Toward the Active Conformation of Insulin

How insulin binds to the insulin receptor has long been a subject of speculation. Although the structure of the free hormone has been extensively characterized, a variety of evidence suggests that a conformational change occurs upon receptor binding. Here, we employ chiral mutagenesis, comparison of corresponding d and l amino acid substitutions, to investigate a possible switch in the B-chain. To investigate the interrelation of structure, function, and stability, isomeric analogs have been synthesized in which an invariant glycine in a β-turn (GlyB8) is replaced by d- or l-Ser. The d substitution enhances stability (ΔΔGu 0.9 kcal/mol) but impairs receptor binding by 100-fold; by contrast, the l substitution markedly impairs stability (ΔΔGu -3.0 kcal/mol) with only 2-fold reduction in receptor binding. Although the isomeric structures each retain a native-like overall fold, the l-SerB8 analog exhibits fewer helix-related and long range nuclear Overhauser effects than does the d-SerB8 analog or native monomer. Evidence for enhanced conformational fluctuations in the unstable analog is provided by its attenuated CD spectrum. The inverse relationship between stereospecific stabilization and receptor binding strongly suggests that the B7-B10 β-turn changes conformation on receptor binding.

Insulin is a small globular protein containing two chains, A (21 residues) and B (30 residues, Fig. 1A). Stored in the ␤ cell as a Zn 2ϩ -stabilized hexamer, the hormone functions as a Zn 2ϩ -free monomer. Although the structure of the free hormone has been well characterized by x-ray crystallography and NMR spectroscopy (Fig. 1B, Refs. 1-3), it is not known how insulin binds to its receptor. Indeed, studies of single chain analogs have suggested that classical structures represent inactive conformations (4,5). What sites in the insulin molecule might function as a structural switch? Here, we employ chiral mutagenesis, comparison of corresponding D and L amino acid substitutions (6), to uncover non-standard structure-function relationships. Our studies focus on the B7-B10 ␤-turn, the hinge between the N-terminal ␤-strand and central ␣-helix of the B-chain. An inverse relationship between protein stability and activity provides a stereospecific signature of induced fit.
Our strategy exploits the right hand side of the Ramachandran plane: substitution of a glycine with positive angle (and hence residing in a region ordinarily "forbidden" to L amino acids) by a D amino acid (7,8). The site of substitution (Gly B8 ; arrow in Fig. 1A and red balls in Fig. 1B) is invariant among mammalian insulins and insulin-like growth factors (9). Gly B8 lies on the surface of a type II ␤-turn comprising residues B7-B10 ( Fig. 1, C and D). This turn contains Cys B7 (part of the canonical A7-B7 disulfide bridge; highlighted in gold in Fig. 1, C and D) and so provides a link between the central B-chain ␣-helix and the A-chain. Substitution of Gly B8 by L amino acids impairs the folding of a single chain insulin precursor (10) and impedes disulfide pairing in insulin chain combination (6). Substitution of a turn-specific D-glycine by a D amino acid would by contrast be expected to enhance the stability of the native fold (7,8). Respective sites of L and D substituents are indicated in blue (the pro-L H ␣ of Gly B8 ; Fig. 1D) or magenta (pro-D H ␣ ). Whereas previous studies of D-and L-Ala B8 insulin analogs established such reciprocal effects on stability, the L analog was observed to misfold; aberrant aggregation prevented structural studies and confounded interpretation of its biological activity (6). Fortuitously, these limitations may be circumvented through the use of D-and L-Ser.
In this article we describe the solution structures of D-and L-Ser B8 analogs of an engineered insulin monomer. Such engineering is required to avoid dimerization and higher order protein assembly (11), which would otherwise limit the feasibility of NMR analysis (12). A monomeric template is provided by DKP-insulin, 2 which contains three amino acid substitutions in the B-chain (Asp B10 , Lys B28 , and Pro B29 ; magenta in Fig. 1A and Ref. 13). The structure of DKP-insulin and its use as a template for study of mutations of interest have previously been described (14,15). Our results demonstrate that whereas both D-and L-Ser B8 -DKP-insulin retain native-like overall folds, the L analog exhibits a profound decrease in thermodynamic stability associated with attenu-ation of CD-derived ␣-helix content. Further, whereas the NOESY spectrum of D-Ser B8 -DKP-insulin is similar to that of DKP-insulin, the spectrum of L-Ser B8 -DKP-insulin exhibits fewer helix-related and long range contacts, presumably because of enhanced nonlocal conformational fluctuations. Such modulation of protein dynamics suggests that stereospecific frustration of the B8 angle is coupled to a global change in the free energy landscape (16). Despite such perturbations, L-Ser B8 -DKP-insulin retains near-native receptor binding affinity whereas the affinity of D-Ser-DKPinsulin is reduced by 100-fold. This inverse relationship between stereospecific stabilization and receptor binding strongly suggests that Gly B8 , and in turn the B7-B10 ␤-turn, changes conformation on receptor binding.

EXPERIMENTAL PROCEDURES
Materials-Human insulin was kindly provided by Eli Lilly and Co. (Indianapolis, IN). All other chemicals were of analytical grade (Fisher Chemicals).
Synthesis of Insulin Analogs-The tetra-S-sulfonate derivative of the human A-chain was obtained by oxidative sulfitolysis (17). B-chain analogs were prepared by solid phase chemical synthesis as described (17). In brief (N-tert-butoxycarbonyl, O-benzyl)-Thr-PAM resin (0.56 mmol/g; Bachem, Inc.) was used as solid support for synthesis of B-chain analogs. The D-Ser B8 -DKP B-chain was prepared by automated solid-phase synthesis using F-moc chemistry (18). For synthesis of the L-Ser B8 -DKP Bchain, a manual double-coupling t-BOC protocol was followed (17,19). D-and L-Ser B8 -DKP-insulin were prepared by chain combination (17,20). Analogs were purified by reverse-phase high performance liquid chromatography (HPLC) as described (17). Whereas the yield of the D analog was at least as high as that of DKP-insulin, the yield of the L analog was reduced by Ͼ10-fold. Predicted molecular masses were confirmed by matrix-assisted laser desorption ionization (MALDI-TOF) mass spectrometry (MS).
Receptor Binding Assays-Relative activity is defined as the ratio of analog to wild-type human insulin required to displace 50% of specifically bound 125 I-human insulin. A human placental membrane preparation containing the insulin receptor (IR) was employed as described (21). In all assays the percentage of tracer bound in the absence of competing ligand was Ͻ15% to avoid ligand-depletion artifacts.
Circular Dichroism-Far-ultraviolet (UV) CD spectra were obtained as described (22). Spectra, acquired with an Aviv spectropolarimeter (Aviv Biomedical, Inc., Lakewood, NJ), were normalized by mean residue ellipticity. Estimates of secondary structure were obtained by deconvolution using an iterative singular value decomposition algorithm as described (23). Samples were dissolved in 10 mM phosphate and 50 mM KCl (pH 7.4) at a protein concentration of ϳ25 M. For equilibrium denaturation studies, samples were diluted to 5 M; guanidine-HCl was employed as denaturant (24). Data were obtained at FIGURE 1. Sequence and structure of insulin. A, sequence of B-chain (top) and A-chain (bottom); arrow indicates invariant Gly B8 (red). Shown above B-chain in magenta are the three substitutions in the monomeric DKP template. B, cylinder models of TR dimer based on crystal structure of zinc insulin hexamers (PDB ID: 1TRZ). The T-state is at left and R-state at right. B-chain ␣-helices are shown in green; the ␣-carbons of Gly B8 are shown as red circles. Three families of hexamers have been characterized, designated T 6 , T 3 R f 3 , and R 6 . The R-state conformation has only been observed within hexamers. C, structure of insulin T-state (stereo pair) showing positions of selected side chains (labeled at left) relative to Gly B8 C ␣ (red) and disulfide bridges (gold; labeled at right). The B-chain is shown in green, and A-chain in black. D, structure of T-state-specific B7-B10 ␤-turn (stereo pair). Main chain of Gly B8 is shown in red; its pro-L and pro-D H ␣ atoms are highlighted in blue and magenta, respectively. 4°C and fitted by non-linear least squares to a two-state model (25).

RESULTS
D-and L-Ser B8 -DKP-insulin exhibit receptor-binding affinities of 1.1 Ϯ 0.1 and 90 Ϯ 6% relative to native human insulin (column 2 in Table 1). Because the relative affinity of DKPinsulin is 161 Ϯ 19 under these conditions (enhanced binding is caused by the Asp B10 substitution (29)), specific effects of the D and L B8 substitutions on affinity relative to the parent monomer are ca. 0.7 and 55%, respectively. The far-UV CD spectrum of D-Ser B8 -DKP-insulin exhibits a slight accentuation of helixspecific features at 196, 208, and 222 nm relative to the spectrum of DKP-insulin (filled circles and solid line, respectively, in Fig. 2A). Conversely, the CD spectrum of L-Ser B8 -DKP-insulin is attenuated at helix-sensitive wavelengths (open circles in Fig.  2A). Because the precision of the measured ellipticities in the range 200 -250 nm is Ϯ1%, these stereospecific differences are significant. Respective values of mean residue ellipticity at 196, 208, and 222 nm are given in Table 2, A. Deconvolution suggests D-specific enhancement, and complementary L-specific attenuation, of mean ␣-helix content (Table 2, B). Such changes may in principle reflect either static structural changes or stereospecific modulation of conformational fluctuations within helical elements (see below). D-and L-Ser B8 -DKP-insulin exhibit marked differences in thermodynamic stability. Whereas the D-Ser B8 analog is more resistant to denaturation in concentrated solutions of guanidine hydrochloride than is DKP-insulin, the L analog is more sensitive (Fig. 2B). Fitting of these CD-detected denaturation curves by a two-state model yields ⌬⌬G u values of 0.9 Ϯ 0.2 kcal/mol (D analog) and Ϫ3.0 Ϯ 0.2 kcal/mol (L) relative to DKP-insulin ( Table 1). The difference in stability between stereoisomers is thus 3.9 Ϯ 0.3 kcal/mol. This difference is significantly greater than would be expected based on chiral inversion within an isolated ␤-strand and so implies non-local effects of the substitution in one or both analogs. a Binding activity is defined by affinity for the human placental insulin receptor relative to human insulin (100%); under these conditions the K d for native insulin is 0.48 Ϯ 0.06 nM. The number of replicates is given in parentheses. b ⌬G u indicates apparent change in free energy on denaturation in guanidine-HCl as extrapolated to zero denaturant concentration by a two-state model (25). c C mid is defined as that concentration of guanidine-HCl at which 50% of the protein is unfolded. d The m value provides the slope in plotting unfolding free energy ⌬G u versus molar concentration of denaturant; this slope is proportional to the protein surface area exposed on unfolding.   a Spectra were obtained at 4°C in 50 mM KCl and 10 mM potassium phosphate (pH 7.4); ͓͔ indicates mean residue ellipticity. Errors in ellipticity measurement do not include possible systematic error in protein concentration. b Fractional deconvolution parameters f ␣ , f ␤ , f , and f rc indicate percent ␣-helix, ␤-sheet, ␤-turn, and random coil, respectively. The basis set employs stably folded globular proteins and may not pertain to flexible polypeptides or molten domains. NMR results suggest that apparent differences in helix content reflect conformational fluctuations rather than static changes in helix end points. 1 H NMR spectra of L-and D-Ser B8 -DKP-insulin at pH 7.4 (spectra a and b in Fig. 2C) exhibit a pattern of resonance line widths and chemical shift dispersion similar to that of DKPinsulin (spectrum c in Fig. 2C). Although differences are apparent in details of these spectra, similar (but not identical) patterns of upfield-shifted methyl resonances (0.00 -0.70 ppm) and downfield-shifted H ␣ resonances (4.8 -5.3 ppm) are observed. The spectra are in each case tractable by standard homonuclear NMR methods (30), permitting near-complete resonance assignment (see supplemental materials). Because analysis of exchangeable amide resonances is incomplete at neutral pH due to base-catalyzed solvent exchange and conformational broadening, additional data were obtained in 20% deuteroacetic acid; in this co-solvent insulin retains a nativelike monomeric fold (31). In accord with estimates of ⌬G u based on guanidine denaturation, the rate of amide proton exchange in 20% deuteroacetic acid and 80% D 2 O is markedly accelerated in the L-Ser B8 analog and retarded in the D-Ser B8 analog (see supplemental material). Quantitative interpretation of protection factors is limited by the absence of baseline exchange rates in this co-solvent.
NMR Studies of the D Isomer-Chemical shifts are essentially identical to those observed in DKP-insulin; significant changes (magnitude Ͼ0.1 ppm) are observed only at neighboring residues Cys B7 and Ser B9 (supplemental material). The novel D-Ser B8 spin system is well resolved (Fig. 3A). Analysis of secondary structure, based on diagnostic strings of d NN , d ␣N , d (i, iϩ3) , and d (i, iϩ4) NOEs, is identical to that of DKP-insulin (supplemental material). The pattern of long range inter-residue NOEs is likewise similar, in each case consistent with structures of T-state crystallographic protomers. Key native-like long range NOEs are observed, for example, between aromatic and aliphatic side chains (Fig. 4, A and B). These include contacts between the side chains of Phe B24 /Leu B15 and Tyr B26 /Val B12 , indicative of native-like B-chain supersecondary structure; between Tyr A19 /Ile A2 , indicative of native-like A-chain supersecondary structure; and between Phe B1 /Leu A13 , His B5 / Ile A10 , Tyr A19 /Leu B15 , and Tyr B26 / Val A3 , indicative of a native-like orientation between chains. The upfield chemical shifts of the Leu B15 methyl resonances, sensitive to the ring current of Phe B24 , are essentially identical in the spectrum of the D-Ser B8 analog (0.13 and 0.54 ppm) and in the spectrum of DKP-insulin (0.15 and 0.54 ppm). The upfield chemical shift of the Val B12 H ␣ resonance, sensitive to the ring currents of Phe B24 and Tyr B26 , is also not significantly perturbed (3.12 ppm in the D-Ser B8 analog versus 3.18 ppm in DKP-insulin; random-coil value 4.18 ppm). Such correspondence of secondary shifts provides evidence of structural similarity given the steep dependence of upfield ring currents on the relative distance and orientation between these side chains.
A summary of inter-residue NOEs is given as a diagonal plot (Fig. 3C). Maintenance of T-state-specific long range interactions by the N-terminal arm of the B-chain is demonstrated by retention of native-like inter-chain NOEs between the B1-A13 and B5-A10 . Two-dimensional NMR identification of D-and L-Ser B8 spin systems and diagonal plot of interresidue NOES. TOCSY spectra of D-Ser B8 -DKP-insulin (A) and L-Ser B8 -DKP-insulin (B) in the region containing the AMX spin system of respective B8 side chains. Spectra (mixing times 55 ms) were observed in D 2 O at 32°C and pD 7.6 (direct meter reading). C and D, diagonal plot of D-Ser B8 -DKP-insulin (C) and L-Ser B8 -DKP-insulin (D) shown inter-residue NOEs, respectively. NOEs between side chains are shown at lower right (open boxes); NOEs between main chain protons or between main chain and side chains are shown at upper left (filled boxes). Red squares in C indicate NOEs consistent with T-state crystal structures but unobserved in the spectrum of L-Ser B8 -DKP-insulin. In C, NOEs a-c indicate (a) set of contacts between the Ile A10 side chain and main chain atoms of Asn B3 , Gln B4 , and His B5 ; (b) contact between the side chains of His B5 and Thr A8 ; and (c) the set of contacts between the Leu B6 side chain and H ␣ of Leu B11 across the ␤-turn. In D, red boxes indicate NOEs present in the L analog but not in the D analog. Respective green and red boxes d and e indicate (d) Cys A11 side chain to H ␣ of Cys A6 ; and (e) contact between H ␤ of Tyr A19 and meta resonance of Phe B24 . side chains; examples are provided by contacts between the side chain of Ile A10 and main chain atoms of Asn B3 , Gln B4 , and His B5 (labeled a in Fig. 3C) and between the side chains of His B5 and Thr A8 (labeled b). These contacts are extended by a network of interchain NOEs involving neighboring main chain and side chain protons in segments A6 -A11 and B3-B7. The orientation of the B9 -B19 ␣-helix relative to the A-chain, as probed by a network of long range NOEs in the hydrophobic core (supplemental material), is also unaffected. NOEs are observed from B8 H N to the methyl resonances of Leu B11 and from B8 H ␣ to ␥ 1 -CH 3 resonance of Val B12 ; these contacts are in accord with corresponding NOEs in DKP-insulin. A novel side chain-specific NOE is observed from the ␤-CH 2 group of D-Ser B8 to the aromatic meta resonance of Tyr B26 (but not to the methyl groups of Val A3 , Val B12 , and Leu B11 , which might also be plausible based on wild-type crystal structures).
NMR Studies of the L Isomer-The L-Ser B8 spin system is likewise well resolved (Fig. 3B), and essentially complete resonance assignment was obtained (supplemental material). Inversion of B8 C ␣ chirality introduces non-local perturbations in NOEs and chemical shifts. Although the overall pattern of interresidue NOEs is native-like, helix-related strings of d NN , d ␣N , d (i, iϩ3) , and d (i, iϩ4) contacts are less complete than those observed in spectra of DKP-insulin or D-Ser B8 -DKP-insulin. In total 51 fewer helix-related NOEs are observed; their attenuation seems consistent with the attenuated CD spectrum of the L analog. Chemical shifts are similar to but distinct from those in D-Ser B8 -DKP-insulin; significant changes (magnitude Ͼ0.1 ppm) are summarized in Table 3. Stereospecific perturbations are observed at adjoining residues Cys B7 and Ser B9 , within contiguous structural elements (⌬␦ 0.22 ppm at Gln B4 H ␣ and 0.16 ppm at Asp B10 H ␤ ), and at transmitted sites (0.19 ppm at Cys B7 H ␤ and 0.15 ppm at the meta resonances of Phe B24 ). Significant chemical-shift differences between D and L analogs occur at residues A7, A10, B4, B5, B9, and B11 at pH 8; a corresponding set of perturbations is seen in 20% deuteroacetic acid (residues A2, A7, A10, B4 -B6, and B10 -B13).
Despite perturbations in chemical shifts and partial attenuation of helix-related NOEs, L-Ser B8 -DKP-insulin retains key native-like long range contacts (Fig. 4C). As in the D-Ser B8 analog, native-like contacts between the side chains of Phe B24 /  Values represent difference in chemical shifts between corresponding resonances in spectra of D-and L-Ser B8 -DKP-insulin at pH 8.0 and 32°C.  13 ppm). Likewise, the upfield chemical shift of the Val B12 H ␣ resonance (3.28 ppm) is less shifted relative to the random coil value (4.18 ppm) than in DKP-insulin (3.18 ppm) or D-Ser B8 -DKP-insulin (3.12 ppm). This subtle trend in secondary shifts correlates with the order of thermodynamic stabilities (⌬G u ; Table 1) and CD-derived helix contents (Table 2, B). We speculate that attenuation of 1 H NMR secondary shifts in the L-Ser B8 analog and attenuation of helixspecific mean residue ellipticities have a common physical origin: conformational fluctuations leading to averaging of ring current shifts in one case and optical chirality in the other.

Residue
Despite such subtle 1 H NMR features, a subset of native-like long range NOEs are retained between A-and B-chains. As in DKP-insulin and D-Ser B8 -DKP-insulin, strong contacts are observed between the side chains of Tyr A19 and Leu B15 . Although NOEs involving the N-terminal arm of the B-chain are less prominent than in the spectrum of the D analog (see below), T-like positioning of the N-terminal segment is retained as indicated by long range NOEs between the side chains of Phe B1 /Leu A13 and His B5 /Ile A10 . The orientation of the canonical B9 -B19 ␣-helix relative to the A-chain is likewise defined by an analogous network of long range NOEs (Fig. 3D  and supplemental material). Nevertheless, the number of long range contacts in the spectrum of the L analog (89 NOEs) is significantly smaller than in the spectrum of the D analog (122 NOEs). NOEs present in the D analog but not in the L analog are highlighted in red in Fig. 3C. Also as in DKP-insulin and D-Ser B8 -DKP-insulin, native-like long range NOEs are maintained within the A-chain between the side chains of Ile A2 and Tyr A19 (Fig. 4C). Similarity of the conformation of the A6 -A11 disulfide bridge is indicated by maintenance of an NOE between Cys A11 H ␤ and Cys A6 H ␣ (labeled d in Fig. 3D). Contacts are also retained between side chains Phe B24 /Leu B15 , Tyr B26 /Val B12 , and Tyr A19 /Leu B15 . These and related NOEs define native-like supersecondary structures within each chain and constrain the orientation between chains. Unlike in DKPinsulin and D-Ser B8 -DKP-insulin, native-like long range NOEs between Val A3 and Leu B11 are not observed.
The L-Ser B8 -specific perturbation extends into the N-terminal arm of the B-chain. Absent or markedly attenuated are native-like inter-chain NOEs between side chains Asn B3 /Ile A10 and His B5 /Thr A8 . Also absent are NOEs from the side chain of Ile A10 to Asn B3 H ␣, ␤ and Gln B4 H N,␣ . Whereas NOEs from the side chain of Cys B7 to Cys A6 H ␣ are also absent, the spectrum of the L analog contains cystine-related NOEs Cys B7 -H ␤ / Cys A7 -H ␣ and Cys B7 -H ␣ /Cys A7 -H N such contacts, not observed in DKP-insulin or the D-Ser B8 analog, indicate an altered or more flexible disulfide linkage. L-Ser B8 -specific NOEs (red in Fig. 3D) also affect contacts between the aromatic ring of Phe B25 and the H ␣ protons of Cys A20 and Asn A21 . Subtle differences in side chain packing are further indicated by a novel NOE between the H ␤ of Tyr A19 and meta resonance of Phe B24 (labeled e in Fig. 3D). Although in each case consistent with crystal structures, such NOEs are not observed in DKP-insulin and indicate an altered orientation of the B24 and B25 side chains. Overall T-like positioning of the N-and C-terminal segments is nonetheless retained in each case as indicated by maintenance of canonical long range NOEs. Attenuation of long range NOEs not directly involving L-or D-Ser B8 provides evidence that stereospecific frustration (or stabilization) of the B7-B10 ␤-turn is coupled to conformational fluctuations elsewhere in the protein. 1 H NMR analysis, when considered in appropriate detail, thus corroborates the apparent attenuation of organized structure implied by CD.
Solution Structures-DG/RMD structures were calculated according to NOE, J-coupling, and hydrogen-bond-related restraints (supplemental material). The total number of restraints employed in calculating the D-and L-Ser B8 ensembles are 631 and 537, respectively (ϳ11-12 restraints per residue). Root-mean-square deviations (RMSDs) are provided as supplemental material. The ensembles verify that the solution structures of D-and L-Ser B8 -DKP-insulin (Fig. 5, A and B, respectively) are similar to the T-like conformation of DKP-insulin (black ribbon, Fig. 5). The L-Ser B8 ensemble exhibits a small shift in the position of the N-terminal A-chain ␣-helix relative to the B-chain. Although analogous structural variation is observed in comparison of crystal structures (supplemental material), it is possible that an L-Ser B8 -specific perturbation is transmitted to the A-chain via the adjoining A7-B7 disulfide bridge. The precisions of the ensemble do not allow atomic scale definition of a discrete pathway of conformational change. Although the precisions of the two ensembles are similar (main chain RMSDs 0.50 Å (D) and 0.61 Å (L) for residues B3-B27 and A2-A20), the D structure is better defined in the neighborhood of B8 and within the B1-B7 arm. B8 Ramachandran angles in the D ensemble are ϭ 60.1°Ϯ 8.1°and ϭ Ϫ150.0°Ϯ 8.2°. These values cluster as expected on the right side of the Ramachandran plot, near the angle of Gly B8 in crystallographic T-state protomers (55.6°Ϯ 2.8°). The angle in the DG/RMD ensemble is also similar to that of Gly B8 in T-state crystallo-graphic protomers (Ϫ131.2°Ϯ 4.2°). The local precision of the L ensemble is not sufficient to determine the B8 conformation. 3 It would be of future interest to define these angles at high resolution through x-ray crystallography.

DISCUSSION
The present study exploits chiral stabilization or destabilization of the B7-B10 ␤-turn by respective D-or L-Ser B8 substitutions to probe structure-function relationships. The substitutions were incorporated into an engineered insulin monomer (13) to circumvent otherwise confounding effects of self-association. Substitution of Gly B8 by D-Ser enhances stability but markedly impairs receptor binding; its substitution by L-Ser markedly impairs stability but is compatible with substantial activity. Such an inverse correlation between stability and activity suggests that the native B7-B10 ␤-turn undergoes a change in conformation in receptor binding.
Use of a D amino acid substitution highlights the power of the total chemical synthesis of proteins to elucidate non-standard structure-activity relationships. Despite the general robustness of insulin chain combination to diverse substitutions, however, low yields were encountered in the synthesis of L-Ser B8 -DKP-insulin; the yield of the D-Ser B8 analog was by contrast at least as high as that obtained in corresponding syntheses of DKP-insulin. The origin of this stereospecific impairment of chain combination is not well understood. Because efficient syntheses of unstable insulin analogs have previously been reported (15,24,32), we imagine that L-Ser B8 imposes a kinetic barrier to disulfide pairing. It is possible that in a reaction intermediate the main chain conformation of residue B8 affects the orientation of Cys B7 and in turn its alignment with Cys A7 . Anomalously low yields have likewise been encountered by Katsoyannis and co-workers (17) on interchange of residues Leu B11 and Val B12 (also near cystine A7-B7) and on substitution of Leu A16 (near cystine A20 -B19; Ref. 33). It would be of interest should this pattern of yields generalize to the efficiency or fidelity of folding of corresponding mutant proinsulins in the endoplasmic reticulum of the pancreatic ␤ cell.
Stereospecific Effects on Stability-CD and NMR spectra of D-Ser B8 -DKP-insulin closely resemble those of the parent DKP-insulin monomer; (14). The pattern of inter-residue NOEs is similar to that of DKP-insulin; differences in chemical shift are confined to the immediate neighborhood of the D side chain. The enhanced thermodynamic stability evident by resistance to denaturation in concentrated solutions of guanidine-HCl is likely to reflect in part stabilization of a nascent native-like B7-B10 turn in the unfolded state, which would reduce the entropic penalty of folding. The increased stability of D-Ser B8 -DKP-insulin (⌬⌬G u 0.9 Ϯ 0.2 kcal/mol) is similar to that observed by Raleigh and co-workers (8) in studies of analogous D-Ala substitutions in unrelated globular domains.
The magnitude of chiral stabilization of D-Ser B8 -DKP-insulin (⌬⌬G u 0.9 Ϯ 0.2 kcal/mol) is less than that of D-Ala B8 -DKPinsulin (⌬⌬G u 1.5 Ϯ 0.1 kcal/mol; 6). This difference may arise in part from residue-specific effects in their respective unfolded states and in part from structural perturbations in the folded state. Among crystal structures in general Ser is more likely than Ala to exhibit positive angles, suggesting that in an unfolded polypeptide D-Ser would be less effective than D-Ala in constraining the angle to the right hand side of the Ramachandran plot. Nonetheless, such subtle residue-specific effects in the unfolded state would seem insufficient to account for the 0.6 kcal/mol difference between the stabilities of D-Ala B8 and D-Ser B8 analogs. Indeed, because complete exclusion of the right side of the Ramachandran plane would be associated with an entropic contribution of only RT ln (2) (i.e. 0.4 kcal/mol at 20°C), any D-Ala/D-Ser difference would be only a small fraction of this.
We favor the hypothesis that D-Ala B8 -DKP-insulin gains additional stability in the folded state from favorable van der Waals interactions between the D-methyl group and surrounding aliphatic and aromatic side chains: D-Ala B8 partially inserts within a local non-polar pocket formed by Val A3 , Leu B11 , Val B12 , and Tyr B26 (6). Such favorable packing is partially disrupted by the polar D-Ser side chain. Effects of local structure may be further modulated by differences in solvation free energy near the B8 D side chain. These considerations indicate that the net effect of D amino acid substitutions in a globular protein is likely to reflect both general chiral restriction of the main chain angle and the residue-specific side chain environment.
Stereospecific Effects of Protein Dynamics-CD and NMR spectra of L-Ser B8 -DKP-insulin seem at first glance to have contradictory implications. On the one hand, the CD spectrum of the L analog is attenuated relative to that of DKP-insulin, suggesting a decrease in helix content. On the other hand, the endpoints of its three helical segments, as defined by strings of local and medium range NOEs, are unchanged. This seeming paradox is resolved by consideration of the physical origins of these respective spectra and by detailed analysis of the network of helix-related NOEs. CD provides an ensemble average of the mean helix content. Such features can be attenuated by either segmental changes in conformation or dynamic perturbations. An example of the former is the segmental unfolding of insulin in response to a cavity-forming mutation Ile A2 3 Ala: this substitution in DKP-insulin leads to disorder of the N-terminal segment of the A-chain (15,22). In this case the attenuated CD helix content is in accord with discrete loss of helix-related NOEs.
A contrasting example of dynamic CD attenuation is provided by the apparent increase in helix content on insulin assembly (23). Although NMR-derived helical endpoints in an isolated monomer are consistent with crystal structures of dimers and hexamers (1), the attenuated helical CD signature of the monomer is likely to reflect enhanced conformational fluctuations. Physical evidence for such fluctuations and their damping on assembly has been provided by analysis of helix-specific Raman vibrational band widths (34). Such damping of conformational fluctuations is likely to underlie the anomalous accentuation of helical CD features observed on substitution of Gly B8 by D-Ser. Although the D analog likewise exhibits identical NMR-defined helical segments, such accentuation implies a non-local coupling between fluctuations in the B8-related ␤-turn and fluctuations in (at least one) ␣-helix. We speculate that D-Ser B8 constrains local unfolding of the adjoining B9 -B19 ␣-helix; transmitted effects to the A-chain cannot be excluded.
Although L-Ser B8 -DKP-insulin retains a native-like overall fold and helical substructure, its NOESY spectrum contains fewer helix-related and long range contacts than are observed in DKP-insulin or the D-Ser B8 analog. In addition, chemicalshift perturbations are observed throughout the molecule. These findings strongly suggest that its attenuated CD spectrum, perturbations essentially mirror image to those of the D analog, reflects conformational excursions leading to transient distortions in helical main chain geometry. Because helical elements in insulin function in receptor recognition (1,21,(35)(36)(37), we imagine that helical destabilization would in itself be expected to impair binding. The substantial receptor binding activity of D-Ser B8 -DKP-insulin may therefore be retained despite such transmitted perturbations.
We suggest that the high affinity of L-Ser B8 -DKP-insulin, a seeming paradox in light of its instability and perturbed CD spectrum, reflects the net result of favorable and unfavorable factors. In this model destabilization of the native B7-B10 ␤-turn would enhance receptor binding whereas transmitted destabilization effects of helices would decrease binding. Although local effects of the D side chain at the hormone-receptor interface cannot be excluded, the 10-fold better binding of D-Ser B8 -DKP-insulin relative to D-Ala B8 -insulin (6) suggests that D-Ser B8 is locally well tolerated. Although it is formally possible that the side chains of D substitutions incur steric clash at the receptor interface, this possibility seems unlikely in light of the 5-fold better binding of a bulky D amino acid side chain (D-para-amino-Phe B8 ) than D-Ala B8 (6). Conversion of this side chain to the photoactivable analog D-paraazido-Phe results in negligible photo-cross-linking to the insulin receptor, 4 in striking contrast to the efficient photo-cross-linking of such insulin derivatives in the classical receptor-binding surface (36 -38). The respective 100-fold and 1000-fold decrements in receptor binding exhibited by D-Ser B8 -DKP-insulin and D-Ala B8 -DKP-insulin are larger than those usually incurred by mutations associated with modest changes in side chain volume (supplemental material).
B8 Chirality and the TR Transition-Although the present study has focused on chiral mutagenesis of an engineered insulin monomer, our results are pertinent to crystallographic studies of zinc-insulin hexamers. Such assemblies exhibit a liganddependent equilibrium among T 6 , T 3 R 3 f , and R 6 hexamers (1, 39). In the T-state as in an engineered monomer in solution (3,14), the B-chain contains an extended N-terminal arm (residues B1-B6) and type II ␤-turn (B7-B10) followed by the central ␣-helix (B9 -B19). In the alternative R-state the B-chain contains an extended ␣-helix (residues B1-B19; Refs. 39 and 40). These structural differences are illustrated in the cylinder models shown in Fig. 1B. 5 Despite the extensive crystallographic characterization of zinc insulin hexamers, the relationship between these structures and the active conformation of insulin has long been the subject of speculation (1,6,31).
The TR transition is accompanied by a change in sign of the angle of Gly B8 from positive (in the T-state-specific ␤-turn) to negative (in the R-state-specific ␣-helix). D amino acids at B8 would thus favor the T-state whereas L substitutions would favor the R-state. Within zinc hexamers chiral substitutions would thus be expected to shift the equilibrium among T 6 , T 3 R 3 f , and R 6 hexamers in one direction or the other. Within the DKP-insulin monomer D-Ser B8 stabilizes a T-like confor-4 K. Huang, B. Xu, P. K. Katsoyannis, and M. A. Weiss, unpublished results. 5 The TR transition is also characterized by a change in the handedness of cystine A7-B7. The sulfur atoms of the latter are exposed in the T-state but buried in a nonpolar crevice in the R-state. We speculate that coupling between the B8 angle and handedness of cystine A7-B7 may account for the low yield of chain combination in synthesis of L-Ser B8 -DKP-insulin.
mation. Although L-Ser B8 destabilizes this conformation, no evidence of an R-like monomeric state is observed. In particular, minor helix-related NOEs in the B1-B8 segment diagnostic of a subpopulation of R-state conformers have not been detected. Further, the attenuated CD spectrum of L-Ser B8 -DKP-insulin is opposite to the increased helix content observed in CD studies of T 3 R 3 f , and R 6 hexamers (41). That similar attenuation occurs in the CD spectrum of L-Ala B8 -DKP-insulin (6) indicates that this effect is unrelated to the relative intrinsic helical propensities of alanine or serine (42).
The very low activities of D-Ser B8 and D-Ala B8 analogs, despite maintenance of a native T-like conformation of enhanced stability, strongly suggest that the T-state must undergo a conformational change on receptor binding. This does not imply, however, that the receptor-bound state of insulin resembles the R-state. It is possible, for example, that the conformation of the B7-B10 ␤-turn changes without an associated R-like helical transition in the B1-B6 segment. Indeed, residues B1-B4 may be deleted without significant change in receptor binding (43). We imagine that the TR transition exploits the intrinsic flexibility of Gly B8 . The structural reorganization of the zinc insulin hexamer may otherwise be unrelated to induced fit of the active monomer. Resolving this issue will require a co-crystal structure of a hormone-receptor complex.
The incomplete correspondence between the TR transition and the mechanism of receptor binding is further illuminated by studies of the single chain analog mini-proinsulin (4). This inactive analog exhibits native assembly and interconversion among T 6 , T 3 R 3 f , and R 6 hexamers (41). Further, its crystal structure reveals that the B29-A1 tether does not constrain or perturb either T or R protomers (4). The inactivity of miniproinsulin thus indicates that aspects of induced fit in the hormone-receptor complex must exceed or be unrelated to the TR transition. We and others have proposed that detachment of the C-terminal B-chain ␤-strand, locked into place within the dimer interface of zinc insulin hexamers, releases a receptor binding arm that exposes an otherwise hidden A-chain surface (31, 37, 44 -46). 6 We thus envisage that both the N-and C-terminal segments of the B-chain reorganize on receptor binding.
In summary, our results strongly suggest that the classical insulin T-state represents an inactive conformation of the hormone. By identifying a critical hinge point at Gly B8 , studies of mirror-image D and L substitutions provide evidence for induced fit on receptor binding. This approach, designated chiral mutagenesis, exemplifies a general strategy for the analysis of protein function through total chemical synthesis. The receptor-bound structure of insulin promises to enable design of novel agonists for the treatment of diabetes mellitus.