Probing the correlation between insulin activity and structural stability through introduction of the rigid A6–A11 bond

The development of fast-acting and highly stable insulin analogues is challenging. Insulin undergoes structural transitions essential for binding and activation of the insulin receptor (IR), but these conformational changes can also affect insulin stability. Previously, we substituted the insulin A6–A11 cystine with a rigid, non-reducible C=C linkage (“dicarba” linkage). A cis-alkene permitted the conformational flexibility of the A-chain N-terminal helix necessary for high-affinity IR binding, resulting in surprisingly rapid activity in vivo. Here, we show that, unlike the rapidly acting LysB28ProB29 insulin analogue (KP insulin), cis-dicarba insulin is not inherently monomeric. We also show that cis-dicarba KP insulin lowers blood glucose levels even more rapidly than KP insulin, suggesting that an inability to oligomerize is not responsible for the observed rapid activity onset of cis-dicarba analogues. Although rapid-acting, neither dicarba species is stable, as assessed by fibrillation and thermodynamics assays. MALDI analyses and molecular dynamics simulations of cis-dicarba insulin revealed a previously unidentified role of the A6–A11 linkage in insulin conformational dynamics. By controlling the conformational flexibility of the insulin B-chain helix, this linkage affects overall insulin structural stability. This effect is independent of its regulation of the A-chain N-terminal helix flexibility necessary for IR engagement. We conclude that high-affinity IR binding, rapid in vivo activity, and insulin stability can be regulated by the specific conformational arrangement of the A6–A11 linkage. This detailed understanding of insulin's structural dynamics may aid in the future design of rapid-acting insulin analogues with improved stability.


The development of fast-acting and highly stable insulin analogues is challenging. Insulin undergoes structural transitions essential for binding and activation of the insulin receptor (IR), but these conformational changes can also affect insulin stability. Previously, we substituted the insulin A6 -A11 cystine with a rigid, non-reducible C‫؍‬C linkage ("dicarba" linkage). A cis-alkene permitted the conformational flexibility of the A-chain N-terminal helix necessary for high-affinity IR binding, resulting in surprisingly rapid activity in vivo.
Here, we show that, unlike the rapidly acting Lys B28 Pro B29 insulin analogue (KP insulin), cis-dicarba insulin is not inherently monomeric. We also show that cis-dicarba KP insulin lowers blood glucose levels even more rapidly than KP insulin, suggesting that an inability to oligomerize is not responsible for the observed rapid activity onset of cis-dicarba analogues. Although rapid-acting, neither dicarba species is stable, as assessed by fibrillation and thermodynamics assays. MALDI analyses and molecular dynamics simulations of cis-dicarba insulin revealed a previously unidentified role of the A6 -A11 linkage in insulin conformational dynamics. By controlling the conformational flexibility of the insulin B-chain helix, this linkage affects overall insulin structural stability. This effect is independent of its regulation of the A-chain N-terminal helix flexibility necessary for IR engagement. We conclude that high-affinity IR binding, rapid in vivo activity, and insulin stability can be regulated by the specific conformational arrangement of the A6 -A11 linkage. This detailed understanding of insulin's structural dynamics may aid in the future design of rapid-acting insulin analogues with improved stability.
Since the discovery of insulin by Banting and Best, insulin therapy has remained the primary treatment for type 1 and late stage type 2 diabetes, being prescribed to effectively lower blood glucose levels (1)(2)(3). Long-acting (e.g. insulin glargine; Lantus, Sanofi) and rapid-acting insulin analogues (e.g. lispro insulin or KP insulin); 5 Humalog, Eli Lilly) currently in clinical use were developed to mimic the physiological basal-bolus insulin profile (recently reviewed by Mathieu et al. (3)). Although these analogues are undoubtedly life-savers for diabetic patients, their pharmacokinetic and pharmacodynamic performance remains suboptimal for some patients (3). Of high priority is the development of a new rapid-acting insulin analogue possessing faster onset of action and greater structural stability. To achieve this, a deeper understanding is required of the structural determinants of insulin necessary for receptor binding and function.
Insulin is a small globular protein synthesized in the pancreatic ␤-cells and secreted as a two-chain polypeptide comprising a 21-residue A chain and a 30-residue B chain. The secondary structure of insulin consists of three ␣-helices, two within the A chain (A1-A8 and A12-A18, respectively) and a single ␣-helix within the central segment of the B chain (B9 -B19) (Fig. 1, A  and B) (4). It is stored in pancreatic ␤-cells as 2Zn-coordinated hexamers that, when released into the blood stream, rapidly dissociate into active monomers (5). The monomeric form then adopts the active conformation and in doing so reveals essential residues within the hydrophobic core for binding (6) but also primes the molecule to the formation of higher-order oligomers (7)(8)(9). This is highlighted in the rapid-acting KP insulin, an analogue possessing an inversion of Pro B28 Lys B29 in the B chain of native insulin to render the molecule essentially monomeric through disruption of the dimer interface ( Fig. 1A) (10,11). As a consequence, KP insulin is rapid acting, but concomitant exposure of core hydrophobic residues means that KP insulin readily forms fibrils (9,12,13). This highlights the fine balance between the competing requirements for stability and conformational plasticity needed for optimal activity.
High-affinity binding to the insulin receptor (IR; a tyrosine kinase receptor (14)) is achieved by interaction through two distinct binding surfaces (1 and 2) on the ligand (6,(15)(16)(17). Insulin site 1-binding residues (also known as the "classical binding site") are also involved in insulin dimerization (18,19), whereas site 2-binding residues overlap with the hexamerforming surfaces within a 2Zn-insulin hexamer ( Fig. 1A) (6,16,20). Recent crystallographic studies of an IR fragment composed of the ectodomain and the ␣CT segment in complex with insulin through site 1 residues provided key insights into the mechanism of insulin-IR interaction (15,21). Key conformational changes in insulin required for effective IR binding were identified. Movement of the C-terminal segment (B24 -B30) of the B chain from the hormone core is required for engagement with the IR ␣CT and L1 domains (22). It is likely that the end of the B chain opens in a zipper-like fashion (23) to allow accessibility to insulin site 1-binding residues that otherwise remain buried within the hydrophobic core (21,22). Evident in our previous findings, insulin also undergoes further conformational change within the first A-chain helix to enable binding. The helix rotates to avoid a steric clash between the first four residue side chains and the ␣CT segment (24). In this conformation, residues A1-A5 form a single ␣-helical turn, whereas residues A3-A9 adopt a wider helix conformation, approximating an (i, i ϩ 5) -helix (24). This conformation is also evident in the IR-insulin crystal structure (PDB entry 4OGA) (21,24) and can be artificially achieved by a mutation of residue B26 that also promotes opening of the B chain (25, 26).
The correct folding and stabilization of the unique threedimensional structure of mature insulin is supported by the three disulfide linkages: two interchain (Cys A7 -Cys B7 and Cys A20 -Cys B19 ) and one intrachain (Cys A6 -Cys A11 ) (Fig. 1, A and B) (5,27). Correct disulfide combination is essential for function (27)(28)(29)(30). The Cys A7 -Cys B7 bond is surface-exposed and holds the N termini of the two chains together. Both Cys A6 -Cys A11 and Cys A20 -Cys B19 linkages are buried within the hydrophobic core. Once folded, the Cys A20 -Cys B19 disulfide is constrained in a fixed configuration and buried deeply within the core, suggesting that its most likely function is to maintain structural stability. The intrachain Cys A6 -Cys A11 cystine, although buried, is relatively flexible and can adopt several different configurations (30 -32). In our recent study, we identified the Cys A6 -Cys A11 linkage as an important modulator of the structural transitions of the N-terminal A-chain helix required for insulin activity. Introduction of two isomeric fixed A6 -A11 dicarba bridges in insulin did not perturb the overall structure but resulted in strikingly different receptor potencies, where the cis-isomer permits high-affinity IR binding and the trans-isomer does not (Figs. 1C and 2A) (24).
Arising from our study of the cis-dicarba insulin was the observation that this analogue promotes a more rapid lowering of blood glucose levels than native insulin. It is also less thermodynamically and chemically stable than native insulin (24). As these properties are reminiscent of the monomeric KP insulin (7,9,33), we investigate here whether the A6 -A11 linkage not only controls A-chain flexibility but also influences the B-chain conformation, leading to the monomeric state required for rapid receptor engagement.
In this study, through a combination of biophysical analyses of cis-dicarba insulin and the monomeric cis-dicarba KP insulin counterpart, we show that, despite its rapid action in vivo and its accelerated fibril formation relative to native insulin, the cis-dicarba insulin is not inherently monomeric. Limited proteolysis studies alluded to an unexpected conformational change in the B-chain helix. Such a link between the A6 -A11 disulfide and the conformational dynamics of the B chain has not been described previously. These findings suggest a key role for the A6 -A11 linkage, not only in regulating A-chain flexibility that primes insulin for receptor engagement, but also in influencing the B-chain conformation and regulating insulin's stability.

Chemical synthesis of dicarba insulins
We aimed to understand the mechanism by which the cisdicarba insulin was apparently rapid acting in vivo by directly comparing its in vitro and in vivo biological and biophysical activities with the monomeric c[⌬ 4 A6,11]-dicarba human lispro insulin (cis-and trans-dicarba KP insulins). The cis-and trans-configured dicarba insulin A chains, in which a CϭC dicarba bond replaces the A6 -A11 intrachain S-S bond (Fig.  1C), were synthesized as described previously using a ringclosing metathesis (RCM) and solid-phase peptide synthesis (SPPS)-catalysis approach (24, 34). The modified dicarba insulin A chains were then combined with requisite insulin B chains to provide cis-and trans-isomers of c[⌬ 4 A6,11]-dicarba human insulin and c[⌬ 4 A6,11]-dicarba KP insulin. The dicarba analogues were purified by reversed-phase HPLC (RP-HPLC) (see Fig. 1 (D and E) and Fig. S1) before being subjected to biological and biophysical analyses.

cis-Dicarba insulin and cis-dicarba KP insulin are equally potent to native insulin in receptor binding and activation
The binding affinities of the dicarba insulin analogues for the IR-B and insulin-like growth factor 1 receptor (IGF-1R) were determined using competition binding assays ( Fig. 2A, Fig. S2, and Table S1). Notably, the restrained A6 -A11 cis-and transdicarba CϭC bonds had the same effect when introduced into the monomeric KP insulin analogue as was previously seen with the cis-and trans-dicarba analogues of human insulin (24). cis-Dicarba insulin and cis-dicarba KP insulin were equipotent to native insulin in binding and activation of both the IR-B (Fig. 2,

A6 -A11 disulfide affects insulin B-chain conformation
A and B) and the IGF-1R (Fig. S1, A and B) (Table S1), suggesting that the restrained cis-dicarba CϭC bond allows both analogues to adopt a conformation that engages with both receptors in a manner similar to insulin. Conversely, trans-dicarba insulin and trans-dicarba KP insulin bind poorly to both the IR-B ( Fig. 2A) and IGF-1R ( Fig. S1A and Table S1), indicating that the trans-dicarba configuration restricts both analogues from forming a high-affinity interaction with these receptors. The trans-dicarba insulins were subsequently excluded in this and further activity assays due to their poor receptor-binding affinities.

The cis-dicarba KP insulin promotes in vitro DNA synthesis and glucose uptake with equal potency to native insulin
Corresponding with its IR-B binding and activation potency, the cis-dicarba KP insulin was equipotent with native insulin in promoting DNA synthesis in L6 rat skeletal myoblast overexpressing IR-A. This is in contrast to the cis-dicarba insulin, which was 5-10-fold less potent than insulin in promoting mitogenic activity (Fig. 2C) (24). The cis-dicarba KP insulin was equipotent with cis-dicarba insulin and insulin in promoting glucose uptake in cultured NIH3T3-L1 adipocytes (Fig. 2D). There is a trend of lower activities for the cis-dicarba insulin and cis-dicarba KP insulin; however, the effect was not significantly different.

The cis-dicarba insulin demonstrates self-association behavior identical to that of native insulin
Analytical ultracentrifugation (AUC) was performed to determine whether the cis-dicarba insulin is monomeric, as suggested by its rapid action in vivo (Fig. 2, E and F) (24). In the presence of Zn 2ϩ , sedimentation equilibrium data for the cisdicarba insulin was fitted to a single species of apparent mass 34,500 Ϯ 400 Da (Fig. 3A), consistent with the expected mass of a 2-Zn 2ϩ human insulin hexamer (34,726 Da). At higher concentrations, the fit was not perfect (reduced 2 ϳ4), suggesting the presence of other high-molecular weight species, consistent with previous observations for native mammalian insulins (35).
In contrast, the shapes of the equilibrium concentration distributions in the absence of Zn 2ϩ were clearly concentration-dependent, implying reversible self-association (Fig.  3B). Accordingly, these data could not be fit as a single species, in sharp contrast to those we have recently described for the strictly monomeric venom insulin of Conus geographus (Fig.  S11) (36). Native insulin shows similar concentration dependence in the shape of its equilibrium concentration distribu-

A6 -A11 disulfide affects insulin B-chain conformation
tions (Fig. S11) and also fails to fit to single-species models, consistent with its expected tendency for self-association. For an initial model-free assessment of the self-association of Zn 2ϩfree cis-dicarba and native insulin, the data were plotted as the square of the radial position scaled by rotor speed ( 2 (r 2 Ϫ r 0 2 )) versus the logarithm of the equilibrium concentration (Fig. 3C). The slope of such a plot is proportional to the weight-average molecular weight of all species present at each point in the cell, and the observed non-linearity confirms the presence of multiple species in the sample. Over much of the accessible concentration range, the slopes of the cis-dicarba and native insulin plots are similar, implying that the self-association behavior of the two insulins are similar. At low concentration (Ͻ10 M), the slope was consistent with that expected for monomeric insulin, suggesting that monomer dominated at these concentrations. The slope increased with increasing concentration, indicating that oligomeric species were dominant over much of the experimental concentration range. Only at the highest concentrations is there evidence of some divergence between the two curves, suggesting the possibility of some difference in the tendency of cis-dicarba insulin to form higher-order oligomers.
Attempts to fit the two data sets to a specific, consistent model of self-association were unsuccessful, due to numerical instabilities in relevant models (37), and perhaps also reflecting the putative subtle difference in higher-order oligomerization. Nonetheless, under the conditions studied, the self-association behavior of the cis-dicarba insulin is qualitatively similar to that of native mammalian insulins. Like native insulin, cis-dicarba insulin is only monomeric in the absence of Zn 2ϩ , and only at low concentration.

cis-Dicarba KP insulin lowers blood glucose levels more effectively and more rapidly compared with the cis-dicarba insulin, KP insulin, and native human insulin
Having established that the cis-dicarba insulin is not monomeric, we sought to compare the in vivo activities of this analogue with cis-dicarba KP insulin, which we can assume is monomeric as per KP insulin. The cis-dicarba KP insulin lowered Dunnett's multiple comparison). C, DNA synthesis in response to increasing concentrations of dicarba insulins is shown as percentage incorporation of [ 3 H]thymidine ( 3 H-Thy) above basal. All data in A-D are the mean Ϯ S.E. n ϭ at least 3 independent experiments. D, glucose uptake stimulated by increasing concentrations of insulin, cis-dicarba insulin, or cis-dicarba KP insulin is expressed as -fold glucose uptake (pmol/min/mg) above basal. ns (not significant), insulin versus cis-dicarba insulin versus cis-dicarba KP insulin (paired t test). E and F, insulin tolerance test in mice fed on a normal diet (chow) diet (E) or on a high-fat diet (F) were administered through intraperitoneal injection (ip) with 0.75 IU/kg insulin, KP insulin, cis-dicarba insulin, or cis-dicarba KP insulin under non-fasting conditions, and tail vein blood glucose was measured via a glucose meter at the indicated times. n ϭ 5-6/group. Blood glucose levels are expressed as change over basal levels (mmol/liter). Chow diet: **, p Յ 0.01, insulin versus cis-dicarba insulin; **, p Յ 0.01, KP insulin versus cis-dicarba KP insulin (paired t test). High fat diet: **, p Յ 0.01, insulin versus cis-dicarba insulin; **, p Յ 0.01, KP insulin versus cis-dicarba KP insulin (paired t test). Significance of the change in blood glucose levels at each time point was also determined by two-way ANOVA followed by Holm-Sidak's multiple comparison test. Chow diet: #, p Յ 0.05, KP insulin versus cis-dicarba KP insulin at t ϭ 60 min. High fat diet: KP insulin versus cis-dicarba KP insulin at t ϭ 30, 45, and 60 min (###, p Յ 0.001) and t ϭ 90 and 120 min (##, p Յ 0.01).

A6 -A11 disulfide affects insulin B-chain conformation
blood glucose levels more effectively and more rapidly compared with native insulin and KP insulin when mice were treated with 0.75 IU/kg insulin or analogue under non-fasting conditions (Fig. 2, E and F). Notably, the cis-dicarba KP insulin was even more effective and rapid acting than the cis-dicarba insulin. The glucose-lowering effect was most evident and significant in insulin-resistant mice fed a high-fat diet (Fig. 2F). The difference in activities between native and cis-dicarba KP insulin, which are both expected to be monomeric, implies that the improved activity of the cis-dicarba analogues is not the result of a change in self-association. This is consistent with the above observation that native and cis-dicarba insulin show similar self-association behavior.

cis-Dicarba insulin and cis-dicarba KP insulin more rapidly form fibrils than native human insulin
Whereas the rapid action of the cis-dicarba insulin compared with insulin is not due to this analogue being monomeric, it must have different biophysical properties from insulin that lead to this difference in biological activity. To explore this further, we next investigated the ability to form fibrils in an atomic force microscopy (AFM) fibrillation assay. The cis-dicarba insulin formed fibrils more rapidly than native insulin, with fibrils first detected after 2 h compared with 6 h for native insulin at the same temperature and concentration (60°C and 1.16 mg/ml, respectively; Fig. 4). Consistent with being monomeric, KP insulin also rapidly formed fibrils, with fibrils first detectable at 2 h. The cis-dicarba KP insulin fibrillation is evident between t ϭ 6 and 8 h. Interestingly, the fibrils formed by both cis-dicarba insulin and cis-dicarba KP insulin appeared shorter, thicker, and of different morphology than those arising from native insulin and KP insulin ( Fig. 4; see cis-dicarba insulin at t ϭ 15 h and cis-dicarba KP insulin at t ϭ 8 h). These observations are consistent with the fact that thioflavin T, a dye commonly used to detect insulin fibrils, did not bind cis-dicarba insulin fibrils (data not shown). Thioflavin T normally binds to insulin fibrils at two sites (between fibers and/or between protofilaments) (38). Our data suggest that whereas cis-dicarba insulin is not inherently monomeric, it is conformationally different from native insulin.

cis-Dicarba insulin and cis-dicarba KP insulin are thermodynamically less stable than native insulin
Next, we compared the thermodynamic stability of the cisdicarba insulin with the cis-dicarba KP insulin. Introduction of a dicarba A6 -A11 bond into KP insulin also led to a decrease in thermal and chemical denaturation stabilities. The far-UV CD spectra (190 -260 nm) of Zn 2ϩ -free cis-and trans-dicarba KP insulin exhibited lower helical content (34 and 17%, respectively) compared with KP insulin (44%) (see Fig. 5A and Table  1). The [] 222 values are directly proportional to the helical content of the proteins. As seen with the cis-dicarba insulin (24), the cis-dicarba KP insulin (Fig. 5B) exhibited smaller [] 222 negative magnitudes measured at 20°C consistent with their lower initial helical content (Table 1). Denaturation induced by increasing temperature (20 -70°C) (Fig. 5B) or guanidine hydrochloride concentrations (0 -8 M) (Fig. 5C) was monitored at a wavelength of 222 nm using CD. The relative thermodynamic stabilities of KP insulins were determined by comparing the relative change of ellipticity with increasing temperature at 1°C intervals (i.e. by comparing the slope of temperature denaturation curves). Native and KP insulins exhibited similar thermal denaturation curves, whereas the cis-dicarba KP insulin exhibited relatively small changes in ellipticity due to its significantly lower starting helical content ( Fig. 5B and Table 1). The cis-dicarba KP insulin is also considerably less stable upon guanidine hydrochloride denaturation, with ⌬G u 0 ϭ 1.71 kcal

A6 -A11 disulfide affects insulin B-chain conformation
mol Ϫ1 , as was seen with the dicarba insulin isomers ( Fig. 5C and Table 1) (24). The effect of introduction of the cis-dicarba bond on thermodynamic stability is striking and much greater than the introduction of the KP mutation (Fig. 5, B and C) (24). The observed reduction in stability is thus independent of being monomeric.

The cis-dicarba A6 -A11 linkage promotes a structural change in the B-chain helix, which leads to instability
To explain why the cis-dicarba peptides are less thermodynamically stable, we performed a limited proteolysis study that allowed us to detect structural differences between the cis-dicarba peptides, native insulin, and KP insulin. This involved RP-HPLC separation of fragments generated by chymotrypsin proteolysis and subsequent MS. First, we observed that cis-and trans-dicarba insulin isomers eluted later (15.8 and 14.6 min, respectively) than native insulin (13.3 min) when separated by RP-HPLC (Fig. 1D). Similarly, the cis-and trans-KP isomers also eluted with delayed retention times (16.3 and 15.1 min, respectively) compared with KP insulin (13.5 min) (Fig. 1E). The relative delay in the retention times of dicarba insulins is an indication of an apparent increase in hydrophobicity compared with the native forms, with cis-dicarba insulins being more surface hydrophobic. This was the first indication that the cis A6 -A11 dicarba linkage induces a more open conformation of insulin than in the native hormone and that it is likely that hydrophobic residues of the core are more exposed.
The enzymatic stability of dicarba insulin analogues was investigated through limited proteolysis by chymotrypsin under non-reducing conditions. The rate and kinetics of proteolysis were monitored through RP-HPLC as described under "Experimental procedures." Our results show that cis-dicarba insulin is significantly more rapidly cleaved by chymotrypsin compared with native insulin (Fig. 6A). At a protein/enzyme ratio of 86:0.08 M, the native insulin remained almost completely undigested after 3 h of proteolysis (Ͼ95% undigested peptide remaining), whereas almost no intact cis-dicarba insulin remained. The monomeric KP insulin is more susceptible to proteolysis than insulin with ϳ65% undigested peptide remaining after 3 h (ϳ30% reduction compared with native insulin). However, the cis-dicarba KP insulin is rapidly cleaved with only ϳ30% of undigested peptide remaining at 3 h (ϳ30% reduction compared with KP insulin). As different batches of chymotrypsin were used in the cis-dicarba insulin and cis-dicarba KP insulin experiments (Fig. 6A versus Fig. S4A), we cannot determine whether the difference in cleavage rates between the two cisdicarba analogues is significant. Clearly, both are much more rapidly cleaved than native insulin and KP insulin, respectively.
Our RP-HPLC chromatograms of a chymotrypsin-cleaved insulin ( Fig. 6B; also see Fig. S3E) and the cis-dicarba insulin Insulin fibrillation is first evident at t ϭ 6 h. After t ϭ 15 h, insulin has formed aggregates, and fibrils are no longer easily detectable. Fibrillation of cis-dicarba insulin is first evident after t ϭ 2 h and is clearly detectable by t ϭ 6 h. Fibrils formed by the cis-dicarba insulin are of a different structure compared with insulin, particularly evident at t ϭ 15 h. Fibrillation of the monomeric KP insulin is evident at earlier time points across a wider incubation range (t ϭ 2-15 h) compared with insulin. At t ϭ 24 h, KP insulin fibrils are no longer easily detectable. Surprisingly, cis-dicarba KP insulin fibrillation is only evident at t ϭ 6 -8 h with a rapid increase in formation of shorter and thicker fibrils at t ϭ 6 h. In summary, the cis-dicarba insulin adopts a different fibrillary pattern compared with insulin with an apparent increased complexity of fibrillary topology. These experiments are representative of n ϭ 4 experiments for insulin, n ϭ 3 for cis-dicarba insulin, n ϭ 5 for KP insulin, and n ϭ 3 for cis-dicarba KP insulin. First detection of fibrils for each analogue is indicated by white arrows. Scale bar (white), 1 m in all images. ( Fig. 6C; also see Fig. S3F) were comparable with previously reported data, which identified four non-reduced metabolites (termed A, B, C, and D) post-cleavage (39). Through the positive ion mode of MALDI mass analysis of the entire digest, we detected native insulin metabolites A, C, and D following the chymotrypsin digest (protein/enzyme ratio of 86:0.08 M; t ϭ 60 min), but metabolite B was not detected (see chromatograms in Fig. 6B and Fig. S3E, MALDI analysis in Fig. S5, and the schematic diagram in Fig. S7). By analysis of individual RP-HPLC fractions, we could assign the different masses to individual peaks (Figs. S5 and S6, tables in bottom panel). Metabolite A eluted ϳ1 min after the undigested peptide (Fig. 6, B and  C), as was seen for all digested insulins (native and dicarba analogues) (labeled as 2 (gray) in Figs. S3 and S4, E-G). Peptides equivalent to metabolite C and metabolite D eluted at t R of ϳ3.5 min and ϳ9.5 min, respectively, for all insulins (see Fig. 6, B, C, and F; labeled as 3 (metabolite C; orange) and 4 (metabolite D; green) in Figs. S3 and S4 (E-G)).

A6 -A11 disulfide affects insulin B-chain conformation
Close comparison of the RP-HPLC chromatograms revealed the presence of an apparently unique metabolite in chymotrypsin digests of cis-dicarba insulins (metabolite E in Fig. 6C; metabolite 5, cis-dicarba KP insulin in Fig. S4F) that was not detected in native insulin (Fig. 6B) or KP insulin (Fig. S4E). Metabolite E was detected in proteolytic samples of cis-dicarba insulin as early as t ϭ 1 h (kinetics of cleavage shown in Fig. S3, C and F). Analysis from MALDI data in negative ion mode identified the cis-dicarba insulin metabolite E from a fractionated and reduced sample (Fig. 6D). As shown in the simplified schematic diagram, Fig. 6F (orange box), the new metabolite is a full-length two-chain peptide (with intact A6 -A11 dicarba bond and A7-B7 and A19 -B20 disulfide bonds) with a single cleavage at the C-terminal end of Tyr B16 of the B chain (indicated in cis-dicarba insulin crystal structure in Fig. 6E, blue circled 1). This new metabolite E was not identified in native insulin ( Fig. 6B and Figs. S3 and S5) or KP insulin (Fig. S4) cleaved with chymotrypsin, suggesting that chymotrypsin is able to access Tyr B16 in cis-dicarba insulin more readily than in insulin or KP insulin.
To understand why only cis-dicarba insulin is cleaved at this new site, we superimposed insulin structures on the active site of chymotrypsin, ensuring that the tyrosine residue that is recognized by the enzyme was localized in the appropriate binding pocket (Fig. 7). Significant steric interactions between the insulin structure and chymotrypsin (Fig. 7B) revealed that it would be necessary for the B-chain helix of insulin to bend significantly to allow Tyr B16 to engage with the chymotrypsin enzyme. Such bending behavior is evident in our previously reported molecular dynamics (MD) simulations of insulin and its cisdicarba analogue (24). It is significant that this bending motion occurs more frequently and persists longer in cis-dicarba insulin (see below). Hydrogen bonds between Val B12 and Tyr B16 , Glu B13 and Leu B17 , and Tyr B16 and Gly B20 must all be broken, allowing the C␣-C␣ distances between Glu B13 and Tyr B16 and between Tyr B16 and Gly B20 to increase from their unperturbed values of ϳ5.5 Å to Ն7 Å (Fig. 7A). This allows Tyr B16 to occupy the chymotrypsin-binding pocket, with the two neighboring loops of the helix wrapping around the binding pocket walls (Fig. 7, compare B versus C). These changes in the B-chain helix structure are concomitant with the twisting of Cys B7 , generally resulting in a decrease of the C␣-C␣ distance across the A7-B7 disulfide linkage and an increase in its torsional strain. The effects of the bulging in the B-chain helix are thereby transmitted to the N-terminal helix of the A chain. This results in the N-terminal end of this latter helix being tilted away from the B-chain helix, although it is important to note that this confor-  Table 1). , ellipticity. B, differences in thermal unfolding were monitored by ellipticity at ϭ 222 nm and show that both the cis-and trans-dicarba insulins are considerably less stable than insulin. C, unfolding in the presence of guanidine hydrochloride (GdnHCl) demonstrates that both cis-dicarba analogues are considerably destabilized compared with insulin. ⌬G values derived from guanidine denaturation studies are listed in Table 1.

A6 -A11 disulfide affects insulin B-chain conformation
mation of the A chain is also seen in the absence of B helix bulging and also in circumstances when Cys B7 is not twisted.
The propensity of the B-chain helices of both insulin and the cis-dicarba insulin to undergo bulging can be assessed through the MD simulation data. In Fig. 8, this bulging is monitored via the Glu B13 -Tyr B16 C␣-C␣ distance (green). The corresponding Cys A7 -Cys B7 C␣-C␣ distances (purple) and the strain in the A7-B7 disulfide linkages (blue) are also shown. It is clear that, although bending of the B-chain helix can occur in both insulin and the cis-dicarba insulin, it occurs far more frequently in the cis-dicarba analogue (Fig. 8, A and B, outlined in black boxes), with the bulging events being more prolonged and showing increased spreading of the loops of the helix (Fig. 7A). As noted previously, the A-chain N-terminal helix is far more labile in the cis-dicarba insulin than in insulin (24). It appears that in the more stable insulin structure, the A6 -A11 disulfide linkage rapidly dampens the bending of the B-chain helix and restores its helical structure. In contrast, this conformational lability of ␣AN in the cis-dicarba insulin allows the bulge in the B-chain helix to form more readily and to persist long enough to allow engagement with and digestion by chymotrypsin. It is important to note that there does not appear to be any correlation between bulging of the B chain and the unwinding motion of the A chain necessary for complexation of the insulin receptor (Fig. S9).
A second important consequence of enhanced B-chain bulging for the cis-dicarba insulin is that the hydrophobic core of the hormone is opened up, resulting in increased exposure to solvent (Fig. 8, red). This is consistent with the longer observed elution time of the cis-analogue in the RP-HPLC experiments (Fig. 1D).
In summary, the limited proteolytic-MS integrated analyses revealed that the kinetics of dicarba insulin proteolysis are different from those of native insulin, leading to the formation of new metabolites. This supports the notion that installation of the intrachain dicarba bridge enhances structural perturbation near Tyr B16 to permit access of chymotrypsin to this site.

Discussion
For the last decade, insulin analogue design has focused on improving insulin efficacy and stability. Ideally, we require new rapid-acting insulin analogues that perfectly mimic the normal rapid onset of bolus insulin action. Desirably, insulin analogues would also be physically and chemically stable during pump delivery or at sites of injection. The "bottleneck" to creating the perfect insulin arises from our incomplete understanding of the relationship between insulin's structure and function, particularly with respect to the fine balance between activity and stability.
Previously, we reported the chemical synthesis of two A6 -A11 dicarba insulin analogues (cis-and trans-dicarba insulins). Using these stereoisomers, we obtained remarkable insight into the previously unexplored function of the insulin A6 -A11 disulfide bond in modulating insulin activity. Unique to these insulin analogues, only cis-dicarba insulin is biologically active, whereas trans-dicarba insulin is inactive. We proposed that the underlying cause of this difference lies in the structural dynamics of the A6 -A11 linkage and that this dictates insulin's ability to transition into its active conformation. We demonstrated that the configuration of the A6 -A11 linkage could modulate insulin's ability to engage with the receptor through its influence on the conformational flexibility of the N-terminal A-chain helix (24).
In the current study, we seek to explain why in vivo the cisdicarba insulin promotes more rapid lowering of blood glucose than native insulin. Taken together with our observation of reduced thermal and chemical denaturation stabilities, we hypothesized that the rapid action might be attributed to the cis-dicarba insulin being monomeric. To address this, we undertook biophysical analyses of the cis-dicarba insulin compared with the cis-isomer of dicarba KP insulin, which we assume is monomeric as per KP insulin.
We first compared receptor binding and biological activity of the dicarba KP insulins with the cis-and trans-dicarba insulins and native insulin. Consistent with the unique biological characteristics of cis-and trans-dicarba isomers of native insulin, the cis-dicarba KP insulin was also equipotent to native insulin (receptor binding, receptor activation, DNA synthesis in myoblasts, and glucose uptake by adipocytes), whereas the transdicarba KP insulin was inactive. This confirms the functional role of the native A6 -A11 cystine bridge. Even in the context of a disrupted dimer interface induced by the B-chain C-terminal KP mutation, the A6 -A11 bond still influences the ability of insulin to engage with the receptor.
Next, we determined that the cis-dicarba insulin is not inherently monomeric. The AUC results clearly showed that the distribution of cis-dicarba insulin into monomeric and dimeric forms under zinc-free conditions was similar to that of native insulin. This was surprising and prompted us to further explore Table 1 Structural

analyses of insulin, KP insulin, and dicarba insulins using CD
Helical content was calculated using the CONTINLL algorithm for deconvolution against the protein database reference set SP43. The program is available on the DICROWEB website (http://dichroweb.cryst.bbk.ac.uk/html/home.shtml). (Please note that the JBC is not responsible for the long-term archiving and maintenance of this site or any other third party hosted site.) ⌬G 0 values were derived from guanidine denaturation studies. NA, not applicable.

A6 -A11 disulfide affects insulin B-chain conformation
the biophysical differences between the cis-dicarba insulin and native insulin that might account for the more rapid action of the cis-dicarba analogue. Further evidence of structural differences between the dicarba insulins and native insulin was provided by our fibrillation assays. The qualitative differences in fibrillation rate and fibril conformation between the cis-dicarba insulin and native insulin are additional indicators of conformational differences in their structures. It is well established that the movement of the B-chain C terminus away from the B-chain helix and hydrophobic core promotes fibrillation, as is seen with KP insulin (Fig. 4) (8,33,40,41). The AUC results show that the cis-dicarba insulin is not inherently monomeric, suggesting that the dimer interface is not disrupted, and hence this is not the source of the increased fibrillation rate of the cis-dicarba insulin. Another key feature of fibril formation is the transition of the A chain N-terminal helix to a ␤-sheet, a process that requires displacement of the A chain away from the B-chain helix and from the hormone core (9,33). Therefore, we postulated that cis-dicarba insulin's increased rate of fibrillation compared with native insulin was likely to be connected to its increased flexibility and altered helicity (more -like) at the N terminus of the A chain.
However, upon further investigation using limited chymotrypsin proteolysis, we were able to detect an unexpected difference in the structure of the B chain between the cis-dicarba insulin and native insulin. The initial cleavage in the cis-dicarba insulin occurs at the C-terminal end of Tyr B16 of the B chain. This site is not the first site of cleavage in native insulin (the product was not detected), suggesting that the enzyme is unable to readily access this site in fully intact native insulin; cleavage at Tyr B16 only occurs after the molecule has been cleaved at other sites. In the cis-dicarba insulin, initial cleavage at Tyr B16 indicates that chymotrypsin can readily access this site, implying that the C-terminal end of the B chain has a tendency to be in a non-native, partially open, or bulged, conformation, thereby allowing enzyme access. Interestingly, this bulged con-  A, overlay of the bent structure (red; MD simulation frame for cis-dicarba insulin) with a reference insulin crystal structure (blue; PDB entry 1MSO, chains C and D). Residues A1-A21 and B9 -B23 are superimposed. Hydrogen bonds in the B-chain helix connecting residues Val B12 and Tyr B16 , Glu B13 and Leu B17 , and Tyr B16 and Gly B20 are broken, allowing the C␣-C␣ distances between Glu B13 and Tyr B16 and between Tyr B16 and Gly B20 to increase from ϳ5.5 to Ͼ7 Å. The bending rotates and compresses the A7-B7 disulfide bond and increases the distance between the N-terminal A-chain and B-chain helices (as measured by the C␣-C␣ distance between residues A6 and B11). B, superimposition of the B-chain helix of the unbent insulin crystal structure (stick model; PDB entry 1MSO, chains C and D) on the active site of chymotrypsin (dark green; PDB entry 4H4F, chymotrypsin in complex with inhibitor eglin C). For proteolysis to occur, Tyr B16 must be recognized by the active site of the enzyme. Superimposition of the backbone atoms of Tyr B16 as well as N and CA of Leu B17 on those of the corresponding residues in the inhibitor reveals that the unbent structure cannot engage correctly with the peptidase; residues B12 and B13 and all residues beyond B18 overlap significantly with the chymotrypsin structure, and the side chain of Tyr B16 cannot sit properly in the active-site cavity. C, superimposition of the B-chain helix of the bent cis-dicarba insulin simulation frame on the active site of chymotrypsin. Bulging of the B-chain helix allows both the B12-B16 and B16 -B20 loops to fit over the surface of the chymotrypsin molecule, with Tyr B16 sitting in the middle of the binding pocket. (Note that in both B and C, the side chains of B16, B17, and chymotrypsin residues 143 and 192 have been rotated to give the best possible engagement of the two molecules.) Loops of chymotrypsin that must move out of the way to allow insulin engagement are shown as dark green ribbons.

A6 -A11 disulfide affects insulin B-chain conformation
formation does not affect the ability of the cis-dicarba insulin to bind the insulin receptor. Previous mutation studies at Tyr B16 (e.g. to His or Ala) highlighted the importance of this residue in receptor binding as well as its involvement in the dimer interface. Interestingly, the Tyr B16 3 Ala insulin mutant behaves as a monomer on size-exclusion chromatography (42).
Previously, MD simulations have captured insulin in both "open" and "wide open" (receptor-bound) states (23). In that study, the open state referred to a zipper-like opening from the end of the B chain. This state was also observed in our MD investigations of native insulin and its cis-dicarba analogue. However, here we additionally observe that simulations of the cis-dicarba insulin show a significantly enhanced propensity for outward bulging of the B-chain helix. This opens up the helix loop between Val B12 and Leu B17 , exposing Tyr B16 for chymotrypsin cleavage (see Fig. 7). Hence, the cis-dicarba linkage causes two fundamental changes to the dynamics of the insulin structure. On the one hand, it increases the mobility of the A-chain N-terminal helix, enhancing its ability to engage favorably with the insulin receptor. On the other, it decreases the stability of the overall structure. This is seen both in the destabilization of the B-chain helix, increasing its susceptibility to chymotrypsin digestion, and in an increase in the solvent exposure of the hydrophobic core of insulin, leaving the hormone vulnerable to degradation via the formation of fibrils.
Recently Wade and co-workers (43) reported that the dicarba substitution of the intra-A-chain disulfide bond of the insulin-like peptide H2 relaxin resulted in significantly reduced stability to enzyme degradation in plasma, despite maintaining the ability to bind relaxin's cognate receptor RXFP1. The mechanisms underlying the instability were not explored. Similar to dicarba insulins, dicarba H2 relaxins also displayed structural differences from the native peptide. We postulate that the role of the intra-A-chain disulfide bond in regulating peptide stability is conserved across the insulin-relaxin superfamily. However, further investigation is required to confirm this.
The rapid action of both cis-dicarba insulin and cis-dicarba KP insulin in lowering blood glucose (compared with native insulin) is most likely explained by the increased mobility of the A-chain N-terminal helix, as observed in our MD simulations. This enhances the ability of cis-dicarba insulin to adopt a conformation consistent with IR engagement. By restraining the A6 -A11 linkage through introduction of a cis-dicarba bond, the bioactive conformation of the molecule is favored, and hence the in vivo metabolism of glucose is promoted. It is as yet unclear whether the accompanying destabilization of the B-chain helix also contributes to the enhanced glucose consumption.
In conclusion, through introduction of a non-reducible A6 -A11 dicarba bridge of fixed configuration into insulin and KP insulin, we reveal the functional and structural roles of this linkage. It not only regulates structural flexibility at the N terminus of the A-chain helix, which is necessary for receptor binding, but also influences the frequency at which the B-chain helix "bulges," thereby increasing insulin mimetic's vulnerability to heat, chemical, and enzymatic degradation.
Our findings suggest that there is potential for the development of ultra-rapid insulin analogues through only minimal manipulation of native insulins or existing insulin analogues. Importantly, this study provides deeper insight into the function of the A6 -A11 bond in regulating the balance between optimal insulin potency and structural stability. We anticipate that this detailed understanding of the structural dynamics of insulin will aid in future design of rapid-acting insulin analogues with improved stability. Defining the determinants of receptor binding (including mechanisms driving necessary Aand B-chain flexibility), stability, and fibrillation will allow us to design analogues that permit high-affinity binding but avoid instability and fibrillation.

Materials
Actrapid insulin was purchased from Novo Nordisk Pharmaceuticals Pty. Ltd. Humalog lispro (KP insulin) was obtained from Eli Lilly Australia. Hybridoma cells expressing Green, r(C␣-C␣) between Glu B13 and Tyr B16 ; this is a direct measure of helix bending. Red, r(C␣-C␣) between residues A6 and B11; a measure of the distance between the N-terminal A-chain and B-chain helices. Purple, r(C␣-C␣) between Cys A7 and Cys B7 ; the length of the A7-B7 interchain linker. Blue, the relative torsional energy of the A7-B7 disulfide linkage. Data are presented as 50 period moving averages to highlight trends. Full data are presented in Fig. S8. Overall averages for the purple and blue traces are shown with dashed and dotted lines, respectively. Bond distances are given in Å, and relative torsional energies are shown in kJ mol Ϫ1 . Bending events in the B-chain helix are outlined in black boxes. Note that each bin on the horizontal axis represents an independent 200-ns simulation; traces are therefore not continuous between bins.

A6 -A11 disulfide affects insulin B-chain conformation
antibodies specific for the IR ␣-subunit (83-7) and the IGF-1R ␣-subunit (24 -31) were a kind gift from Prof. K. Siddle (44 -46). [ 3 H]Thymidine and Eu-PY20 were purchased from PerkinElmer Life Sciences. Sequencing grade chymotrypsin was purchased from Promega. The sodium salt of iodoacetic acid (IAA), HPLC grade acetonitrile (CH 3 CN), and TFA and DTT were purchased from Sigma-Aldrich. Iodoacetamide was a product of Bio-Rad. All antibodies were purchased from Cell Signaling Technology (Danvers, MA) unless specified. The matrices for MALDI-TOF MS (sinapinic acid and ␣-cyano-4hydroxycinnamic acid) were products of Bruker Daltonics (Leipzig, Germany). All other reagents used were analytical grade.

Chemical synthesis of dicarba insulins
The synthesis of c[⌬ 4 A6,11]-dicarba insulin (cis-and transdicarba insulins) (24) and c[⌬ 4 A6,11]-dicarba KP insulins (cisand trans-dicarba KP insulins) were essentially the same. Synthesis of dicarba A chain was achieved through interrupted SPPS-catalysis and RCM procedures (24, 34). Construction of insulin and KP insulin B-chain was achieved through microwave-accelerated SPPS. The monocyclic A-B conjugates were prepared by combination of the dicarba A chains with the insulin (or KP insulin) B chain under basic conditions, resulting in spontaneous oxidation of the liberated free thiol groups generating cis-and trans-dicarba insulins (or cis-and trans-dicarba KP insulins). The details of the synthesis method for the dicarba KP insulins are provided in the supporting material. The synthesis method for dicarba insulins has been described previously (24).

Immunocaptured receptor-binding assay
IR-A, IR-B, and IGF-1R binding was measured essentially as described by Denley et al. (47). Human IR isoform A (IR-A), isoform B (IR-B), and IGF-1R were solubilized from R Ϫ IR-A, R Ϫ IR-B, and P6 cells, respectively. Briefly, cells were serumstarved in serum-free medium (SFM) containing 1% BSA for 4 h before lysis in ice-cold lysis buffer (20 mM HEPES, 150 mM NaCl, 1.5 mM MgCl 2 , 10% (v/v) glycerol, 1% (v/v) Triton X-100, 1 mM EGTA, and 1 mM phenylmethylsulfonyl fluoride, pH 7.5) for 1 h at 4°C. Lysates were centrifuged for 10 min at 2,200 ϫ g, and then 100 l of lysate was added per well to a white Greiner Lumitrac 600 96-well plate previously coated with anti-IR antibody 83-7 or anti-IGF-1R antibody 24-31 (250 ng/well in bicarbonate buffer, pH 9.2). Approximately 500,000 fluorescent counts of europium-labeled insulin (Eu-insulin, prepared inhouse) were added to each well along with increasing concentrations of unlabeled competitor in a final volume of 100 l and incubated for 16 h at 4°C. Wells were washed four times with 20 mM Tris, pH 7.4, 150 mM NaCl, and 0.1% (v/v) Tween 20 (TBST). Then 100 l/well DELFIA enhancement solution (PerkinElmer Life Sciences) was added. After 10 min, time-resolved fluorescence was measured using 340-nm excitation and 612-nm emission filters with a BMG Lab Technologies Polarstar fluorometer (Mornington, Australia). Assays were performed in triplicate in at least three independent experiments.

Kinase receptor activation assay
IR-A, IR-B, and IGF-1R phosphorylation was detected essentially as described by Denley et al. (47). Briefly, R Ϫ IR-A, R Ϫ IR-B, or P6 cells (5 ϫ 10 4 cells/well) were plated in a 96-well flat-bottom plate and grown overnight at 37°C in 5% CO 2 . Cells were serum-starved for 4 h before treatment with insulin or dicarba insulins in 100 l of SFM, 1% BSA for 10 min or in a time course (0, 2, 5, 8, 12, 20, and 30 min) at 37°C and 5% CO 2 . Cells were lysed with ice-cold lysis buffer containing 2 mM Na 3 VO 4 and 100 mM NaF, and receptors were captured onto white Greiner Lumitrac 600 96-well plates precoated with anti-IR antibody 83-7 or anti-IGF-IR antibody 24-31 (250 ng/well) (44) and blocked with TBST/0.5% BSA. Following overnight incubation at 4°C, the plates were washed three times with TBST. Phosphorylated receptor was detected by incubation with Eu-PY20 (76 ng/well) at room temperature for 2 h. Wells were washed four times with TBST, and time-resolved fluorescence was detected as described above. Assays were performed in triplicate in at least three independent experiments.

DNA synthesis assay
DNA synthesis was carried out as described by Gaugin et al. (51). Briefly, hIR-A L6 myoblasts were plated in a 96-well flatbottom plate (1.5 ϫ 10 4 cells/well) and grown overnight at 37°C in 5% CO 2 . Cells were starved in SFM/1% BSA for 4 h before treatment with increasing ligand concentrations for 18 h in SFM/1% BSA. The cells were incubated with 0.13 Ci/well [ 3 H]thymidine for 4 h, shaken for 2 h with 50 l of disrupting buffer (40 mM Tris, pH 7.5, 10 mM EDTA, and 150 mM NaCl), and then harvested onto glass fiber filters (Millipore) using a MICRO 96 TM Skatron harvester (Molecular Devices). The filters were counted in a Wallac MicroBeta counter (PerkinElmer Life Sciences). Assays were performed in triplicate in at least three independent experiments.

A6 -A11 disulfide affects insulin B-chain conformation
confluence and were then differentiated into adipocytes as described (52). Glucose uptake in response to insulin and dicarba insulin analogues was measured essentially as described (53). Briefly, 3T3-L1 adipocytes were serum-starved in SFM/1% BSA for 4 h, washed twice with Krebs-Ringer phosphate buffer (KRP; 12.5 mM HEPES, 120 mM NaCl, 6 mM KCl, 1.2 mM MgSO 4 , 1 mM CaCl 2 , 0.4 mM Na 2 HPO 4 , and 0.6 mM Na 2 HPO 4 (pH 7.4)) containing 1% BSA, and incubated for 15 min at 37°C. Insulin or insulin analogues were added at decreasing concentrations (from 100 to 0.3 nM) for 30 min at 37°C. For the final 10 min, 2-deoxyglucose uptake was initiated by the addition of 50 M cold deoxyglucose and 1 Ci of [ 3 H]deoxyglucose per well. The assay was terminated by rapidly washing the cells three times with ice-cold KRP. Cells were solubilized in 0.5 M NaOH, 0.1% SDS, and 3 H content was determined by scintillation counting. Non-specific 2-deoxyglucose uptake was determined in the presence of 50 M cytochalasin B.

Insulin tolerance test
Eight-week-old C57BL6 male mice were fed either a standard rodent chow diet containing (w/w) 77% carbohydrate, 20% protein, and 3% fat from Ridley AgriProducts (Pakenham, Victoria, Australia) or a high-fat diet containing (w/w) 57% carbohydrate, 19% protein, and 15% fat from Specialty Feeds (SF08-044, Glen Forrest, Western Australia, Australia) for 12 weeks. Mice (5 or 6 mice per group) were injected intraperitoneally with 0.75 IU/kg insulin or insulin analogues under non-fasting conditions, and tail vein blood glucose was measured via a glucometer at the indicated times (54). Experimental procedures were carried out in accordance with the protocols approved by the Austin Health Animal Ethics Committee (AEC 2011/04396). Native insulin (Actrapid) and KP insulin (Humalog) were administered as formulated peptide diluted to the correct dose in PBS, pH 7.4 (semi-formulated). Both cis-dicarba insulin and cis-KP dicarba insulin were dissolved in 10 mM HCl and diluted to the correct dose in PBS, pH 7.4 (non-formulated). Subsequent to this experiment, we found no difference between semi-formulated KP insulin and non-formulated KP insulin in an insulin tolerance test (Fig. S10), demonstrating that the vehicles had no effect on the insulin tolerance test outcomes.

Sedimentation equilibrium AUC
Analytical ultracentrifugation was conducted at 20°C using a Beckman XLI analytical centrifuge in 12-mm path-length cells. cis-Dicarba insulin was diluted from a 10 mg/ml stock in 10 mM HCl into 100 l of 10 mM sodium phosphate, 100 mM NaCl, pH 7.4, with or without 0.2 mM ZnCl 2 to a final concentration of 100 or 300 g/ml. Radial concentration distributions were measured by absorbance at 230 nm. Sedimentation equilibrium was established at 25,000 and 40,000 rpm, as assessed by sequential absorbance scans 2 h apart. Data at both speeds were jointly fit to a single ideal sedimenting species or to various models of self-association in SEDPHAT (55), using values of solution density and solute partial specific volume estimated from composition using SEDNTERP (56), neglecting any effect of the dicarba modification. Mass conservation was applied as a constraint on the fits by floating the position of the bottom of the cell. Reported errors describe the precision of the fit at 0.68 confidence level, estimated from Monte Carlo simulations as implemented in SEDPHAT.

AFM
Analysis of fibrillation was monitored by AFM essentially as described (57). Briefly, lyophilized insulin, insulin lispro, and dicarba isomers were resuspended in 200 mM KCl-HCl in Milli-Q water, pH 1.6, at 200 M (1.16 mg/ml) and incubated at 60°C with gentle agitation. Samples (5 l) were taken at different times, diluted 1:10 in the same buffer, and immediately frozen. 30 l of the insulin solution was dropped onto a freshly cleaved mica substrate. Once dried, the sample was washed dropwise with Milli-Q water and then dried with a gentle stream of dry nitrogen. Images of the protein aggregates were recorded with a Multimode Nanoscope IV atomic force microscope (Veeco Instruments, Santa Barbara, CA), operating in Tapping Mode. Rigid cantilevers with resonance frequencies of 325 kHz and equipped with silicon tips (HQ:NSC15, Mikromasch, Sofia, Bulgaria) with nominal spring constant of 40 newtons/m and a nominal tip diameter of 16 nm were used. Typical scan size was 5 m (512 ϫ 512 points), and the scan rate was 1-2 Hz. The scanner was calibrated in the x, y, and z axes using silicon calibration grids (Bruker model numbers PG (1-m pitch, 110-nm depth) and VGRP (10-m pitch, 180-nm depth)). Images were analyzed using the Nanoscope analysis program version 1.40.

RP-HPLC
Purification of insulin analogues was performed on a Vydac C 4 analytical column (214TP5210; 5 m, 2.1 ϫ 100 mm, 300 Å) connected to the Agilent 1260 Infinity Quaternary LC system. Buffer A contained 0.1% aqueous TFA. Buffer B contained 80% CH 3 CN in 0.08% TFA. Peptides were eluted using a linear gradient of 20 -25% CH 3 CN for 5 min, followed by a 25-38% CH 3 CN gradient over 13 min at a flow rate of 0.5 ml/min. UV detection was at 280 nm, and peak areas were used for quantitation of peptides.

CD
CD was carried out as described previously (24). Briefly, CD spectra were recorded on a Jasco J-815 CD spectrometer. Spectra were from 260 to 190 nm with a 1.0-nm step size using a 1.0-s response time and 1.0-nm bandwidth in a quartz cuvette with a 0.1-cm path length. Insulin and insulin analogues were incubated in 10 mM phosphate buffer (pH 7.4) to a concentration of 0.22 mg/ml (38 M). Spectra were background-corrected by subtraction of the spectrum of buffer alone. Temperature denaturation was achieved by automated thermal control, increasing by 2°C/min at 1°C intervals. Samples were diluted to 10 M for equilibrium denaturation studies in guanidine hydrochloride (1-8 M where MRW is the protein mean weight ((atomic mass units/ Da)/number of residues)), P is path length (cm), and Conc. is

A6 -A11 disulfide affects insulin B-chain conformation
protein concentration in mg/ml. [] 222 is the molar ellipticity per residue at wavelength 222 nm.

In-solution time-course limited proteolysis assay
Limited proteolysis analyses were performed in 100 mM Tris-HCl, pH 7.5, using sequencing grade chymotrypsin at a final protease/peptide ratio of 1:50 or 1:250 (w/w), incubated at 37°C. The final concentration of peptide was 0.5 mg/ml (86 M). The proteolytic reactions were analyzed on a time-course basis by sampling 10 l of proteolytic mixture (equivalent to 5 g of insulin) into 100 l of 0.5% (v/v) TFA at t ϭ 0, 1, 2, and 3 h. Digested samples were analyzed and fractionated (fraction size: 0.25 or 0.5 ml) using RP-HPLC and lyophilized. The extents of proteolysis were analyzed as follows.

Preparation of proteolyzed insulin samples for MS
Lyophilized fractions were resuspended in 20 l of 100 mM ammonium bicarbonate (NH 4 HCO 3 ). Half of the volume was kept separately as non-reducing samples for MS analysis. The other half was reduced in 5 mM DTT for 45 min at 56°C and alkylated with 14 mM iodoacetamide (for positive mode) or IAA (for negative mode) for 30 min in the dark at room temperature.

MALDI-TOF MS
The molecular weights of intact insulin analogues and the proteolytically cleaved metabolites were identified using a MALDI-TOF Autoflex TM III mass spectrometer (Bruker Daltonics, Leipzig, Germany) equipped with an Nd:YAG laser (wavelength, 355 nm). Either the entire cleavage mixture or isolated peaks separated by RP-HPLC were subjected to mass spectral analysis using combinations of positive or negative ion with linear or reflectron mode. Insulin digest samples intended for mass analysis using positive mode were desalted using Zip-Tip C 18 (Millipore, Billerica, MA), eluted with 2 l of 0.1% TFA/CH 3 CN (70:30, v/v). Samples intended for negative mode analysis were desalted and eluted with no TFA. 1 l of each sample was spotted onto the MTP 384 ground steel target plate (Bruker Daltonics) and air-dried, and then 1 l of sinapinic acid matrix was spotted on top of each dried sample. Calibrations were performed using Bruker Daltonics' Peptide Calibration Standard II and Protein Calibration Standard I according to the manufacturer's instructions.

Molecular dynamics
Previously, MD simulations of insulin and its cis-dicarba analogue were performed, and the different behaviors of the N-terminal A-chain helix were analyzed in detail (24). Here, these simulation data are further analyzed to give insight into the effect of the A6 -A11 linkage on insulin structural stability. Details of the simulation and analysis protocols are provided in our previous report (24) and in the supporting material. In brief, the AMBER14 program package (58) in conjunction with Amber ff14SB force field parameters (59) (along with calculated parameters for the dicarba linkage) were used to perform 200-ns room temperature MD simulations for insulin and its dicarba isomers based on four different starting structures derived from high-resolution X-ray crystallography experiments. Twelve independent 200-ns simulations were performed for insulin, and 16 simulations were performed for the cis-dicarba insulin. Simulation frames recorded at 200-ps intervals were used for the analysis in this paper (1,000 frames/simulation). Data were processed and analyzed using CPPTRAJ (60), along with custom computer scripts. The torsional strain in the A7-B7 disulfide linkages was estimated based on the data reported previously (61,62). Insulin structures were superimposed on the crystal structure of a chymotrypsin-inhibitor complex (PDB entry 4H4F) using Swiss-PDBViewer (63) and VMD (64).

Statistical analyses
Statistical analyses of receptor binding, receptor activation, and DNA synthesis assays were performed using a two-way ANOVA with Dunnett's multiple comparison. Insulin tolerance test and glucose uptake assay data were analyzed with a paired t test. Significance was accepted at p Ͻ 0.05.