The Structure of a Mutant Insulin Uncouples Receptor Binding from Protein Allostery

The zinc insulin hexamer undergoes allosteric reorganization among three conformational states, designated T6, \batchmode \documentclass[fleqn,10pt,legalpaper]{article} \usepackage{amssymb} \usepackage{amsfonts} \usepackage{amsmath} \pagestyle{empty} \begin{document} \(\mathrm{T}_{3}\mathrm{R}_{3}^{\mathrm{f}}\) \end{document}, and R6. Although the free monomer in solution (the active species) resembles the classical T-state, an R-like conformational change is proposed to occur upon receptor binding. Here, we distinguish between the conformational requirements of receptor binding and the crystallographic TR transition by design of an active variant refractory to such reorganization. Our strategy exploits the contrasting environments of HisB5 in wild-type structures: on the T6 surface but within an intersubunit crevice in R-containing hexamers. The TR transition is associated with a marked reduction in HisB5 pKa, in turn predicting that a positive charge at this site would destabilize the R-specific crevice. Remarkably, substitution of HisB5 (conserved among eutherian mammals) by Arg (occasionally observed among other vertebrates) blocks the TR transition, as probed in solution by optical spectroscopy. Similarly, crystallization of ArgB5-insulin in the presence of phenol (ordinarily a potent inducer of the TR transition) yields T6 hexamers rather than R6 as obtained in control studies of wild-type insulin. The variant structure, determined at a resolution of 1.3Å, closely resembles the wild-type T6 hexamer. Whereas ArgB5 is exposed on the protein surface, its side chain participates in a solvent-stabilized network of contacts similar to those involving HisB5 in wild-type T-states. The substantial receptor-binding activity of ArgB5-insulin (40% relative to wild type) demonstrates that the function of an insulin monomer can be uncoupled from its allosteric reorganization within zinc-stabilized hexamers.

Insulin is a small globular protein containing two chains, A (21 residues) and B (30 residues). In pancreatic ␤-cells, the hormone is stored as Zn 2ϩ -stabilized hexamers, which form microcrystalline arrays within specialized secretory granules (1). The hexamers dissociate upon secretion into the portal circulation, enabling the hormone to function as a zinc-free monomer. Structure-function relationships have been inferred from patterns of sequence conservation (2) and extensively probed by mutagenesis (2)(3)(4)(5)(6)(7)(8)(9)(10). A variety of evidence suggests that insulin undergoes a change in conformation on binding to the insulin receptor (IR) 2 (11). A model for induced fit is provided by the TR transition, a long range allosteric reorganization of zinc insulin hexamers (12). The structural basis of the TR transition has been extensively investigated by x-ray crystallography (13)(14)(15). In this paper, we investigate the relationship between such allosteric reorganization and biological activity. Experimental design exploits the contrasting structures of classical hexamers to introduce an electrostatic block to the TR transition.
The TR transition encompasses three families of zinc insulin hexamers, designated T 6 , T 3 R 3 f , and R 6 ( Fig. 1). Interconversion among these families is regulated by ionic strength (13,16) and the binding of small cyclic alcohols (14). NMR studies have established that the structure of an insulin monomer in solution resembles the crystallographic T-state (Fig. 2, left) (17)(18)(19). The TR transition is characterized by a change in the secondary structure of the B-chain (Fig. 2, right). 3 Although extensive contacts stabilize the hexamer, within an individual protomer the TR transition is associated with a loss of interchain contacts. Such R-state features have been observed only in zinc hexamers; monomeric R-like conformers have not been detected in solution by NMR (18 -20). Dodson and colleagues (14) have nonetheless proposed that an insulin monomer may adopt an R-like conformation upon receptor binding. Such induced fit could extend the potential receptor-binding surface of insulin by exposing side chains otherwise inaccessible in the T-state. Indirect evidence favoring this hypothesis has been obtained by nonstandard mutagenesis (21,22). D-Amino acid substitutions at Gly B8 (black circles in Fig. 2) markedly impair receptor binding in association with a shift in the conformational equilibrium * This work was supported, in whole or in part, by National Institutes of Health among T 6 , T 3 R 3 f , and R 6 hexamers favoring the T-state. The low biological activity of such analogs (reduced by 10 2 to 10 3 -fold) (21) has been ascribed to thermodynamic stabilization of a native-like but inactive T-state conformation (22).
The present study focuses on the role of residue B5, conserved as His among eutherian mammals. Previous studies have shown that His B5 contributes to the foldability of human proinsulin but is not required for biological activity (22). Experimental design exploits the contrasting environments of His B5 in wild-type structures: on the T 6 surface but within an intersubunit crevice in R-containing hexamers. In the T-state, His B5 lies within the N-terminal extended strand (Fig. 2, left, filled red circle). In the T 6 hexamer (Fig. 3A, red side chains), His B5 is surrounded in part by water molecules (blue spheres); the imidazole ring packs against an interchain crevice near Ile A10 and the solvent-exposed A7-B7 disulfide bridge (shown in stereo view in Fig. 3C). 4 Participation of His B5 in the R-specific ␣-helix ( Fig. 2, right, red circle) causes its side chain to move from the T 6 surface to pack within a trimer interface; the B5 side chain does not participate in zinc coordination or the immediate binding of phenol. The structural environment of His B5 within an R-specific trimer interface is shown in Fig. 4 (see also supplemental Figs. S1 and S2).
Our study has two parts. We first describe pH-dependent 1 H NMR studies of the wild-type R 6 hexamer. R-specific burial of His B5 is shown to be associated with a marked reduction in its side chain pK a . These results in turn predict that a positive charge at B5 would destabilize the associated R-specific trimer interface. To test this prediction, a human insulin analog was prepared in which His B5 was substituted by Arg, a variant observed in some hystricomorph mammals, fish, and reptiles (23). 5 Although the variant hormone retains substantial activity, Arg B5 -insulin is unable to undergo the TR transition in solution. Similarly, crystals of Arg B5 -insulin, grown under conditions leading to formation of wild-type R 6 hexamers, contain only T 6 hexamers. The variant structure, determined by molecular replacement at a resolution of 1.3 Å, closely resembles the wild-type T 6 hexamer. The side chains of His B5 and Arg B5 , despite their differences in size and shape, pack within corresponding interchain crevices and participate in part in analogous networks of van der Waals interactions and hydrogen bonds. Despite such similarities, Arg B5insulin is less stable than the wild-type protein, as probed by chemical denaturation studies, presumably due either to loss of weakly polar interactions associated with the imidazole ring (24) or to less efficient packing of the linear Arg B5 side chain within the crevice.
Together, the properties of Arg B5 -insulin uncouple the mechanism of receptor binding from the choreography of conformational changes in the classical TR transition. Impairment of hexamer reorganization by introduction of a charged side chain, reminiscent of classical mutations in hemoglobin that impair its cooperativity (25)(26)(27)(28), demonstrates the utility of "electrostatic engineering" in studies of protein allostery.

EXPERIMENTAL PROCEDURES
Synthesis of Insulin Analogs-Human insulin and Asp B28 -insulin were obtained from Lilly and Novo-Nordisk (Bagsvaerd, Denmark), respectively. Synthesis of Arg B5 -insulin was performed as described (9,10); see also the supplemental materials.
Receptor Binding Assays-The human IR was expressed and purified as described (29). Competitive IR binding assays were performed by a microtiter plate antibody capture assay. Microtiter strip plates (Nunc Maxisorb) were incubated overnight at 4°C with anti-FLAG IgG (100 ml/well of a 40 g/ml solution in phosphate-buffered saline). Washing, blocking, receptor binding, and competitive binding assays with labeled and unlabeled peptides were performed as described (29). Binding data were analyzed by a two-site sequential model with homologous or heterologous labeled and unlabeled ligands to obtain dissociation constants. The percentage of tracer bound in the absence of competing ligand was Ͻ15% to avoid ligand depletion artifacts.
Visible Absorption Spectroscopy-To probe the TR transition of Co 2ϩ -substituted insulin hexamers, the d-d optical absorption bands of Co 2ϩ (a characteristic feature of a tetrahedral complex) (30,31) were monitored as described (32). Solutions contained 0.2 mM insulin or insulin analog in a buffer consisting of 0.07 mM CoCl 2 and 50 mM phenol in 50 mM Tris-HCl (pH 8). Absorption spectra were obtained in the presence of 0.8 M NaSCN; the thiocyanate anion contributes a fourth ligand (in addition to three symmetry-related His B10 side chains) to the coordination of each axial metal ion and so enhances the d-d band intensity (30,31). 1 H NMR pH Titrations-Spectra were obtained at 700 MHz with a high sensitivity cryogenic probe (Bruker Biospin, Inc., Billerica, MA). Free histidine (Sigma-Aldrich, St. Louis, MO), human insulin (calculated pI 5.4), or Asp B28 -insulin (calculated pI 4.9) were first dissolved in 500 l of 99.0% D 2 O containing 10 mM deuterated Tris-HCl (pH* 8.5, direct meter reading; Isotec, Miamisburg, OH) at respective sample concentrations of 10 mM, 1 mM, and 1 mM. After ϳ12 h of hydrogen-deuterium amide proton exchange at room temperature, samples were lyophilized and redissolved in 500 l of 99.9% D 2 O containing 50 mM deuterated phenol (Sigma-Aldrich); 0.33 mM ZnCl 2 was then added to the protein solutions to form phenol-stabilized R 6 hexamers. The apparent pD of the samples (designated pH*; uncorrected for isotope effects) was in each case measured with a microelectrode (Ingold 6030; Toledo-Mettler, Columbus, OH) at 40°C and adjusted to pH* 8.3 with DCl or NaOD. 1 H NMR spectra were acquired at 40°C; chemical shift values are shown relative to 2,2-dimethyl-2-silapentane-5-sulfonic acid (DSS). Successive pH* values were obtained by adding an aliquot of dilute DCl; serial spectra were obtained at each pH* tested between pH* 8.5 and 2.0. Titration curves were fitted to a modified Henderson-Hasselbalch equation using Kaleida-Graph, Synergy Software (Reading, PA) by nonlinear least squares analysis as follows, in which ␦ obs is the chemical shift observed at each pH* value, and ␦ AHϩ and ␦ A are the chemical shifts of the protonated and deprotonated histidines, respectively. The relationship, pK a H ϭ 0.929 ϫ pK a H* ϩ 0.42 (Eq. 2) was applied to convert the apparent pK a estimates obtained in D 2 O to corresponding values in H 2 O as described (33). In each panel, the positions of residues B5 and B8 (C a atoms) are shown as filled red and black circles, respectively. Dimerization is mediated by an anti-parallel ␤-sheet (central ribbons) and nonpolar interactions between central B-chain ␣-helices. In the T state conformation, the A-chain contains an N-terminal ␣-helix (residues A1-A8) followed by a noncanonical turn, second helix (A12-A18), and C-terminal segment (A19 -A21); the B-chain contains an N-terminal segment (residues B1-B6), type IIЈ ␤-turn (B7-B10), central ␣-helix (B9 -B19), type I ␤-turn (B20 -B23), and C-terminal ␤-strand (B24 -B28), extended by less well ordered terminal residues B29 and B30. In the R-state, the N-terminal portion of the B-chain participates in a single long ␣-helix. The resulting B1-B19 ␣-helix (or B3-B19 in the frayed R f state) projects from the globular core of the protomer to make extensive hexamer contacts, including formation of a specific phenol-binding pocket at a trimer interface (14). The side chain of His B5 packs within an R-state-specific subunit interface in T 3 R 3 f and R 6 hexamers; the side chain neither contacts a zinc ion nor contacts a bound phenolic ligand.
Circular Dichroism-Far-UV CD spectra were obtained as described (34). Spectra, acquired with an Aviv spectropolarimeter (Aviv Biomedical Inc., Lakewood, NJ), were normalized by mean residue ellipticity. Samples were dissolved in 10 mM potassium phosphate (pH 7.4) and 50 mM KCl at a protein concentration of ϳ25 M. ZnCl 2 was added to provide 2.2 zinc ions/insulin hexamer. For equilibrium denaturation studies, samples were diluted in the same buffer to 5 M; guanidine HCl was employed as denaturant (34). Data were obtained at 4°C and fitted by nonlinear least squares to a two-state model (35).
X-ray Crystallography-Crystals were grown by hanging drop vapor diffusion under intended R 6 conditions (see supplemental materials). Data were collected from single crystals mounted in a rayon loop and flash-frozen to 100 K. Reflections from 30.6 to 1.3 Å were measured with a CCD detector system using synchrotron radiation at the Advanced Light Source (Lawrence Berkeley National Laboratory, Berkeley, CA). Data were processed with the program DTREK. Crystals belong to space group R3 with unit cell parameters a ϭ b ϭ 81.39 Å, c ϭ 34.00 Å, ␣ ϭ ␤ ϭ 90, and ␥ ϭ 120°. These dimensions are typical of T 6 crystals under cryoconditions. The structure was determined by molecular replacement using the CNS suite of programs (36). A model was obtained using the native T 2 dimer (Protein Data Bank identifier 4INS) following removal of all water molecules and zinc ions. A translationfunction search was performed using coordinates from the best solution for the rotation function following analysis of data between 15.0 and 4.0 Å resolution. Rigid body refinement using CNS, employing overall anisotropic temperature factors and bulk solvent correction, yielded respective values of 0.31 and 0.30 for R and R free for data between 19.2 and 3.0 Å resolution. Between refinement cycles, 2F o Ϫ F c and F o Ϫ F c maps were calculated using data to 3.0 Å resolution; zinc ions were then built into the structure using the program O (37). The geometry was monitored with PROCHECK (38); zinc ions and water molecules were built into the difference map as the refinement proceeded. Calculation of omit maps (of particular importance in relation to the N-terminal segment of the B-chain) and further refinements were carried out using CNS (36), which implements maximum likelihood torsion angle dynamics and conjugate-gradient refinement. X-ray diffraction and refinement statistics are provided in supplemental Table S1.

RESULTS
1 H NMR Studies of the R 6 Hexamer-We first describe pHdependent 1 H NMR studies of the wild-type R 6 zinc insulin hexamer as a probe of the electrostatic environment of His B5 . Due to isoelectric precipitation of the wild-type hexamer near pH 6, these studies were extended through the use of a variant R 6 hexamer with lower isoelectric point (Asp B28 -insulin) (39,40). Crystallographic studies have established that Asp B28 -insulin (a fast acting analog in clinical use for treatment of diabetes mellitus) forms R 6 hexamers that (with the exception of a local distortion at the dimer interface) are essentially identical to those of wild-type insulin. Human insulin contains two histidine residues (positions B5 and B10). Whereas His B10 mediates the binding of axial zinc ions in each family of insulin hexamer, the environment of His B5 varies among T-and R-states.
Although 1 H NMR spectra of T 6 and T 3 R 3 f hexamers are not tractable, the spectrum of the phenol-stabilized R 6 hexamer exhibits high resolution features and has been well characterized (supplemental Fig. S3) (41). To probe the pK a values of the histidine residues, 1 H NMR spectra were obtained in D 2 O at 40°C as a function of pD; reference spectra of free histidine were also obtained. (A temperature of 40°C was chosen to enhance resonance line widths relative to room temperature with retention of native pattern of chemical shifts (41)). Whereas on the surface of a T 6 hexamer, His B5 would be expected to exhibit a pK a between 6 and 7 (in accord with past NMR studies of engineered T-like monomers and dimers) (18,42), 6 its burial within an R 6 interface predicts a shift of the pK a to lower values. By contrast, because His B10 is bound by zinc, its imidazole ring is unavailable for protonation, so its resonances would not be affected by changes in pH under conditions of stable self-assembly. 1 H NMR spectra of wild-type insulin and Asp B28 -insulin were obtained below pH* 3.8 and above either pH* 5.8 (wild type) or pH* 5.3 (Asp B28 variant). Spectra were not obtained within respective pH* ranges 3.8 -5.8 and 3.8 -5.3 (direct meter reading in D 2 O buffer at 40°C) due to reversible pH-Dependent isoelectric precipitation (39). Although the R 6 hexamer is highly soluble under neutral and basic conditions, the wild-type and variant protein solutions each exhibited a pretransition from clear to hazy near pH* 6.8. Below pH* 4, the solutions reclarified as protonation of His B10 leads to release of the axial zinc ions and disassembly of the hexamers to form soluble zincfree insulin dimers and monomers. pH-dependent aggregation and then disassembly of the zinc insulin hexamer at low pH* precludes conventional assessment of side-chain pK a values by continuous pH* titration. Crystallographic and NMR studies under acidic conditions have nonetheless demonstrated retention of a native-like fold, including in the packing of the His B5 imidazole ring along the surface of the A-chain (43)(44)(45)(46).
Despite these experimental limitations, upper bounds to pK a values were estimated from the apparent pH* dependence of side-chain imidazole C 2 H or C 4 H 1 H NMR resonances under accessible conditions (Fig. 5). Because measurements were obtained in D 2 O at elevated temperature in a buffer with a marked pH-dependent pK a (Tris⅐DCl), care was taken with interpretation of direct pH meter readings (see "Experimental Procedures"). In accord with standard values, spectra of free His (Fig. 5A) yielded essentially identical estimates of its pK a from analysis of either ring resonance (6.10 Ϯ 0.07) on correction for solvent isotope effects (Fig. 5B). 1 H NMR spectra of wild-type and Asp B28 R 6 hexamers exhibit similar but not identical patterns of chemical shift dispersion, as expected from comparison of their crystal structures (47). In such spectra, the C 4 H resonance of His B5 is upfield of its position in zinc-free insulin (and of its random coil value), in each case well resolved in one-dimensional spectra (Fig. 5A, red asterisks). Strikingly, in each case, this resonance exhibits essentially no change in chemical shift at accessible pH* values of Ͼ6. Because a pK a of 6.0 would predict significant changes in C 4 H chemical shift between pH* values 6.5 and 7.0, the actual pK a of His B5 must be lower. Despite the unavailability of chemical shift values between intermediate pH* values due to isoelectric precipitation (see above), upper bounds on the His B5 pK a may be estimated by simulation as follows. (i) The limiting chemical shift of His B5 C 4 H on protonation of the imidazole ring at pH 2.0 is assumed to be similar to that observed in the zinc-free dimer (wild-type insulin) or monomer (Asp B28 -insulin). (ii) At intermediate pH* values, 3.0 and 4.0 the same chemical shift was assumed (Fig. 5B, open circles), forcing the putative pK a to be Ͼ4.0 to enable conservative estimate of an upper bound. The resulting curve fitting suggests that the pK a of His B5 6 NMR studies of an engineered T-like insulin monomer (DKP-insulin) near neutral pH indicate that the pK a of His B5 is near 7.0 (Q. X. Hua and M. A. Weiss, unpublished results) (18). NMR studies of an engineered T 2 -like dimer (Asp B9 -insulin) likewise yield an estimate of the B5 pK a of between 6.8 and 7.0 (42). Because the environment of His B5 on the surface of a T 6 hexamer is similar to that in a monomer or dimer, these measurements suggest that, unlike an R 6 hexamer, a T 6 hexamer can accommodate either charged or neutral side chain at B5.  JULY 25, 2008 • VOLUME 283 • NUMBER 30

Insulin Function and the TR Transition
in the wild-type R 6 hexamer is less than 5.1; this upper bound is tightened to less than pH 4.8 upon analysis of the Asp B28 variant. Although these bounds depend on the above two assumptions, our central observation, that the C 4 H resonance of His B5 in R 6 hexamers does not depend on pH in the range 6 -8, immediately predicts that a positively charged side chain at B5 would be unfavorable in this conformational state. Characterization of Arg B5 -insulin-If competence to undergo the formal TR transition in zinc insulin hexamers were subject to evolutionary selection, then the above 1 H NMR studies suggest that positively charged side chains might be disallowed. Inspection of the protein sequence data base nevertheless suggests that His B5 may functionally be replaced by Arg; indeed, Arg is observed as the only alternative to His. A variant of human insulin containing this positively charged side chain was therefore prepared. The binding affinity of Arg B5 -insulin for the IR is 40% relative to that of wild-type human insulin (Fig. 6A).
Thermodynamic Stability-The stability of Arg B5 -insulin was inferred from CD-detected denaturation studies (Fig. 6B).
Analysis of fractional unfolding as a function of guanidine concentration permitted extraction of the free energy of unfolding (⌬G u ) by a two-state model (35). Whereas the stability of wildtype insulin under these conditions is 4.4 kcal/mol, that of Arg B5 -insulin is 3.6 kcal/mol. The decrease in free energy (⌬⌬G u 0.8 Ϯ 0.1 kcal/mol) is less than that observed in studies of an Ala B5 analog (⌬⌬G u 1.7 Ϯ 0.1 kcal/mol) (22). The decreased stability of Arg B5 -insulin is more modest than that observed on substitution of Val A3 by Thr (⌬⌬G u 1.3 Ϯ 0.1 kcal/ mol) in which a polar ␤-OH makes unfavorable interactions within an analogous nonpolar crevice elsewhere in the protein (48).
Protein Allostery-The capacity of Arg B5 -insulin to undergo the TR transition in solution was assessed in cobalt-substituted hexamers by optical absorption spectroscopy (Fig. 7B). This method exploits a change in the coordination geometry of the axial metal ions from octahedral in each hexamer-related T-state trimer to tetrahedral in each hexamer-related R-state trimer (Fig. 7A). Whereas no such bands are observed in an FIGURE 6. Receptor binding and stability assays. A, human insulin and Arg B5 -insulin exhibit slightly reduced affinity compared with human insulin for the IR. Competition binding assay in which fractional bound 125 I-labeled insulin (trace) is plotted versus the log-concentration of unlabeled wild-type insulin (wt; squares) or analog (triangles). The relative affinity of Arg B5 -insulin is ϳ40%. B, CD-detected protein denaturation studies demonstrate that Arg B5insulin is less stable than wild-type insulin; fractional change in ellipticity is plotted as a function of concentration of guanidine HCl at 4°C. Analysis by a two-state model (35) implies ⌬G u values of 3.6 kcal/mol (Arg B5 -insulin) and 4.4 kcal/mol (wild type).

FIGURE 7. Substitution of His B5 by Arg impedes the TR transition in solution.
A, molecular structure (stereo pairs) of zinc-binding sites in representative T 6 hexamer and R 6 hexamer. Whereas the T 3 -related site is predominantly octahedral, the R 3 -related site is tetrahedral. This difference has been exploited to monitor the TR transition in Co 2ϩ -substituted hexamers in wildtype insulin (wt) and among insulin analogs (32,72). T 3 R 3 f hexamers contain one octahedral (T3) and one tetrahedral (R 3 f ) metal-ion binding site. B, optical absorption studies of Co 2ϩ -substituted insulin hexamers in the presence of 50 mM phenol exhibit characteristic d-d transitions (a signature of R-specific tetrahedral metal ion coordination) between 525 and 650 nm in wild-type R 6 hexamer (upper trace); by contrast, this feature is markedly attenuated by the B5 substitution (lower trace). C and D, CD studies of zinc insulin hexamers in the absence (solid lines) and presence (open squares or circles) of 50 mM cyclohexanol, an inducer of the TR transition. The spectra suggest that this ligand enhances the ␣-helix content of wild-type insulin hexamers (C) but has no significant effect on the CD spectrum of Arg B5 -insulin hexamers (D). octahedral coordination site, the tetrahedral-specific d-d transitions of the bound Co 2ϩ ions therefore provide an intrinsic probe of the TR transition. As expected, wild-type insulin exhibits an intense tetrahedral signature between 500 and 650 nm in the absorption spectrum of the phenol-stabilized R 6 hexamer (Fig. 7B, upper trace). By contrast, this spectroscopic feature is markedly attenuated or absent in Arg B5 -insulin (lower trace).
Extension of these studies from cobalt insulin hexamers to zinc hexamers utilized CD as a probe of an R-state-specific increase in ␣-helix content; this feature is due to a conformational change in the N-terminal segment of the B-chain (Fig. 2). Whereas wild-type insulin exhibits an increase in the magnitude of helix-associated mean residue ellipticity at 208 and 222 nm upon the addition of cyclohexanol (an inducer of the TR transition whose low absorbance at these wavelengths, unlike phenol, is compatible with far-UV CD studies) (Fig. 7C), the CD spectrum of Arg B5 -insulin is unaltered upon the addition of this cyclic alcohol (Fig. 7D). CD spectra of wild-type insulin and the analog in the absence of cyclohexanol are nonetheless similar, suggesting that the substitution does not lead to a gross structural perturbation. In accord with the results of optical spectroscopy, the 1 H NMR spectrum of Arg B5 -insulin under the high phenol conditions that yield high resolution spectra of wild-type (or Asp B28 ) R 6 hexamers (see above) instead exhibits broad and poorly resolved resonances (see supplemental Fig.  S3), similar to those observed in the spectrum of zinc insulin hexamers in the absence of phenol or other R-specific ligands (30).
Crystal Structure-Arg B5 -insulin was crystallized under conditions that lead to crystallization of wild-type insulin as phenol-stabilized R 6 hexamers. In accord with the above spectroscopic studies, Arg B5 -insulin crystallized instead as a T 6 zinc hexamer. A classical lattice was observed in which one T 2 dimer (with protomers designated molecules 1 and 2) defines the asymmetric unit; the hexamer is generated by crystallographic symmetry (see supplemental Fig. S4). The dimensions of the unit cell (see "Experimental Procedures") are inconsistent with standard T 3 R 3 f and R 6 crystal forms. 7 The structure (refined at a resolution of 1.3 Å) includes all residues in each chain and 129 water molecules per protein dimer. 2F o Ϫ F c electron density maps, illustrating the region near Arg B5 in molecules 1 and 2, are shown in Fig. 8B (left and right panels, respectively); continuous density is observed throughout each side chain. The final model exhibits an R-factor of 0.206 with an R free of 0.232. No density is observed suggestive of bound phenol molecules. The two protomers of Arg B5 -insulin are similar to each other and to those of wild-type insulin in T 6 hexamers (see supplemental Fig. S5 and Tables S2 and S3). No significant changes are observed in secondary structure, chain orientation, mode of assembly, or structure of Zn 2ϩ -binding sites (see supplemental Fig. S6). Comparison between the variant protomers and wild-type T 6 -state structures yields root mean square differences (excluding B5) of 0.85 Å (main chain) and 1.50 Å (side chains). These values are similar to those obtained in pairwise comparison between T-state structures of wild-type insulin in different crystal forms (see supplemental Tables S2 and S3). The two axial Zn 2ϩ -binding sites are each well defined without evidence of multiple ligand conformations. Ligation is mediated in each case by three symmetry-related His B10 side chains with zincnitrogen distances of 2.04 Å (see supplemental Fig. S6).
The environment of Arg B5 is similar but not identical to that of His B5 in T-state protomers. In the wild-type and variant T 6 hexamers, the B5 side chain lies at the protein surface at an interface between A-and B-chains ( Fig. 9A and 10A, left-hand  panels). The main-chain dihedral angles (, ) of Arg B5 are (Ϫ98°, 125°) in molecule 1 and (Ϫ94°, 129°) in molecule 2. These values fall within the range of dihedral angles adopted by His B5 in multiple wild-type T 6 hexamers (see supplemental Table  S4). 8 The terminal guanidinium group of Arg B5 protrudes into solvent. The B5-related crevice is less efficiently filled by the linear methylene portion of Arg B5 (Fig. 10A, right) than by the planar imidazole ring of His B5 (Fig. 9A, right). Whereas classical cavities within cores of globular proteins are lined predominantly by nonpolar side chains, the shallow B5-related crevice, in large part open to solvent, is lined by a combination of polar and nonpolar atoms. A complex and asymmetric electrostatic environment is created by (i) the bordering A7-B7 and A6 -A11 disulfide bridges, (ii) the orientation of the A1-A8 ␣-helical dipole, and (iii) individual main-chain amide and carbonyl functional groups; neighboring side chains and bound water molecules in the respective structures are shown in Figs. 9B and 10B. The lower stability of Arg B5 -insulin (relative to wild-type insulin) may reflect more favorable electrostatic interactions by an aromatic heterocycle within this shallow asymmetric pocket.
The packing of Arg B5 is constrained by its local packing in the B5-related crevice and possibly also by hexamer-hexamer contacts in the crystal lattice. 9 Whereas the guanidinium NH 2 groups of Arg B5 interact only with bound water molecules, an interchain hydrogen bond is formed within each protomer from B5 N ⑀ H (Fig. 10B, pink) to the carbonyl oxygen of Cys A7 (Fig. 10B, red ball; see also Fig. 3D and supplemental Fig. S7). (Although the former hydrogen atom is not visualized, the positions of neighboring heavy atoms strongly favor such hydrogen bonding; the nitrogen-oxygen distance in molecule 1 is 2.70 Å, and the N-H⅐⅐⅐O bond angle is 137.3°.) This inferred hydrogen bond recapitulates an interchain hydrogen bond in wild-type insulin from the imidizole N ⑀2 H of His B5 (Fig. 9B, purple) to the carbonyl oxygen of Cys A7 (Fig. 9B, red ball; N ⑀2 -O distance 2.41 Å and bond angle 143.2°in 2-Zn molecule 1) (2). In molecule 2, the N ⑀ H of Arg B5 forms bifurcating hydrogen bonds to the car- 7 The dimensions of the unit cell in themselves suggest the presence of a T 6 hexamer as T 3 R 3 f , and R 6 crystals are ordinarily associated with different unit cell dimensions (T 3 R 3 f : a ϭ 80.1 Å, c ϭ 37.8 Å (Protein Data Bank code 1TRZ); R 6 : a ϭ 61.2 Å, b ϭ 61.6 Å, c ϭ 48.0 Å, ␣ ϭ 90°, ␤ ϭ 110.5°, ␥ ϭ 90°( Protein Data Bank code 1ZNJ)). 8 In the 4INS structure of the T 6 insulin hexamer, B5 main-chain dihedral angles are (Ϫ85°, 117°) in molecule 1 and (Ϫ114°, 108°) in molecule 2; values in other T 6 structures are provided as supplemental material. 9 The corresponding distance in molecule 1 is 5.0 Å; this difference between molecules 1 and 2 appears to be due to a hexamer-hexamer contact in the crystal lattice. Such hexamer-hexamer contacts appear to alter the conformation of the A9 -A11 loop in molecule 2, enabling formation of the bifurcated hydrogen bond between B5 N ⑀ H and the A9 carbonyl oxygen. JULY 25, 2008 • VOLUME 283 • NUMBER 30

JOURNAL OF BIOLOGICAL CHEMISTRY 21205
bonyl oxygens of both Cys A7 (distance 2.80 Å and angle 129.2°) and Ser A9 (distance 3.03 Å and angle 105.1°). Hydrogen bonds and van der Waals contacts made by Arg B5 in molecules 1 and 2 are given in supplemental Table S5; other lattice contacts are described in supplemental Table S6. Formation of analogous interchain hydrogen bonds by Arg B5 and His B5 is of interest in relation to the failure of Met B5 to support insulin chain combination (above), thus highlighting the likely contribution of specific side-chain functional groups to foldability. We envisage that formation of B5-related interchain hydrogen bond(s) in an oxidative folding intermediate would facilitate alignment of neighboring reactive A7 and B7 thiolate moieties as a kinetic aid to disulfide pairing (see "Discussion").

DISCUSSION
Allosteric assembly of globular proteins has long provided a model for the transmission of conformational change (13, 49 -51). Might the TR transition of insulin foreshadow the mechanism of receptor binding? This intriguing possibility, first suggested by the late D. C. Hodgkin and colleagues (2,14), has been raised anew by studies of nonstandard insulin analogs (21,52).
Interpretation of crystal structures of insulin is limited by its state of assembly as the hormone binds to the IR as a monomer.
Although studies of truncated insulin analogs have revealed only T-like features (18,19,53,54), a variety of theoretical and experimental studies indicate that the monomer is flexible (41,46,(55)(56)(57), including at the N-terminal segment of the B-chain (residues B1-B8). Stabilization of its T-state conformation has been achieved by substitution of Gly B8 by D-amino acids. Because in the T-state Gly B8 resides in a ␤-turn, with positive angle in a region of the Ramachandran plane unfavorable to L-amino acids, D-analogs exhibit both greater thermodynamic stability and partial impairment of the TR transition (21,52). Remarkably, however, such stabilized analogs exhibit very low activities. Although these observations suggest that the T-state represents an inactive conformation and indeed that B8 functions as a site of induced fit, such studies do not specify the overall features of the bound state, in particular whether the N-terminal segment of the B-chain reorganizes as an ␣-helix as in the hexamer-specific TR transition (13,14).
As a further test of the relationship between the TR transition and biological activity, the present study has exploited a species variant (Arg) at position B5. We chose to focus on this substitution in human insulin because of the structural environment of His B5 in wild-type hexamers. In the T-state, the imidazole ring packs within a solvated crevice at the edge of the hexamer surface (2), whereas in the R-state the side chain packs at an interface between dimers. Although this interface is in part solvated, NMR studies of the pK a of His B5 suggested that the R-specific (or R f -specific) trimer interface would be destabilized by the uncompensated positive charge of an Arg B5 substituent. In accord with this prediction, our spectroscopic results demonstrated that substitution of His B5 by Arg impedes the TR transition in solution. Further, crystals of Arg B5 -insulin grown under conditions leading to stabilization of wild-type R 6 hexamers (i.e. in the presence of phenolic ligands) were observed to contain only T 6 hexamers. A similar block to the TR transition has been recently reported in a review article of an analog containing substitutions of Asn B3 by Lys and of Lys B29 by Glu B29 (insulin glulisine) (58).
The crystal structure of Arg B5 -insulin as a variant T 6 hexamer is essentially identical to that of the wild-type T 6 hexamer, including analogous interactions by the variant B5 side chain. Because Arg B5 -human insulin binds well to the IR (with affinity ϳ40% relative to wild type), our results collectively demonstrate that competence to undergo the TR transition in a hex- amer is not required for the biological activity of an insulin monomer. We discuss these results in relation to protein stability. Because of the recent clinical finding of a mutation at B5 causing neonatal diabetes mellitus due to protein misfolding (59), we also discuss our findings in relation to the nascent folding of proinsulin in the endoplasmic reticulum of the pancreatic ␤-cell.
Determinants of Insulin Stability-His B5 contributes to the thermodynamic stability of insulin: its substitution by Ala in an engineered monomer yields a nativelike structure whose unfolding free energy (⌬G u ), as inferred from chemical denaturation studies, is reduced from 4.9 to 3.2 kcal/mol at 4°C (22). This decrement (⌬⌬G u 1.7 Ϯ 0.1 kcal/mol relative to DKPinsulin) is presumably due to a local packing defect in the B5-related interchain crevice and possible changes in solvation. In Arg B5 -insulin, a nativelike pattern of contacts in this crevice is maintained, including interchain hydrogen bonds and a neighboring network of bound water molecules. These interactions constrain the orientation of the N-terminal segment of the B-chain against the A chain and in part define the environment of the solvent-exposed A7-B7 disulfide bridge.
The stability of Arg B5 -insulin as a zinc-free monomer is also lower than that of wild-type insulin; the decrement (⌬⌬G u 0.8 Ϯ 0.1 kcal/ mol) is less marked than that of an Ala B5 analog. The intermediate stability of Arg B5 -insulin suggests that, relative to Ala B5 , the variant side chain in part recapitulates the native packing of His B5 but that its net contribution is less favorable than that of an imidazole ring. Favorable electrostatic interactions may also be present as the Arg B5 guanidinium group (with its positive charge) projects into solvent near the C-terminal (and hence negative) end of the A1-A8 ␣-helical dipole axis. In addition, the N ⑀ H group of Arg B5 donates a hydrogen bond to the carbonyl oxygen of Cys A7 and perhaps (in molecule 2) a bifurcating second hydrogen bond to the carbonyl oxygen of Ile A10 . Although these interactions recapitulate native contacts by His B5 , the proximal aliphatic portion of the Arg B5 side chain differs from His in shape, size, and electrostatic properties. We speculate that the packing of His B5 within the interchain crevice uniquely optimizes an asymmetric pattern of weakly polar interactions not satisfied by Arg or other side chains. Determinants of Foldability-Just as Arg B5 partially reverts the thermodynamic instability of an Ala B5 analog, this species variant also partially restores the efficiency of disulfide pairing in the course of chain combination. Whereas chemical synthesis of an Ala B5 insulin analog encountered very low yields, the yield of Arg B5 -insulin was reduced by less than 3-fold. This improvement appears due to the guanidinium group, since an attempted synthesis of Met B5 -insulin failed to yield detectable product. A similarly profound block to chain combination was observed upon substitution of Gly B8 by any L-amino acid (21). Such impediments seem remarkable in light of the general utility of chain combination in the synthesis of diverse insulin analogs since its development in 1966 (60).
Specific blocks to chain combination presumably reflect conformational events in a critical oxidative folding intermediate. FIGURE 9. Environment of His B5 in wild-type T-state protomer. A, left, space-filling model showing the B5 side chain (green) at the protein surface (green arrow). The A-chain is shown in light gray, and the B-chain is otherwise shown in dark gray. The position of the His B5 side chain (green sticks) is shown relative to the PyMolgenerated electrostatic surface in corresponding orientation; the color bar (below the structure) indicates calibration from negative (Ϫ15 kT/e in red) to positive (ϩ15 kT/e in blue). B, stereo representation of side-chain packing in the B5-related crevice. The His B5 side chain is shown in green with the N ⑀ atom highlighted in purple. The carbonyl oxygen of Cys A7 is shown as a red ball; other carbonyl oxygen atoms are shown as red sticks. A-and B-chain residues are otherwise shown in gray and black, respectively. Adjoining bound water molecules are shown in blue spheres, and the solvent-exposed A7-B7 disulfide bridge is shown in gold. The corresponding structural features of Arg B5 -insulin are shown in Fig. 10.
We envisage that the conformation of Gly B8 and interchain packing of His B5 are each integral to stabilization of the nascent B7-B10 ␤-turn and in turn to formation of the A7-B7 disulfide bridge. The biological relevance of these observations is supported by the recent finding of mutations in the insulin causing neonatal diabetes mellitus (61). Remarkably, the emerging clinical data base of such mutations includes substitutions at B5 and B8 (59). Cell biological studies have shown that Ala B5 and diverse other side chains, including the diabetes-associated mutation Asp B5 (59), impair the folding of proinsulin in the endoplasmic reticulum of a transfected human cell line (22).
Because insulin chain combination and the in vivo folding of proinsulin reflect kinetic processes, small decrements in the thermodynamic stability of the products (the insulin analogs) cannot in themselves account for relative yields. Relative to Ala B5 , therefore, how can Arg B5 enhance both stability and efficiency of disulfide pairing? We imagine that allowed side chains at B5 both participate in nascent interchain interactions preceding formation of cystine A7-B7 and stabilize the subsequent disulfide-bridged species, including the mature hormone, once folding has been achieved. That Arg B5 is allowed among vertebrate insulin sequences suggests that its nativelike contacts in the B5-related crevice, including the directional interchain hydrogen bonds observed in the present crystal structure, would productively guide disulfide pairing in vivo. Efforts to test this prediction in cell culture are planned.
Why is His B5 (rather than Arg B5 ) conserved among eutherian mammals and the predominant residue among other classes of vertebrates? It seems unlikely that this preference reflects a physiological advantage of 2-fold higher receptor-binding affinity, since the activities of vertebrate insulins vary by more than 10-fold, and small decrements in affinity are readily compensated by correlated changes in clearance rates (62,63). We suggest instead that an evolutionary constraint is imposed by requirements of protein folding and trafficking in the ␤-cell. Interchain interactions by His B5 in a nascent proinsulin polypeptide may enhance its folding efficiency in the endoplasmic reticulum (as it does in chain combination in vitro), in turn protecting the ␤-cell from toxic protein misfolding (64,65). Importantly, the genetic link between misfolding of proinsulin and human ␤-cell dysfunction (described above in relation to neonatal diabetes) (61) suggests that the foldability of proinsulin has imposed a powerful constraint on the evolution of allowed insulin sequences.
Conclusion-The TR transition of insulin provides a model of protein allostery. The classical nomenclature T and R reflects an analogy between the transmission of conformational change in insulin and the TR transition of hemoglobin, associated with the cooperative binding and release of three gases (O 2 , CO, and NO) (49,50,66). Whereas we have characterized a novel substitution in insulin that blocks its TR transition, biochemical studies of variant human globin chains associated with defective oxygen transport have identified clinical mutations that perturb cooperativity. Interestingly, two such clinical mutations also involve substitution of His by Arg; these occur near side chain (green) at the protein surface (green arrow). The A-chain is shown in light gray, and the B-chain is otherwise shown in dark gray. The position of the Arg B5 side chain (green sticks) is shown relative to the PyMol-generated electrostatic surface in corresponding orientation; color bar (below the structure) indicates calibration from negative (Ϫ15 kT/e in red) to positive (ϩ15 kT/e in blue). B, stereo representation of side-chain packing in the variant B5-related crevice. The Arg B5 side chain is shown in green, with the N ⑀ atom highlighted in pink. The carbonyl oxygen of Cys A7 is shown as a red ball; other carbonyl oxygen atoms are shown as red sticks. A-and B-chain residues are otherwise shown in gray and black, respectively. Adjoining bound water molecules are shown in blue spheres, and the solvent-exposed A7-B7 disulfide bridge in gold. Corresponding structural features of wild-type insulin are shown in Fig. 9.
An essential difference between insulin and classical TR assemblies is that insulin functions as a monomer. Although the zinc insulin hexamer is of biological relevance as a storage vehicle, the relevance of its structure to the mechanism of receptor binding has long been the subject of speculation. Substitutions at subunit interfaces of the insulin hexamer that modulate its allosteric properties may thus be unrelated to the mechanism of monomer binding to the IR (i.e. monomeric Arg B5 -insulin can form the R-state and therefore bind to the IR, but it cannot stabilize the R-state in the hexamer due to the high intrinsic pK a value for residue Arg B5 ). Arg B5 -insulin retains substantial receptor-binding affinity (40% relative to wild-type human insulin), strongly suggesting that competence of an insulin analog to undergo the formal TR transition is not necessary for biological activity. We may regard such substitutions as extrinsic to the conformational repertoire of the monomer. The small decrement in activity of Arg B5 -insulin is presumably due to a change in the electrostatic or steric contours of the receptorbinding surface (see supplemental Fig. S8).
Does it remain possible that on receptor binding the active insulin monomer exhibits R-like features? This possibility is indeed suggested by studies of an intrinsic probe of the conformational repertoire: stereospecific modulation of the B8 angle by corresponding D-or L-amino acid substitutions (21,52). Substitution of Gly B8 by D-amino acids favors the T-state by a mechanism unrelated to subunit interfaces and yet markedly impairs receptor binding. In light of these and the present results, we propose that induced fit of the insulin monomer and the hexameric TR transition exploit corresponding sites of flexibility in the molecule but as distinct processes. Identification of such sites has direct relevance to structural determinants of foldability (21) and hence the newly recognized genetics of neonatal diabetes mellitus (61). How induced fit enables this ancestral hormone to bind to and trigger the insulin receptor represents a major unsolved problem in structural biology.