Supramolecular Protein Engineering

Bottom-up control of supramolecular protein assembly can provide a therapeutic nanobiotechnology. We demonstrate that the pharmacological properties of insulin can be enhanced by design of “zinc staples” between hexamers. Paired (i, i+4) His substitutions were introduced at an α-helical surface. The crystal structure contains both classical axial zinc ions and novel zinc ions at hexamer-hexamer interfaces. Although soluble at pH 4, the combined electrostatic effects of the substitutions and bridging zinc ions cause isoelectric precipitation at neutral pH. Following subcutaneous injection in a diabetic rat, the analog effected glycemic control with a time course similar to that of long acting formulation Lantus®. Relative to Lantus, however, the analog discriminates at least 30-fold more stringently between the insulin receptor and mitogenic insulin-like growth factor receptor. Because aberrant mitogenic signaling may be associated with elevated cancer risk, such enhanced specificity may improve safety. Zinc stapling provides a general strategy to modify the pharmacokinetic and biological properties of a subcutaneous protein depot.

Supramolecular chemistry envisages the construction of novel materials and nanoscale devices ranging from molecular sensors to stimulus-responsive polymers (1,2). Spatial organization may be achieved either by templating (design from the top down) or through self-assembly of molecular components (bottom up). Connective tissue provides an example of the bottom-up design of a biomaterial based on hierarchical self-as-sembly of the collagen triple helix. Can such general principles be exploited in pharmacology? The present study explores the application of protein engineering (3) to the supramolecular chemistry of a therapeutic subcutaneous depot. Bottom-up control is accomplished by the pH-dependent binding of metal ions within and between self-assembled structures.
A model is provided by insulin, a protein containing two chains, A (21 residues) and B (30 residues) (Fig. 1A) (4). The hormone is protected from misfolding in ␤-cells by Zn 2ϩ -stabilized assembly and microcrystallization (5). Storage hexamers dissociate on secretion, enabling insulin to function as a dilute zinc-free monomer. Zn 2ϩ -stabilized hexamers are also employed in pharmaceutical formulations to delay insulin degradation (6). Treatment of diabetes mellitus increasingly employs insulin analogs with altered pharmacokinetics (7). Such pharmacokinetic "tuning" has enabled more effective glycemic control. Two classes of analogs extend the properties of wild-type formulations. More rapid acting analogs are designed to limit self-assembly (8) or accelerate disassembly (9). Longer acting analogs are designed to promote self-assembly in the subcutaneous depot (10) or retard disassembly (11,12). These classes each exploit an inverse relationship between the size of a subcutaneous protein complex and rate of capillary absorption (8,9,12).
Although insulin is highly soluble at pH values Ͼ7 or Ͻ5, between these limits the protein reversibly undergoes isoelectric precipitation (13). This property underlies the design of long acting analog insulin glargine (the active ingredient of Lantus) (12). Its B-chain is extended by paired Arg residues (Fig. 1A, top, red) whose positively charged side chains shift the pI from 5.8 to 7.0 (11,12). Lantus is formulated at pH 4 as a clear unbuffered solution that, following subcutaneous injection, undergoes pH-dependent precipitation to form a slow release depot (11). Insulin glargine exhibits increased cross-binding to the type 1 insulin-like growth factor receptor (IGF-1R), 3 a homolog of the insulin receptor (IR) (12,14). Augmented mitogenic signaling inhibits apoptosis and can drive the proliferation of cancer cells (14). The impaired specificity of the analog has thus raised safety concerns (15). A recent clinical case control study of Lantus reported a dose-dependent increase in IGF-1R-related malignancies (16). This statistical analysis and its clinical implications have provoked intense debate. 4 We present a new strategy for design of long acting depots: engineered "zinc staples" between protein assemblies. The essential idea envisaged a zinc-dependent switch between an insoluble subcutaneous depot and a soluble metabolic signal. To implement this strategy, paired His substitutions (Glu A4 3 His and Thr A8 3 His; Fig. 1A, bottom) were introduced at positions (i, iϩ4) of an ␣-helix (Fig. 1B), thus recapitulating part of a Zn 2ϩ -binding site (17). Such paired His elements are ubiquitous among classical zinc fingers (supplemental Fig. S1). Positions A4 and A8 were chosen as sites of modification to avoid structural perturbation, preserve activity (18), and maintain classical zinc assembly (4,19). Choice of this A-chain ␣-helix was further motivated by potentially favorable effects of its modification on the ratio of IR:IGF-1R binding (20,21). Zincstapled insulin hexamers exemplify a structure-based approach to the design of a subcutaneous protein depot. The stringent receptor selectivity and supramolecular assembly of the analog may enhance the safety and efficacy of insulin therapy.
Synthesis of Insulin Analog-Variant insulin chains were prepared by solid-phase synthesis (22). Wild-type S-sulfonate B-chain derivatives were obtained by oxidative sulfitolysis of insulin (23). Insulin analogs were prepared by chain combination and purified as described (22,23). The paired His A-chain substitutions were also incorporated into engineered monomer DKP-insulin (His B10 3 Asp, Pro B28 3 Lys, and Lys B29 3 Pro) (24). Yields were similar to that of wild-type insulin. Predicted molecular masses were confirmed by mass spectrometry.
Biochemical Assays-The effects of the substitutions on pI relative to insulin and insulin glargine were measured by native isoelectric-focusing gel electrophoresis (supplemental Methods). Solubility was tested in dilute HCl (pH 4.0) in the presence of 0.52 mM ZnCl 2 , 185 mM glycerol, and 25 mM metacresol and on dilution into a buffered solution of 10 mM Tris-HCl (pH 6 -9) and 140 mM NaCl by the method of DiMarchi and co-workers (20) (supplemental Methods).
Crystallography-Crystals were obtained by hanging-drop vapor diffusion at room temperature in the presence of a 1:1.7 ratio of Zn 2ϩ :protein monomer and a 3.5:1 ratio of phenol:protein monomer in Tris-HCl (25). Diffraction was observed using synchrotron radiation at a wavelength between 1.0000 and 1.2398 Å at the Advanced Light Source (beamline 4.2.2; Berkeley, CA); crystals were flash-frozen to 100 K. The lattice contained one TR f dimer per asymmetric unit. The structure was determined by molecular replacement (supplemental Methods). Of the 86 residues in the refined model of the TR f dimer in the asymmetric unit (excluding 8 Gly, 2 Pro, and 6 end residues), 79 residues (92%) lie in the most favored Ramachandran region, and 7 residues (8%) lie in generously allowed regions.
1 H-NMR Spectroscopy-Native self-assembly was prevented by "DKP" substitutions as described (24). Spectra of [His A4 , His A8 ]DKP-insulin were acquired in D 2 O at 700 MHz and 25°C. The analog was made 0.5 mM in 10 mM deuterated Tris-DCl (pD 7.6, direct meter reading) in the presence or absence of equimolar ZnCl 2 .
Receptor Binding Assays-IR (isoform B) and IGF-1R binding assays were performed by a microtiter plate antibody capture assay (supplemental Methods) (21). Wild-type insulin, insulin glargine, and IGF-I provided controls. Data were analyzed by non-linear regression using a two-site sequential model (21). The percentage of tracer bound in the absence of competing ligand was Ͻ15% to avoid ligand depletion artifacts. Wild-type A-and B-chains are shown in black and green, and disulfide bridges (A6 -A11, A7-B7, and A20 -B19) are shown as black lines. Glargine contains a 2-residue B-chain extension (Arg B31 and Arg B32 ) and substitution Asn A21 3 Gly (upper panel, red). Endogenous subcutaneous proteases may slowly remove one or both Arg residues, in part alleviating its augmented mitogenicity (31). The present analog contains substitutions Glu A4 3 His and Thr A8 3 His (lower panel, red). Long acting analog insulin detemir (Levemirா, Novo-Nordisk) operates by attachment of an albumin-binding element (not shown) (32). B, ribbon model of insulin monomer depicting the portion of putative zinc-binding site (violet) formed by His A4 and His A8 (red) at the external surface of A1-A8 ␣-helix. A-and B-chain ribbons are shown in black and green. C and D, structures of wild-type (C) and variant (D) T 3 R 3 f insulin hexamers. The two axial zinc ions within each hexamer are aligned at the center (violet), coordinated by trimer-related His B10 side chains (light gray). The variant hexamer contains three non-classical zinc ions at T 3 trimer surface (D, magenta spheres). Shown in red are side chains of His A4 , His A8 , and the third His A4Ј from the adjoining hexamer. In each case, the A-chains are shown in black, and B-chains are in green or blue (R f -specific B1-B8 ␣-helix). The wild-type structure was obtained from the Protein Data bank (entry 1TRZ). E, 2F o Ϫ F c electron density map (stereo pair contoured at 1 ) showing a novel zinc-binding site formed by His A4 and His A8 in the T-state protomer. Distorted tetrahedral coordination is completed by residue A4Ј, which belongs to an R f -state protomer in adjoining hexamer.
Rodent Assay-Male Lewis rats (mean weight ϳ300 g) were rendered diabetic by streptozotocin (26). The effects of insulin analogs on blood glucose concentration following subcutaneous injection were assessed using a clinical glucometer (Hypoguard Advance Micro-Draw meter) in relation to wild-type insulin or buffer alone (16 mg of glycerin, 1.6 mg of meta-cresol, 0.65 mg of phenol, and 3.8 mg of sodium phosphate (pH 7.4); Lilly diluent). Wild-type insulin and [His A4 , His A8 ]insulin were made zinc-free in the above buffer. [His A4 , His A8 ]insulin and insulin glargine were also dissolved in dilute HCl (pH 4) containing a 5.2:1 ratio of ZnCl 2 :insulin monomer, 25 mM metacresol, and 185 mM glycerol. Rats were injected subcutaneously with 3.44 nmol of insulin or insulin analogs (ϳ12-13.7 nmol) in 100 l of buffer per rat (for wild-type insulin, this corresponds to 2 units/kg). For neutral zinc-free formulations, blood was obtained from the tail every 10 min from 0 -90 min. For acidic zinc-containing formulations, blood was obtained at times 0, 1, 2, 4, 6, 10.8, and 24 h.

RESULTS AND DISCUSSION
The pI of [His A4 , His A8 ]insulin was found to be 6.6. Although highly soluble at pH 7.4 in the absence of zinc ions, the analog precipitated at Zn 2ϩ :insulin molar ratios above 0.4. Solubility was largely regained at pH values Ͼ8.5 at room temperature, presumably due to deprotonation of N-terminal ␣-amino groups. Such precipitation requires native protein assembly as 1 H-NMR studies of an engineered monomeric analog (24) indicated that the paired His element does not bind equimolar Zn 2ϩ in this context (supplemental Fig. S2). Lattice contacts between wild-type hexamers in insulin crystals provided a model for structural mechanisms of assembly-dependent zinc binding. The proximity of A4 and A8 side chains across such contact surfaces suggested that tetrahedral zinc-binding sites could be formed between successive [His A4 , His A8 ]insulin hexamers. Novel A4 -A8-mediated zinc coordination might therefore be possible near sites of bound interfacial water molecules in wild-type crystal lattices ( Fig. 2A, blue spheres, shown in stereo in supplemental Fig. S3A).
To test this hypothesis, we determined the crystal structure of [His A4 , His A8 ]insulin. Crystals were grown in the presence of Zn 2ϩ and phenol to yield T 3 R 3 f hexamers (27). The structure was obtained by molecular replacement at 1.9 Å resolution ( Table 1). The mode of hexamer assembly of the analog (Fig. 1D) is identical to that of wild-type insulin (Fig. 1C). Respective conformations of T and R f protomers are essentially identical to those of wild-type insulin (supplemental Fig. S4 and supplemental Table S1). No transmitted perturbations occur at receptor-binding surfaces (4).
Wild-type and variant hexamers each contain two axial zinc ions, one per T 3 and R 3 f trimer (Fig. 1, C and D, overlaid violet  spheres). Coordination at each site is mediated by trimer-related His B10 side chains with distorted tetrahedral geometry (Fig. 2, C and D, light gray at center of hexamers, and supplemental Table S2). Stereo views of the variant TR f dimer in the asymmetric unit and of the zinc-binding surface of the T protomer are provided in supplemental Fig. S5. In the R 3 f trimer, the fourth ligand is a chloride ion; in the T 3 trimer, this site (more exposed than in the R 3 f trimer) exhibits partial occupancy by either a chloride ion or a bound water molecule (supplemental Fig. S6). These features are consistent with wildtype structures (27). As is also observed in wild-type crystals grown under similar conditions, the R 3 f trimer contains three bound phenol molecules (not shown). The A4 and A8 substitutions thus do not block the TR transition (22), a classical model for the reorganization of insulin on receptor binding (4).
The variant T 3 R 3 f hexamer contains three additional trimerrelated zinc ions at the T-state surfaces (Fig. 1, B and D, magenta spheres; see also supplemental Fig. S5). These novel zinc ions are coordinated in part by His A4 and His A8 at an interfacial site. Representative electron density at the peripheral zinc-binding site defines a distorted tetrahedral site (Fig. 1E, supplemental Table S2). Coordination is completed by a chloride ion and a "stapled" His A4 side chain belonging to an R f protomer of an adjoining hexamer (Fig. 1E, labeled A4Ј, and Fig.  2B, brown arrows). Views of the opposing T and R f faces of adjoining hexamers are shown in Fig. 2C (90°rotated from the orientation shown in Fig. 2B). Binding of the chloride ion is stabilized by a network of three water molecules bound to the R f protomer (Fig. 2D, blue spheres in stereo pairs); His A8 in R f is displaced from the zinc-binding site. The three non-classical zinc ions thus bridge the T 3 and R f 3 trimers of adjacent hexamers (Fig. 2, B-D, magenta spheres, and in stereo in supplemental Fig. S3B), in part displacing water molecules ordinarily bound at the wild-type interface ( Fig. 2A, blue spheres). N-Zn 2ϩ bond distances and angles are similar to those of the axial metal ion-binding sites (supplemental Table S2). Sidechain conformations of His A4 and His A8 differ between T and R f protomers (supplemental Table S3).
Studies of hormone binding to IR and IGF-1R were undertaken to assess relative affinities and receptor binding selectivity (Fig. 2E and supplemental Table S4). Ligands were characterized as zincfree monomers. Relative to the binding of human insulin to IR and IGF-IR (Fig. 2E, solid and dotted green lines, respectively), insulin glargine (solid and dotted blue lines) exhibits 2-fold reduced affinity for IR and 3-fold enhanced affinity for IGF-1R. By contrast, [His A4 , His A8 ]insulin exhibits native-like affinity for IR (Fig. 2E, solid red line) but 6-fold reduced affinity for IGF-1R (dotted red line shifted to right). Thus, although the receptor binding selectivity of insulin glargine is impaired by ϳ6-fold, that of [His A4 , His A8 ]insulin is enhanced by 7.5 (Ϯ2.5)-fold. This represents an improvement of Ͼ30-fold relative to insulin glargine.
The potency and duration of action of [His A4 , His A8 ]insulin were tested in diabetic rats in relation to insulin glargine (Fig. 2F). Glycemic control by long acting insulin analogs in rodents (5-10 h) is less prolonged than in humans (18 -24 h), presumably due to smaller depot sizes (28). [His A4 , His A8 ]insulin and insulin glargine were dissolved (like Lantus) in dilute HCl (pH 4.0) with a Zn 2ϩ : insulin ratio of 5.2:1. The time course and extent of glycemic control were similar on injection of the two analogs (Fig. 2F, red and  blue lines). A rapid acting control was provided by zinc-free human insulin in Lilly diluent (Fig. 2F, green line). Because the rats ate only at night, the effects of daytime insulin injections were influenced by diurnal fasting; controls were provided by injection of diluent alone (Fig. 2F, brown line). Control studies were also undertaken of [His A4 , His A8 ]insulin in neutral zinc-free Lilly diluent; its time course was similar to that of wild-type insulin control (not shown). Zinc-free glargine was not tested at neutral pH due to its sparing solubility.
Isoelectric precipitation of [His A4 , His A8 ]insulin at pH 7.4 is due to the combined effects of two electrostatic changes: removal of a wild-type negative charge by substitution Glu A4 3 His and the binding of three interfacial zinc ions per hexamer. Although insulin glargine contains 12 additional positive charges per hexamer, the net change in the formal charge of the [His A4 , His A8 ] hexamer may be between 9 and 12, depending on the fractional Cl Ϫ occupancy at the fourth coordination site and indirect effects of protein  modifications at other titratable sites (such as the ␣-amino group of Gly A1 and imidazole ring of His B5 ). Although the present crystal structure provides a model for zinc-bridged self-association surfaces, the number and distribution of bound zinc ions may differ in a non-crystalline aggregate formed on injection. In the future, this issue may be addressed by biochemical characterization of the initial subcutaneous depot.

CONCLUDING REMARKS
The present study demonstrates the potential of interfacial zinc-binding sites, introduced by design (17,29), to modify the pharmacokinetics of a protein in a subcutaneous depot. Such bottom-up control of assembly illustrates general principles of supramolecular chemistry and their application to nanobiotechnology (1,2).
Optimal treatment of diabetes mellitus often requires combined use of fast and slow acting insulin analogs (7,30). The principle of isoelectric precipitation underlies the prolonged action of insulin glargine, an analog in broad clinical use (Lantus) (11,12,30). Although slow release of zinc-free monomers enables basal metabolic regulation, concerns have been raised regarding IGF-1R-mediated mitogenicity (14) and possible dose-dependent cancer risk (15,16). [His A4 , His A8 ]insulin achieves slow release by formation of an insoluble hexamer stapled by interfacial zinc ions. Because the zinc-free monomer discriminates between IR and IGF-1R more stringently than wild-type insulin or existing analogs, this or related "second generation" analogs may enhance the safety of insulin therapy. Zinc stapling of insulin exemplifies a general strategy to modify the pharmacokinetic and biological properties of a subcutaneous protein depot. The engineering of novel lattice contacts in protein crystals can thus enable control of supramolecular assembly as a therapeutic protein nanotechnology.