Enhancing the Activity of a Protein by Stereospecific Unfolding

A central tenet of molecular biology holds that the function of a protein is mediated by its structure. An inactive ground-state conformation may nonetheless be enjoined by the interplay of competing biological constraints. A model is provided by insulin, well characterized at atomic resolution by x-ray crystallography. Here, we demonstrate that the activity of the hormone is enhanced by stereospecific unfolding of a conserved structural element. A bifunctional β-strand mediates both self-assembly (within β-cell storage vesicles) and receptor binding (in the bloodstream). This strand is anchored by an invariant side chain (PheB24); its substitution by Ala leads to an unstable but native-like analog of low activity. Substitution by d-Ala is equally destabilizing, and yet the protein diastereomer exhibits enhanced activity with segmental unfolding of the β-strand. Corresponding photoactivable derivatives (containing l- or d-para-azido-Phe) cross-link to the insulin receptor with higher d-specific efficiency. Aberrant exposure of hydrophobic surfaces in the analogs is associated with accelerated fibrillation, a form of aggregation-coupled misfolding associated with cellular toxicity. Conservation of PheB24, enforced by its dual role in native self-assembly and induced fit, thus highlights the implicit role of misfolding as an evolutionary constraint. Whereas classical crystal structures of insulin depict its storage form, signaling requires engagement of a detachable arm at an extended receptor interface. Because this active conformation resembles an amyloidogenic intermediate, we envisage that induced fit and self-assembly represent complementary molecular adaptations to potential proteotoxicity. The cryptic threat of misfolding poses a universal constraint in the evolution of polypeptide sequences.

How insulin binds to the insulin receptor (IR) 2 is not well understood despite decades of investigation. The hormone is a globular protein containing two chains, A (21 residues) and B (30 residues) (Fig. 1A). In pancreatic ␤-cells, insulin is stored as Zn 2ϩ -stabilized hexamers (Fig. 1B), 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. The monomer is proposed to undergo a change in conformation upon receptor binding (2). In this study, we investigated a site of conformational change in the B-chain (Phe B24 ) (arrow in Fig.  1A). In classical crystal structures, this invariant aromatic side chain (tawny in Fig. 1B) anchors an antiparallel ␤-sheet at the dimer interface (blue in Fig. 1C). Total chemical synthesis is exploited to enable comparison of corresponding D-and Lamino acid substitutions at this site, an approach designated "chiral mutagenesis" (3)(4)(5). In the accompanying article, the consequences of this conformational change are investigated by photomapping of the receptor-binding surface (6). Together, these studies redefine the interrelation of structure and activity in a protein central to the hormonal control of metabolism.
The structure of an insulin monomer in solution resembles a crystallographic protomer ( Fig. 2A) (7)(8)(9). The A-chain contains an N-terminal ␣-helix, non-canonical turn, and second helix; the B-chain contains an N-terminal segment, central ␣-helix, and C-terminal ␤-strand. The ␤-strand is maintained in an isolated monomer wherein the side chain of Phe B24 (tawny in Fig. 2A), packing against the central ␣-helix of the B-chain, provides a "plug" to seal a crevice in the hydrophobic core (Fig. 2B). Anomalies encountered in previous studies of insulin analogs suggest that Phe B24 functions as a conformational switch (4, 7, 10 -14). Whereas L-amino acid substitutions at B24 generally impair activity (even by such similar residues as L-Tyr) (15), a seeming paradox is posed by the enhanced activities of nonstandard analogs containing D-amino acids (Table 1) (10 -12).
Why do D-amino acid substitutions at B24 enhance the activity of insulin? In this study, we describe the structure and function of insulin analogs containing L-Ala or D-Ala at B24 (Fig. 2, C and D). Our studies were conducted within an engineered monomer (DKP-insulin, an insulin analog containing three substitutions in the B-chain: Asp B10 , Lys B28 , and Pro B29 ) to circumvent effects of self-assembly (16). Whereas the inactive L-analog retains a native-like structure, the active D-analog exhibits segmental unfolding of the B-chain. Studies of corresponding analogs containing either L-or D-photoactivable probes (L-para-azido-Phe B24 or D-para-azido-Phe B24 (L-or D-Pap B24 ), obtained from photostable para-amino-Phe (Pmp) precursors (17)) demonstrate specific cross-linking to the IR. Although photo-contacts map in each case to the N-terminal domain of the receptor ␣-subunit (the L1 ␤-helix), higher cross-linking efficiency is achieved by the D-probe. Together, this and the following study (6) provide evidence that insulin deploys a detachable arm that inserts between domains of the IR.
Induced fit of insulin illuminates by its scope general principles at the intersection of protein structure and cell biology. Protein evolution is enjoined by multiple layers of biological selection. The pathway of insulin biosynthesis, for example, successively requires (a) specific disulfide pairing (in the endoplasmic reticulum), (b) subcellular targeting and prohormone processing (in the trans-Golgi network), (c) zinc-mediated protein assembly and microcrystallization (in secretory granules), and (d) exocytosis and rapid FIGURE 1. Sequence and structure of insulin. A, sequences of the B-chain (upper) and A-chain (lower) with disulfide bridges as indicated. The arrow indicates invariant Phe B24 . The B24 -B28 ␤-strand is highlighted in blue. B, crystal structure of the T 6 zinc insulin hexamer (Protein Data Bank code 4INS): ribbon model (left) and space-filling model (right). The B24 -B28 ␤-strand is shown in blue, and the side chain of Phe B24 is highlighted in tawny. The B-chain is otherwise dark gray; the A-chain, light gray; and zinc ions, magenta. Also shown at the left are the side chains of His B10 at the axial zincbinding sites. C, cylinder model of the insulin dimer showing the B24 -B26 antiparallel ␤-sheet (blue) anchored by the B24 side chain (tawny circle). The A-and B-chains are shown in light and dark gray, respectively. The protomer at the left is shown in the R-state, in which the central ␣-helix of the B-chain is elongated (B3-B19 in the frayed R f protomer of T 3 R f 3 hexamers and B1-B19 in the R protomer of R 6 hexamers). The three types of zinc insulin hexamers share similar B24 -B26 antiparallel ␤-sheets as conserved dimerization elements. FIGURE 2. Role of Phe B24 in an insulin monomer. A, shown is a cylinder model of insulin as a T-state protomer. The C-terminal B-chain ␤-strand is shown in blue, and the Phe B24 side chain is shown in tawny. The black portion of the N-terminal A-chain ␣-helix (labeled buried) indicates a hidden receptorbinding surface (Ile A2 and Val A3 ). B, the schematic representation of insulin highlights the proposed role of the Phe B24 side chain as a plug that inserts into a crevice at the edge of the hydrophobic core. C and D, whereas substitution of Phe B24 by L-Ala (C) would only partially fill the B24-related crevice, its substitution by D-Ala (D) would be associated with a marked packing defect. An alternative conformation, designated the R-state, is observed in zinc insulin hexamers at high ionic strength (74) and upon binding of small cyclic alcohols (75) but has not been observed in an insulin monomer. disassembly of insulin hexamers (in the portal circulation), in turn enabling binding of the monomeric hormone to target tissues (1). Each step imposes structural constraints, which may be at odds. This study demonstrates that stereospecific pre-detachment of a receptor-binding arm enhances biological activity but impairs disulfide pairing and renders the hormone susceptible to aggregation-coupled misfolding (18). Whereas the classical globular structure of insulin and its self-assembly prevent proteotoxicity (3,19), partial unfolding enables receptor engagement. We envisage that a choreography of conformational change has evolved as an adaptative response to the universal threat of toxic protein misfolding.

EXPERIMENTAL PROCEDURES
Synthesis of Insulin Analogs-Human insulin was obtained from Lilly and Novo Nordisk (Copenhagen, Denmark). Synthesis of variant B-chains, insulin chain combination, and protein purification were performed as described (Refs. 14 and 20; see also supplemental "Experimental Procedures"). B-chain analogs each contained three "DKP" substitutions to prevent self-association of insulin (His B10 3 Asp, Pro B28 3 Lys, and Lys B29 3 Pro) (8,16). Whereas the yield of L-Pmp B25 -DKP-insulin was similar to that of DKPinsulin, the yield of L-Ala B24 -DKP-insulin was reduced by 2-fold; the yields of D-Ala B24 -DKP-insulin and D-Pmp B24 -DKP-insulin were reduced by 4-fold. Precursor Pmp analogs were converted to their azido form as described (17,20). Pmp/Pap analogs were prepared with a B1-linked biotin tag to enable detection by an avidinbased reagent (NeutrAvidin, Pierce).
Receptor Binding Assays-Activities of insulin analogs were evaluated by a competitive displacement assay using a human placental membrane preparation (21); cross-binding to the type I insulin-like growth factor receptor was evaluated as described (22). The percentage of tracer bound in the absence of competing ligand was in each case Ͻ15% to avoid ligand depletion artifacts. Assays were performed in triplicate.
Circular Dichroism-Far-UV CD spectra were obtained as described (23). Spectra were normalized by mean residue ellipticity. Samples were dissolved in 10 mM potassium phosphate (pH 7.4) and 100 mM KCl at a protein concentration of ϳ25 M. For equilibrium denaturation studies, samples were diluted in the same buffer to 5 M; guanidine HCl was employed as denaturant (23). Data were obtained at 25°C and fitted by nonlinear least squares to a two-state model (24). 1 H NMR Spectroscopy-Spectra were obtained at 600 and 800 MHz in aqueous solution at pH 7-8 at 25 and 32°C and in 20% deuterioacetic acid at 25°C as described (4,25). Distance geometry and restrained molecular dynamics calculations were performed as described (20).
Photocross-linking Studies-Wheat germ-agglutinin-purified IR ectodomain and holoreceptor were prepared as described (26,27); methods are provided under supplemental "Experimental Procedures." Photocross-linking of biotin-labeled Pap analogs to the isolated ectodomain and solubilized IR (isoform B) was induced by UV irradiation (17); photoproducts were characterized by SDS-PAGE and Western blotting as described (20,26,27). Gels were probed with NeutrAvidin to detect the B1-linked biotin tag and with polyclonal antiserum recognizing the N-terminal segment of the IR ␣-subunit (designated IR ␣ -N; Santa Cruz Biotechnology). Control experi-ments verifying specificity were performed to demonstrate competition between binding of the Pap analogs and native ligands (human insulin and insulin-like growth factor I) (27).
Insulin Fibrillation-Proteins were made 60 M in degassed phosphate-buffered saline (10 mM phosphate and 140 mM NaCl at pH 7.4) with 0.1% sodium azide in pre-sterilized glass vials with airtight sealed caps (Allergy Laboratories Inc.). Vials were rocked on a BD Biosciences nutator at ϳ60 rpm as described (28). At successive times, aliquots were withdrawn with a sterilized singleuse syringe. Aliquots were added to a thioflavin T solution for fluorescence assay. Emission spectra were collected in 1-cm quartz cuvettes from 470 to 500 nm following excitation at 450 nm; the integration time was 1 s. Assays were performed in duplicate (D-Ala B24 -DKP-insulin) or triplicate (Asp B10 -des-octapeptide-(B23-B30)-insulin, DKP-insulin, and its L-Ala B24 analog).

RESULTS
Analogs of DKP-insulin containing L-amino acid substitutions at B24 ("L-analogs") exhibit reduced receptor-binding affinities (relative to the parent monomer), whereas D-analogs exhibit enhanced affinities ( Fig. 3 and Table 2). No disproportionate changes were observed in the extent of low affinity cross-binding to the type I insulin-like growth factor receptor. The far-UV CD spectra of L-Ala B24 -DKP-insulin and D-Ala B24 -DKP-insulin are each similar to that of DKP-insulin (supplemental Fig. S1A). Protein denaturation studies (supplemental Fig. S1B) indicated that the L-and D-Ala B24 analogs are less stable than the parent monomer, with similar decrements in unfolding free energy (⌬⌬G u ϭ 0.8 Ϯ 0.2 and 0.7 Ϯ 0.2 kcal/ mol, respectively) ( Table 2 and supplemental Table S1).
Solution Structures-The 1 H NMR spectra of D-and L-Ala B24 analogs each exhibit resonance line widths (in aqueous solution at neutral pH) similar to those of DKP-insulin (8,16), suggesting that the analogs are likewise monomeric. The variant spectra exhibit attenuated secondary shifts consistent with the absence of the Phe B24 ring current (29); complete resonance assignment was in each case obtained (supplemental Tables  S2-S5). D-and L-analogs exhibit native-like patterns of helixrelated nuclear Overhauser effects (NOEs) within canonical ␣-helical segments (residues A2-A7, A13-A19, and B9 -B19) and the B7-B10 ␤-turn ( The aromatic resonances of Phe B25 and Tyr B26 provide probes for the packing of this segment against the ␣helical subdomain. Whereas the L-analog exhibits a native-like pattern of long-range NOEs (Fig. 4A), few such contacts are observed in the D-analog (Fig. 4B). Of particular importance is the L-specific maintenance of an NOE between Tyr B26 and the methyl resonances of Ile A2 and Val A3 in the N-terminal A-chain ␣-helix (cross-peaks a, k (B26-A2), and e (B24-A3) in Fig. 4, A and C; see also supplemental Table  S6). Attenuation of these and related NOEs in D-Ala B24 -DKP-insulin provides evidence for ste-  3 ; y, B1-H ⑀ /B6-H␦ 2 CH 3 ; z, B1-H ⑀ /B6-H␦ 1 CH 3 ; ␣, B1-H ⑀ /A13-H␦ 2 CH 3 ; ␤, B1-H ⑀ /A13-␦ 1 CH 3 ; and ␥, B1-H ⑀ /B18-␥CH 3. B, corresponding spectrum of the D-Ala B24 analog. Cross-peaks are as follows: a Relative affinity is based on studies of binding to a human placental membrane preparation (21). Numbers in parentheses refer to replicates. The hormone-receptor dissociation constant under assay conditions is (5.0 Ϯ 0.4) ϫ 10 Ϫ10 M. b Thermodynamic stability is based on CD-detected guanidine denaturation studies at 25°C (24). c The enhanced stability of DKP-insulin relative to insulin is due to substitution His B10 3 Asp (70). This substitution blocks the trimer interface and provides a more favorable N-Cap (i.e. the N-terminal residue of an ␣-helix) for the central ␣-helix; it is also responsible for the enhanced activity of the analog (16,71). The two substitutions in the C-terminal ␤-strand (Pro B28 3 Lys and Lys B29 3 Pro), which impair classical dimerization (72), are destabilizing (73).  Fig. 4, A and C), at neutral pH, the corresponding NOEs are absent in the D-analog. The NOESY spectrum of D-Ala B24 -DKP-insulin in an organic co-solvent (20% deuterioacetic acid), although otherwise native-like, exhibits weak B26-related NOEs (cross-peaks (c) and (d) in Fig. 4D) in association with weak non-native B25-B12 and B25-B16 NOEs (arrow), suggesting transient contacts by either aromatic ring within a disordered C-terminal segment.
Structures were calculated by distance geometry and restrained molecular dynamics on the basis of ϳ700 restraints (15 restraints/residue); statistical parameters are provided in supplemental Table S7. The structure of DKP-insulin (Fig. 5B, Phe B24 in black) recapitulates the conformation of crystallographic protomers and so provides a native base line (Fig. 5A) (8). The analogs exhibit marked differences in the B-chain. Whereas the B-chain of the L-analog recapitulates a native-like (but less well defined) U-shaped supersecondary structure (Fig.  5C), residues B20 -B30 in the D-analog are essentially unrestrained (Fig. 5D). In the distance geometry and restrained molecular dynamics ensemble, the latter segment is depicted by a near-random conformational sampling; such modeling may overestimate the extent of conformational excursions.
L-and D-analogs each retain native-like ␣-helical subdomains. Subtle differences are nonetheless observed between the ␣-helical orientations (Fig. 5, C and D). Whereas almost all helix-related and interhelical NOEs are shared between analogs, a small number of differences are observed (Fig. 4, C and D); these appear unrelated to the relative extent of resonance overlap. Such differences presumably reflect transmitted effects of the B24 substitutions on segmental structure and dynamics. Because aliphatic side chains in the ␣-helical subdomain (Ile A2 , Val A3 , Ile B11 , Val B12 , and Leu B15 ) ordinarily engage the C-terminal B-chain segment, the D-analog exposes associated hydrophobic surfaces. Mutagenesis suggested that these surfaces participate in receptor binding (30). Photocross-linking Studies-B1-biotin-labeled L-and D-photoactivable B24 derivatives of DKP-insulin were prepared by chemical synthesis; the corresponding derivative at B25 was prepared as a control (17,26). Fidelity of synthesis and conversion of photostable Pmp (amino) precursors to Pap (azido) derivatives were verified by electrospray mass spectrometry. Pap analogs and the IR ectodomain (Fig. 6A) were mixed at 1:1 stoichiometry at protein concentrations (ϳ200 nM) Ͼ100-fold higher than the weakest dissociation constant of the photostable precursors; the predominant species is thus expected to be a 1:1 bimolecular complex. A short UV exposure time (20 s at 254 nm) yielded essentially complete photolysis or cross-linking (17,26). Whereas the L-para-azido-Phe B24 analog was found to exhibit a photocross-linking efficiency (Fig. 6B,   ing efficiency was at least 1.5-fold higher (lane 8). Similar results were obtained using the holoreceptor (see below). Control blots demonstrated equal loading of the ectodomain (Fig.  6B, middle panel) and insulin derivative (lower panel). As expected, in the absence of the ectodomain, photocross-linked complexes were not observed either before (lanes 1, 5, and 9) or after (lanes 2, 6, and 10) UV irradiation; covalent complexes were likewise not formed in the presence of the ectodomain but in the absence of irradiation (lanes 3, 7, and 11).
Mapping of photoproducts by limited proteolysis was undertaken in corresponding holoreceptor complexes (26). The structural basis of such mapping within the "inverted V" (␣␤) 2 -dimer is illustrated in Fig. 7 (A-D). The results indicate that D-Pap B24 photocross-links to the L1 domain of the IR ␣-subunit (Fig. 7) as found previously for the L-para-azido-Phe B24 derivative (26). The mapping strategy exploited exposed basic side chains near the junction of L1 and the cysteine-rich (CR) domain (magenta arrowhead in Fig. 7A); the positions of these side chains are peripheral to the putative hormone-binding interface of the ectodomain (Fig. 7, C and D). Limited tryptic cleavage of the photocross-linked complexes yielded a predominant 31-kDa glycosylated fragment (Fig. 7E) recognized by antiserum against the N-terminal 20 residues of the ␣-subunit (IR ␣ -N) (Fig. 7F). Upon deglycosylation, the apparent mass of this fragment is 20 kDa, indicating that it contains the L1 domain (residues 1-158) and at most a small portion of the CR domain.
Insulin Fibrillation-Partial unfolding of an insulin monomer at elevated temperatures (Ͼ50°C) promotes formation of amyloid (Fig. 8A) (31)(32)(33)(34). Such fibrillation occurs via a nucleation-growth mechanism with a lag phase (leading to an amyloidogenic nucleus) and log phase (leading to formation of mature fibrils) (18). Analogous nucleation-growth mechanics underlie deposition of pathological amyloid (19), suggesting that the susceptibility of insulin to fibrillation provides a model for proteotoxicity. Partial thermal unfolding of insulin leads to destabilization of the C-terminal segment of the B-chain (purple dashed line in Fig. 8B) (35). Because this partial fold is reminiscent of the conformation of D-Ala B24 -DKP-insulin, we investigated whether the analog might exhibit a foreshortened lag time. Fibrillation was monitored by thioflavin T fluorescence upon gentle agitation of the protein solution in phosphate-buffered saline at pH 7.4 at 37°C. Under such conditions, DKP-insulin forms fibrils in 12.4 Ϯ 2.5 days. The lag time of the L-Ala B24 analog is reduced by ϳ40% (7.0 Ϯ 0.7 days); that of the D-Ala B24 analog is more markedly accelerated (between 2 and 3 days).
These observations demonstrate that L-and D-Ala B24 -DKPinsulin are each more susceptible to fibrillation than the parent monomer. Although this in accord with their reduced global stabilities, the difference between L-and D-analogs correlates with the extent to which hydrophobic surfaces (concealed in native insulin by the C-terminal B-chain segment) are aberrantly exposed. Accelerated fibrillation of the D-analog is in accord with the enhanced amyloidogenicity of insulin analogs containing C-terminal B-chain deletions (18). Under the present conditions, the truncated analog Asp B10 -des-octapeptide(B23-B30)-insulin (obtained by tryptic removal of the C-terminal segment) exhibits a lag time of 4.3 Ϯ 1.5 days. Although this fragment (shorn of its receptor-binding arm) is without biological activity, NMR studies indicate maintenance of a native-like ␣-helical domain similar to that of D-Ala B24 -DKP-insulin.

DISCUSSION
Crystal structures of insulin have been determined at atomic resolution, enabling detailed analyses (30,36,37). Studies of single-chain insulin analogs nonetheless suggest that such structures depict inactive conformations (supplemental Fig.  S3) (38, 39). Single-chain insulin analogs contain short con-  1-4), B24 (D-chirality) (lanes 5-8), or B24 (L-chirality) (lanes 9 -12) treated without (odd-numbered lanes) or with (evennumbered lanes) UV irradiation. Analogs have a biotin tag at B1. After crosslinking, reduction with dithiothreitol (upper and middle panels), SDS-PAGE separation, and blotting onto nitrocellulose membranes, cross-linked adducts were probed with alkaline phosphatase-conjugated NeutrAvidin (NAv; upper panel). Control blots probed with anti-IR ␣-subunit antibody (IR ␣ -N; middle panel; after dithiothreitol reduction) demonstrate equal amounts of IR. Similarly, control blots probed with NeutrAvidin (lower panel; without dithiothreitol reduction) demonstrate equal amounts of insulin analog. Control lanes, with any of the three Pap derivatives, a photocross-linked band was not detected in the absence of the IR and UV irradiation (lanes 1, 5, and 9), in the absence of the IR and in the presence of UV irradiation (lanes 2, 6, and 10), or in the presence of the IR and in the absence of UV irradiation (lanes 3, 7, and 11).
necting segments that tether the C terminus of the B-chain to the N terminus of the A-chain (38 -43). Although structurally well tolerated, tethers of less than four residues markedly impair receptor binding (8,40,44). Activity is partially restored with connecting domains of six or more residues (39,45,46). Similar length-dependent effects have been observed on interposing chemical cross-linking reagents between the B-chain (using the ⑀-amino group of Lys B28 ) and the N terminus of the A-chain (47). These findings suggest that the C-terminal B-chain ␤-strand (residues B24 -B30; highlighted in blue in Fig. 1) reorganizes to contact the IR (7,11,38). In this study, we have reported that stereospecific detachment of the C-terminal segment of the B-chain enhances receptor binding; D-and L-photoprobes at the site of substitution cross-link efficiently to the IR. The accompanying article provides evidence for the insertion of the B-chain ␤-strand between domains of the IR (6).
Conformational Switch-Evidence for a receptor-directed conformational change at B24 has been previously provided by structural studies of Gly B24 analogs (7,13). This non-conservative substitution leads to only a small decrease in activity (11,12,15) and yet is associated with segmental or local destabilization of the B-chain (7,13). The extent of perturbation has been the subject of debate. 3 That the packing of the B24 -B28 segment in classical structures represents an inhibitory conformation is suggested by the marked disparity between the effects of substitutions in the intact hormone versus truncated analogs. Whereas substitution of L-Phe B24 by D-Phe in native insulin enhances activity, for example, the same substitution in the truncated analog des-pentapeptide(B26 -B30)-insulin-amide (itself equipotent with insulin) (48,49) markedly impairs receptor binding (50). Converse findings have been observed upon substitution of Tyr B26 by D-Ala: low activity in native insulin but enhanced activity in des-tetrapeptide(B27-B30)-insulin-amide (51). Receptor-directed reorganization of insulin between closed (inactive) and open (active) conformations provides a coherent framework for interpretation of these findings.
Detachment of residues B20 -B30 in D-Ala B24 -DKP-insulin augments the accessibility of nonpolar side chains in the classical receptorbinding region (52); these include Ile A2 , Val A3 , Val B12 , and Tyr B16 in the ␣-helical domain and Phe B24 , Phe B25 , and Tyr B26 in the C-terminal segment itself. Substitutions at these sites markedly impair the activity of insulin (15,30,47,53) despite (in several cases) negligible non-local structural perturbations (9,14,20,54,55). Compelling examples are provided by substitution of Ile A2 by allo-Ile (i.e. inversion of ␤-carbon chirality) (14,47,56) and substitution of Val A3 by Leu, a clinical mutation associated with diabetes mellitus (56 -58). We suggest that upon binding of the wild-type hormone to the IR, the insulin monomer undergoes analogous segmental detachment and receptor-dependent refolding, likewise enabling close engagement of underlying nonpolar surfaces. Direct evidence is provided by synthetic photoscanning in the accompanying article (6). We imagine that the activity of D-Ala B24 -DKPinsulin, although high, would be even higher but for a trade-off: the benefit provided by destabilization of an inhibitory conformation is offset by the cost of refolding the C-terminal B-chain segment upon receptor binding. Similarly, because truncated analogs lack part of the inhibitory segment, the low affinity of D-Phe B24 -des-pentapeptide(B26 -B30)-insulin-amide (see above) (50) is presumably due to the cost of refolding incurred in the absence of an offsetting benefit. The low activity of L-Ala B24 -DKP-insulin (despite partial destabilization of its C-terminal arm) may reflect local perturbation of the B24-receptor contact, as packing of L-Phe B24 against the L1 ␤-helix ordinarily acts as a trigger for further conformational change (11,15).
Other Sites of Induced Fit-Evidence that the T-state-specific B7-B10 ␤-turn undergoes a change in conformation upon receptor binding has also been obtained through chiral mutagenesis. D-Amino acid substitutions within the turn (mimicking the positive dihedral angle of Gly B8 ) stabilize insulin but markedly impair receptor binding (3). L-Amino acid substitutions at B8 are destabilizing but can be compatible with high  activities (4). The photocross-linking studies presented in the following article suggest that such impairment is unlikely to represent steric clash by the introduced D-side chain and so indicates an additional site of induced fit (6). Induced fit of the N-terminal segment of the B-chain is in accord with classical allostery among zinc insulin hexamers (the TR transition; see Fig. 8 in the accompanying article (6)) (30). Indeed, in this transition, the conformation of Gly B8 changes from D-like ( Ͼ 0) to L-like ( Ͻ 0). Furthermore, long-range coupling between sites of conformational change at B8 and B24 would be consistent with a subtle feature of the TR transition: the salient R-state-specific change in conformation of the N-terminal segment of the B-chain (including in the sign of the B8 angle) is associated with slight separation of the C-terminal B-chain ␤-strand from the A-chain, in turn breaking a T-state-specific hydrogen bond between Phe B25 -H N and Tyr A19 -CϭO (30). Thus, although frank detachment of the ␤-strand is constrained by assembly (i.e. by the dimer-related ␤-sheet shared among T 6 , T 3 R f 3 , and R 6 structural families) (30), conformational variability among zinc insulin hexamers foreshadows molecular mechanisms of induced fit upon receptor binding at both ends of the B-chain.
Determinants of Disulfide Pairing-Insulin chain combination (air oxidation of isolated A-and B-chains) (59) yields native disulfide pairing and thus provides a peptide model of proinsulin folding (60). The reaction is under kinetic control and limited by off-pathway reactions (formation of cyclic A-and B-chains, B-chain polymers, and B-chain fibrils) (23,61,62). Although chain combination is generally robust to amino acid substitutions (23), we have found that substitution of Phe B24 by D-or L-Ala caused a 4-or 2-fold reduction in yield, respectively. The reactions exhibited proportionate increases in side products rather than formation of insulin disulfide isomers (61,62). Such impaired yields are unlikely to reflect decreased end-product stability because D-and L-Ala B24 -DKP-insulin (although less stable than DKP-insulin) are as stable as wild-type insulin ( Table 2). The absence of correlation between synthetic yield and thermodynamic stabilities among these and other analogs (23) suggests that Phe B24 plays a kinetic role in directing disulfide pairing. In accord with this hypothesis, the B24 aromatic ring participates in nascent native-like B-chain supersecondary structure in partially folded insulin analogs lacking either cystine A6 -A11 or A7-B7, constructed as models of oxidative folding intermediates (8,63,64). Of the three native disulfide bridges, Phe B24 is close to only cystine A20-B19.
We propose that in an initial stage of chain combination, Phe B24 -related hydrophobic clustering favors the alignment of Cys B19 and Cys A20 for disulfide pairing. Upon substitution of Phe B24 by D-or L-Ala, such clustering is presumably less efficient. It would be of interest to investigate whether B24 substitutions likewise affect the foldability of proinsulin in mammalian cells (25). 4 The importance of structural interactions flanking disulfide bridges for the cellular folding of proinsulin is highlighted by an emerging data base of mutations causing diabetes as discussed in the following article (6). As a consequence of induced fit, such interactions may be dispensable in the mature hormone once folding has been achieved (25). Foldability thus imposes an evolutionary constraint that is, in principle, independent of receptor binding.
Self-assembly and Proteotoxicity-The classical "closed" conformation of insulin mediates its hexameric assembly (Fig. 8A,  left), which in ␤-cells provides a stable storage form of the hormone within secretory granules (1). Such assembly is proposed to protect the ␤-cell from proteotoxicity due to aggregationcoupled misfolding (1,3) and in turn to formation of fibrils (Fig.  8B, right) (18). Insulin fibrillation is delayed by native self-assembly and promoted by partial unfolding (33). Such fibrillation leads to canonical cross-␤-structure (as observed in diverse pathological amyloid deposits) and so may be regarded as an assay for toxic protein misfolding (19).
Phe B24 packs within the dimer-related ␤-sheet and so has a dual function in self-assembly and receptor binding (30). We have found that substitution of Phe B24 by L-or D-Ala reduces the lag time prior to fibrillation (relative to DKPinsulin). Although these analogs exhibit similar global stabilities, the onset of fibrillation is more rapid in the D-Ala analog. We envisage that segmental unfolding and detachment of the C-terminal segment of the B-chain expose underlying hydrophobic surfaces that in turn participate in formation of an amyloidogenic nucleus (Fig. 8A) (34,35). That these or related surfaces participate in receptor binding suggests that the susceptibility of the wild-type hormone to fibrillation, apparently conserved among vertebrate insulins (18), is intrinsic to its mechanism of action (28). The interrelation between conformational distortion of the insulin monomer and its receptor-bound structure is illustrated in schematic form in Fig. 8A (middle).
An inhibitory extension of the B-chain may have evolved to reduce the propensity of an active monomer to undergo aggregation-coupled misfolding with its potential proteotoxicity (65). This hypothesis is supported by the increased susceptibility of truncated insulin analogs to fibrillation (18). Similarly, substitution of flanking Gly B23 by Ala (5), a perturbation of the B20 -B23 ␤-turn that presumably redirects the C-terminal segment, accelerates fibrillation to an extent similar to that of L-Ala B24 . 5 Conversely, immobilization of the C-terminal segment of the B-chain in its inhibitory conformation within inactive single-chain insulin analogs (40,41) confers marked protection against fibrillation (18,28). These principles may provide a foundation for the design of thermal fibrillation-resistant single-chain analogs for clinical use in challenged regions of the developing world (46).
Although of major concern in pharmaceutical chemistry (18), insulin fibrillation rarely occurs in vivo 6 and is not ordi-narily present in human amyloid deposits. As a seeming paradox, the absence of pathological fibrillation may highlight its underlying importance in the evolution of insulin-like sequences. Diverse site-directed mutations in insulin promote fibrillation in vitro, whereas protective substitutions are seldom encountered (33,34). Like the dog that did not bark in the nighttime, 7 the general avoidance of amyloidogenic variants among vertebrate insulin sequences implies a selective advantage to their exclusion. Resistance to fibrillation defines a biological constraint that is in principle independent of receptor binding.
Concluding Remarks-The total chemical synthesis of proteins facilitates incorporation of nonstandard amino acids as probes of structure-activity relationships (66). Of particular interest is the inversion of chirality at potential sites of conformational change (3,5,67). Whereas D-amino acid substitutions may be employed to stabilize a ␤-turn (testing its contribution to folding or function) (4), within an ␣-helix or ␤-strand such substitutions are likely to be destabilizing. In this study, we have exploited such stereospecific perturbation to illuminate the role of an invariant aromatic side chain (Phe B24 ) in folding, activity, and misfolding. We propose that the structure of the distorted D-analog foreshadows the wild-type mode of receptor recognition and so enables enhancement of biological activity. In the following article, this model is tested by systematic photocross-linking studies (6). The resulting "photoscan" provides evidence that detachment of the C-terminal segment of the B-chain leads to its insertion between receptor domains.
Insulin has long provided a model for analysis of protein structure and function. The results of this study of chiral substitutions at B24 suggest that induced fit upon receptor binding represents the molecular adaptation of a globular protein to multiple competing biological constraints. The invariance of Phe B24 among vertebrate insulins is presumably enjoined by its complementary roles in folding, protective self-assembly, and signaling. We envisage that partial unfolding extends the accessible receptor-binding surface but also exacerbates the risk of non-native aggregation. The striking resemblance between the proposed active conformation of insulin and the partial fold of an amyloidogenic intermediate rationalizes the general susceptibility of vertebrate insulins to fibrillation. The complex conformational life cycle of insulin from biosynthesis to receptor binding thus highlights the cryptic role of toxic misfolding as a universal constraint in the evolution of polypeptide sequences.