The A-chain of Insulin Contacts the Insert Domain of the Insulin Receptor

The contribution of the insulin A-chain to receptor binding is investigated by photo-cross-linking and nonstandard mutagenesis. Studies focus on the role of ValA3, which projects within a crevice between the A- and B-chains. Engineered receptor α-subunits containing specific protease sites (“midi-receptors”) are employed to map the site of photo-cross-linking by an analog containing a photoactivable A3 side chain (para-azido-Phe (Pap)). The probe cross-links to a C-terminal peptide (residues 703-719 of the receptor A isoform, KTFEDYLHNVVFVPRPS) containing side chains critical for hormone binding (underlined); the corresponding segment of the holoreceptor was shown previously to cross-link to a PapB25-insulin analog. Because Pap is larger than Val and so may protrude beyond the A3-associated crevice, we investigated analogs containing A3 substitutions comparable in size to Val as follows: Thr, allo-Thr, and α-aminobutyric acid (Aba). Substitutions were introduced within an engineered monomer. Whereas previous studies of smaller substitutions (GlyA3 and SerA3) encountered nonlocal conformational perturbations, NMR structures of the present analogs are similar to wild-type insulin; the variant side chains are accommodated within a native-like crevice with minimal distortion. Receptor binding activities of AbaA3 and allo-ThrA3 analogs are reduced at least 10-fold; the activity of ThrA3-DKP-insulin is reduced 5-fold. The hormone-receptor interface is presumably destabilized either by a packing defect (AbaA3) or by altered polarity (allo-ThrA3 and ThrA3). Our results provide evidence that ValA3, a site of mutation causing diabetes mellitus, contacts the insert domain-derived tail of the α-subunit in a hormone-receptor complex.

tivable amino acid derivative in each case enables efficient and specific photo-cross-linking to the IR (11,14,15), providing support for a potential role in receptor binding.
Introduction of photoactivable amino acids at specific sites in a protein by chemical synthesis provides a powerful method to map interaction surfaces (16). A limitation can arise, however, if the photo-probe is significantly larger than the native side chain it replaces, as in such cases the probe may extend beyond the native contact surface (17). Thus, although para-azido-Phe (Pap) 6 is similar in size to Phe or Tyr and so provides a commensurate probe at aromatic sites in insulin (such as Tyr B16 , Phe B24 , and Phe B25 ) (14,18,19), extension of this strategy to sites containing smaller side chains may be confounded by probe-dependent nonnative interactions. Substitution of Val A3 by Pap, for example, is predicted to alter the surface topography of insulin; modeling suggests that the photo-probe protrudes beyond the A3-related crevice (Fig.  1B). Whereas the side chain of Val A3 is contained within this crevice (Fig.  1A), the azido group of Pap A3 (red ball in Fig. 1B) is predicted to project 3-5 Å beyond the hormone surface. An analogous limitation can arise in structure-activity studies of analogs containing small-to-large substitutions. Because Leu A3 protrudes beyond the molecular dimensions of Val A3 , for example, the very low activity of insulin Wakayama could reflect steric clash at the hormone-receptor interface even if Val A3 itself were not directly engaged (Supplemental Material). This possibility is not excluded by the native-like crystal structure of Leu A3 -insulin as a zincstabilized hexamer (20). Conversely, nonlocal structural perturbations may confound use of large-to-small substitutions as exemplified by segmental destabilization of the A1-A8 ␣-helix on substitution of Val A3 by Gly or Ser (21,22).
We describe here the contribution of Val A3 to the structure, stability, and function of insulin. The two methyl groups of Val A3 line the floor of an inter-chain crevice (red and yellow balls in Fig. 1A). Structure-activity relationships are first investigated through studies of insulin analogs containing A3 side chains similar in size to Val as follows: ␣-aminobutyric acid (Aba), Thr, and allo-Thr (Fig. 2). Although none of these side chains would be predicted (in the absence of nonlocal tion but exhibiting normal glucose tolerance (86). Because substitutions at B24 (but not at A3 or B25) impair the efficiency of insulin chain combination (P. G. Katsoyannis, personal communication), one possibility is that the folding of Ser B24 -proinsulin engenders an increased propensity for endoreticular stress in the ␤-cell (87,88). 6 The abbreviations used are: Pap, para-azido-phenylalanine; Aba, ␣-aminobutyric acid; DG, distance geometry; DKP-insulin, insulin analog containing three substitutions in the B-chain (Asp B10 , Lys B28 , and Pro B29 ); DTT, dithiothreitol; 125 I-Tyr A14 -labeled insulin, insulin labeled with 125   conformational change) to protrude from the crevice (Fig. 1C), on engagement with the IR the variant side chains might perturb the packing or polarity of the interface. To address these issues, the substitutions were introduced within an engineered monomer of high activity (DKP-insulin; see Ref. 23) to facilitate spectroscopic studies without confounding effects because of self-assembly. Although none of these substitutions perturbed the efficiency of disulfide pairing in insulin chain combination, each analog exhibits lower thermodynamic stability, presumably because of altered packing or solvation of the variant side chains. 1 H NMR studies nonetheless demonstrate retention of a native-like A3 crevice without protrusion of the substituted side chains (as in Leu A3 -insulin) or transmitted conformational changes (as in Gly A3 or Ser A3 analogs; see Refs. 21,22). Despite such structural similarities, receptor binding activities are lower than that of the parent monomer (DKP-insulin) but higher than that of Leu A3 -insulin (24). These studies suggest that Val A3 functions directly in receptor binding.
The second part of this study focuses on photo-cross-linking. Because our initial studies of isosteric analogs suggested that Val A3 directly contacts the IR, we sought to map the site of photo-cross-linking between Pap A3 -DKP-insulin and the IR ␣-subunit. Through the use of engineered midi-receptors (25) containing specific protease sites (26), evidence is presented that Pap A3 contacts a 17-residue segment at the extreme C ter-minus of the ␣-subunit. Derived from the insert domain (ID), the same segment (residues 704 -718 in the IR B-isoform; designated ␣CT) was shown previously to be a site of photo-crosslinking between a truncated Pap B25 analog of insulin and the holoreceptor (14). This segment contains several conserved aromatic and aliphatic residues and so may provide a nonpolar docking site for Val A3 and Phe B25 . The importance of ␣CT in hormone binding has been demonstrated by Ala-scanning mutagenesis of the holoreceptor (27,28).
This study thus brings together total chemical synthesis with receptor engineering to define a key feature of the hormonereceptor complex. Nonstandard mutagenesis of insulin extends the repertoire of conventional site-directed mutagenesis, whereas construction of model midi-receptors facilitates biochemical characterization of photo-cross-linked complexes. These complementary approaches highlight the contribution of the A-chain to the function of insulin at a site of clinical mutation causing diabetes mellitus. To our knowledge, these findings represent the first characterization of a contact between the insulin A-chain and a cognate binding element of the insulin receptor.

EXPERIMENTAL PROCEDURES
Synthesis of Insulin Analogs-The general protocol for solidphase synthesis is as described previously (29). In brief, 4-methylbenzhydrylamine resin (0.6 mmol amine/g; Bachem, Inc.) was used as solid support for synthesis of A-chain analogs; N-tertbutoxycarbonyl-O-benzyl-threonine-PAM resin (0.56 mmol/g; Bachem, Inc.) was used for synthesis of a B-chain analog containing substitutions His B10 3 Asp, Pro B28 3 Lys, and Lys B29 3 Pro (designated the DKP B-chain). A manual doublecoupling protocol was followed (30). The p-NH 2 moiety of para-amino-Phe (Pmp) was protected by a 2-chlorobenzyloxycarbonyl group, which is stable during peptide synthesis. The Pmp A3 -A-chain also contained a biotin group attached to the ⑀-amino group of D-Lys A1 , a well tolerated substitution for Gly A1 (20,31). Chain recombination employed an S-sulfonated DKP-B-chain and S-sulfonated variant A-chains (ϳ2:1 by weight) in 0.1 M glycine (pH 10.6) in the presence of dithiothreitol (30). Analogs were isolated from the combination mixture (32) and purified as described (33) with essentially native yields. Electrospray mass spectrometry (MS) in each case gave expected values and in the photoactivable derivative verified conversion of Pmp to Pap (14). As a control, a photoactivable analog was similarly prepared containing Pap B16 , which contacts the N-terminal L1 domain of the IR ␣-subunit (19); this analog contains biotin linked to the B1 ␣-amino group. Purities were in each case Ͼ98% as evaluated by analytical reversephase high performance liquid chromatography. MS revealed no anomalous molecular masses.
CD Spectroscopy-CD spectra were obtained using an Aviv spectropolarimeter (34). Samples contained 25-50 M DKPinsulin or A3 analogs in 50 mM potassium phosphate (pH 7.4); samples were diluted to 5 M for denaturation studies.
Thermodynamic Modeling-Guanidine denaturation data were fitted by nonlinear least squares to a two-state model as described (35). In brief, CD data (x), where x indicates the concentration of denaturant, were fitted by a nonlinear least squares program according to Equation 1, where x is the concentration of guanidine and where A and B are base-line values in the native and unfolded states. Base lines were approximated by pre-and post-transition (36). The m values obtained in fitting the variant unfolding transitions are lower than the m value obtained in fitting the wild-type unfolding curve (Supplemental Material). To test whether this difference and apparent change in ⌬G u might be a consequence of an inability to measure the CD signal from the fully unfolded state, simulations were performed in which the data were extrapolated to plateau CD values at higher concentrations of guanidine; essentially identical estimates of ⌬G u and m were obtained.
NMR Spectroscopy-1 H NMR spectra were obtained at 600 and 800 MHz under the following three conditions: (i) in 50 mM potassium phosphate (pH 7.0), (ii) in 20% deuteroacetic acid (pH 1.9) at 25°C, and (iii) in D 2 O at pH 8.0 and 32°C as described (37,38). This range of pH and temperature conditions enables resonances that overlap in one spectrum to be resolved in another. In addition, use of acidic pH in a co-solvent facilitates analysis of amide resonances, some of which are attenuated at pH 7.0 as a result of base-catalyzed solvent exchange. The protein concentration was 1 mM in each case. Resonance assignments were based on two-dimensional NOESY (mixing time 80 and 200 ms), total correlated spectroscopy (mixing time 30 and 55 ms), and double-quantum filtered correlated spectroscopy spectra.
Molecular Modeling-Structures were calculated by distance geometry and restrained molecular dynamics (DG/RMD) using X-PLOR (39) as described (19). Geometries were monitored with PROCHECK (40); solvent-accessible areas were obtained by using X-PLOR, and molecular cavities were calculated by using SURFNET (41). Rigid-body models of Pap A3 -and Leu A3insulin analogs were built using InsightII (Accelrys Inc., San Diego) and X-PLOR based on the structure of a T-state crystallographic protomer (2-Zn molecule 1; Protein Data Bank code 4INS).
Biological Assays-Two types of receptor-binding assays were performed. Determinations were performed with 3-6 replicates; values are reported as mean Ϯ S.D. (Table 1). The percentage of tracer bound in the absence of competing ligand was less than 15% to avoid ligand-depletion artifacts. (i) Activities of Thr A3 -DKP-insulin and DKP-insulin were first measured relative to 125 I-Tyr A14 -labeled human insulin using a human placental membrane preparation containing the IR (column 2 in Table 1) (33,42,43). This assay is limited by underestimation of the affinities of super-active analogs (such as DKP-insulin (33,44)) but is included to allow comparison with previous studies of related analogs using this assay (45)(46)(47). Relative activity is defined as the ratio of analog to human insulin required to displace 50% of specifically bound 125 I-insulin. (ii) To evaluate Aba A3 -, Thr A3 -, and allo-Thr A3 -DKP-insulin (column 3 in Table 1), a cell culture-based assay was employed relative to 125 I-Tyr A14 -labeled DKP-insulin as described (44,48). Use of the parent monomeric analog as tracer circumvents uncertainties because of the base-line activity of DKP-insulin. Relative activity is defined as ratio of dissociation constants obtained by nonlinear regression curve fitting. IM-9 human lymphocytes were grown in suspension as described (49) in RPMI media supplemented with 2 mM L-glutamine, 10% fetal calf serum, and 100 units/ml of penicillin/streptomycin (Invitrogen) at 37°C in a 5% CO 2 humidified atmosphere. Procedures are described elsewhere (44). In both types of assays data were corrected for nonspecific binding (amount of radioactivity remaining membrane-associated in the presence of 1 M human insulin). To evaluate the negative cooperativity, a cell culture-based assay was employed (44,50). A dose-response curve for negative cooperativity was established by measuring extent of dissociation of prebound labeled insulin by increasing concentrations of analogs, after allowing 30 min for dissociation in a 40-fold dilution (50).
Engineered Midi-receptors-An IR-derived midi-receptor (25,51) was constructed by PCR amplification and ligation of DNA fragments derived from human IR cDNA (isoform B). This ␣-subunit construct encodes the N-terminal IR signal peptide, domains L1-CR-L2-FnIII-1 (residues 1-601), and the C-terminal insert domain-derived segment (650 -719 (25). The C terminus was extended to contain a Myc epitope (EQKLI-SEEDL), permitting Western blotting by anti-Myc antiserum (Santa Cruz Biotechnology, Inc., Santa Cruz, CA). To facilitate mapping of photoproducts, two such constructs were designed, each containing specific internal cleavage sites for a protease derived from the tobacco etch virus (TEV; consensus recognition site ENLYFQ2G with cleavage after Q); the wild-type IR ␣-subunit lacks consensus TEV cleavage sites. (i) Midi-receptor T1 contains a single TEV site inserted following residue 601; cleavage thus liberates a C-terminal polypeptide of 12 kDa (81 residues (IR-(650 -719) plus a 10-residue c-Myc tag) plus the mass of the photo-cross-linked insulin A-or B-chain. (ii) Midireceptor T2 contains two consecutive TEV sites inserted after residue 702; complete cleavage in this case liberates a C-terminal peptide of mass 3 kDa (28 residues; IR-(703-719) plus c-Myc tag) plus the mass of the photo-cross-linked insulin chain. 7 cDNA inserts were verified by DNA sequencing.
Expression of Midi-receptors-The above cDNAs were inserted into mammalian expression vector pcDNA 3.1 (Invitrogen). Constructs were transiently expressed by transfection into 293H cells using Lipofectamine 2000 reagent (Invitrogen). Media were collected 72 and 144 h post-transfection and concentrated 7-fold by ultrafiltration (Amicon Ultra 15; Millipore). The concentrated protein solution was adjusted to contain 50 mM Hepes (pH 7.4). After addition of sodium azide to a final concentration of 0.02%, the recombinant receptor fragments were stored at 4°C. Binding of 125 I-Tyr A14 -insu-lin by midi-receptors was assayed in duplicate by polyethylene glycol precipitation as described (52).
Photo-cross-linking-Purified IR ectodomain or midi-receptors were incubated overnight with either 125 I-labeled or biotinylated photoactive insulin derivatives at a hormone concentration of 100 -200 nM at 4°C with gentle shaking. Solutions were in each case transferred to a Costar assay plate (Corning Glass) for UV irradiation (20 s at 254 nm) using a Mineralight lamp (model UVG-54, UVP, Upland, CA); the distance from the source was 1 cm. For analysis of ectodomain photoproducts after UV irradiation, covalent complexes were reduced with 2% ␤-mercaptoethanol or 100 mM dithiothreitol (DTT) and analyzed by SDS-PAGE.
Mapping of Hormone-Receptor Photoproducts-To characterize midi-receptor complexes, products were immunoprecipitated with anti-Myc antibody (c-Myc (A-14); Santa Cruz Biotechnology) and eluted with a free Myc peptide. 1 unit of TEV protease (Invitrogen) was then added to the eluate, and the digestion allowed to proceed overnight at 30°C; TEV-digested products were resolved by SDS-PAGE. To detect protease-digested 125 I-labeled photoadducts, gels were fixed for 20 min in 10% acetic acid and 25% isopropyl alcohol (v/v), dried onto Whatman 3MM paper, and exposed to x-ray film (Kodak Biomax MS) or a phosphor screen (Packard Cyclone). For detection of protease-digested biotinylated photoadducts, proteins were blotted onto a nitrocellulose membrane and probed with NeutrAvidin (Pierce) or a polyclonal antiserum directed against the N-terminal 20 residues of the ␣-subunit (N-20, Cruz Biotechnology; herein designated IR␣N).

RESULTS
Our study has two parts. Structure-activity relationships are first described among three analogs of DKP-insulin containing isosteric (Thr and allo-Thr) or smaller (Aba) substitutions at A3. Photo-cross-linking studies of Pap A3 -DKP-insulin to engineered midi-receptor fragments of the IR ␣-subunit are then described. The midi-receptors, whose hormone-binding properties are similar to those of the IR ectodomain (25, 51), contain specific TEV protease sites to facilitate mapping of photocross-linking sites.
Structure-Activity Relationships at Position A3-Receptorbinding affinities of Thr A3 -DKP-insulin, allo-Thr A3 -insulin, and Aba A3 -DKP-insulin are each lower than that of DKP-insulin ( Table 1). Effects of Aba A3 and allo-Thr A3 are consistent with previous studies of these substitutions in a native B-chain context (53,54), validating use of the monomeric DKP template. The activity of Aba A3 -DKP-insulin (7 Ϯ 1% relative to DKP-insulin) is significantly less than that of the Thr A3 -DKPinsulin (22 Ϯ 3%) but is similar to that of allo-Thr A3 -DKPinsulin (5 Ϯ 1%). No disproportionate changes in negative cooperativity were observed among the A3 analogs.
To probe possible conformational changes, structures were determined by two-dimensional 1 H NMR spectroscopy in relation to the parent monomer (23). 1 H NMR spectra in each case exhibit chemical shift dispersion similar to that of DKP-insulin (Supplemental Material). Although patterns of chemical shifts are generally similar (data not shown), small changes are in each a Receptor binding relative to human insulin was measured as described by using a human placental membrane-receptor binding assay (33,43). This assay underestimates activities of super-active analogs. The tracer was 125 I-labeled human insulin. b Receptor binding was measured using an IM-9 lymphocyte assay as described (44); the tracer was 125 I-labeled DKP-insulin. c The present results relate to previously published values as follows: Shoelson et al. (81) reported the activity of DKP-insulin relative to insulin as 200%. Chen and Feng (53) reported the activity of Thr A3 -insulin as 50% relative to insulin and allo-Thr A3 -insulin as 7.6% (54). The number of replicates is in each case given in parentheses; uncertainties represent Ϯ one S.D. d There is a discrepancy in the literature regarding Thr A3 -insulin analogs; a relative affinity of 13% was reported by Tager and co-workers (6), whereas a value of 50% was reported by Feng and co-workers (53). The binding activity of Thr A3 -DKP-insulin in a human placental membrane assay (80 Ϯ 13%) must be considered in relation to the affinity of DKP-insulin. In the placental membrane assay, the apparent activity is ϳ160% (column 2 in Table I), but this assay is known to underestimate activities of analogs whose affinities are significantly greater than wild-type insulin (33,44). Accordingly, affinities were re-measured in relation to 125 I-labeled DKP-insulin in IM-9 lymphocytes (column 3), an assay in which the affinities of super-active insulin analogs correlate well with their potencies in lipogenesis in adipocytes (33,44). e In this assay the affinity of DKP-insulin is 3-fold higher than that of insulin (14,18,19). Relative to DKP-insulin, the activity of Thr A3 -DKP-insulin is 22 Ϯ 3% (average and standard deviation of six replicates); this range is in accord with the above estimate of 80 Ϯ 13% relative to wild-type insulin in the placental membrane assay when corrected for the 3-fold enhanced binding of DKP-insulin. f ND indicates not determined. NOVEMBER 30, 2007 • VOLUME 282 • NUMBER 48 case observed. These are widely distributed in the structure, suggesting subtle reorganization of the core. Chemical shifts of key core side chains (Ile A2 , Leu A16 , Tyr A19 , Leu B11 , Phe B26 , and Leu B15 ) are nonetheless similar to those of DKP-insulin, indicating that no marked changes occur in orientations of neighboring aromatic rings and their associated ring currents (Tyr A19 , Phe B24 , Phe B25 , and Tyr B26 ). Particularly prominent is the upfield shift of the Leu B15 spin system (outlined in each panel of Fig. 3), a characteristic feature of B-chain supersecondary structure ascribed to the ring current of Phe B24 (55). In addition, the pattern of amide protection from solvent exchange in freshly prepared D 2 O solutions is in each case similar to that observed in DKP-insulin. The majority of such protection occurs within ␣-helices, and their maintenance is consistent with retention of native-like far-UV CD spectra.

Contact between Insulin and the Insulin Receptor
Patterns of inter-residue NOEs among the analogs closely resemble the pattern of NOEs in DKP-insulin, including framework contacts between conserved side chains (Phe B24 /Leu B15 , Tyr B26 /Val B12 , and Tyr A19 /Ile A2 ). In two-dimensional NMR spectra, the ␥ 1 -and ␥ 2 -methyl resonances of Val A3 are resolved between 0.8 and 0.9 ppm; NOEs are observed selectively from the ␥ 1 -methyl group to the meta ring resonance of Tyr A19 (cross-peak a in Fig. 3A) and to the ortho resonance of Tyr B26 (crosspeak b). Attenuation of related NOEs from these aromatic protons to the ␥ 1 -methyl group of Val A3 indicates that conformation of the wild-type side chain is well defined, constrained (as in crystal structures) by its packing within the inter-chain crevice. Similarly, comparison of Thr A3 -and allo-Thr A3 analogs demonstrates that inversion of ␤-carbon chirality is associated with corresponding "inversion" of specific long range NOEs between the single remaining A3 ␥-methyl resonance and the aromatic resonances of Tyr A19 and Tyr B26 (corresponding resonances a, b, and c in Fig. 3, B and C). These NOEs, which are in accord with contacts made in wild-type insulin by the side chain of Val A3 , aid in determining the respective side chain conformations of Thr A3 -and allo-Thr A3 . Analogous A3-related contacts are observed in the NOESY spectrum of Aba A3 -DKP-insulin (cross-peaks a and b in Fig. 3D). A summary of NOEs involving the wild-type or variant A3 side chains is given in supplemental Table S2; reference distances are provided based on the crystal structure of a variant zinc KP-insulin hexamer (i.e. bearing substitutions Pro B28 3 Lys, and Lys B29 3 Pro as in DKP-insulin) (56).
Structures were calculated by DG/RMD, starting from random coordinates on the basis of 655 (Aba A3 ), 649 (Thr A3 ), and 656 restraints (allo-Thr A3 ; Supplemental Material). Mean distance-and dihedral-angle violations are in each case Ͻ0.04 Å and 0.55 o , respectively. There are no distance violations Ͼ0.2 Å and no dihedral violations Ͼ3 o . Ensembles are shown in Fig. 4, C-E, in relation to crystal structures of native insulin (Fig. 4A) and the NMR-derived model of parent DKP-insulin (Fig. 4B). Analysis of Ramachandran maps by PROCHECK indicates that among these three ensembles, Ͼ70% of residues are in the most-favored region, 18 -25% are in additionally allowed regions, and 2-5% are in a generously allowed region; no residues are observed in disallowed regions. Packing of core side chains is in each case similar to that of DKP-insulin and consistent with crystal structures of insulin dimers and hexamers (Fig. 5). The variant cores contain no new cavities detectable by Surfnet calculations (41). The ␤-OH groups of Thr A3 and allo-Thr A3 are nonetheless associated with distinct changes in the electrostatic potential of the crevice, whereas the absence of one methyl group in Aba A3 is associated with a slight alteration in the topography of the crevice floor. The distinct side chain packing schemes of Thr A3 and allo-Thr A3 are in accord with stereospecific differences in their thermodynamic stabilities (Table 1); although each is less stable than DKP-insulin, the greater stability of the allo-Thr A3 analog relative to Thr A3 -DKP-insulin may reflect an unfavorable electrostatic interaction between the ␤-oxygen of Thr A3 and the electron cloud of Tyr B26 , which appears to be relieved on chiral inversion.
The absence of significant nonnative A3-related features is consistent with (a) maintenance of native-like overall structures and (b) the similar shapes and volumes of Val, Thr, allo-Thr, and Aba. Side chain-specific solvent accessibilities were calculated and compared with those of native insulin (Supplemental Material). As expected based on maintenance of native-like overall folds, the A3 main chain remains inaccessible in each analog, whereas side chain accessibilities are low as in DKP-insulin. Thr A3 and Aba A3 are associated with a decrease in the precision of the adjoining side chain of Glu A4 (Fig. 5, C and D); this is because of attenuation of inter-residue NOEs between H ␤ of Glu A4 and H ␣ of Thr A8 in the two analogs. Because Glu A4 itself does not make an important contribution to receptor binding (5, 57), decreased receptor binding by the present analogs is likely to reflect a direct A3-related perturbation of the hormone-receptor interface (see "Discussion"). Although the structure of insulin is proposed to change on receptor binding (12,58,59), the proximity of the A3-related crevice to the classical receptor-binding surface (A8, B16, and B24 -B26) and a nonclassical second site in a T-state protomer (A13 and B17 (60)) is shown in Fig. 6C.
Photo-cross-linking Studies-The IR is a modular protein containing two ␣and two ␤-subunits linked by disulfide bonds (Fig. 6, A and B). The ␣-subunits bind insulin, whereas the ␤-subunits contain the cytoplasmic tyrosine kinase domain (59). The isolated ectodomain ((␣␤ ⌬ ) 2 ) retains specific insulin binding activity (although with 10-fold lower affinity (61, 62)). 8 We have shown previously that insulin derivatives containing Pap at positions A3 or B16 exhibit efficient photo-cross-linking to the IR (19,31). Although the affinity of 8 The reduced affinity of the isolated ectodomain for insulin is presumably because of enhanced flexibility between the two ␣␤ ⌬ subunits in the absence of tethering by the transmembrane ␣-helices; native affinity is restored on fusion of the ectodomain ␤ ⌬ fragment to either a coiled coil (1) or dimeric F c domains (90).   NOVEMBER 30, 2007 • VOLUME 282 • NUMBER 48

JOURNAL OF BIOLOGICAL CHEMISTRY 35343
para-amino-Phe A3 -DKP-insulin is reduced by 30-fold relative to insulin (and so the corresponding azido analog would also be expected to exhibit reduced binding), under our experimental conditions the concentrations of the hormone and receptor are Ͼ10 3 -fold higher than the wild-type dissociation constant, thus permitting study of even low-affinity insulin analogs.
To characterize further the photo-cross-linking properties of Pap A3 -and Pap B16 -insulin derivatives, we employed the isolated ectodomain and engineered midi-receptors. Photo-crosslinking of Pap A3 and Pap B16 to the isolated ectodomain was first investigated. Following reduction of the covalent insulin-IR complexes by DTT, a photoadduct between the ␣-subunit and biotin-tagged insulin derivatives can readily be detected by NeutrAvidin (NAv, upper panel of Fig. 6D); the mobility of this band is essentially identical to that of the free ␣-subunit (molecular mass 110 kDa) as detected by Western blot (designated IR␣N, lower panel of Fig. 6D). Extension of our previous domain mapping strategy by tryptic and chymotryptic diges-tion established that, as found previously in photo-cross-linked holoreceptor complexes, Pap B16 contacts the N-terminal L1 domain (19), whereas Pap A3 is likely to contact a site within a C-terminal 14-kDa fragment of the ␣-subunit (molecular mass 20 kDa with glycosylation). 9 More precise mapping of the Pap A3 -specific photoproduct was obtained using midi-receptors T1 and T2 (Fig. 7, A and B). These glycosylated proteins contain the first four domains of the ␣-subunit (L1-CR-L2-FnIII 1 ; residues 1-601), the ␣-subunit portion of the insert domain (ID; residues 650 -719), and a C-terminal Myc tag (gray bars in Fig. 7, A and B). To facilitate mapping studies, T1 and T2 contain specific TEV protease sites (Fig. 7, A and B, red bars; see "Experimental Procedures"). After transfection of these constructs into 293H cells, immunoblot analysis demonstrated secretion of a 260-kDa truncated IR-derived dimer in the medium. The midi-receptors could readily be reduced by ␤-mercaptoethanol to 130-kDa monomers. The medium was observed to exhibit specific insulin binding activity; the IC 50 value (in the range of 0.1-2.0 nM) is consistent with the results of Brandt and co-workers (25,51). 10 Pap A3 -DKP-insulin cross-links to T1 and T2 with efficiencies similar to that observed in studies of the holoreceptor (1st and 3rd lanes in Fig. 7C). After TEV digestion and subsequent reduction, the photoadducts were observed to migrate at 18 kDa (T1) and ϳ5 kDa (T2) (red asterisks, 2nd and 4th lanes in Fig. 7C). Similar results were obtained either on probing by NeutrAvidin or by autoradiography following 125 I labeling of the Pap derivatives. Control studies of Pap B16 -DKP-insulin demonstrated compa-9 Independent mapping of the Pap A3 photoproduct was limited by the absence of a corresponding antiserum directed against the C-terminal portion of the ␣-subunit. Preliminary assignment to the ID was suggested by the similarity between the proteolytic digestion patterns of Pap A3 and Pap B25 photo-cross-linked complexes, yielding a C-terminal 14-kDa polypeptide containing part of FnIII-2 and the ␣-specific portion of the ID (31). This assignment depended on the assumption that the predominant site of photo-cross-linking by Pap B25 -DKP-insulin is within ␣CT (14). 10 The IC 50 of midi-receptor T1 was estimated to be 0.11 and 0.15 nM in two independent assays; the IC 50 of T2 was 1.8 and 1.9 nM. Decreased insulin binding by T2 may reflect interference by the inserted pair of TEV sites adjacent to ␣CT. The observed affinities are nonetheless higher than that reported by Kristensen et al. (51) in studies of a midi-receptor lacking engineered TEV sites (K d 4.4 nM).  (20)) or the L1 domain (B16 adduct (19)).
rable efficiencies of photo-cross-linking to T1 and T2. Following TEV digestion, however, the Pap B16 -derived photoadducts migrated at 100 kDa (T1, right-hand lane in Fig. 7D) and 110 kDa (T2; data not shown). These mobilities are consistent with a B16-binding site in the L1 domain as shown previously (19).

DISCUSSION
Recent progress in the crystallographic analysis of the IR has revealed that the ectodomain forms an inverted V structure (1). Although approximate insulin-binding sites are suggested by the positions of the dimer-related L1 domains, the structure does not contain bound insulin. Modeling of the insulin-bind-ing site has been limited by the absence of interpretable electron density for the insert domain, presumably because of its disorder in the free receptor (1). Despite considerable efforts, independent crystals of a hormone-receptor complex have not been obtained. In the absence of direct crystallographic information, structure-function relationships in insulin have been inferred by mutagenesis and chemical modification (5,6). In this study we have focused on the role of Val A3 , an invariant residue in the A-chain and site of clinical mutation associated with human diabetes mellitus (63). A related mutation has been identified in human insulin-like growth factor-I causing developmental abnormalities (64).
Contribution of Val A3 to Structure and Stability-Val A3 belongs to the N-terminal A-chain ␣-helix (residues A1-A8). Among crystal structures of insulin this helix is more variable than the other ␣-helices (A12-A18 and B9 -B19; T state) both in overall orientation and internal geometry (5). In solution the A1-A8 segment, unlike the other ␣-helices, lacks protected amide resonances in D 2 O solution, providing evidence for conformational fluctuations leading to the breakage of main chain hydrogen bonds (65). The A1-A8 sequence (GIVEQCCT) is notable for ␤-branched amino acids of low intrinsic helical propensity at positions 2 and 3 (Ile A2 and Val A3 ) and an unfavorable C-cap residue at position 8 (Thr A8 ) (66 -68). In accord with the low intrinsic helical propensity of the A1-A8 sequence, this ␣-helix undergoes segmental unfolding on substitution of Ile A2 by Ala (69) or on pairwise substitution of cystine A6 -A11 by Ala or Ser (23,34). Its helical conformation is stabilized by packing of Ile A2 and Val A3 and by the intra-A-chain disulfide bridge.
The A1-A8 ␣-helix orients the side chain of Val A3 to pack within a crevice between A-and B-chains. This crevice is in part nonpolar, but its physicochemical properties are made complex by the aromatic rings of Tyr A19 and Tyr B26 (whose electron clouds are associated with an asymmetric distribution of partial charges (70)) and by solvation at the mouth of the crevice. The present set of analogs permits the contribution of the A3-related crevice to the stability of insulin to be assessed. Although these analogs each retain native-like overall structures, they exhibit reduced stability ( Table 1). The decreased stability of Aba A3 -DKP-insulin is remarkable in view of the increased intrinsic helical propensity of Aba relative to Thr; substitution of a ␤-branched side chain by an unbranched near-isostere would be expected to enhance stability by ϳ0.5 kcal/mol (71)(72)(73), and yet the substitution impairs stability by 0.9 Ϯ 0.2 kcal/ mol. 11 This decrement highlights the distributed importance of packing efficiency within the A3-related crevice; removal of a ␥-methyl group presumably leaves a destabilizing gap in the floor of the crevice, in turn altering its dynamics and solvation. 11 The net instability of the Aba A3 analog (0.9 kcal/mol plus a helical propensity term of ϳ0.5 kcal/mol) is thus more than double that expected based on the change in side chain volume (ϳ21-29 Å 3 ), corresponding to a decrease in stability of 0.45-0.64 kcal/mol according to an empirical correlation developed by Matthews and co-workers (22 cal/mol/Å 3 (91)). The volume of Aba was estimated based on the known volumes (V) of 20 standard amino acids (92): calculations V Aba ϭ V Ala ϩ (V Leu Ϫ V Val ) or V Aba ϭ V Gly ϩ (V Val Ϫ V Ala ) yields respective estimates of 118.6 and 111.5 Å 3 . Given the V Val ϭ 140 Å 3 , a cavity of size 21-29 Å 3 is thus predicted by rigid-body modeling on substitution of Val by Aba. The insulin derivative was labeled with both 125 I and biotin; gel at left (10% Tricine) was visualized by 125 I-autoradiography, and gel at right (10 -20% Trisglycine) was probed by NeutrAvidin (NAv). Minus and plus signs at bottom of gels indicate before (Ϫ) or after (ϩ) digestion by TEV protease. D, control study of Pap B16 -DKP-insulin relative to A3 derivative. Products of photo-crosslinking to T2 were visualized by autoradiography following SDS-PAGE (8 -16% Tris-glycine) before (Ϫ) and after (ϩ) TEV digestion. Whereas A3 probe cross-links to C-terminal TEV fragment (red asterisk), Pap B16 -DKP-insulin cross-links to the large N-terminal fragment containing L1-CR-L2-FnIII 1 domains (ϳ110 kDa).
Evidence for such local dynamic changes in Aba A3 -DKP-insulin is provided by a reduction in the density of the inter-residue NOE network near A3. Because the intrinsic helical propensities of Thr and allo-Thr are similar to that of Val (66 -68), the decreased stabilities of Thr A3 -and allo-Thr A3 -DKP-insulin are also likely to reflect the importance of nonpolar packing in the A3-related crevice as in each analog a penalty would be incurred for complete or partial burial of a polar moiety. The decrements differ in severity, 0.8 Ϯ 0.2 (allo-Thr A3 ) versus 1.3 Ϯ 0.2 kcal/mol (Thr A3 ). The difference in stability on chiral inversion (⌬⌬G u 0.5 Ϯ 0.2) may reflect two factors. First, whereas the ␤-OH moiety of Thr A3 is completely buried, that of allo-Thr A3 is partially exposed. Thus, Thr A3 -DKP-insulin would be expected to incur a greater hydrophobic transfer free-energy penalty than allo-Thr A3 -DKP-insulin (74). Second, respective orientations of the ␤-OH group differ relative to the aromatic rings of Tyr A19 and Tyr B26 ; the internal ␤-OH moiety of Thr A3 abuts the cloud of Tyr B26 and is near that of Tyr A19 . We speculate that this is an example of how protein stability can be modulated by weakly polar interactions (70). In particular, insulin contains conserved aromatic rings (including tyrosines at A19 and B26) whose position and orientation, maintained in the analogs, provide a geometric framework for packing of neighboring aliphatic side chains. These rings exhibit partial positive charges at their edges and partial negative charges at each face (70,75,76).
The relative stabilities of A3 analogs may be influenced by our choice of DKP-insulin as a template. Although this template enables effects of substitutions on a monomer to be distinguished from confounding effects on self-association, one of the DKP substitutions (Pro B28 3 Lys) may in principle alter the structure or dynamics of the A3-related crevice. In wild-type insulin the B28 pyrrolidine ring contributes to the nonpolar crevice lining and in turn to the environment of Val A3 . Although the methylene chain of Lys B28 in part recapitulates this surface, the uniquely constrained shape of a proline is absent. It would be of future interest to investigate the thermodynamic consequences of A3 substitutions in wild-type insulin, including assessment of any changes in dimerization and higher order assembly. It would be of further interest to test whether crevice packing contributes not only to the stability of insulin once folded but also to the efficiency of folding of proinsulin, a process that might be influenced by the relative stabilities of partially folded conformations (77). Whereas the biosynthesis of insulin Wakayama in the pancreatic ␤-cell is apparently unperturbed (4), mutations elsewhere in proinsulin that block its folding in the endoplasmic reticulum are associated with permanent neonatal-onset diabetes mellitus (78).
Contribution of Val A3 to Receptor Binding-The biological importance of Val A3 is suggested by the clinical association between the Leu A3 variant and diabetes mellitus (insulin Wakayama (24)). Yet the extremely low activity of this variant is anomalous among mutant insulins (5). Whereas substitutions at the classical receptor-binding surface ordinarily impair affinity by 5-100-fold, the activity of Leu A3 -insulin is reduced by 500-fold. Leu A3 -insulin nonetheless retains a native-like crystal structure as an R 6 zinc insulin hexamer (20). Such a marked reduction in binding unaccompanied by structural dis-tortions suggests that protrusion of the larger A3 side chain beyond the A3-related crevice may incur steric clash at the hormone-receptor interface, possibly leading to transmitted perturbations at neighboring receptor contacts. This model left open the possibility that the side chain of Val A3 itself is near the receptor but not directly engaged.
To address this possibility, we have chosen to study A3 substitutions of similar size to Val and hence unlikely to protrude from the crevice. Use of a monomeric template circumvents a structural issue inherent in crystallographic studies of zinc insulin hexamers: that self-assembly may obscure possible conformational perturbations through imposition of a shared framework of subunit interfaces. This issue is pertinent to A3 substitutions as the A3-related crevice adjoins the classical dimer interface. That A3 substitutions may result in nonlocal structural perturbations in a zinc-free monomer is suggested by previous spectroscopic studies of Gly A3 and Ser A3 analogs (21,22). In each case evidence has been obtained that the A1-A8 helix is distorted. The low activities of these analogs (0.1 and Ͻ2%, respectively) may thus reflect disruption of multiple hormone-receptor contacts.
Our results demonstrate that substitution of Val A3 by Aba, Thr, or allo-Thr (unlike substitution by Ser or Gly) is structurally well tolerated, and yet (in accord with past studies (6,53,54)) impairs receptor binding. The extent of impairment is 4 -20-fold, values typical of mutations at the surface of insulin (5). We envisage that such decreased binding reflects perturbed side chain interactions between A3 and a cognate binding pocket of the IR. Of particular interest is the stereospecific difference in activity between Thr A3 -DKP-insulin (affinity ϳ20% in the IM-9 lymphocyte assay relative to DKP-insulin) and allo-Thr A3 -DKP-insulin (5%). Although the stability of the allo-Thr A3 analog and the local dynamics of its A3-related crevice are less perturbed than these features of Thr A3 -DKP-insulin (see above), the allo-Thr A3 analog is 4-fold less active than the Thr A3 analog. These results suggest that the pro-S ␥-methyl group of Val A3 docks against the nonpolar surface intolerant of a hydroxyl substituent, whereas the pro-R ␥-methyl group may reside at a less selective microenvironment. 12 Removal of the offending ␤-OH moiety from allo-Thr A3 -DKP-insulin (yielding the nonstandard Aba A3 analog) results in only a small improvement in binding (this small difference may in fact be within experimental error); the absence of the polar perturbation may largely be offset by introduction of a destabilizing packing defect analogous to that observed in the free hormone.
Direct engagement of Val A3 at the receptor interface may require a change in conformation of the B-chain to expose the A3-related crevice. Such exposure could be associated with detachment on receptor binding of the overlying C-terminal B-chain ␤-strand (residues B24 -B28 (12,58)). The detachment model is supported by the anomalous properties of several insulin analogs, including the very low activity of mini-proinsulin, a single-chain analog containing a peptide bond between Lys B29 12 Distinct receptor environments surrounding the A3 side chain were previously suggested by functional differences between an Ile A3 analog (relative activity 11%) and the corresponding allo-Ile A3 analog (activity 18% (6)). This near 2-fold stereospecific difference between S-and R-steric probes (Ile and allo-Ile (6)) is less marked than the 4-fold stereospecific difference between R-and S-polarity probes (Thr and allo-Thr (54)).
and Gly A1 (58,79), and the enhanced activities of analogs containing D-amino acids at position B24 (80,81). Detachment of the C-terminal B-chain ␤-strand would also expose the side chain of Ile A2 , rationalizing the 50-fold loss of activity exhibited by allo-Ile A2 analogs (6) despite maintenance of native-like structures (45,82). Packing of the ␤-branched side chains of Ile A2 and Val A3 at the receptor interface may underlie their otherwise puzzling conservation within an ␣-helix.
Evidence that Val A3 directly contacts the IR motivated efforts to map the site of cross-linking between a photo-reactive probe at A3 and the receptor ␣-subunit. Previous studies of a holoreceptor complex established that Pap A3 -DKP-insulin photocross-links to a putative C-terminal fragment whose molecular mass, estimated by SDS-PAGE following deglycosylation, is 14 kDa (ϳ110 -130 amino acids). Assignment as a C-terminal fragment was based on a correspondence between proteolytic footprints of Pap A3 -DKP-insulin and Pap B25 -DKP-insulin (19,20). Proteolytic footprints of the holoreceptor-derived photo-products are essentially identical to those derived from the isolated ectodomain.
Here, we have exploited "midi-receptors" (active dimeric fragments of the ␣-subunit in the absence of the ␤-subunit; see Ref 51) to obtain a rapid and simple mapping protocol. To this end, we have introduced engineered protease TEV site(s) within the midi-receptor without loss of hormone binding activity. This assay provides evidence that Pap A3 -DKP-insulin forms a photoadduct with the 650 -719 segment (construct T1) and the 703-719 segment (construct T2). The A3 probe thus contacts the same C-terminal peptide (␣CT) as does a Pap B25 probe (14). Interestingly, B25 is also a site of clinical mutation causing diabetes mellitus (insulin Chicago (4)). The sequence of ␣CT (KTFEDYLHNVVFVPRPS) contains a mixture of hydrophobic and hydrophilic amino acids; residues critical to hormone binding (boldface in sequence above) have been identified by Ala-scanning mutagenesis (28,83). These residues may either contact insulin or mediate interactions between ␣CT and other domains of the receptor ␣-subunit. The residues in ␣CT that contribute to the binding sites for residues A3 and B25 are presumably aromatic or nonpolar.
The structure of the IR ectodomain (Fig. 8, A and B; see Ref. 1,84), provides a framework for understanding insulin binding. Unfortunately, interpretable continuous electron density was lacking for portions of FnIII-2 and the insert domain (dashed red lines in Fig. 8A). Intriguingly, an unassigned region of poor electron density (comprising ϳ20 residues) was observed adjoining the hormone-binding surface of the L1 domain (green in Fig. 8C) (1). The present results support assignment of this density to ␣CT. A dual structural role of ␣CT, simultaneously docking against L1 and the bound insulin, would rationalize the large number of critical residues identified by Ala-scanning mutagenesis and the severity of their impact on hormone binding (28,83).
The putative proximity of L1 and ␣CT would also rationalize how consecutive photoreactive probes in insulin (at positions B24 and B25) can cross-link to these respective domains of the IR (18). These L1 and ␣CT domains could belong to the same ␣-subunit (a cis-model) or to dimer-related ␣-subunits (a transmodel). Recent mutagenesis results support a trans-model (52). In this regard it is noteworthy that the present substitutions at . Individual domains L1, CR, L2, FnIII-1, FnIII-2, and FnIII-3 are shown in gray, black, blue, red, purple, and dark blue, respectively. Missing or discontinuous electron density (residues 655-755; IR isoform A) is depicted in schematic form by red dashed lines (not intended to represent actual conformation). B, crystal structure of inverted V ectodomain dimer. C, L1 domain (gray ribbon) in relation to unassigned and discontinuous electron density (green), potentially from ␣CT; density may represent ϳ20 residues of insert domain. Residues in L1 critical to hormone binding (as inferred from Ala-scanning mutagenesis (93) are shown as red stick. C was kindly provided by C. W. Ward.
A3 cause no disproportionate perturbations in negative cooperativity, whereas this phenomenon is impaired by substitutions at B25 (85). Because the mechanism of negative cooperativity presumably involves binding of one or more additional insulin molecules to already occupied receptors, this difference between A3 and B25 analogs suggests structural differences in how insulin binds to such additional sites. Whether this in turn implies corresponding differences in a singly occupied high affinity complex (as in a trans model) is unclear. In either cis-or trans-models, the absence of interpretable electron density for the C-terminal portion of the ␣-subunit suggests that this region is less well ordered in the absence of insulin binding. We therefore imagine that the insert domain undergoes a disorderto-order transition on insulin binding, i.e. both insulin and its receptor exhibit induced fit. Hormone-dependent folding of the IR may contribute to the mechanism of signal transduction, leading to activation of the cytoplasmic tyrosine kinase domains.
Concluding Remarks-This study has demonstrated the importance of a conserved side chain (Val A3 ) to the structure, stability, and function of insulin. Our strategy has combined structural studies with receptor-based mapping of photocross-linked adducts. Use of isosteric analogs as probes of structure-activity relationships circumvents possible confounding effects of larger side chains (protrusion from the A3-related crevice) or smaller side chains (nonlocal structural perturbations). The properties of these analogs demonstrate that Val A3 , despite its low intrinsic helical propensity, contributes to the stability of the free hormone through its packing in an inter-chain crevice. Photo-cross-linking studies suggest that analogous packing occurs at the receptor interface, mediated by the C-terminal tail of the ␣-subunit. Receptor binding is impaired by the present isosteric substitutions and is incompatible with Leu A3 (insulin Wakayama), leading to diabetes mellitus. Because Val A3 is not accessible in the free hormone, a conformational change in the overlying B-chain may be required for its engagement. Extension of the present strategy to other sites in insulin may enable construction of a model of the hormone-receptor complex. Such a model may enable design of novel insulin agonists for the treatment of diabetes mellitus.