Structural Insights into the Design of Nonpeptidic Isothiazolidinone-containing Inhibitors of Protein-tyrosine Phosphatase 1B*

Structural analyses of the protein-tyrosine phosphatase 1B (PTP1B) active site and inhibitor complexes have aided in optimization of a peptide inhibitor containing the novel (S)-isothiazolidinone (IZD) phosphonate mimetic. Potency and permeability were simultaneously improved by replacing the polar peptidic backbone of the inhibitor with nonpeptidic moieties. The C-terminal primary amide was replaced with a benzimidazole ring, which hydrogen bonds to the carboxylate of Asp48, and the N terminus of the peptide was replaced with an aryl sulfonamide, which hydrogen bonds to Asp48 and the backbone NH of Arg47 via a water molecule. Although both substituents retain the favorable hydrogen bonding network of the peptide scaffold, their aryl rings interact weakly with the protein. The aryl ring of benzimidazole is partially solvent exposed and only participates in van der Waals interactions with Phe182 of the flap. The aryl ring of aryl sulfonamide adopts an unexpected conformation and only participates in intramolecular π-stacking interactions with the benzimidazole ring. These results explain the flat SAR for substitutions on both rings and the reason why unsubstituted moieties were selected as candidates. Finally, substituents ortho to the IZD heterocycle on the aryl ring of the IZD-phenyl moiety bind in a small narrow site adjacent to the primary phosphate binding pocket. The crystal structure of an o-chloro derivative reveals that chlorine interacts extensively with residues in the small site. The structural insights that have led to the discovery of potent benzimidazole aryl sulfonamide o-substituted derivatives are discussed in detail.

Type II diabetes, also known as non-insulin-dependent diabetes mellitus, is characterized by a deficiency in insulin signaling despite normal or greatly elevated levels of insulin. Although the cause of insulin resistance is unknown, inhibiting enzymes that negatively regulate the signaling pathway may restore insulin responsiveness. Protein-tyrosine phosphatase 1B (PTP1B), 2 a key negative regulator of insulin signaling, has emerged as an attractive target for the treatment of type II diabetes. PTP1B directly inactivates the insulin receptor (IR) by dephosphorylating tyrosine residues in the regulatory domain (1,2) and its overexpression inhibits IR signaling (3). The most convincing evidence for its potential use as a therapeutic target, however, is that PTP1B knock-out mice (4) and those injected with an antisense oligonucleotide (5, 6) displayed increased insulin sensitivity.
A number of PTP1B inhibitors containing highly charged non-hydrolyzable phosphonate mimetics have been reported (7)(8)(9)(10)(11)(12)(13). Poor membrane permeability, however, has limited their development as drug candidates. The lack of inhibitors with suitable physicochemical properties can be attributed to the need to interact with the highly positively charged primary phosphate binding pocket and the surrounding flat, solventexposed region.
We have recently reported the structure-based design of a novel isothiazolidinone (IZD) heterocyclic phosphonate mimetic and its potential use as an inhibitor of PTP1B when incorporated into a peptide (14). Our lead compound 6 has an IC 50 value of 190 nM, it is selective against SHP 1 and 2 (IC 50 Ͼ 100 M), and it is nine times more potent than the equivalent difluoromethylphosphonic acid (DFMP) derivative 11 ( Table  2). The IZD heterocycle is a unique phosphonate mimetic because it interacts extensively with the phosphate binding loop, but only possesses a single delocalized negative charge (15). We had hoped that the latter feature would confer improved membrane permeability compared with the highly charged phosphonate and carboxylate moieties found in most inhibitors. Unfortunately, IZD peptide 6 is not cell permeable. We rationalized that the lack of permeability may be due in part to the polar character of the peptidic scaffold and not solely the IZD heterocycle and that replacing it with less polar moieties might improve permeability. The discovery of nonpeptidic IZD-containing inhibitors with IC 50 values in the low nanomolar range and cellular activity in an IR phosphorylation assay has recently been described (16). We now report the structural insights that have led to the design of these inhibitors.

EXPERIMENTAL PROCEDURES
Protein Expression, Purification, and Biochemical Assay-The protein was expressed and purified as reported previously (15). Briefly, two constructs of the catalytic domain of human PTP1B (435 amino acids) were expressed in Escherichia coli: 1-321 and 1-(His 6 )-298. PTP1B-(1-321) was initially purified in two steps using Q-Sepharose and Phenyl-Sepharose Fast Flow resins (GE Healthcare), and PTP1B 1-(His 6 )-298 was initially purified using Talon metal affinity resin (BD Bioscience). Column fractions containing PTP1B were then injected onto a HiLoad Superdex 75 26/60 Prep Grade column (Amersham Biosciences). Fractions containing the pure protein were concentrated to 10 mg/ml and stored at Ϫ20°C. Enzymatic assays were performed as described in the literature (15,17).
Cryodata collections were performed by transferring crystals stepwise at 2-min intervals into cryoprotectant solutions of mother liquor containing 10 and 20% glycerol and then flash cooling them in a nitrogen stream at Ϫ180°C. Diffraction data were collected with one crystal per dataset on a Rigaku/MSC RAXIS IVϩϩ imaging system mounted on a Rigaku/MSC MicroMax TM -007 rotating anode microfocus generator (CuK ␣ ) operating at 40 kV and 20 mA and equipped with a 0.3-mm cathode, 0.5-and 0.03-mm double collimator, and a Blue Confocal Max-Flux ® optical system. Diffraction intensities were integrated and scaled using CrystalClear (Rigaku/MSC).
Structure Determination and Refinement-The structures of PTP1B 1-(His 6 )-298/1 and PTP1B-(1-321) in complex with 3 and 4 were solved by molecular replacement using the protein coordinates from the Protein Data Bank (PDB) (18) entry code 1EEO as the search model and the program CNX (Accelrys). All rotation and translation searches yielded single consistent solutions for data between 15 and 4 Å. The resulting models were subjected to rigid body refinement followed by simulated annealing (using torsion angle dynamics), positional, and temperature factor refinements using CNX. The structures of PTP1B-(1-321) in complex with 2 and 5 were solved by difference Fourier methods using the relevant starting model. The conformations of the bound inhibitors were unambiguously determined after one or two cycles of refinement. The final models were generated after several cycles of model building and refinement using QUANTA (Accelrys) and CNX, respectively. Only the (S)-IZD stereoisomers of 3, 4, and 5 were observed in the crystal structures even though diastereoisomeric mixtures were used in the crystallization experiments. Solvent molecules were added by visual inspection of electron density maps using X-SOLVATE (Accelrys). Crystallographic data and refinement statistics are listed in Table 1.

RESULTS AND DISCUSSION
Survey of the PTP1B Active Site-A wealth of structural data has been reported ever since the crystal structure of PTP1B was first determined in 1994 (19). One of the most interesting structures is PTP1B in complex with the IR kinase activation segment (20), which revealed that the large, negatively charged substrate interacts with multiple positively charged sites in the protein (A, B, C, and E) (Fig. 1). To help us rank the sites in terms of suitability for drug design, we compared their structural characteristics, biological relevance, and impact on ligand binding affinity.
Structural Characteristics and Biological Relevance of Ligand Binding Sites-The A site is the catalytic pocket of the enzyme where phosphotyrosine (Tyr(P)) residues of the IR kinase activation peptide are dephosphorylated. The importance of interacting with this site is also highlighted by the fact that Tyr(P) accounts for about 53% of the buried surface area of an epidermal growth factor receptor hexapeptide (21). The pocket is ϳ9 Å deep and 10 Å wide and can be divided into two regions (Fig.  2). The upper half of the pocket is located near the surface of the protein and contains hydrophobic residues that interact with the aryl ring of Tyr(P). The lower half is located near the posi- The A site is the positively charged catalytic phosphate binding pocket; the B site is a noncatalytic secondary phosphate binding pocket; the C site is a large flat region that can accommodate large negatively charged substituents; the D site is a small pocket adjacent to the A site; and the E site is a highly solvent exposed region located over the flap (Phe 182 ) of the protein. Arrows indicate that all inhibitors reported in this paper bind in the A site and extend into at least one additional site. Trifluoroacylamide, benzothiazole, and aryl sulfonamide substituents bind in the C site just over the A-C border; ortho-Cl binds in the D site; salicylate binds in the B site; and benzimidazole binds at the A-E border.
tive end of the helix dipole of ␣4 and contains the signature motif common to all members of the PTP family (22). This unique motif is a well defined loop with all of its backbone nitrogens pointing toward the center. This arrangement creates a polar surface complementary to the phosphate anion of Tyr(P) just above the catalytic thiolate of Cys 215 , where phosphonate oxygens hydrogen bond to the backbone nitrogens. The three-dimensional shape and partially solvent-shielded nature of the A site provide the possibility of forming extensive interactions with a small molecule, and interacting with Asp 48 on the outer edge of the pocket has been shown to increase selectivity (23). Furthermore, ligand binding can displace several tightly bound waters that interact with the phosphate binding loop (15). The only disadvantage of building into the A site is the need for inhibitors to possess a highly complimentary polar surface to avoid burying unsolvated polar residues.
The B site, a noncatalytic secondary phosphate binding pocket, is larger (13 ϫ 20 Å), shallower (Ͻ4 Å), and more solvent exposed than the A site (Fig. 2). Although these characteristics are not conducive to binding small nonpolar ligands, the advantage of interacting with this site is the possibility of simultaneously increasing affinity and specificity. For example, phosphorylating Tyr 1163 of the monophosphorylated (Tyr(P) 1162 ) IR kinase activation segment increases binding affinity by 70-fold (20). The lack of a side chain at Gly 259 , located at the center of the pocket, is thought to be the reason why the site can accommodate the bulky phosphate group of Tyr(P) 1163 . Any other residue at this location would sterically hinder phosphate binding. Other desirable design features are its close proximity to the A site, the possibility of displacing tightly bound waters (15), and the opportunity to interact with hydrophobic residues (Val 49 , Phe 52 , Ile 219 , and Met 258 ).
The C site does not appear to have any biological relevance other than binding Thr 1160 and Glu 1159 of the IR kinase activation peptide. The pocket, which shares Tyr 46 and Asp 48 with the A site, is large, completely solvent exposed, and extremely flat except for Arg 47 and Lys 41 (Fig. 2). The lack of binding determinants is consistent with the fact that C site substituents are often disordered. Determining the bound conformations of these substituents is also complicated by the fact that they often participate in crystal packing interactions. For example, the benzyl substituent of compound 1 (Table 1) stacks with the side chain of Glu 167 from a symmetry related molecule. We therefore consider the C site to be less suitable than the A and B sites for structure-based design techniques, which often require accurate predictions of bound conformations of substituents.
The D site does not appear to have a biological function and it is the only site that does not interact with the IR kinase activation peptide. It is a small narrow pocket, partially shielded from solvent, and lined with polar and charged residues: Tyr 46 , Glu 115 , Lys 120 , Asp 181 , and Ser 216 (Fig. 2). The most desirable aspect of building into this site is the possibility of increasing potency with only a small increase in molecular weight. Small substituents (one or two atoms) attached ortho to the phosphonate on the aryl ring of Tyr(P) can interact extensively with D site residues.
Finally, the E site is a flat solvent-exposed region located over the flap and adjacent to the B site. It has no known biological function except for binding Arg 1164 of the IR kinase activation peptide, which appears to stabilize the closed conformation of the flap through -stacking interactions with Phe 182 . Inhibitors that mimic these interactions may benefit from a slow off-rate. Extending beyond the flap, further into the E site, is probably not a good idea because this region, like the C site, contains very few binding determinants.
Preliminary SAR-The K i value for the minimal unit of p-difluorophosphonomethyl phenylalanine (pF 2 PMP), which binds in the A site, is only 2.5 mM compared with 180 nM when it is integrated into the epidermal growth factor receptor peptide (residues 1012-1017) (24,25). This observation highlights the importance of interacting with sites other than the primary phosphate binding pocket. To obtain an initial estimate of the FIGURE 2. Ligand binding sites of the PTP1B active site. The A site is 9-Å deep (from Cys 215 to Phe 182 ) and 10 Å wide (from Tyr 46 to Gln 262 ). Its lower half contains the polar phosphate binding loop (Cys 215 -Arg 221 ) and the catalytic Cys 215 , and its upper half contains hydrophobic residues (Y46, V49, F182, A217, I219, and Q262) that interact with the aryl ring of Tyr(P). Note the site is shielded from solvent when the flap (Asp 181 and Phe 182 ) closes over the pocket upon ligand binding (the inhibitor 4 was omitted for clarity). The B site is larger (13 ϫ 20 Å) and shallower (Ͻ4 Å) than the A site and contains several polar and hydrophobic residues (Wat5 is a water at the center of the site). The C site is highly solvent exposed and completely flat except for Lys 41 and Arg 47 . The D site is a small narrow pocket, shielded from solvent, and accessible from the A site. Note the pocket exists only when the flap is closed.
importance of binding to the A and C sites, we solved the crystal structure of PTP1B/1 and measured the IC 50 values for various DFMP derivatives ( Table 2). The bound conformation of 1 is very similar to that of other reported DFMP structures (Fig. 3). The DFMP moiety adjacent to the C-terminal primary amide anchors the inhibitor in the A site by hydrogen bonding to the phosphate binding loop and side chain of Arg 221 . Based on the IC 50 values of 7 and 8, this proximal DFMP increases potency by 82,000-fold. The C-terminal amide (R2 substituent), which hydrogen bonds to the carboxylate of Asp 48 , increases potency by 27-fold based on a comparison of the IC 50 values of 1 and 9. The rest of the inhibitor extends into the C site where the benzyl at position R3 binds over the C-terminal amide and the distal pF 2 PMP binds between the side chains of Lys 41 and Arg 47 . A comparison of the IC 50 values of 1, 10, and 11 indicates that the C site substituents pF 2 PMP and R 3 -benzyl increase potency by 1000-and 25-fold, respectively. The importance of the distal pF 2 PMP is surprising given the fact that it is fairly disordered (average temperature factor is 40 Å 2 ) and it only interacts with Lys 41 and Arg 47 , which are also disordered and solvent exposed. Based on this preliminary SAR, the A site DFMP is clearly the most important substituent followed by the C site pF 2 PMP, R2-amide, and R3-benzyl. Substituents that bind in the B site can also significantly increase potency. As mentioned above, phosphorylating Tyr 1163 increases the binding affinity of the monophosphorylated IR kinase peptide by 70-fold; and the hydroxyl group of salicylate, which hydrogen bonds to Arg 24 and Arg 254 , increases the potency of an oxalylaminobenzoic acid derivative by 90-fold (26). Ligands that bind to the D site are also responsible for a wide range of increases in potency. For example, bromine and chlorine attached ortho to the phosphonate on the aryl ring of pF 2 PMP increase potency by about 5-and 16-fold, respectively (27), and o-carboxyl attached to a 2-carboxymethoxybenzoic acid derivative increases potency by 130fold (28). Finally, two ligands that bind at the A-E border, the side chain of Arg 1164 of the IR kinase activation peptide (20), and the indole ring of a 2-(oxalyl-amino)-thiophene-3-carboxylic acid derivative (29), interact with Phe 182 of the flap through -stacking interactions; their contributions to binding, however, have not been reported.
Based on the structural analysis and preliminary SAR described above, the A site is clearly the most promising drug binding site, despite its highly polar character, followed by the D, B, C, and E sites. Although interacting with the C site can significantly increase potency, we downgraded its importance due to the lack of binding determinants. Our attempts to improve the cell permeability of our lead IZD-peptide 6 ( Table  2) therefore consisted of eliminating its peptidic nature by replacing the C-terminal amide with a less polar substituent, replacing the N terminus with a small nonpeptidic C site substituent, and building into the D site.
Designing Nonpeptidic IZD-based Inhibitors-Compound 6 was first modified by replacing the C-terminal primary amide with a benzimidazole ring in an attempt to hydrogen bond to the carboxylate of Asp 48 , an important interaction based on our preliminary SAR (Fig. 3). The resulting inhibitor was ϳ10-fold more potent than the primary amide (16). The crystal structure of PTP1B/2 confirms that the benzimidazole ring hydrogen bonds to Asp 48 and participates in van der Waals (VDW) interactions with Phe 182 , at the A-E border (Fig. 4a). The increase in potency is therefore due to favorable VDW interactions as well as an increase in conformational entropy, because the benzimidazole ring can hydrogen bond equally well to Asp 48 in two orientations related by 180°. The structure also explains the lack of SAR in the 5-position and the loss of activity in the 4-position, because 5-substituents simply project into solvent and 4-substituents probably clash with Phe 182 or Asp 48 depending on the orientation of the ring. Unsubstituted or 5-substituted benzimidazoles were therefore subsequently used to identify suitable replacements for the remainder of the peptide scaffold.  Design and SAR of C Site Substituents-Nonpeptidic substituents were designed to replace the entire C site substituent up to the amide NH adjacent to the benzimidazole ring. Retaining the NH group would preserve the second hydrogen bond previously observed between the peptide scaffold and the carboxylate of Asp 48 (Fig. 3). Furthermore, new substituents would project just beyond the A-C border and interact with nonpolar regions of Tyr 46 and Arg 47 . Three nonpeptidic substituents were rapidly discovered: benzothiazoleamine, trifluoroacylamide, and aryl sulfonamide (16). Encouragingly, the derivatives have similar or smaller molecular weights and are nearly equipotent to 6. Optimizing the derivatives for potency, however, proved to be very difficult.
The benzothiazoleamine 2 appeared to be a promising lead with an IC 50 value of 240 nM; however, substitutions to the aryl ring of benzothiazole reduced potencies by 1-5-fold (Table 3). The crystal structure of PTP1B/2 reveals that the amine NH hydrogen bonds to Asp 48 and the benzothiazole ring binds between the side chains of Tyr 46 and Arg 47 (Fig. 4a). Although electron density for the planar benzothiazole ring is fairly symmetric, we believe the sulfur atom faces the protein because of potential VDW interactions with Tyr 46 , Arg 47 , and Asp 48 and . Crystal structures of PTP1B in complex with inhibitors containing nonpeptidic C site substituents. a, PTP1B/2 reveals that the aryl ring of benzothiazoleamine is almost completely solvent exposed; the thiazole ring is in VDW contact with Tyr 46 , Arg 47 , and Asp 48 and its NH hydrogen bonds to water; and the amine NH hydrogen bonds to the carboxylate of Asp 48 . The two small red spheres are water molecules that form a bridge between the benzothiazole and benzimidazole rings. b, PTP1B/3 reveals that the fluorines of trifluoroacylamide hydrogen bond to the backbone NH (via a water molecule) and guanidinium group of Arg 47 and participate in VDW interactions with Tyr 46 , Arg 47 , and Asp 48 ; the carbonyl group points into solvent; and the amide NH hydrogen bonds to Asp 48 . c, PTP1B/4 reveals that the aryl ring of the aryl sulfonamide stacks over the benzimidazole ring and does interact with the protein; and one sulfone oxygen hydrogen bonds to the backbone NH of Arg 47 via a water molecule, the other participates in a two-water bridge to benzimidazole. Finally, note the two hydrogen bonds to Asp 48 in each structure: one from benzimidazole and the other from the NH of the point of attachment for C site substituents; also note that the conformations of the IZD-phenyl template and benzimidazole ring are independent of the C site substituent.

TABLE 3 SAR of C site benzothiazoleamine substituents
because the nitrogen, on the opposite side of the ring, is solvent exposed and hydrogen bonds to water. Interestingly, a twowater bridge connects the benzothiazole and benzimidazole rings, perhaps stabilizing the bound conformation of the inhibitor. The most important observation with regards to the drug design process, however, is that the aryl ring of benzothiazole is almost completely solvent exposed. This observation explains why our attempts to improve potency by substituting positions 4, 5, and 6 were unsuccessful.
The trifluoroacylamide 3 is one of our most potent (IC 50 ϭ 110 nM) C site amide derivatives. This was an unexpected result because the IC 50 value of the methyl derivative 24 was ϳ24-fold less active (Table 4). The structure of PTP1B/3 reveals that the trifluoroacylamide moiety binds just over the A-C border where the trifluoro group is in VDW contact with Tyr 46 , Arg 47 , and Asp 48 (Fig. 4b). One of the fluorines also hydrogen bonds directly to the guanidinium group of Arg 47 and another to the backbone NH of Arg 47 via a water molecule. These interactions not only explain the high potency of the trifluoro derivative but also the loss of activity when fluorines are replaced with nonpolar substituents. Apart from the obvious loss of electrostatic interactions, replacing a single fluorine with methyl (25) or phenyl (26) or replacing the entire trifluoro group with an isopropyl (27) probably causes varying degrees of steric clash with Arg 47 . In the latter case, potency can be partially recovered but only after significantly increasing molecular weight by replacing the isopropyl with phenethyl (28), which probably interacts with the hydrophobic region of Arg 47 .
Aryl sulfonamides are our most potent C site substituents ( Table 5). The initial result that the unsubstituted aryl sulfonamide 36 was 5-fold more potent than the methylsulfonamide derivative (IC 50 ϭ 590 nM) was very encouraging; however, no obvious SAR trends were observed for a variety of substitutions to the aryl ring. Compound potencies were evenly distributed around the IC 50 value of compound 36, with a maximum variation of only 3-fold in each direction. The most potent inhibitor 29 is substituted with 3-trifluoromethyl and 4-bromo and has an IC 50 value of 40 nM, whereas the least potent 42 is substituted with 2-methyl and has an IC 50 of 300 nM. The crystal structure of PTP1B/4 reveals that the aryl ring of aryl sulfonamide does not interact significantly with the protein but instead

Nonpeptidic Isothiazolidinone-based Inhibitors of PTP1B
stacks over the benzimidazole ring (Fig. 4c). This unusual conformation appears to be stabilized by a two-water bridge that links the sulfone to the benzimidazole ring; a similar interaction is also present in the benzothiazoleamine 2 crystal structure (described above ; Fig. 4a). The most important inhibitor-protein interactions are two hydrogen bonds from the sulfonamide and benzimidazole NHs to the carboxylate of Asp 48 and a third between a sulfonamide oxygen and the backbone NH of Arg 47 (via a water molecule). Because these hydrogen bonds are also likely present in the PTP1B/methylsulfonamide derivative complex, we believe the 5-fold increase in potency for the unsubstituted aryl sulfonamide 36 is due primarily to intramolecular -stacking interaction of its aryl ring with the benzimidazole ring. Furthermore, the Ϯ3-fold variation in IC 50 values observed for various substituted aryl sulfonamides is probably a result of the stabilizing and destabilizing effects of the substituents on -stacking interactions. The unexpected conformation of the aryl sulfonamide ring therefore not only explains the increased potency of the series but also the lack of useful SAR because substituents on the aryl ring project into solvent.
As expected, optimizing nonpeptidic C site substituents is not trivial even when the crystal structures of lead compounds are available. Structural data clearly indicates that the benzothiazoleamine, trifluoroacylamide, and aryl sulfonamide substituents are not suitable candidates for optimization, as they are highly solvent exposed. Furthermore, the lack of binding determinants in the C site makes it even more difficult to optimize the substituents because their bound conformations are not easily predicted. The unexpected conformation of the aryl sulfonamide ring is a good example of what can happen when building into the flat, shallow C site, the aryl ring stacks on top of the benzimidazole ring instead of interacting with the protein. Fortunately, the original hits represent a significant reduction in peptidic character and retain key binding interactions. Aryl sulfonamides were selected as the preferred C site substituents because they are the most potent derivatives: the IC 50 values for 2 (benzothiazoleamine), 3 (trifluoroacylamide), and 4 (aryl sulfonamide) are 240, 110, and 59 nM, compared with 190 nM for peptide 6.
Design and SAR of D Site Substituents-Halides and small linear substituents attached ortho to the IZD heterocycle on the aryl ring of the IZD-phenyl template were synthesized first because o-Br had been reported to increase the activity of DFMP derivatives by 5-16-fold (27). The IC 50 value for the unsubstituted IZD benzimidazole aryl sulfonamide 36 is 120 nM; o-CN, -F, -CH 3 , -Cl, -acetylene, and -Br substituents increase potency by 3.9-, 3.5-, 2-, 1.6-, 1-, and 0.7-fold, respectively ( Table 5). The crystal structure of PTP1B in complex with the o-Cl derivative 5 confirms that the chlorine atom binds in the D site, less then 4 Å from Tyr 46 and Ser 216 and just over 4 Å from the side chain of Phe 182 (Fig. 5). Its close proximity to residues in the narrow D site may explain why halides and linear substituents interact more favorably than branched substituents. The structure also reveals that o-Cl is only 3.5 and 3.1 Å from the CH and one of the sulfone oxygens of the heterocycle, respectively. This suggests that ortho halides may have differing effects on the conformational energy of the bound IZD-phenyl template depending on their VDW radii. Small halides would stabilize the observed low energy orthogonal orientation of the template, whereas large halides would clash with the heterocycle and increase the conformational energy of the bound template. The latter may explain why o-F and -CN derivatives are the most potent with IC 50 values of about 30 nM, whereas o-Br is the least active with an IC 50 of 170 nM. Note, however, that the requirement for small unbranched D site substituents is not a general observation for all Tyr(P) mimetics because o-Br and o-carboxylic acid substituents increase the potency of DFMP (27) and 2-carboxymethoxybenzoic acid (28) derivatives by 5-16-and 130-fold, respectively.
Extending IZD-based Inhibitors into the B Site-The crystal structure of PTP1B in complex with the potent (IC 50 ϭ 21 nM) tool compound 5 (Table 1) confirms that the IZD-phenyl template can project substituents into the B site (Fig. 5). The salicylate moiety hydrogen bonds to Arg 24 and Arg 254 and participates in VDW interactions with Met 258 . A future strategy for optimizing potency and reducing the polar surface area of the inhibitor might include rigidifying the linker, increasing interactions with the aryl ring of aryl sulfonamide, and replacing the salicylate with a less polar substituent.
Conclusion-Structural analysis of the PTP1B active site and preliminary SAR have led us to conclude that the most important ligand binding site for drug design is the primary phosphate binding pocket, despite its highly polar character, followed by the D, B, C, and E sites. Based on this information, our design strategy for improving cell permeability of our lead IZDpeptide 6 consisted of eliminating the peptide scaffold in three FIGURE 5. Crystal structure of PTP1B/5. The ortho-Cl substituent of the inhibitor binds in the D site where it participates in VDW interactions with Tyr 46 , Ser 216 , Asp 181 , and Phe 182 and it is only 3.1 and 3.5 Å from one of the sulfone oxygens and the CH of the heterocycle, respectively. The IZD-phenyl template and aryl sulfonamide bind in previously observed conformations, and salicylate binds in the B site where it is in VDW contact with Met 258 and hydrogen bonds to Arg 24 and Arg 254 . The linker that spans the A and B sites is well defined in the electron density even though it does not interact extensively with nonpolar residues. major steps. First, the C-terminal primary amide was replaced with a benzimidazole ring, which hydrogen bonds to the carboxylate of Asp 48 and participates in VDW interactions with the side chain of Phe 182 . Second, the entire peptidic C site substituent was replaced with three shorter nonpeptidic moieties: benzothiazoleamine, trifluoroacylamide, and aryl sulfonamide. Although the three substituents appeared to be promising leads, optimizing for potency proved to be very difficult due to a lack of useful SAR. Crystal structures of PTP1B in complex with the derivatives reveal that the substituents are not suitable candidates for further optimization. The aryl rings of benzothiazoleamine and aryl sulfonamide are highly solvent exposed, and the three fluorine atoms of trifluoroacylamide interact extensively with the protein and thus replacing one of them with a larger moiety probably causes a steric clash that disrupts interactions involving the other two. Fortunately, the nonpeptidic leads are more potent than 6 (except for the benzothiazoleamine derivative) and represent a significant reduction in peptidic character. Aryl sulfonamide derivatives were selected, without further modification, as the preferred C site series simply because they are the most potent compounds. Third, halides and small linear substituents were attached ortho to the IZD heterocycle on the aryl ring of the IZD-phenyl template. Based on the crystal structure of an o-Cl derivative, it appears that o-fluoro and o-cyano are the most potent substituents because they compliment the small, narrow D site and preorganize the IZD-phenyl template for binding. Combining the best substituents from each design step has produced one of our most potent inhibitors (IC 50 ϭ 10 nM): (S)-IZD-phenyl 5-Clbenzimidazole substituted with a C site m-(CF 3 )-aryl sulfonamide and a D site o-fluoro. But more importantly, the D site o-CH 3 derivative 45 (Table 5) has recently been reported to have cellular activity in an IR phosphorylation assay (16). This result provides further evidence that IZD-containing compounds are a promising new class of PTP1B inhibitors.