nonpolar interactions are quite important since replacing the phenyl ring by a pyridyl group resulted in much poorer binding (24). Interestingly, to avoid steric clashing with the N-phenyl ring, the S-ethyl group in SENPITU complex adopts a different conformation from that observed in the S-ethylisothiourea complex structure. In the SENPITU complex the bridging carbon in the ethyl group makes a tight contact (3.1 Å) with the carbonyl oxygen of Pro-336, whereas in S-ethyl-isothiourea it is the terminal carbon of the ethyl group that is in the proximity of Pro-336 (Fig. 2).
With reasonably tight binding to all three isoforms of NOS, SENPTIU becomes a useful scaffold to further explore isoform selectivity because substituents can be readily added to various positions on the phenyl ring. Although compounds with substituents on all three positions (2-, 3-, and 4-) of the phenyl ring showed similar binding affinity, the 4-position substituents have a much greater effect on isoform selectivity (24). Among a few dozen compounds, which have been characterized, medium-sized apolar substituents at the 4-position of the phenyl ring significantly elevate the K i value toward iNOS by one order of magnitude while leaving K i to eNOS unchanged. The derivative with a 4-trifluoromethyl (4-CF 3 ) substituent exhibits a 115-fold selectivity for nNOS over iNOS. By examining the structure of the respective NOS active sites, it is clear that the 4-position substituent would make direct contact with the heme propionates. Support for this view stems from the much poorer inhibitory potency observed when the 4-position has a carboxylate substituent creating a repulsive clash with the heme propionates. However, exactly how these contacts with

Structure of Isothioureas Complexed to eNOS
heme propionates lead to isoform selectivity will remain difficult to envision until the structures of the selective inhibitor bound to different NOS isoforms become available.
Not surprisingly, the amidino analogue of N-phenyl-isothiourea, N-phenyl-acetamidine (Fig. 3), also proved to be a good scaffold for SAR studies (26). It is slightly different from the N-phenyl-isothiourea in that substituents on the 3-(meta-), rather than 4-(para-), position of the phenyl ring of acetamidine showed better selectivity toward nNOS. Among the functional groups tested, 3-aminomethyl was the best choice because it presumably hydrogen bonds to a heme propionate. Once again, when the amidino methyl group is replaced with a slightly bulkier nonpolar moiety, such as 2-furanyl or 2-thienyl, the potency of the inhibitor improved dramatically (26). This is because the amidino methyl group does not make any direct contact with the protein whereas 2-furanyl or 2-thienyl can most likely protrude into the hydrophobic pocket defined by Pro-336, Val-338, and Phe-355 (Fig. 2) (18). However, the 2-pyridyl group causes a decrease in binding affinity indicating the incompatibility of a more polar group in this hydrophobic protein environment lining one side of the active site.
Extended N-Substituted Bisisothioureas-Crystal structures of both endothelial (5,16) and inducible NOS's (15-17) revealed a deep, wide open access channel that connects the bulk solvent to both the heme and pterin binding sites. In a substrate-bound form, the amino acid moiety of L-Arg is confined within a pocket, which is not part of the substrate access channel but specifically recognizes the L-enantiomer of Arg. As we have seen in the present work, any non-amino acid-based inhibitor that does not have a functional group capable of binding in the chirality-specific pocket must extend out into the space leading toward the open channel.
To test if bulky ligands can still bind to the active site by taking advantage of the wide open channel, we collected data with crystals grown in the presence of a bisisothiourea with a long 14 carbon chain bridging between N-positions of two ureido groups (1,14-BITU, Fig. 3). Unfortunately, although the structure has been refined to 2.0 Å, the electron density for the ligand is only well defined for the anchoring the thioureido group along with four carbon atoms of the linking alkyl chain (Fig. 1) while the rest of the molecule is disordered. Even though the long alkyl chain is not visible, this experiment shows that a bulky ligand with a chain as long as 14 carbons can still reach the active site. As long as the ligand does have a NOS active site anchoring group such as guanidino, amidino, or ureido functionality, it will be able to bind in the active site with the open channel of NOS accommodating the rest of the inhibitor. This information provides insights into possible ways for designing additional classes of isoform-selective inhibitors.
The long chain bisisothioureas with a chain length of 12-14 carbons showed higher inhibitory potency compared with the ones with shorter chain length. 4 This implies that 1,14-BITU or 1,12-BITU may have additional favorable contacts with the protein. Using the first ureido group as a structure-known anchor point, 1,14-BITU with both ureido groups and its 14methylene spacer were modeled into the access channel. The second ureido at the end of alkyl chain extends out along the long access channel. The model indicates that residues Glu-271 and Asp-480 provide a potential site of interaction for the ureido group of long chain bisisothioureas (Fig. 5). Interestingly these residues are on the surface of the eNOS heme domain structure and at the periphery of the entrance to the substrate access channel. Both residues are also in the vicinity of Trp-482, the last residue in the heme domain prior to the calmodulin binding motif. If, in holo-NOS, either calmodulin or a portion of the reductase domain is positioned close to the mouth of the substrate access channel, the long chain bisisothioureas might possibly reach out to interact with new partners. The unique feature of dual ureido groups makes this class of inhibitors competitive not only with the substrate binding but also potentially with the enzyme's electron transfer partner or allosteric effector. In fact, 1,14-BITU, unlike simpler S-alkyl isothioureas, was observed to inhibit electron transfer reactions of the reductase domain (27).
Amino Acid Substitutions along the Substrate Access Channel-Crystal structures of all three NOS heme domains (5, 15-17) (nNOS 2 ) show a high level of amino acid conservation and striking structural similarity in the immediate vicinity of the substrate binding site, which limits the use of direct contacts in the active site as a basis for isoform-selective inhibitor design. The structures we have described here and elsewhere 5 provide some modest insights into the possibility of achieving isoform selectivity by exploiting differences along the substrate access channel. Shown in Table II are amino acid variations that are potentially useful for the inhibitor design. For instance, Ser-248 in eNOS (Ser-477 in nNOS) is replaced by Ala-262 in iNOS. A functional group from the ligand, e.g. amino or carboxylate, that is capable of making a hydrogen bond with the Ser side chain may exhibit better selectivity toward eNOS (or nNOS). Similarly, the substitution of Leu-107 in eNOS (Leu-337 in nNOS) by Thr-121 in iNOS could be utilized in designing an iNOS-specific inhibitor by introducing a hydrogen bonding partner from the ligand to Thr-121.
A high degree of primary sequence variation exists also in the region between the conserved Ser residue (334 in rat nNOS, 104 in bovine eNOS, and 119 in human iNOS), which makes a hydrogen bond with H 4 B, and the first ␣-helix (␣1) in NOS. Unfortunately, this region constitutes a surface loop that is disordered in both the eNOS and nNOS crystal structures. In the human iNOS structure this loop is ordered and lies along the substrate access channel (17). The amino acid substitutions on this loop can, therefore, be utilized in isoform-selective inhibitor design. Lys-123 in iNOS protrudes into the substrate access channel. This residue is replaced by Arg-109 in eNOS and Ser-339 in nNOS and should be reachable by the ligand, assuming the loop in eNOS or nNOS adopts a conformation similar to that observed in iNOS.  acid differences along the access channel may provide a useful approach. BN 80933, a bulky nNOS selective inhibitor described by Chabrier et al. (28), possesses an N-phenyl-thiopheneamidine scaffold (Fig. 3) that anchors at the substrate guanidino binding site. An antioxidant moiety, vitamin E or Trolox, is linked to the scaffold by a piperidine molecule. The bulky antioxidant moiety must be accommodated by the open channel. An attempt to model the molecule in the eNOS heme domain structure reveals a few potential contacts that may explain its isoform selectivity. The carbonyl group between the piperidine and vitamin E moiety could be in the vicinity of Ser-248, the vitamin E ring system being in contact with Leu-107 while the hydroxyl group from vitamin E can reach as far as Asp-480. These three contact points are exactly where the amino acid substitutions occur from the constitutive NOS (eNOS and nNOS) to iNOS (Table II), which could be the reason for the selection against iNOS. Selectivity for nNOS over eNOS presumably dependents on other, unfavorable interactions with eNOS not present in nNOS. Another similar example is the nNOS-specific inhibitor, ARL17477 (Fig. 3) (29) and its analogues (30), which have an identical NOS active site binding scaffold as that in BN 80933 but with a chlorophenyl group attached to it through a small NH-containing alkyl chain. Again, the bulky functional group could provide isoform selection by interacting with amino acids along the substrate access channel. Similarly, a series of Nnitroarginine-containing dipeptide amide compounds developed in Silverman's laboratory showed impressive isoform selectivity (31). Considering the length of the dipeptides, the second residue should be able to extend into the substrate access channel. The nitroarginine moiety provides good binding affinity to the NOS active site while the second residue imposes the selectivity.

Reported Bulky NOS Inhibitors That Lend Support to This
Conclusion-The high degree of structural similarity in the active sites among the three NOS isoforms provides the rationale for the lack of isoform selectivity among simple amino acid-based or other small isothiourea-or amidine-based inhibitors. However, bulky inhibitors often do exhibit better selectivity. Based on our current state of structural information, we can conclude that a good isoform-selective inhibitor consists of three components. The first is a structural scaffold that provides a guanidino, amidino, or ureido group that donates hydrogen bonds to the glutamate located in the NOS active site. A small hydrophobic group such as alkyl or thienyl is often part of the scaffold to provide additional non-polar interaction with the protein surroundings opposite to the location of the glutamate residue involved in H-bonds to guanidino of thioureido nitrogen atoms. The second is an isoform-selectivity conferring functional group bearing hydrogen bonding capability that can reach into the substrate access channel remote from the active site. The functional group imposes isoform selection by taking advantage of amino acid substitutions along the channel. The third is a linker serving as a spacer between the scaffold and the functional group. The linker should have the appropriate length and flexibility to allow the functional group to reach hot spots along the channel for isoform-specific interactions.