Proinsulin Is Refractory to Protein Fibrillation

Insulin is susceptible to fibrillation, a misfolding process leading to well ordered cross-β assembly. Protection from fibrillation in β cells is provided by sequestration of the susceptible monomer within zinc hexamers. We demonstrate that proinsulin is refractory to fibrillation under conditions that promote the rapid fibrillation of zinc-free insulin. Proinsulin fibrils, as probed by Raman microscopy, are nonetheless similar in structure to insulin fibrils. The connecting peptide, although not well ordered in native proinsulin, participates in a fibril-specific β-sheet. Native insulin and proinsulin exhibit similar free energies of unfolding as inferred from guanidine denaturation studies: relative amyloidogenicities are thus not correlated with global stability. Strikingly, the susceptibility of proinsulin to fibrillation is increased by scission of the connecting peptide at single sites. We thus propose that the connecting peptide constrains a large scale conformational change in the misfolded protein. A tethering mechanism is proposed based on a model of an insulin protofilament derived from electron-microscopic image reconstruction. The proposed relationship between cross-β assembly and protein topology is supported by studies of single-chain analogs (mini-proinsulin and insulin-like growth factor I) in which foreshortened connecting peptides further retard fibrillation. In addition to its classic function to facilitate disulfide pairing, the connecting peptide may protect β cells from toxic protein misfolding in the endoplasmic reticulum.

Insulin is susceptible to fibrillation, a misfolding process leading to well ordered cross-␤ assembly. Protection from fibrillation in ␤ cells is provided by sequestration of the susceptible monomer within zinc hexamers. We demonstrate that proinsulin is refractory to fibrillation under conditions that promote the rapid fibrillation of zinc-free insulin. Proinsulin fibrils, as probed by Raman microscopy, are nonetheless similar in structure to insulin fibrils. The connecting peptide, although not well ordered in native proinsulin, participates in a fibril-specific ␤-sheet. Native insulin and proinsulin exhibit similar free energies of unfolding as inferred from guanidine denaturation studies: relative amyloidogenicities are thus not correlated with global stability. Strikingly, the susceptibility of proinsulin to fibrillation is increased by scission of the connecting peptide at single sites. We thus propose that the connecting peptide constrains a large scale conformational change in the misfolded protein. A tethering mechanism is proposed based on a model of an insulin protofilament derived from electron-microscopic image reconstruction. The proposed relationship between cross-␤ assembly and protein topology is supported by studies of single-chain analogs (mini-proinsulin and insulin-like growth factor I) in which foreshortened connecting peptides further retard fibrillation. In addition to its classic function to facilitate disulfide pairing, the connecting peptide may protect ␤ cells from toxic protein misfolding in the endoplasmic reticulum.
Insulin is a small globular protein containing two chains, A (21 residues) and B (30 residues). Stored in the pancreatic ␤-cell as a Zn 2ϩstabilized hexamer, the hormone functions as a Zn 2ϩ -free monomer. Insulin is the product of a single-chain precursor, proinsulin, in which a connecting region (35 residues) links the C-terminal residue of B chain (residue B30) to the N-terminal residue of the A chain ( Fig. 1A) (1). Although the structure of proinsulin has not been determined, a variety of evidence indicates that it consists of an insulin-like core and disordered connecting peptide (Fig. 1B) (2). Formation of three specific disulfide bridges (A6 -A11, A7-B7, and A20 -B19; Fig. 1B) is coupled to oxidative folding of proinsulin in the rough endoplasmic reticulum (ER). 3 Proinsulin assembles to form soluble Zn 2ϩ -coordinated hexamers shortly after export from ER to the Golgi apparatus (2,3). Endoproteolytic digestion and conversion to insulin occurs in immature secretory granules followed by morphological condensation (3,4). Crystalline arrays of zinc insulin hexamers within mature storage granules have been visualized by electron microscopy (EM) (4). Assembly and disassembly of native oligomers is thus intrinsic to the pathway of insulin biosynthesis, storage, secretion, and action ( Fig. 2) (2,5).
Insulin readily misfolds in vitro to form a prototypical amyloid (6,7). Unrelated to native assembly, fibrillation occurs via an amyloidogenic partial fold (Fig. 1C) (8,9). Factors that accelerate or hinder fibrillation have been extensively investigated in relation to pharmaceutical formulations (10). Zinc-free insulin is susceptible to fibrillation under a broad range of conditions (11) and is promoted by factors that impair native dimerization and higher order self-assembly (10,11). Physiological protection in ␤ cells is provided by the formation of Zn-insulin hexamers (11). 4 Although the structure of an insulin fibril is not well characterized at atomic resolution, a model of a protofilament has been proposed based on cryo-EM image reconstruction (12). This model envisages a parallel two-layered cross-␤ assembly in which the N terminus of the A chain and C terminus of the A chain lie on opposite sides. The protofilament is remarkable for a near-complete reorganization of ␣-helical structure (residues A1-A8, A12-A18, and B9 -B19; Fig. 1B) into ␤-sheet. Analogous models have been proposed for the Alzheimer-associated A␤ fibril and are supported by high resolution solid-state NMR studies (13,14).
In this report we investigate the structure of proinsulin fibrils and the effect of the connecting peptide on the kinetics of fibrillation. As observed in studies of insulin fibrils (15)(16)(17), Raman studies demonstrate that proinsulin can undergo a large scale conformational transition from ␣-helix to ␤-sheet. Whereas in native proinsulin the connecting peptide is largely disordered, analysis of amide Raman bands suggests that in fibrils the connecting peptide in part participates in a ␤-sheet. An associated transition is observed in the conformations of the disulfide bridges, which are similar to those observed in insulin fibrils. Despite such similarities, proinsulin is markedly less susceptible to fibrillation than is insulin (10): the lag time prior to fibrillation is prolonged by 15-fold at acidic pH. Moreover, at neutral pH proinsulin misfolds to form a ␤-sheet-rich amorphous precipitate on a time scale 4-fold slower than that of insulin fibrillation. The relative resistance of proinsulin to fibrillation and its less regular mode of aggregation at neutral pH are not due to enhanced thermodynamic stability as insulin and proinsulin exhibit similar free energies of unfolding (⌬G u ). Effects of protein structure and stability on amyloidogenicity were further investigated through comparative studies of single-chain and two-chain analogs. Split proinsulin analogs, products of single-site scission within the connecting peptide, are more susceptible to fibrillation than is the intact protein; by contrast, foreshortening of the connecting peptide (green segments in Fig. 1, A and 1B) (18 -20) markedly retards the fibrillation of mini-proinsulin (10). Such effects are likewise uncorrelated with stabilities. Enhanced fibrillation is also observed on cleavage of insulin-like growth factor I (IGF-I, an homologous single-chain protein), suggesting a general relationship between the kinetics of fibrillation and protein topology. Together, these studies suggest that the connecting region hinders fibril formation by functioning as a topological tether. We propose that such tethering interferes with a critical step in the mechanism of cross-␤ assembly and so protects the ␤ cell from proinsulin misfolding and its associated proteotoxicity (21).

EXPERIMENTAL PROCEDURES
Materials-Human insulin, proinsulin, [32,33]-split proinsulin, and IGF-I were kindly provided by Eli Lilly and Co. (Indianapolis, IN). Porcine mini-proinsulin (with dipeptide linker AK between B30 and A1) (20) was the gift of Dr. You-ming Feng (Shanghai, China). Endoproteinase Lys-C was purchased from Wako, Japan. Porcine pancreatic carboxypeptidase B was obtained from Roche Applied Science. Thioflavin T (ThT) was obtained from Sigma. All other chemicals were of analytical grade (Fisher Chemicals).
Preparation and Purification of Split Analogs-Proinsulin or IGF-I was made 1 mg/ml in 50 mM Tris-HCl (pH 8.5). Limited proteolysis of proinsulin was obtained with Lys-C at 37°C for 25 min using a ratio of 1:10,000 (w/w); complete digestion of IGF-I was obtained at 37°C for 24 h using ratio 1:50 (w/w). Aliquots were taken at successive times for reverse-phase high-performance liquid chromatography and SDS-PAGE. Reactions were stopped by adding trifluoroacetic acid to a final concentration of 1% and analyzed by reversed-phase high-performance liquid chromatography (C8 column, Vydac, Inc., Hesperia, CA). Solvent A was 0.1% aqueous trifluoroacetic acid, and solvent B was 0.1% trifluoroacetic acid in methanol. The elution gradient was 52-60% B over 60 min; the flow rate was 1 ml/min with detection wavelength 215 nm. Fractions were manually collected and lyophilized; molecular masses were confirmed by matrix-assisted laser desorption ionization (MALDI-TOF) mass spectrometry (MS). A physiological processing intermediate, des- [31,32]-split proinsulin, was generated by digesting [32,33]-split proinsulin with porcine carboxypeptidase B (molar ratio, 1:1000) in Tris-HCl buffer (100 mM, pH 8.0) at 37°C for 30 min. Digestion was stopped by adding trifluoroacetic acid to a final concentration of 1%. The analog was purified by reversed-phase high-performance liquid chromatography (C8 column, Vydac, Inc.); its predicted molecular mass was confirmed by MALDI-TOF MS.
Mass Spectrometry-MALDI-TOF MS employed a pulsed nitrogen laser source ( ϭ 337 nm) and acceleration voltage of 28 kV operated in the linear mode (Voyager Biospectrometry Workstation, PerSeptive Biosystems, Framingham, MA). Samples were mixed with an equal volume of a saturated matrix solution (␣-cyano-4-hydroxycinnamic acid in 70% acetonitrile and 0.3% trifluoroacetic acid). The mixture (1 l) was applied onto the laser target probe and air-dried. 100 -200 spectra were obtained per sample. To simplify MS analysis, two-chain analogs were reduced (by addition of 10 mM dithiothreitol at 37°C for 30 min) and alkylated to yield separate chains. Alkylation was effected by addition of The asterisk indicates foreshortened AK linker in mini-proinsulin (green). The color scheme is as in panel A. C, proposed pathway of insulin fibrillation via partial unfolding of monomer (11,39,76). The native state is protected by classic self-assembly (far left). Disassembly leads to equilibrium between native-and partially folded monomers (open triangle and magenta trapezoid, respectively). This partial fold may unfold completely as an off-pathway event (open circle) or aggregate to form a nucleus en route to a protofilament (far right). A model of the intermediate is shown in Fig. 7A. FIGURE 2. Pathway of insulin biosynthesis, storage, and secretion. A, nascent proinsulin folds as a monomer in ER wherein zinc-ion concentration is low; in Golgi apparatus zinc-stabilized proinsulin hexamer assembles, which is processed by cleavage of connecting peptide to yield mature insulin. Zinc-insulin crystals are observed in secretory granules. B, on metabolic stimulus, zinc-insulin crystals are released into portal circulation (pink) and disassociate in steps to liberate the functional monomer.
25 mM iodoacetic acid in the dark. After 15 min, 10 mM dithiothreitol was added to quench unreacted iodoacetic acid.
Circular Dichroism-Samples were dissolved in either 10 mM phosphate and 100 mM KCl (pH 7.4) or 0.01 N HCl (pH 2.0) at a protein concentration of 25 M. To remove particulate matter and protein aggregates, samples were filtered (0.22 M, Satorius, Goetlingen, Germany). Spectra, acquired with an Aviv spectropolarimeter (Aviv Biomedical, Inc., Lakewood, NJ), were normalized by mean residue ellipticity and ellipticity per mole of protein. For equilibrium denaturation studies samples were diluted to 5 M; guanidine-HCl was employed as denaturant (22). Data were obtained at 4°C.
Thermodynamic Modeling-Guanidine denaturation data were fitted by non-linear least squares to a two-state model as described previously (23). In brief, CD data (x), where x indicates the concentration of denaturant, were fitted by a non-linear least-squares program according to Equation 1, where x is the concentration of guanidine hydrochloride and where A and B are baseline values in the native and unfolded states. These baselines were approximated by pre-and post-transition Fitting the original CD data and baselines simultaneously circumvents artifacts associated with linear plots of ⌬G as a function of denaturant according to ⌬G 0 (x) ϭ ⌬G H 2 O 0 ϩ m o x (for reviews see Refs. 23 and 24).
Preparation of Fibrils-Native samples were prepared immediately prior to fibrillation. Proteins were made 60 M in each of the following conditions: (i) degassed phosphate-buffered saline (PBS, 10 mM phosphate, 140 mM NaCl) at pH 7.4 (mimicking the intracellular cytosolic pH near neutrality (25,26)), (ii) pH 5.5 (mimicking the mildly acidic pH in immature and mature secretory granules (26,27)), and (iii) 0.01 N HCl (0.01 N HCl, 150 mM NaCl, pH 2.0). All samples contain 0.1% sodium azide. (Control studies demonstrated that sodium azide has no effect on the kinetics of fibrillation.) Samples at pH 2.0 (fibril-promoting condition) were incubated in Eppendorf tubes at 65°C without agitation. Samples in PBS were incubated in pre-sterilized glass vials with airtight sealed caps (Allergy Laboratories, Oklahoma City, OK). Vials were rocked on a BD Biosciences tilting platform (BD Microbiology Systems) at ϳ60 rpm. At successive times aliquots were withdrawn with a sterilized single-use syringe. Aliquots were added to a ThT solution for fluorescence assay. The isolated insulin B-chain (as a bi-S-sulfonate derivative, Eli Lilly and Co.) was made 0.5 mM in degassed glycine buffer (0.1 M, pH 9.0). Dithiothreitol was added to a final concentration of 1.5 mM (50% excess relative to thiol groups in B chain). The sample was introduced into a sealed quartz cuvette and stirred at 25 rpm; fibrillation was monitored by light scattering at 360 nm at 25°C (9).
Fluorescence Spectroscopy-ThT was made 1 mM in double-distilled water and stored at 4°C in the dark. To monitor fibrillation, 10-l aliquots obtained at indicated time points were mixed with 3 ml of ThT assay buffer (5 M ThT in 50 mM Tris-HCl (pH 7.5) and 100 mM NaCl). Fluorescence measurements were performed using an Aviv spectrofluorometer in 1-cm quartz cuvettes. Emission spectra were collected from 470 to 500 nm following excitation at 450 nm; the integration time was 1 s. ThT in buffer without protein was used as baseline. The fibrillation lag time is defined as the time required to observe 2-fold enhancement in thioflavin T (ThT) emission. Control studies of the time course of ThT fluorescence in rela-tion to optical scattering (360 nm), and EM morphology indicate that such initial enhancement reflects formation of amorphous aggregates and immature fibrils in accord with past reports (28 -30). The threshold of 2-fold-enhanced ThT fluorescence is followed by a rapid increase in turbidity associated with elongation of mature fibrils and a further 15-fold increase in ThT fluorescence (see supplemental materials).
Transmission Electron Microscopy-Aliquots (10 l) were deposited on Formvar-coated 400-mesh copper grids (Electron Microscopy Sciences, Hatfield, PA) for 5 min. Excess solution was adsorbed to filter paper. Grids were washed three times with distilled water and three times with filtered 1% uranyl acetate for negative staining. Stained grids were allowed to dry for 20 min at room temperature. Specimens were observed with a Jeol 1200EX transmission electron microscope operating with an accelerating voltage of 80 kV.
Raman Microscopy and Spectroscopy-Studies in solution were conducted using a high light grasp spectrograph based on a Holospec f/1.4 (Kaiser Optical) (31). Typical conditions were 752 nm, 1-watt excitation, and 5-min acquisition time. An excitation frequency of 752 nm (Kr ϩ ) was used, because excitation at 647.1 nm gave rise to a high background. Sample volume was 50 l. A buffer spectrum with a waterbending mode at 1645 cm Ϫ1 was subtracted from each amide I profile. Solid samples were placed in the focused laser beam of a Raman microscope system (32). Ten milliwatts of 647.1 nm excitation (krypton laser) were used to generate Raman scattering with acquisition time 1 min. Spectra were normalized relative to the 1003 cm Ϫ1 Phe band as an internal standard (17); Phe is absent in the connecting peptide. Following normalization, small differences were sometimes observed in tyrosine bands at 850 and 830 cm Ϫ1 , presumably due to slight pH drift following lyophilization (33).

Overview
Our studies are presented in three parts. In part I Raman spectroscopy is employed to compare the structures of insulin and proinsulin in respective native states and as fibrils. In Part II kinetic features of fibrillation are investigated in relation to chain topology: single-chain and two-chain analogs are utilized as probes. In Part III differences in fibrillation lag times are assessed in relation to native-state structures and stabilities.
Fibrillation was monitored by ThT fluorescence under three conditions: (a) on gentle agitation in PBS (pH 7.4) at 37°C and (b) in a quiescent solution of dilute HCl (pH 2.0) at 65°C. PBS provides a model for fibrillation of insulin in pharmaceutical formulations; the present solutions are made zinc-free to accelerate fibrillation. Acidic conditions are classic (6,7,35,36) and widely employed in structural studies (12,16,17). Fibrillation of insulin is more rapid under acidic conditions, presumably due to the absence of native self-association (37) and destabilization of the monomeric fold (38). Under these conditions both insulin and proinsulin form fibrils (Fig. 3, A and B) (39). In PBS insulin forms Topological Protection of Proinsulin from Cross-␤ Assembly DECEMBER 23, 2005 • VOLUME 280 • NUMBER 51 JOURNAL OF BIOLOGICAL CHEMISTRY 42347 fibrils (Fig. 4A), whereas proinsulin slowly forms an amorphous precipitate in which linear structures are not well resolved (Fig. 4B). The fibrillation or amorphous precipitation of proinsulin is markedly slower than that of insulin (TABLE ONE). Under either condition, the isolated C-peptide (a 31-residue random-coil peptide released from proinsulin after proteolytic conversion (40 -42)) is highly soluble and refractory to fibrillation (lag time Ͼ 60 days; TABLE ONE).

Structure of a Proinsulin Fibril
Insulin and proinsulin were investigated by Raman spectroscopy (i) in their respective native states (Fig. 5, A and B) and (ii) as insoluble fibrils or precipitates (air-dried on glass coverslips; Fig. 5, C and D). Spectra of insulin and proinsulin in their native conformations provide a foundation for analysis of fibrils and amorphous precipitates. The native proteins were characterized in PBS solution (Fig.  5A) and as lyophilized powders (Fig. 5B); signal-to-noise was higher in powder spectra due to reduced background scattering. In accord with past studies by other spectroscopic methods (15,43,44), 5 the Raman spectrum of native proinsulin suggests that it consists of a native-like insulin fold and a largely unstructured connecting peptide.
Raman Analysis of Native Proinsulin-Amide bands provide probes of secondary structure, expected to be predominantly ␣-helical in insulin and a combination of ␣-helix and random coil in proinsulin. (Because the anti-parallel ␤-sheet in the native insulin dimer comprises only four residues per protomer, its contribution is expected in each case to be small and difficult to resolve.) The spectra are in accord with these expectations. In each case amide III bands provide evidence for a mixture of helical and disordered structure; in proinsulin these bands appear near 1285 (shoulder) and 1250 cm Ϫ1 (trace a in Fig. 5A), whereas in the spectrum of insulin (trace b) corresponding bands appear near 1285 (shoulder) and 1246 cm Ϫ1 . Additional evidence of a major helical component is provided in each case by the intensity near 940 cm Ϫ1 . Further insight is provided by respective amide I bands at 1659 cm Ϫ1 (proinsulin) and 1655 cm Ϫ1 (insulin). Their widths are similar and anomalously broad: 55 cm Ϫ1 (proinsulin) and 52 cm Ϫ1 (insulin). Such broadening reflects in part heterogeneity of helical conformations 6 and in part the presence of non-helical regions (17): in each spectrum a ␤-strand signature appears as a small shoulder near 1680 cm Ϫ1 .
The relationship between insulin and proinsulin may be probed by analysis of the Raman difference spectrum (trace c in Fig. 5A). Evidence that the connecting region is disordered is provided by the broad amide I difference feature at 1668 cm Ϫ1 and broad amide III difference features at 1262 and 1250 cm Ϫ1 . Bands arising from C-S stretch modes in cystines (the broad features near 665 and 725 cm Ϫ1 ), and S-S stretch modes of disulfide bridges (490 -560 cm Ϫ1 ) are similar in native insulin and proinsulin in solution; the difference spectrum shows only weak differences near 515 cm Ϫ1 . Similar amide I and amide III difference features were observed on comparison of the powder spectra (Fig. 5B). Their enhanced resolution and sensitivity enable observation of difference features in the C-S and disulfide regions (trace cЈ in Fig. 5B). Residual intensities near 667 and 740 cm Ϫ1 and near 520 cm Ϫ1 suggest subtle changes in -C-S-and -S-S-dihedral angles between insulin and proinsulin in the powder state. It is not known whether such features indicate a subtle reorganization of core packing (in accord with small changes in NMR chemical shifts (44)) or are due to lyophilization. On lyophilization, Raman bands are in general sharper throughout the spectrum (e.g. comparison of traces a and aЈ in Fig. 5, A and B, respectively) but located at similar wavenumbers, suggesting that a broader range of conformations is sampled in solution than in powders. Analogous sharpening generally occurs in protein crystals and is ascribed to encagement of the proteins in the lattice (17,32).
Raman Microscopy of Proinsulin Fibrils-A global conformational change is apparent in the spectrum of proinsulin fibrils (grown at pH 2 and 65°C) (Fig. 5D). As in the spectrum of insulin fibrils (trace eЈ in Fig.  5D) (17,36), the spectrum of proinsulin fibrils (trace dЈ) exhibits an overall simplification due to more uniform and well ordered residue environments. Of particular interest are prominent amide III and amide I bands at 1228 and 1672 cm Ϫ1 , respectively. Insulin fibrils grown at the same conditions exhibit corresponding bands at 1227 and 1672 cm Ϫ1 . The extremely narrow width of the amide I band (24 cm Ϫ1 ) indicates a well ordered ␤-sheet (17), similar to the width observed in the spectrum of insulin fibrils (27 cm Ϫ1 ). 7 These similarities indicate that the secondary structure of proinsulin fibrils resembles that of insulin fibrils in their predominance of well ordered ␤-sheet. In addition, the amide III region of each spectrum contains a weak shoulder near 1251-1254 cm Ϫ1 , indicating residual disorder.
As in control studies of the native states, further insight is obtained by analysis of the proinsulin-insulin difference spectrum (trace f Ј in Fig.  5D). Prominent differences features at 1672 cm Ϫ1 in the amide I region and at 1227 cm Ϫ1 in the amide III region suggest that a portion of the connecting peptide participates in a significant ␤-sheet structure and so is not disordered as in native proinsulin. Indeed, the extremely narrow width of the amide I difference feature (24 cm Ϫ1 ; see above) indicates that this portion of the connecting region is unusually well ordered. In addition, the amide III difference feature at 1255 cm Ϫ1 suggests that proinsulin fibrils contain a disordered region not present in insulin fibrils, presumably also due to another portion of the connecting peptide. Unlike native proinsulin or insulin, the disulfide modes in insulin fibrils are close in frequency and sharp, indicating that the three bridges occupy similar and well ordered environments. These modes in proinsulin fibrils are almost degenerate. The difference spectrum contains 7 Amide I widths represent the average of 19 measurements from five independent samples; the standard error is Ϯ2 cm Ϫ1 .

Fibrillation propensities of insulin, proinsulin, and proinsulin analogs
Values represent the mean lag time (in days) until ThT emission is first enhanced by more than two-fold. Initial protein concentrations were 60 M unless otherwise specified. Numbers of replicants are given in parentheses. Pro. c C-peptide-like product (residues 30 -64) and des-͓Thr B30 -Arg A0 ͔-insulin were obtained following Lys-C digestion of proinsulin. d ND, not determined. e Porcine insulin precursor (53 residues; 20) (see Fig. 1). f Split IGF-I consists of residues 1-27 and 28 -65 and so is missing part of D domain (residues 66 -70).  DECEMBER 23, 2005 • VOLUME 280 • NUMBER 51 only two weak features at 534 and 513 cm Ϫ1 . The similar secondary structures and disulfide environments of insulin and proinsulin fibrils, as indicated by Raman microscopy, are consistent with similar overall modes of cross-␤ assembly.

Topological Protection of Proinsulin from Cross-␤ Assembly
Raman Analysis of an Amorphous Precipitate-Whereas insulin forms well ordered fibrils in PBS (pH 7.4), proinsulin forms an amorphous precipitate that nonetheless exhibits comparably enhanced ThT fluorescence. Spectra of proinsulin precipitates (trace d in Fig. 5C), bona fide proinsulin fibrils (above), and insulin fibrils formed under these conditions (trace e) exhibit overall similarities, a conformational conversion to a predominance of ␤-sheet 8 but also differences in the overall sharpness and small changes in positions of diagnostic bands. Although the positions of the amide I and amide III bands are in each case similar, the amide I width of the insulin fibrils grown in PBS (21 cm Ϫ1 , trace e) is significantly narrower than the amide I width of proinsulin precipitates (28 cm Ϫ1 , trace d). The broader spectrum of the proinsulin precipitate suggests structural heterogeneity. The PBS difference spectrum, calculated between the proinsulin precipitate and insulin fibrils (trace f in Fig.  5C), exhibits an amide I feature at 1664 cm Ϫ1 with a shoulder at 1685 cm Ϫ1 and broad amide III feature at 1239 cm Ϫ1 . Together, these features indicate that the connecting region contains both ␤-sheet and residues in a polyproline-II-like conformation (45). The spectrum of proinsulin precipitates nonetheless contains a sharp and uniform disulfide band at 516 cm Ϫ1 . We speculate that the amorphous precipitate contains local domains of cross-␤ structure but that the connecting peptide is less well ordered and so impairs their propagation in a single mode of long range linear assembly.

Relationship between Fibrillation Kinetics and Chain Topology
Protein fibrillation is characterized by a lag phase (prior to detectable increases in ThT fluorescence or optical scattering) and an elongation phase (11). The lag phase is associated with aberrant aggregation of amyloidogenic partial fold(s) to form a nucleus (46,47). Our studies focus on the duration of the lag phase. Insulin and proinsulin each exhibit a significant lag phase whose duration depends on conditions and the properties of surfaces in contact with the solution (8,10). Insulin and proinsulin exhibit significant differences in lag times. Whereas zinc-free insulin made 60 M in PBS forms fibrils in 3-4 days on gentle rocking at 37°C, proinsulin begins to precipitate only after 15 days (TABLE ONE) to give a predominance of amorphous aggregates (Fig.  4B). Unlike these globular proteins, the isolated B chain (an unfolded peptide) exhibits an immediate burst phase at 25°C. B-chain fibrillation is 50% complete after ϳ5 h and reaches a plateau after ϳ15 h, yielding a gel with strong ThT fluorescence and well defined fibrils in accord with previous reports (Fig. 4F and Supplemental Material) (48 -50). The absence of a B-chain lag period highlights the importance of unfolding in the initial stages of fibrillation of globular proteins (8).
Studies in PBS-Although the connecting region of protein is unfolded in native proinsulin, its presence alters the topology of the protein. Whether the contrasting susceptibilities of proinsulin and insulin for ␤-sheet-associated aggregation reflects their topologies (one chain or two chains) was investigated following introduction of single proteolytic cleavages in the connecting peptide. Three such analogs were tested; respective sites of cleavage were introduced between residues 64 -65 (i.e. at the CA junction; [64,65]-split proinsulin), between residues 32-33 (at the BC junction; [32,33]-split proinsulin), and between residues 29 -30 (between the terminal two residues of the B domain; [29,30]-split proinsulin). 9 Of these analogs, [64,65]-split proinsulin precipitates from solution with lag time (5 days) similar to that of insulin, whereas the other two analogs exhibit lag times intermediate between insulin and proinsulin. Although ThT fluorescence is enhanced, transmission electron microscopic images in each case reveal a predominance of amorphous aggregation (Fig. 4, C-E). Mini-proinsu-lin, a single-chain analog more tightly constrained than proinsulin, resists precipitation for Ͼ100 days; solutions remain clear and do not exhibit enhanced ThT fluorescence.
Studies under Acidic Conditions-Insulin fibrillation is markedly accelerated at acidic pH and elevated temperature; indeed, upon mild agitation at pH 2.0 and 60°C fibrils form within a few hours (9). Fibrillation in quiescent acidic solution at 65°C yield a similar trend in lag times as is observed in PBS but with overall faster rates. Whereas insulin fibrils are obtained in 4 -6 h, the mean lag time for proinsulin is Ͼ3 days; EM images in each case revealed fibrils (Fig. 3, A and B). The three split analogs also formed fibrils (Fig. 3, D-F) with lag times intermediate between those of insulin and proinsulin (18 -30 h). As in PBS, [64,65]split proinsulin exhibits the lag time most similar to that of insulin. Non-covalent inclusion of the free C-peptide to form an equimolar solution of insulin and C-peptide did not affect the lag time (TABLE ONE), suggesting that covalent attachment is required for its modulation. Also as in PBS, foreshortening of the connecting peptide yields a single-chain analog refractory to fibrillation: although mini-proinsulin precipitates from solution after ϳ15 days, this aggregate is amorphous and yields only weak ThT fluorescence, suggesting that cross-␤ structure is absent (51). 10 The generality of the relationship between chain topology and lag time was investigated in studies of IGF-I, a single-chain protein with homologous A and B domains (sequence identity Ͼ50% relative to insulin A and B chains). Disulfide pairing corresponds to that of insulin and proinsulin. Relative to proinsulin, IGF-I contains a foreshortened connecting region (12 residues; C domain) and a unique C-terminal extension of the A domain (the D domain; residues 62-70). The crystal structure of IGF-I is similar but not identical to that of insulin 11 ; the C domain is in part ordered and in part disordered (52,53). In PBS IGF-I exhibits slow precipitation (ϳ26 days) with amorphous morphology (not shown). Under acidic conditions, IGF-I exhibits a lag time (ϳ5 days) longer than that of proinsulin (ϳ3 days); EM indicates formation of fibrils (Fig. 3G). Lys-C digestion of IGF-I liberates part of the C terminus of D domain (residues 66 -70) and cleaves the C domain between residues 27 and 28 to yield a split and truncated analog. This less-tethered analog forms fibrils (Fig. 3H) with a lag time (ϳ2 days) shorter than that of native IGF-I, consistent with the properties of split proinsulin analogs.
Studies under Mildly Acidic Conditions-Fibrillation studies were extended to pH 5.5, to mimic the pH in the trans-Golgi network and immature secretory granule (26,27). The fibrillation rates of proinsulin and two split analogs yielded similar trends as observed under neutral and fibril-promoting conditions (intact proinsulin and split analogs [64,65] and [29,30] exhibit lag times of 17, 9, and 14 days, respectively). To enhance the relevance of these results to the physiology of the ␤ cell, we also tested des- [31,32]-split proinsulin, the most abundant intermediate in vivo. 12 Remarkably, this intermediate is more resistant to fibril-lation than is proinsulin (lag time, Ͼ18 days), presumably due to enhanced solubility associated with a decrease in the pI of the protein from 5.2 to 4.7.

Relationship between Fibrillation and Thermodynamic Stabilities
The native conformations of insulin, proinsulin, mini-proinsulin, and split proinsulin analogs were investigated by CD spectroscopy and CDdetected studies of denaturant-induced unfolding (Fig. 6). The spectra in each case indicate that the analogs are well folded at neutral pH (Fig.  6A) and under acidic conditions (Fig. 6C). Although proinsulin and its split analogs exhibit attenuated mean residue ellipticity at helix-sensitive wavelengths, this is a consequence of normalization and can be accounted for by inclusion of the random-coil connecting segment. Alternative normalization of the spectra per molecule demonstrates that ␣-helix contents are in each case similar (Fig. 6, B and D). In accord with past studies (43, 54) and the above Raman analysis, the CD difference spectrum (proinsulin minus insulin) suggests that proinsulin consists of an insulin-like moiety and disordered connecting peptide (see supplemental materials). The spectrum of mini-proinsulin (blue open circles in Fig. 6, A and C) is likewise similar to that of insulin in accord with crystallographic and NMR studies of related analogs (18,19). CD spectra of the split analogs indicate that no significant structural perturbations are associated with proteolytic cleavage of the connecting region. Analogous results were obtained at 4, 37, and 65°C (see supplemental materials). IGF-I and its split analog likewise exhibit similar helix content.
Thermodynamic stabilities at neutral and acidic pH were investigated by CD-monitored guanidine denaturation (Fig. 6, E and F); free energies of unfolding (⌬G u ) were inferred from a two-state model (TABLE  TWO) (23). The stabilities of insulin, proinsulin, and IGF-I are consistent with previous reports (54 -56). The results demonstrate that relative lag times are not correlated with relative stabilities. Despite their FIGURE 6. CD studies. Spectra of proinsulin and analogs at acidic and neutral pH: A and B, far-UV spectra of proinsulin analogs in PBS; C and D, spectra in 0. 01 N HCl (pH 2.0). In each case the spectrum of insulin is represented in black-filled circles, proinsulin in blue-filled circles, [32,33]-split proinsulin in red triangles, [29,30]-split proinsulin in green diamonds, [64,65] marked differences in fibrillation properties, insulin, proinsulin, and split proinsulin analogs exhibit similar stabilities. 13 Indeed, proinsulin is somewhat less stable than insulin, despite its resistance to fibrillation. IGF-I and mini-proinsulin are likewise less stable than insulin under either condition (TABLE TWO) and yet are more refractory to fibrillation (TABLE ONE). Such lack of correlation is likely to be fundamental to the mechanism of fibrillation (Fig. 1C). Because guanidine-induced denaturation is essentially complete, ⌬G u values probe free-energy differences relative to the unfolded state (an off-pathway event in the kinetic scheme of fibrillation) rather than relative to amyloidogenic partial fold(s) proposed to mediate nucleus formation (Fig. 1C).

DISCUSSION
Diverse human diseases are caused by protein misfolding and aberrant aggregation. Examples of critical medical importance include systemic amyloidoses, neurodegenerative diseases, and prion-related encephalopathies (57,58). In such pathological processes proteins lose their native structures and undergo cross-␤ assembly to form fibrils (59,60). Whereas some amyloidogenic proteins are natively unfolded (59), others (such as immunoglobulin light chains, ␤-microglobulin, and lysozyme) are globular proteins (59,(61)(62)(63). The latter contain multiple disulfide bridges, whose irregular spacing may serve as natural topological constraints (12). Indeed, reduction of the disulfide bridges in lysozyme or ␣-lactalbumin markedly accelerates fibrillation (64,65). It is not known whether the structures of such fibrils at atomic dimensions resemble those comprising the oxidized proteins.
Fibrillation of globular proteins is proposed to be mediated by an amyloidogenic partial fold and not the completely unfolded state (Fig.  1C). Despite similar morphologies at EM resolution, polypeptides can exhibit different structural modes of cross-␤ assembly, differing in protofilament orientation (parallel or anti-parallel), registry, and packing (13, 66 -68). Molecular inheritance of such features may underlie the phenomenon of prion strains (68 -70). Insulin exemplifies many of the general features of amyloidogenic globular proteins. A small globular protein containing three disulfide bridges, the insulin monomer is exquisitely susceptible to fibrillation at elevated temperature and on agitation. Although rarely observed in vivo, fibrillation of insulin has long complicated its manufacture and use in the treatment of diabetes mellitus (10). We have chosen to investigate the molecular basis of insulin fibrillation as a model for an important class of amyloidogenic globular proteins and in relation to ER stress in the ␤ cell (21,71,72). Our results demonstrate how the susceptibility of a protein to aggregation-coupled misfolding may be modulated by the intrinsic properties of polypeptide segments and by overall topological features.
Topological Restriction on Insulin Fibrillation-An emerging concept in diabetes research posits the role of the unfolded-protein response and ER stress in ␤-cell viability and apoptosis (21). Motivated by the Akita mouse (in which a mutation in Cys A7 blocks proinsulin folding, leading to the accumulation of electron-dense deposits in the ER) (73, 74) and unfolded-protein response-related knock-out mice (21), this perspective emphasizes the potential contribution of proinsulin misfolding to ␤-cell exhaustion in Type II diabetes mellitus. It is not known whether such potential misfolding leads to toxic aggregates, protofilament-like assemblies, or mature fibrils. Pancreatic islets are ordinarily protected from insulin misfolding and fibrillation by the formation of stable Zninsulin hexamers and their storage as microcrystals in the secretory granules of ␤ cells (2,75). Such protection is not available to proinsulin in the ER wherein the concentration of zinc ions is very low. We speculate that in this organelle the topology of proinsulin and the solubility of the connecting peptide together provide analogous kinetic protection.
The structure of an insulin-or proinsulin fibril is not well characterized on the atomic scale. A variety of evidence nonetheless suggests that separation of the N-terminal region of the A chain from the C-terminal region of the B chain, termini ordinarily in close proximity in native structures, enhances fibrillation (8 -10). Such separation is consistent with a model of an insulin partial fold, recently proposed based on NMR studies at high temperature (central panel of Fig. 7A) (9). Here, we have provided evidence that a topological constraint, the tethered connecting regions of proinsulin, IGF-I, and mini-proinsulin, protects these proteins from fibrillation. Proinsulin and IGF-I can form fibrils, but the process is significantly slower than that observed on cleavage of their respective connecting regions.
Proinsulin at neutral pH forms an amorphous ␤-sheet-rich precipitate rather than a predominance of linear fibrils. Because classic fibrils are accessible to proinsulin under acidic conditions, we imagine that its amorphous precipitation reflects a competing kinetic pathway wherein amyloidogenic nuclei do not efficiently propagate in a single mode of assembly. Rather, multiple parallel pathways of aggregation supervene to form a heterogeneous collection of ␤-sheet-rich microstructures. The structural relationship between fibrillation and ␤-sheet-rich amorphous precipitation poses an interesting issue. Given the complexity of protein sequences and the multiple possible ways of aligning potential ␤-strands, formation of a highly ordered fibril with constant registry seems remarkable. Indeed, 13 C solid-state NMR studies of insulin fibrils obtained either from PBS or under acidic conditions demonstrate a single well ordered conformation within a protofilament. 14 Variation between the registry of successive ␤-strands might occur in amorphous precipitates and so preclude linear elongation. If so, atomic-scale features of such precipitates (amorphous only on the length scale of EM) may in part resemble those of classic fibrils.
Unlike proinsulin and IGF-I, mini-proinsulin in PBS forms neither fibrils nor ␤-sheet-rich precipitates during the 100 days of monitoring. We imagine that its short connecting peptide is effectively incompatible with a cross-␤ structure. Such incompatibility may be structural (i.e. the only available mode of cross-␤ assembly is disallowed by the short tether length) or kinetic: an alternative mode of cross-␤ assembly exists but is inaccessible on the time scale of "ordinary" precipitation. The resistance of mini-proinsulin to fibrillation challenges the view that amyloid is the universal ground state of polypeptides irrespective of sequence (46). How can this exception be rationalized? A molecular model of an insulin fibril has been proposed based on cryo-EM image reconstruction (Fig. 7, B and C) (12). This model envisages a parallel two-layered cross-␤ assembly in which the N terminus of the A chain and C terminus of the A chain lie on opposite sides. 15 Such extended and parallel packing, reminiscent of an NMR-based model of A-␤ amyloid proposed by Tycko and coworkers (13), has important implications for proinsulin. The connecting peptide would constrain the large scale conformational change required to splay the chain termini and thereby hinder formation of a parallel two-layered ␤-sheet. In the case of mini-proinsulin, the foreshortened connecting region would make this conformational change impossible. Although details of this model may not represent the actual structure of an insulin protofilament, its overall features are consistent with the present findings. We anticipate that future studies of insulin fibrils by solid-state NMR and Raman spectroscopy will enable refinement of this model and test its applicability to proinsulin and related single-chain analogs.
Concluding Remarks-In this study the connecting region in proinsulin (35 residues) has been shown to hinder fibrillation. This effect is likely to reflect in part the high solubility and intrinsic resistance of the connecting peptide to fibrillation. Whereas the length of the connecting peptide in proinsulin allows considerable segmental flexibility in the native state, Raman analysis suggests that less conformational freedom is available in a cross-␤ assembly. We further envisage that the connectivity of the protein imposes an intrinsic barrier to fibrillation. This barrier may hinder cross-␤ assembly of IGF-I (with its shorter C domain; 12 residues) by a mechanism unrelated to solubility. Evidence for the importance of tether length is provided by the almost complete protection of mini-proinsulin (which contains only a dipeptide linker) from fibrillation. Such protection is uncorrelated with the thermodynamic stability of the native state.
The present studies thus suggest that the susceptibility of a globular protein to aggregation-coupled misfolding can be modulated both by protein topology and by the physicochemical properties of discrete polypeptide segments. Whereas zinc-free insulin readily forms fibrils under diverse conditions, proinsulin and physiological processing intermediates are markedly less susceptible. We propose that such resistance to amor-14 A. Petkova, M. A. Weiss, and R. Tycko, manuscript in preparation. 15 In an insulin protofilament a ␤-bend was constructed in the A chain by Saibal and colleagues (12) to accommodate the intrachain A6 -A11 disulfide bond. The two inter-chain disulfide bonds (cystines A7-B7 and A20 -B19) are important in directing the registry of inter-chain hydrogen bonds.  (9) and unknown nucleus are shown at the center. Proposed low resolution model of insulin fibril (12) is shown at far right. In the model of partial fold, cylinders and the arrow indicate substructure (residues B9 -B26 and A16 -A20). Dashed lines indicate disordered regions; disulfide bridges are indicated by black balls (sulfur atoms). The color scheme follows that of the dimer at far left. B, schematic representation of stacked insulin backbones in protofibril as proposed by Saibil and colleagues (12). The cross-␤ unit consists of alternatively stacked A and B chains (red and blue, respectively) to form in-register parallel ␤-sheets (segments A1-A6, A12-A21, B1-B6, and B12-B30). The yellow arrow indicates direction of long axis of fibril. C, an individual insulin molecule in a protofibril as viewed down long axis; disulfide bridges are in yellow. Note extended non-native polypeptide conformations with central bend.
phous precipitation and fibrillation protects the pancreatic ␤ cell prior to sequestration of the protein within native Zn-coordinated assemblies.