Molecular basis for insulin fibril assembly

Magdalena I. Ivanovaa, Stuart A. Sieversa, Michael R. Sawayaa, Joseph S. Wallb, and David Eisenberga,1
aHoward Hughes Medical Institute, UCLA-DOE Institute for Genomics and Proteomics, Los Angeles CA 90095-1570; and bBiology Department, Brookhaven National Laboratory, Upton, NY 11973-5000

Contributed by David S. Eisenberg, September 11, 2009 (sent for review August 7, 2009)

In the rare medical condition termed injection amyloidosis, extracellular brils of insulin are observed. We found that the segment of the insulin B-chain with sequence LVEALYL is the smallest segment that both nucleates and inhibits the brillation of fulllength insulin in a molar ratio dependent manner, suggesting that this segment is central to the cross- spine of the insulin bril. In isolation from the rest of the protein, LVEALYL forms microcrystalline aggregates with brillar morphology, the structure of which we determined to 1 resolution. The LVEALYL segments are stacked into pairs of tightly interdigitated -sheets, each pair displaying the dry steric zipper interface typical of amyloid-like brils. This structure leads to a model for brils of human insulin consistent with electron microscopic, x-ray ber diffraction, and biochemical studies.
amyloid bril structure

ne of the roughly 25 disorders categorized as amyloid diseases (1) is injection amyloidosis. In this rare condition, full-length insulin molecules are found in fibrillar form at the site of frequent insulin injections (24). These insulin fibrils formed in vivo display the defining characteristics of amyloid aggregates (5), including binding the dye Congo red with apple-green birefringence, an elongated, unbranched fibrillar morphology (4), nucleation-dependent polymerization, and the cross- x-ray diffraction pattern (68). Recently, serum samples from patients with Parkinsons disease have been found to display an autoimmune response to insulin oligomers and fibrils (9), possibly indicating the presence of insulin aggregates in this disease as well. Insulin also forms amyloid-like fibrils in vitro, which are promoted by elevated temperatures, low pH, and increased ionic strength (4, 10). In addition, insulin fibril formation has been a limiting factor in long-term storage of insulin for treatment of diabetes. Thus, better understanding of insulin fibrillation could lead to safer handling and more cost-effective storage of insulin. Because insulin offers the structural simplicity of two short polypeptide chains constrained by one intramolecular and two intermolecular disulfide bonds, models of the structure of insulin fibrils have been numerous. Upon fibrillation, the molecule of insulin undergoes structural changes from a predominantly -helical state to a -sheet rich conformation. The fibrillar -sheets have been described as either parallel (1113) or antiparallel (1416). An early model was based on the crystal packing of a despentapeptide insulin molecule (with residues B26B30 removed) (4, 17). More recently, Jimenez et al. (18) used cryo-electron microscopy images to propose that upon fibril assembly the -helical insulin molecules undergo conformational transition into a flat -sheetrich state. Still more recently, from x-ray solution scattering and ab initio modeling, Vestergaard et al. (19) proposed that insulin fibrils are formed by primarily -helical oligomers. The first atomic-level view of the interactions between segments of insulin which may be part of fibrillar spine came from single crystal structures of the fibrilforming peptide segments LYQLEN (residues A13A18) and VEALYL (residues B12B17) (16). Previous biophysical studies suggest that the B chain, or a segment of it, may be the primary determinant of insulin fibrillation. For example, equimolar amounts of the peptide RRRRRRLVEALYLV (containing residues B11-B17 of the B
18990 18995 PNAS November 10, 2009 vol. 106 no. 45

chain) can attenuate insulin fibrillation (20). In addition, the point mutations H10D and L17Q in the B chain of insulin prolong the lag phase of insulin fibrillation, further supporting the importance of this segment in fibril formation (21). Also, exposing this fibril-prone segment by truncating the C-terminal five residues of the B chain increases the propensity of insulin for fibril formation (4, 17). Other studies have shown that the A chain also contributes to insulin fibrillation. Both A chain and B chain can form fibrils on their own (22, 23), and seeds of A chain or B chain can nucleate the fibrillation of full length insulin (22). In addition, it was reported that segments as short as six residues from either A chain (residues A13A18) or B chain (residues B12B17) can form fibrils by themselves (24). The same segments from A chain (residues A13A19) and B chain (residues B9B19) were found to be protected against hydrogen exchange when insulin was incubated in conditions favorable for fibril formation (25). Based on these findings, it seems that a segment from the A chain may also be involved in the formation of the spine of insulin fibrils. Much evidence has accumulated in support of the view that only specific segments of amyloidogenic proteins form the spine of amyloid fibrils. Peptide segments, as short as three to four residues, can form fibrils by themselves in vitro (2630). Furthermore, these segments from the spine of the fibril often act as inhibitors of fibrillation of their parent proteins. For example, prion peptide (residues 113120) and modified peptides from A (residues 1620) and IAPP (residues 2227) can attenuate fibril formation of the full-length protein (3135). Our work suggests that the B-chain segment LVEALYL is the main contributor to the spine formation of fibrils of full-length insulin. The crystal structure of the segment LVEALYL provides a molecular view of the structural organization of the spine of insulin fibrils and was used to build a model of fibrils of the full-length protein. The model is supported by the x-ray fiber diffraction of insulin fibrils and scanning-transmission electron microscopy (STEM) analysis of the morphology of insulin fibrils. Results
Insulin Segments Form Fibrils and Alter Lag Time of Insulin Fibril Formation. The rationale behind our experiments is that segments

that form fibrils on their own and alter the rate of fibril formation when co-incubated with full-length protein are likely to participate in the spine of full-length fibrils. Based on previous studies (24), there are two segments of insulin that form fibrils in isolation from the rest of the protein; these are VEALYL from the B chain and LYQLEN from the A chain. To ascertain whether either or both of these segments form the spine of insulin fibrils, we studied their effects on the rate of
Author contributions: M.I.I., S.A.S., and D.E. designed research; M.I.I., S.A.S., and M.R.S. performed research; M.I.I., M.R.S., and J.S.W. contributed new reagents/analytic tools; M.I.I., S.A.S., and M.R.S. analyzed data; and M.I.I., S.A.S., M.R.S., and D.E. wrote the paper. The authors declare no conict of interest. Data deposition: The coordinates of the crystal structure of LVEALYL have been deposited with the Protein Data Bank, www.pdb.org (PDB ID code 3HYD; structure factors PDB ID code 3HYD-SF).
1To

whom correspondence should be addressed. E-mail: david@mbi.ucla.edu.

This article contains supporting information online at www.pnas.org/cgi/content/full/ 0910080106/DCSupplemental.

www.pnas.org cgi doi 10.1073 pnas.0910080106

Fig. 1. An eight-residue segment from the insulin B chain accelerates and inhibits insulin bril formation. (A) Amino acid sequence of insulin. Segment SLYQLENY of the A chain is dark red. Segment LVEALYLV of the B chain is dark blue. Disulde bonds are colored in yellow. (B) SLYQLENY of the A chain, when added to the reaction mixture, does not affect the rate of full-length insulin bril formation. Fibrillation was monitored as a function of time by measuring ThT uorescence. Full-length insulin starts to form brils after 8 10 h, and its conversion is complete in 20 h. (C) Fibrillation assay showing LVEALYLV from the B chain inhibits insulin bril formation at a concentration 10 times less than the concentration of insulin. LVEALYLV also accelerates bril formation when present at concentrations 25 40 times less than the concentration of insulin. Note that neither SLYQLENY (A chain) nor LVEALYLV (B chain) uoresce in the presence of ThT. All points represent the mean value of at least four replicates, with error bars representing the SD. (D) Electron micrographs showing that SLYQLENY and LVEALYL aggregates are brillar in morphology (E). LVEALYLV in equimolar ratios inhibits insulin bril formation. Many brils were observed in the micrograph of the sample of full-length insulin taken 48 h after the beginning of brillation assay (left). In contrast, there were only a few brils in the sample of insulin incubated at equimolar ratio with LVEALYLV (right). (Scale bars, 400 nm.)

fibrillation of full-length insulin. Both of these segments are located between the two interchain disulfide bonds of full-length insulin (Fig. 1A). In our first experiments, we tested the effect on insulin fibril formation of nine six-residue segments with sequences spanning the A and B chain segments between the two interchain disulfide bonds of insulin [Supporting Information (SI) Table S1]. None of the nine six-residue segments affected the fibrillation of full-length insulin. These results suggest that the spine of insulin fibrils is longer than six residues. Searching for the smallest segment of insulin that affects the kinetics of fibrillation of full-length insulin, we synthesized two eight-residue segments: SLYQLENY (A chain) and LVEALYLV (B chain) (Fig. 1 A). Both SLYQLENY and LVEALYLV form fibrils in solution (Fig. 1D). As seen in Fig. 1B, the A chain segment SLYQLENY does not affect insulin fibrillation. In contrast, the B chain segment LVEALYLV slows insulin fibrillation when incubated in molar concentrations down to 10 times less than that of insulin (Fig. 1C). The strongest inhibitory effect of LVEALYLV is observed when incubated in equimolar concentrations with insulin. The lag time of fibril formation with LVEALYLV at a 1:1 molar ratio is more than 10 times that of insulin alone (Fig. 1C). In addition, we observed few fibrils in electron micrographs of the sample of insulin with LVEALYLV (1:1) compared with the densely packed fibrils observed in the sample containing only insulin (Fig. 1E). We also observed that LVEALYLV nucleated fibril formation when incubated in molar concentrations 40 times less than insulin (Fig. 1C). Thus, depending on the concentrations used, LVEALYLV either inhibits or nucleates fibril formation. Given that the six-residue segments do not affect the rate of insulin fibril formation and that the eight-residue segment LVEALYLV does, we wanted to know whether either of the two seven-residue peptides within LVEALYLV affected fibril formation. The two seven-residue peptides LVEALYL and VEALYLV formed fibrillar aggregates (Fig. 2A and Fig. S1 A). However, only LVEALYL inhibited fibril formation of fulllength insulin (Fig. 2C and Fig. S1B), and there were only sparse
Ivanova et al.

fibrils present in the sample of insulin incubated with LVEALYL (Fig. 2B). Also, depending on its molar ratio relative to that of insulin, LVEALYL, like LVEALYLV, either delays or nucleates fibril formation (Fig. 2C).
Structure of LVEALYL. The segment LVEALYL forms microcrystals diffracting to 1 resolution, and we were able to determine
BIOPHYSICS AND COMPUTATIONAL BIOLOGY

Fig. 2. LVEALYL from B chain is the smallest segment that can alter the rate of insulin bril formation. (A) Electron micrograph of brillar aggregates of B-chain LVEALYL. (B) Electron micrograph of sample taken from equimolar solution of insulin and LVEALYL after 48 h from beginning of assay. Note that there are only sparse bril-like aggregates. (C) Fibrillation assay showing that B-chain LVEALYL accelerates insulin bril formation when added to the reaction mixture at low concentrations, but inhibits insulin bril formation at higher concentrations. (Scale bars, 400 nm.)

PNAS

November 10, 2009

vol. 106

no. 45

18991

Y16

C
V12

V12

Y16

Fig. 3. Atomic structure of B-chain segment LVEALYL, which forms bril-like microcrystals. (A) View perpendicular to bril axis showing pair of -sheets formed by LVEALYL molecules. Crystal needle length runs vertical in this orientation. (B) View down bril axis showing one layer of interdigitated pair of LVEALYL molecules, which interlock tightly to form the dry steric zipper interface. Pairs of extended -strands of LVEALYL are stacked in register upon each other, so this gure may be thought of as a projection of two -sheets, each containing some 100,000 layers. Note that this dry steric zipper interface is devoid of water molecules (shown in yellow). (C) Packing of LVEALYL molecules in crystal, viewed down bril axis as in (B). Molecules forming the dry steric zipper interface are purple and are separated by water molecules from the next sheets in gray. Thus in LVEALYL crystals one side of each sheet faces water molecules (wet interface).

its fibril-like structure. Extended strands of LVEALYL pack in register into parallel -sheets, which run the entire length of the needle crystal (Fig. 3A). Alternating side chains L, E, L, and L extend toward a second, identical -sheet (Fig. 3B). The same set of side chains extending from the second sheet intermesh with those from the first, forming a dry, highly complementary interface of the type termed steric zipper (16). Each pair of sheets forming the dry interface packs against two other identical steric zippers, separated by wet interfaces containing six water molecules (Fig. 3C). The dry interface of the LVEALYL steric zipper buries a larger surface area and displays higher surface complementarity than does the steric zipper of the structure of VEALYL, which we determined earlier (16).
Mass per Unit Length of Insulin Fibrils. A measurement that tells much about the molecular structure of a fibril is its mass per unit length (MPL), which can be determined by STEM. The STEM observations were made on insulin fibrils grown under the same conditions used in the fibril kinetic assays. The most common MPL values of the narrowest fibrils clustered around a mean of 2.85 0.35 kDa/ (Fig. 4A). This MPL value of 2.85 0.35 kDa/ corresponds well with the value of 2.47 kDa/ expected for two insulin molecules (Mr 2 5808 Da) per 4.7 rise of an amyloid fibril (2 5.81 kDa/4.7 2.47 kDa/). Other common MPL values clustered around two and three times this value, suggesting that these fibrils consist of two and three of the narrowest measured filaments. Also observed was a clustering around an MPL value of about four times that of the narrowest measured filament. As seen in the Fig. 4B, the morphologies of the fibrils with different MPLs are quite similar with nearly equal cross-over distances. For example, the cross-over distances of fibrils with MPL values of 2.85 kDa/ and 5.43 kDa/ is 1,200 . This value was used in building our model of insulin fibrils shown in Fig. 5. This cross-over distance of 1,200 implies that each layer of the steric zipper of the fibril spine is not only separated by 4.7 from the layer below but is also rotated by 0.71 ([4.7 /2400 ] 360).

tions than insulin LVEALYL accelerates it. This dual mode of action of the B-chain segment LVEALYL on insulin fibril formation suggests that this segment is important for the formation of the spine of the insulin fibrils. In contrast, segments from the A chain lack this dual mode of action. The experiments represented in Fig. 1B show that the eight-residue A-chain segment SLYQLENY does not affect the rate of insulin fibril formation. Apparently this segment, although bound to the B-chain through two disulfide bonds, is peripheral to the spine. In short, our experiments on fibrillation indicate that the spine

Discussion
Insulin Segments Forming the Spine of the Fibril. The work presented here shows that the seven-residue B-chain segment LVEALYL can either delay or accelerate insulin fibril formation in a molar ratiodependent manner. In equimolar ratios with insulin, LVEALYL inhibits fibrillation, and in much lower concentra18992 www.pnas.org cgi doi 10.1073 pnas.0910080106

Fig. 4. STEM measurements of MPL of insulin brils. (A) MPL value of the most abundant brils was measured as 2.85 0.35 kDa/, which is comparable to MPL value of 2.47 kDa/ of the insulin bril model shown in Fig. 5. The insulin bril model contains two molecules of insulin per 4.7- layer. MPL values of the thicker brils correspond to brils with four, six, and eight molecules of insulin per 4.7- layer. Histogram was produced by using a binning window of 0.25 kDa/. (B) Electron micrographs of insulin brils representing the MPL of the bril populations shown in (A). Note that brils with various MPL values display similar morphologies and cross-over distances, suggesting that particles with larger MPL values comprise brils with the smallest MPL value of 2.85 kDa/. (Scale bars, 100 nm.)

Ivanova et al.

Native insulin

LVEALYL

Model of the fibril

L17 L11

L11 L17

Fig. 5. Fibril model of insulin. (Left) Native structure of the insulin dimer (PDB code 1GUJ). A and B chains of insulin molecule are shown in pale red and pale blue, respectively. The LVEALYL segment, which forms the spine of the bril, is in dark blue. The SLYQLENY segment, from the A chain, which forms auxiliary sheets to the spine of the bril, is in dark red. Disulde bonds are shown in yellow. (Middle) View down bril axis of four -sheets of crystal structure of B chain LVEALYL. The two sheets forming the dry steric zipper interface are in blue. Water molecules are shown as green spheres. (Right) View of bril model, looking down bril axis. One layer of bril model is made by stretching both monomers of native insulin (left) in a horizontal direction, converting the deep blue helix of the B chain and the deep red helix of the A chain into extended -strands. These extended -strands are given the conformations of the four chain segments of the crystal structure shown in the middle. Thus the spine of the bril consists of a dry steric zipper formed by the mating of the central two LVEALYL strands from the B chains of the two insulin molecules, plus two outer strands from the A chains of the two molecules.

of insulin fibrils includes the segment of the B chain between the interchain disulfide bonds, and model building of the fibril was based on this basic observation. In addition, the A chain is covalently bound to the B chain via two interchain disulfide bonds and contributes peripherally to the fibril. The crystal structure of LVEALYL shows how the A chain can be accommodated at the periphery of the spine, as explained below.
Dual Action of LVEALYL on the Rate of Fibrillation. Our observation

compete for sites on the nucleus with full insulin molecules and can slow fibril growth.
Model for the Insulin Fibril, Based on the LVEALYL Crystal Structure.

that the segment LVEALYL is an effective inhibitor of insulin fibrillation, and yet when present in low molar ratios to insulin can accelerate fibril formation, has precedents. In two separate studies, 1-antichymotrypsin has been observed to both accelerate and inhibit -amyloid (A ) fibrillation. For example, 1-antichymotrypsin, when incubated with A peptides at ratios lower than 1:100, accelerates A fibrillation (36). However, when the 1-antichymotrypsin to A 140 ratio is raised to 1:10, fibril formation is inhibited (37). Apoliprotein E4 (Apo E4) is another example in which fibrillation of A peptide is both inhibited and accelerated. In contrast with LVEALYL and 1-antichymotrypsin, the apo E4 dual effect on A fibrillation is reversed: Apo E4 inhibits fibrillation when incubated at concentrations lower than those that accelerate fibril formation (36, 38, 39). The dual effect of action of apo E4 and 1-antichymotrypsin on fibrillation of A was reported in separate studies. This study shows that a short peptide segment can also nucleate and inhibit fibril formation. The molecular mechanism of the dual effect is unknown. One possibility is that, at low concentrations, LVEALYL nucleates the spine of full-length insulin fibrils more rapidly than do insulin molecules. This is reasonable because the side chains of the segment LVEALYL are exposed, and hence a small group of molecules can interact with each other by intermeshing their side chains, thereby forming a nucleus. In full-length insulin molecules, there must be conformational changes for the LVEALYL side chains of the segment to be exposed and to interact with each other. Once nuclei are formed, full insulin molecules can add to the growing fibril. At high concentrations, LVEALYL can
Ivanova et al.

Taken together, the x-ray fiber diffraction pattern of aligned insulin fibrils and MPL measurements suggest important features of fibrils of full-length insulin. The x-ray fiber diffraction pattern reveals that the fibrils consist of -sheets running parallel to the fibril axis with their -strands running perpendicular (13). Thus, the 4.7 reflection in the x-ray diffraction pattern (Fig. 6A) corresponds to the distance between the strands in the -sheets. The two broad reflections centered at 11.7 and 9.0 on the equator (horizontal axis) correspond to the distances between the sheets in the insulin fibril. These two

Fig. 6. Comparison ofobserved and calculated bril diffraction patterns of insulin brils. (A) Cross- x-ray diffraction pattern of oriented insulin brils. On the meridian (vertical axis), there is one strong reection at 4.7 , corresponding to separation of strands within each -sheet. The weaker reections on the equator (horizontal axis) are at 9.0 and 11.7 , arising from separations between -sheets. (B) Simulated bril diffraction pattern calculated from the model of the insulin bril of Fig. 5. An excellent agreement of the diffraction pattern of the model with the observed pattern is noticeable, particularly the 4.7 reection on the meridian and the 9.0 and 11.7 reections on the equator.

PNAS

November 10, 2009

vol. 106

no. 45

18993

BIOPHYSICS AND COMPUTATIONAL BIOLOGY

reflections might imply that there are two pairs of sheets with different sheet-to-sheet distances, which form the fibrils. The MPL measurements suggest that there are two insulin molecules per 4.7 along the fibril (Fig. 4A). Based on the findings above, the atomic structure of the B chain segment LVEALYL was used as the molecular basis for the steric zipper spine of the insulin fibril. Insulin fibrils are primarily composed of -sheets (40), which implies that LVEALYL undergoes a conversion from its native -helical structure (shown in dark blue in Fig. 5, left panel) into an extended -strand, as in the crystal structure of LVEALYL (shown in dark blue in Fig. 5, middle panel). In our insulin fibril model, two strands of LVEALYL from two insulin molecules interdigitate to form one layer of the steric zipper spine of the fibril. Thus, in the fibril of full-length insulin, LVEALYL retains its conformation from the crystal structure (Fig. 5, right panel). The covalent constraints of the disulfide bonds linking the B chain to the A chain (41) force the A chain also to convert to an extended -strand upon extension of the LVEALYL -helix of the B chain (Fig. 5, right panel). The crystal packing of LVEALYL molecules also provides a structural template for the extended A chain segment LYQLENY. Thus, this A chain segment adopts the backbone conformation of the -strands (colored in gray Fig. 5, middle panel) which lies astride the -strands forming the dry steric zipper interface. These two outer -sheets of the spine of insulin fibril are formed by two stacks of LYQLENY segments of the A chain (Fig. 5, right panel). The two B chain LVEALYL -sheets, which form the dry steric zipper interface, lie between the two sheets formed by LYQLENY. Because LYQLENY contains a Tyr residue in the second position, this side chain superimposes on a Tyr from LVEALYL in the crystal structure, preserving the kissing tyrosine interaction observed across the wet interface of the crystal of LVEALYL (Fig. 5, middle panel) in the fibril model of full-length insulin. These four -strands complete the model of the fibril spine. The remaining segments of the A and B chains outside of the fibril spine were modified from their native structure to fit within the constraint of two molecules per 4.7 layer. Furthermore, to be compatible with the 1,200 cross-over distance observed in insulin fibrils, each layer of our model is given a left-hand twist of 0.71 with respect to the layer below. The result is the fibril model shown in Fig. 5 (right panel). In summary, the spine of our model contains four -sheets, the inner pair forming a steric zipper that superimposes on our crystal structure of LVEALYL from the B chain. The outer two -sheets are formed from the segment LYQLENY of the A chain, also found to contribute to the fibril spine. At the periphery of the model, the N- and C-termini retain the native-like structure of the insulin molecule. A similar model, with a steric zipper spine and native-like structure on the periphery, was proposed for a designed amyloid of ribonuclease A (42). A comparison of our model with the density of the 3D cryo-EM reconstruction of insulin fibrils, determined by Jimenez et al. (18), is given in Fig. S2. The overall shape of the electron density is captured well in most parts by our model (Fig. S2, right). The parts of the model that do not fit the density map are outside the spine of the fibril and possibly disordered, which could explain why they are not observed in the low-resolution density map. Moreover, the calculated MPL of insulin fibrils, including only the residues from the core of the fibrils, is within experimental error of the value obtained by Jimenez et al. (18) (for more details, see SI Text). Thus, we can conclude that our model agrees with the cryo EM density (18).
18994 www.pnas.org cgi doi 10.1073 pnas.0910080106

Assessment of Our Insulin Fibril Model. The model has been built to

be consistent with several structural and biochemical observations. First, its structure with two insulin molecules per 4.7 layer of the fibril is consistent with our MPL measurements (Fig. 4). It was also found that three dimers of insulin comprise the fibril precursors (43). Thus the fibril growth proceeds via stacking of these precursors along the fibril axis, suggesting that insulin fibrils are formed of dimers repeating along the length of the fibrils in agreement with our MPL measurements. Second, the simulated fiber diffraction pattern (Fig. 6B) computed from our model corresponds well to the observed cross- pattern of oriented insulin fibrils (Fig. 6A). Of note, two equatorial reflections, which arise from sheet-to-sheet interactions, are present in both the simulated and observed diffraction patterns at 11.7 and 9.0 . (Fig. 6). Third, our previous finding that crystals of LVEALYL can accelerate fibril formation of insulin (16) implies that the crystal structure of the peptide resembles the structural organization of the fibril spine. This is the basis for our use of the structure of LVEALYL as a template for the spine of the full-length insulin fibrils. Fourth, the model is consistent with previous findings (4, 17, 20, 21) and our observation that the B chain is important for fibril formation: the B chain forms fibrils independently and can also delay and accelerate fibril formation. Fifth, our model incorporates the LYQLENY segment of the A chain on the periphery of the spine. This is consistent with finding that segments from the A chain (residues A13 A19) and B chain (residues B9 B19) are protected against proton exchange (25). Sixth, inclusion of the Glu residues of LVEALYL and LYQLEN in the spine of our model may explain the more rapid fibrillation of insulin at pH 2.5, where the Glu is expected to be largely uncharged, than at neutral pH, where it carries a negative charge. Thus our model for the insulin fibril is supported by much structural and biochemical data. Materials and Methods
Polymerization Assays. Insulin was purchased from Sigma-Aldrich. Peptides were purchased from CS Bio and Celtek Bioscience Peptides. Insulin brillation assays were performed with 0.25 mM insulin in 50-mM glycine buffer pH 2.5. All reactions were performed with four or more replicates. The progress of the reaction was monitored by ThioavinT (ThT) uorescence (for more information, see the SI Text). Electron Microscopy. For details on EM, see the SI Text. Crystallization of LVEALVL. Crystals of LVEALYL grew in a hanging drop after mixing 2 l of 1.82-mM peptide with 1 l crystallization solution containing 20% MPD/0.1 M sodium citrate pH 5.5 at room temperature. X-Ray Crystallographic Data Collection and Processing. X-ray diffraction data sets were collected at the Swiss Light Source beamline X10SA, equipped with a MAR CCD detector. Data were collected in 5 wedges at a wavelength of 0.97645 using a 5 m beam diameter. For more details see Table S2 and SI Text. Preparation of Oriented Samples and X-Ray Diffraction. For details on preparation of oriented samples and x-ray diffraction, see the SI Text Model Construction. The model of the insulin bril was built directly from the crystal structure of the segment LVEALYL (insulin B chain residues 1117). Model building was performed with the graphics program O (44). This crude model was energy minimized using the program CNS (45) with van der Waals, electrostatic, and hydrogen bonding terms (46) (for more details, see the SI Text). Simulation of Fibril Diffraction. Simulated bril patterns of bril models were produced by cylindrical averaging of the single crystal diffraction intensities (for more information, refer to the SI Text).

Ivanova et al.

STEM Sample Preparation and Data Processing. The samples were prepared at the STEM facility at the Brookhaven National Laboratory (for more information, see the SI Text and Figs. S3 and S4). ACKNOWLEDGMENTS. We thank Drs. Andrew D. Miranker, Lukasz Salwinski, Ruben Diaz-Avalos, and Ivaylo Dinov for discussion; Dr. Martha Simon and Beth Lin of Brookhaven National Laboratory for making STEM measurements and preparing specimens; Dr. Helen Saibil for providing a

cryo-EM reconstruction of insulin brils; Prof. Ehmke Pohl and the staff at the Swiss Light Source beamline X10SA for assistance in data collection; and the staff at the Advanced Photon Source beamline 24-ID-E. We thank the National Science Foundation, the Department of Energy/Ofce of Biological and Environmental Research, the National Institutes of Health, and Howard Hughes Medical Institute for support. S.A.S. was supported by a University of California Los Angeles National Science Foundation Integrative Graduate Education and Research Traineeship. BNL STEM is supported by Department of Energy/Ofce of Health and Environment Research.

1. Westermark P, et al. (2005) Amyloid: Toward terminology clarication. Report from the Nomenclature Committee of the International Society of Amyloidosis. Amyloid 12:1 4. 2. Dische FE, et al. (1988) Insulin as an amyloid-bril protein at sites of repeated insulin injections in a diabetic patient. Diabetologia 31:158 161. 3. Storkel S, Schneider HM, Muntefering H, Kashiwagi S (1983) Iatrogenic, insulindependent, local amyloidosis. Lab Invest 48:108 111. 4. Brange J, Andersen L, Laursen ED, Meyn G, Rasmussen E (1997) Toward understanding insulin brillation. J Pharm Sci 86:517525. 5. Westermark P (2005) Aspects on human amyloid forms and their bril polypeptides. FEBS J 272:59425949. 6. Astbury WT, Dickinson S (1935) The x-ray interpretation of denaturation and the structure of the seed globulins. Biochem J 29:23512360. 7. Geddes AJ, Parker KD, Atkins ED, Beighton E (1968) Cross-beta conformation in proteins. J Mol Biol 32:343358. 8. Sunde M, Blake C (1997) The structure of amyloid brils by electron microscopy and x-ray diffraction. Adv Protein Chem 50:123159. 9. Wilhelm KR, et al. (2007) Immune reactivity towards insulin, its amyloid and protein S100B in blood sera of Parkinsons disease patients. Eur J Neurol 14:327334. 10. Ahmad A, Uversky VN, Hong D, Fink AL (2005) Early events in the brillation of monomeric insulin. J Biol Chem 280:42669 42675. 11. Bouchard M, Zurdo J, Nettleton EJ, Dobson CM, Robinson CV (2000) Formation of insulin amyloid brils followed by FTIR simultaneously with CD and electron microscopy. Protein Sci 9:1960 1967. 12. Nettleton EJ, et al. (2000) Characterization of the oligomeric states of insulin in self-assembly and amyloid bril formation by mass spectrometry. Biophys J 79:1053 1065. 13. Burke MJ, Rougvie MA (1972) Cross-beta protein structures. I. Insulin brils. Biochemistry 11:24352439. 14. Yu NT, Jo BH, Chang RC, Huber JD (1974) Single-crystal Raman spectra of native insulin. Structures of insulin brils, glucagon brils, and intact calf lens. Arch Biochem Biophys 160:614 622. 15. Turnell WG, Finch JT (1992) Binding of the dye congo red to the amyloid protein pig insulin reveals a novel homology amongst amyloid-forming peptide sequences. J Mol Biol 227:12051223. 16. Sawaya MR, et al. (2007) Atomic structures of amyloid cross-beta spines reveal varied steric zippers. Nature 447:453 457. 17. Brange J, Dodson GG, Edwards DJ, Holden PH, Whittingham JL (1997) A model of insulin brils derived from the x-ray crystal structure of a monomeric insulin (despentapeptide insulin). Proteins 27:507516. 18. Jimenez JL, et al. (2002) The protolament structure of insulin amyloid brils. Proc Natl Acad Sci USA 99:9196 9201. 19. Vestergaard B, et al. (2007) A helical structural nucleus is the primary elongating unit of insulin amyloid brils. PLoS Biol 5:e134. 20. Gibson TJ, Murphy RM (2006) Inhibition of insulin brillogenesis with targeted peptides. Protein Sci 15:11331141. 21. Nielsen L, Frokjaer S, Brange J, Uversky VN, Fink AL (2001) Probing the mechanism of insulin bril formation with insulin mutants. Biochemistry 40:8397 8409. 22. Devlin GL, et al. (2006) The component polypeptide chains of bovine insulin nucleate or inhibit aggregation of the parent protein in a conformation-dependent manner. J Mol Biol 360:497509. 23. Hong DP, Fink AL (2005) Independent heterologous brillation of insulin and its B-chain peptide. Biochemistry 44:1670116709. 24. Ivanova MI, Thompson MJ, Eisenberg D (2006) A systematic screen of beta(2)microglobulin and insulin for amyloid-like segments. Proc Natl Acad Sci USA 103:4079 4082.

25. Tito P, Nettleton EJ, Robinson CV (2000) Dissecting the hydrogen exchange properties of insulin under amyloid bril forming conditions: A site-specic investigation by mass spectrometry. J Mol Biol 303:267278. 26. Balbirnie M, Grothe R, Eisenberg DS (2001) An amyloid-forming peptide from the yeast prion Sup35 reveals a dehydrated beta-sheet structure for amyloid. Proc Natl Acad Sci USA 98:23752380. 27. Lopez de la Paz M, Serrano L (2004) Sequence determinants of amyloid bril formation. Proc Natl Acad Sci USA 101:8792. 28. Tenidis K, et al. (2000) Identication of a penta- and hexapeptide of islet amyloid polypeptide (IAPP) with amyloidogenic and cytotoxic properties. J Mol Biol 295:1055 1071. 29. Reches M, Porat Y, Gazit E (2002) Amyloid bril formation by pentapeptide and tetrapeptide fragments of human calcitonin. J Biol Chem 277:3547535480. 30. Tjernberg L, Hosia W, Bark N, Thyberg J, Johansson J (2002) Charge attraction and beta propensity are necessary for amyloid bril formation from tetrapeptides. J Biol Chem 277:43243 43246. 31. Tjernberg LO, et al. (1996) Arrest of beta-amyloid bril formation by a pentapeptide ligand. J Biol Chem 271:8545 8548. 32. Findeis MA, et al. (1999) Modied-peptide inhibitors of amyloid beta-peptide polymerization. Biochemistry 38:6791 6800. 33. Gilead S, Gazit E (2004) Inhibition of amyloid bril formation by peptide analogues modied with alpha-aminoisobutyric acid. Angew Chem Int Ed Engl 43:4041 4044. 34. Kapurniotu A, Schmauder A, Tenidis K (2002) Structure-based design and study of non-amyloidogenic, double N-methylated IAPP amyloid core sequences as inhibitors of IAPP amyloid formation and cytotoxicity. J Mol Biol 315:339 350. 35. Chabry J, Caughey B, Chesebro B (1998) Specic inhibition of in vitro formation of protease-resistant prion protein by synthetic peptides. J Biol Chem 273:1320313207. 36. Ma J, Yee A, Brewer HB, Jr, Das S, Potter H (1994) Amyloid-associated proteins alpha 1-antichymotrypsin and apolipoprotein E promote assembly of Alzheimer betaprotein into laments. Nature 372:9294. 37. Eriksson S, Janciauskiene S, Lannfelt L (1995) Alpha 1-antichymotrypsin regulates Alzheimer beta-amyloid peptide bril formation. Proc Natl Acad Sci USA 92:2313 2317. 38. Evans KC, Berger EP, Cho CG, Weisgraber KH, Lansbury PT, Jr (1995) Apolipoprotein E is a kinetic but not a thermodynamic inhibitor of amyloid formation: Implications for the pathogenesis and treatment of Alzheimer disease. Proc Natl Acad Sci USA 92:763 767. 39. Wisniewski T, Castano EM, Golabek A, Vogel T, Frangione B (1994) Acceleration of Alzheimers bril formation by apolipoprotein E in vitro. Am J Pathol 145:1030 1035. 40. Nielsen L, Frokjaer S, Carpenter JF, Brange J (2001) Studies of the structure of insulin brils by Fourier transform infrared (FTIR) spectroscopy and electron microscopy. J Pharm Sci 90:29 37. 41. Nettleton EJ (1998) Ph.D. thesis. (Oxford University, Oxford). 42. Sambashivan S, Liu Y, Sawaya MR, Gingery M, Eisenberg D (2005) Amyloid-like brils of ribonuclease A with three-dimensional domain-swapped and native-like structure. Nature 437:266 269. 43. Nayak A, Sorci M, Krueger S, Belfort G (2009) A universal pathway for amyloid nucleus and precursor formation for insulin. Proteins 74:556 565. 44. Jones TA, Zou JY, Cowan SW, Kjeldgaard M (1991) Improved methods for building protein models in electron density maps and the location of errors in these models. Acta Crystallogr A 47:110 119. 45. Brunger AT, et al. (1998) Crystallography & NMR system: A new software suite for macromolecular structure determination. Acta Crystallogr D Biol Crystallogr 54:905 921. 46. Fabiola F, Bertram R, Korostelev A, Chapman MS (2002) An improved hydrogen bond potential: Impact on medium resolution protein structures. Protein Sci 11:14151423.

Ivanova et al.

PNAS

November 10, 2009

vol. 106

no. 45

18995

BIOPHYSICS AND COMPUTATIONAL BIOLOGY

Supporting Information
Ivanova et al. 10.1073/pnas.0910080106
SI Text
Polymerization Assays. Insulin was purchased from Sigma-

Aldrich. Peptides were purchased from CS Bio Company and Celtek Bioscience Peptides. Insulin fibrillation assays were performed with 0.25 mM insulin in 50 mM glycine buffer, pH 2.5. Lyophilized insulin was freshly dissolved and filtered through 0.2- m filters to make 0.5 mM or 0.33 mM stock solutions. Reaction mixtures were incubated at 37 C and shaken at 1,000 rpm in an Eppendorf Thermomixer R (Fisher Scientific) or in a Torrey Pines Incubator/Shaker at 37 C shaken at level 6 ( 650 rpm). All reactions were performed with four or more replicates using 200 l sample volume in black, 96-well, optical bottom NUNC plates (Fisher Scientific). The progress of the reaction was monitored by ThioflavinT (ThT) fluorescence: 3- l aliquots of sample were removed at the specified times and mixed with 200 l of 10 M ThT in 10 mM Tris pH 8.0. Emission fluorescence at 482 nm was monitored using excitation 444 nm with the bottom read mode of a Spectra Max M5 plate reader (Molecular Devices). Six readings were averaged with cutoff of 475 nm. In our figures, the mean values of ThT fluorescence are plotted with error bars showing the standard deviation.
X-Ray Crystallographic Data Collection and Processing. X-ray diffraction data sets were collected at Swiss Light Source beamline X10SA, equipped with a MAR CCD detector. Data were collected in 5 wedges at a wavelength of 0.97645 using a 50 12 m beam. The crystals were cryo-cooled (100 K) for data collection. All data were processed and reduced using Denzo/ Scalepack from the HKL suite of programs (1). In Table S2, the statistical findings of data processing and atomic refinement are summarized. The LVEALYL structure was determined using the molecular replacement program Phaser (2) with the structure of VEALYL (PDB code: 2OMQ) as a search model. Crystallographic refinement was performed with the program REFMAC (3). The model was built with the graphics program COOT (4). The geometric quality of the model was assessed with the programs PROCHECK (5) and WHATIF (6). Segment structures were illustrated using the program PyMOL (7). Statistical findings of data collection and processing are summarized in Table S2. Preparation of Oriented Samples and X-Ray Diffraction. Insulin fibrils

was used to calculate the structure factors from the coordinates of the fibril model. A program was written to calculate the cylindrical average of the squared structure factors and to simulate the fibril diffraction. Reflection intensities with the same index k were averaged together in rings of thickness 0.005 1. The cylindrically averaged reciprocal lattice was allowed to precess 180 around the b cell direction in 2 increments. with precession angles ranging from 0 to 20 in 2 increments. The x-ray beam is aligned perpendicular to the fibril axis. During the simulation, whenever a ring crosses the sphere of reflection, the intensity of the ring is recorded in area with polar coordinates: 2,000 pixels in radius (0.000167 1 per pixel) and 360 pixels (1 per pixel) in .
Fibril Model Construction. The model of the insulin fibril was built directly from the crystal structure of the segment LVEALYL (insulin B chain residues 1117). The spine contains two -sheets composed of parallel, in-register LVEALYL -strands running perpendicular to the fibril axis (Figs. 3 and 5). The sheets are related by a crystallographic 21-screw axis coaxial with the fibril axis, running midway between the sheets. The remainder of the insulin molecule was molded around the LVEALYL spine, with the requirement that the native disulfide bonds be retained in the fibril model (Cys6A-Cys11A, Cys7A-Cys7B, and Cys20ACys19B) (8). The close packing of two LVEALYL peptides every 4.7 along the fibril axis placed strong constraints on the conformation of the remaining insulin molecule. Accompanying the two B chains (containing LVEALYL) are two covalently attached A chains every 4.7 along the fibril a total of 1 full insulin molecule on either side of the fibril axis. To accommodate all of these atoms in a thin (4.7 ) layer, most of the -helical content of chain A was removed, leaving a flattened strand. Furthermore, the orientation of the flattened A chain strand was constrained to be parallel to LVEALYL by virtue of its two intermolecular disulfide bonds, one at each end, that is, N terminus to N terminus (Cys7A-Cys7B), and C terminus to C terminus (Cys20A-Cys19B) (Fig. 5). Because LVEALYL is extended in a -strand conformation, Cys7B and Cys19B are 34 apart. Hence, chain A was also extended into a -strand so that Cys7A and Cys20A could span the distance required to maintain these two intermolecular disulfide bonds. The central portion of chain A (residues 1319; corresponding to the sequence LYQLENY) was packed against the outward-facing side chains of the LVEALYL spine (Val12, Ala14, and Tyr16). By inspection, it was noted that a hydrogen bond between Tyr16 of LVEALYL and Tyr14 of LYQLENY side chains could be modeled to resemble a hydrogen bond observed in the LVEALYL peptide structure, between Tyr16 and its symmetry mate) (Fig. 5, middle). In total, there are four sheets in our fibril model, two from the A chain (LYQLENY) and two from the B chain (LVEALYL) (Fig. 5). Residues presumed to be disordered were not included in the model, specifically chain A residues 15 and chain B residues 16 and 2030. The remaining segments of the A and B chains outside of the fibril spine were modified from their native structure to fit within the constraint of two molecules per 4.7 layer. The exceptions are the disulfide-bonded residues C7A-C7B and C6A-C11A; maintaining good geometry for these residues requires that their dimensions along the fibril axis be greater than the 4.7 allowed per layer. These segments were oriented slightly differently in alternating layers to allow for their non -strand conformation.
1 of 9

(1 ml of 0.25-mM insulin in 50 mM glycine buffer pH 2.5 incubated at 37 C with 100 rpm shaking for three days) were centrifuged at 10,000 g for 3 min. The pellet was washed twice with water (pH 2.5 adjusted with HCl), resuspended in 5 l water, and placed between two fire-polished silanized glass capillaries (1 mm apart) and dried at room temperature. X-ray diffraction data of the samples were collected on beamline 24-ID-E at the Advanced Photon Source (APS) at the Argonne National Laboratory. The sample to film distance was 150 mm. Data were collected on a Quantum315 detector using 0.97918.
Simulation of Fibril Diffraction. Simulated fibril patterns of fibril models were produced by cylindrical averaging of the single crystal diffraction intensities. The fibril was placed in a cell with unit cell dimensions a c 200 b 2,400 90, where the fibril axis was oriented along the unit cell dimension b. The length 2,400 corresponds to the length of the fibril required to complete 360 (pitch) and was measured from the STEM images of insulin fibrils. The CCP4 program SFALL
Ivanova et al. www.pnas.org/cgi/content/short/0910080106

All model building was performed with the graphics program O (9). A twist of 0.71 around the fibril axis was introduced into successive 4.7 layers of the spine to simulate the observed cross-over distance 1,200 of the fibrils. This crude model was then energy minimized using the program CNS (10) with van der Waals, electrostatic, and hydrogen bonding terms (11). The model of insulin fibrils can be viewed at: http://www.doembi.ucla.edu/ sawaya/jmol/insulin/ and coordinates can be obtained at: http://www.doe-mbi.ucla.edu/ sawaya/jmol/fibrilmodels/.
Comparison of Model to Low-Resolution Density Map from 3D Reconstruction of Cryo-EM Images of Insulin Fibrils. We compared our

the STEM facility at the Brookhaven National Laboratory, using a scanning pixel size of (20 )2. TMV particles with MPL of 13.10 kDa/ were added to all samples as an internal calibration standard. The calculated electron dose in the STEM images was 10 e-/2 for a 20- pixel size.
STEM Data Processing. Measurements from the STEM dark-field

model to fibril images by a superposition of the model on a 25 resolution cryo-EM projection of a two-protofilament fibril model (12) (Fig. S2). The cross-section of our model appears somewhat wider than the projection of Jimenez et al. (12). Parts of the model that do not fit the density map are not modeled and are probably disordered, and are thus not seen in the lowresolution density map. If we recalculate the MPL value including only the 28 residues from the core of the fibrils (A6 to A21 and B7 to B19), we obtain 1.1 insulin molecules per 4.7 repeat (2 molecules [28 core residues/51 residues] 1.1) Based on the the MPL value obtained by cryoEM (12), there are between 1.08 and 1.28 insulin molecules per repeat in the cryoEM map shown in Fig. S2. Thus, based on these calculations, our MPL estimate of insulin fibrils is close in value to the MPL determined by Jiminez et al. (12).
Electron Microscopy. Negatively stained specimens for transmission electron microscopy (TEM) were prepared by applying 5 l of sample on hydrophilic 400 mesh carbon-coated formvar support films mounted on copper grids (Ted Pella, Inc.). The samples were allowed to adhere for 3 min, rinsed twice with distilled water, and stained for 1 min with 1% uranyl acetate (Ted Pella, Inc.). Grids were examined in a Hitachi H-7000 microscope. STEM Sample Preparation. The samples were prepared at the STEM facility at the Brookhaven National Laboratory. Thin (3 nm) carbon substrates were prepared by carbon arc evaporation onto freshly cleaved, single crystal rock salt, floated onto a water surface, and picked up from above by touching to a 2.3 mm titanium grid covered with holey film, retaining a drop of water. First, 2 l of 100 g/ml tobacco mosaic virus (TMV) were injected into a droplet of water covering the grids and allowed to adhere for 1 min. After four washings, 2 l of insulin fibrils (prepared as described in the polymerization assay) were injected into the droplet and allowed to adhere for 1 min. The grid was washed then 10 times with water. Each grid was pinched between two pieces of filter paper and immediately plunged into liquid nitrogen slush; the grid was then transferred to an ion-pumped freeze dryer, freeze dried overnight by gradually warming to 80 C, and transferred under vacuum to the STEM. STEM Data Collection. The dark-field images of unstained, freezedried specimens were recorded for MPL measurements (13) at
1. Otwinowski ZMW (1997) Processing of x-ray diffraction data collected in oscillation mode. Methods Enzymol 276:307326. 2. McCoy AJ, Grosse-Kunstleve RW, Storoni LC, Read RJ (2005) Likelihood-enhanced fast translation functions. Acta Crystallogr D Biol Crystallogr 61:458 464. 3. Murshudov GN, Vagin AA, Dodson EJ (1997) Renement of macromolecular structures by the maximum-likelihood method. Acta Crystallogr D Biol Crystallogr 53:240 255. 4. Emsley P, Cowtan K (2004) Coot: model-building tools for molecular graphics. Acta Crystallogr D Biol Crystallogr 60:2126 2132. 5. Laskowski RA, MacArthur MW, Moss DS, Thornton JM (1993) PROCHECKa program to check the stereochemical quality of protein structures. J Appl Cryst 26:283291.

images of unstained samples were made with the program Pcmass29 (14) and as described by Diaz-Avalos et al. (15). Pcmass29 subtracts the background from the images automatically and provides a measurement of electron scattering proportional to the mass inside a box. Unencumbered segments of fibril and TMV images were measured using a 6 nm rod (Model 24: boxes 160 400 ) for segment sizes: 2.85 0.35 kDa/, 9 nm rod (Model 25: box size 160 300 ) for particle sizes: 5.43 0.70 kDa/ and Amyloid (Model 61: box size 200 400 ) to measure particle sizes: 8.12 0.55 kDa/ and 10.08 1.04 kDa/. Reliable absolute scaling of the electron scattering mass requires an internal calibration standard in the images. For each STEM image of the freeze-dried, unstained specimens, we collected as many measurements as possible from the TMV particles present (total of 348 sampling points). Then the average MPL of half of the measured TMV particles was calculated to be 12.12 0.53 kDa/. The scaling factor used to compute the MPL values of insulin fibrils was calculated by dividing the known TMV value of 13.10 kDa/ by the average MPL value of TMV particles of half of the measured samples (13.10 kDa//12.12 0.53kDa/). To test how well the known MPL value of TMV particles is recovered, the scale factor was used to compute the average MPL value of the remaining half of TMV measurements (Fig. S3A). STEM measurements were additionally validated by measuring the mass values of the earth worm hemoglobin which was previously determined by Zhu et al. (16) to be 3.75 MDa (Fig. S3B and C).
Statistical Modeling, MPL, and Classification of Insulin Fibrils. Expectation maximization (17) was used to determine the number of mixtures as well as to determine the weight coefficients of the Gaussian mixture of MPL values. Then the normality of the distributions of the MPL values modeled by each of the four Gaussians was assessed using QQ-normal probability plots (Fig. S4) (18) and that Gaussian mixture model fitting was applied to describe the different populations of MPL values. Thus, the insulin fibrils were grouped into four different populations using four component Gaussian mixture models to the histogram of the MPL values of the insulin fibrils using the Statistic Online Computational Resource Modeler (http://www.socr.ucla.edu/ htmls/SOCR Modeler.html) (18). The Kolmogorov-Smirnoff (K-S) test (19) was used to assess the quality of the four Gaussian mixture model fit to the MPL values (n 763). To compute this test, we simulated 763 points according to the Gaussian mixture model which yielded the K-S test statistics D 0.0262 corresponding to a P value 0.953. This result indicates the four Gaussian mixture model is a good approximation to the process generating the measured MPL values of the insulin fibrils.
6. Vriend G, Sander C (1993) Quality control of protein models: Directional atomic contact analysis. J Appl Cryst 26:47 60. 7. DeLano WL (2002) (DeLano Scientic, San Carlos, CA). 8. Nettleton EJ (1998) Ph.D. thesis. (Oxford University, Oxford). 9. Jones TA, Zou JY, Cowan SW, Kjeldgaard M (1991) Improved methods for building protein models in electron density maps and the location of errors in these models. Acta Crystallogr A 47:110 119. 10. Brunger AT, et al. (1998) Crystallography & NMR system: A new software suite for macromolecular structure determination. Acta Crystallogr D Biol Crystallogr 54:905921. 11. Fabiola F, Bertram R, Korostelev A, Chapman MS (2002) An improved hydrogen bond potential: Impact on medium resolution protein structures. Protein Sci 11:14151423.

Ivanova et al. www.pnas.org/cgi/content/short/0910080106

2 of 9

12. Jimenez JL, et al. (2002) The protolament structure of insulin amyloid brils. Proc Natl Acad Sci USA 99:9196 9201. 13. Wall JS, Hainfeld JF (1986) Mass mapping with the scanning transmission electron microscope. Annu Rev Biophys Biophys Chem 15:355376. 14. Wall JS, Simon MN (2001) Scanning transmission electron microscopy of DNA-protein complexes. Methods Mol Biol 148:589 601. 15. Diaz-Avalos R, King CY, Wall J, Simon M, Caspar DL (2005) Strain-specic morphologies of yeast prion amyloid brils. Proc Natl Acad Sci USA 102:1016510170.

16. Zhu H, et al. (1996) Assembly of the gigantic hemoglobin of the earthworm Lumbricus terrestris. Roles of subunit equilibria, non-globin linker chains, and valence of the heme iron. J Biol Chem 271:3000730021. 17. Tohka J, et al. (2007) Genetic algorithms for nite mixture model based voxel classication in neuroimaging. IEEE Trans Med Imaging 26:696 711. 18. Dinov ID, et al. (2006) LONI visualization environment. J Digit Imaging 19:148 158. 19. Lampariello F (2000) On the use of the Kolmogorov-Smirnov statistical test for immunouorescence histogram comparison. Cytometry 39:179 188.

Ivanova et al. www.pnas.org/cgi/content/short/0910080106

3 of 9

Fig. S1. B-chain segment VEALYLV of insulin does not affect rate of insulin bril formation. (A) Micrograph of the needle-like aggregates formed by VEALYLV. (B) Fluorescence assay showing that VEALYLV, in contrast to LVEALYL, does not change the kinetics of insulin bril formation when added to the reaction mixture in equimolar ratio with insulin. Scale bar, 400 nm.

Ivanova et al. www.pnas.org/cgi/content/short/0910080106

4 of 9

Fig. S2. Comparison of the insulin bril model proposed in this paper to a 25 resolution density (gray) calculated from cryo-EM image reconstruction of two-protobril particles by Jimenez et al. (12). The parts of our model unaccounted by the density can possibly be explained by the loss of these features in this low-resolution density map.

Ivanova et al. www.pnas.org/cgi/content/short/0910080106

5 of 9

Fig. S3. Testing the correction factor used to compute the MPL values of the insulin brils, shown in Fig. 4A. The correction factor used in the MPL calculations of the insulin brils was determined using half of the measured masses of tobacco mosaic virus (TMV) particles. (A) To test its preciseness, this correction factor was applied to the remaining half of TMV mass measurements, giving a Gaussian distribution centered at 13.21 0.54 kDa/. This value is in good agreement with 13.10 kDa/, which is the mass of TMV particles. Our mass measurements were also validated by determining the mass of the well studied earthworm hemoglobin. First, the correction factor was determined based on 560 mass measurements of TMV particles from 40 micrographs. (B) The distribution of MPL values of the remaining half of TMV sampling points gave a Gaussian distribution centered at 13.10 0.81 kDa/, (C) To determine the mass value of earthworm hemoglobin, the correction factor calculated from the TMV mass measurements shown in (B) was applied to the 725 mass measurements for the earth worm hemoglobin. This yielded a Gaussian curve centered at 3.62 0.23 MDa, which corresponds well to the previously determined value of 3.75 MDa (16).

Ivanova et al. www.pnas.org/cgi/content/short/0910080106

6 of 9

Fig. S4. QQ-Normal plots of MPL measurements of the standard particle TMV (A) and the insulin bril (B). Note that the QQ-Normal plot of the standard particle TMV is linear, which leads to the assumption of the normality of the standard particle distribution. In contrast, the QQ-Normal plot of MPL values of insulin brils are not linear, indicating the multimodal behavior of the data leading to the use of mixture model tting.

Ivanova et al. www.pnas.org/cgi/content/short/0910080106

7 of 9

Table S1. Sequences of nine six-residue-long peptides used in insulin bril formation assays
A chain A12: SLYQLE A13: LYQLEN A14: YQLENY B chain B8:GSHLVE B9:SHLVEA B10:HLVEAL B11:LVEALY B12:VEALYL B13:EALYLV

None of these segments changed the rate of insulin bril formation.

Ivanova et al. www.pnas.org/cgi/content/short/0910080106

8 of 9

Table S2. Statistics of crystal data collection and atomic renement

Data Space group Cell dimensions a,b,c () alpha,beta,gamma,() Resolution () Rmerge (%)(linear/square) I/sigma(I) Completeness (%) Redundancy Renement Resolution () No. reections Rwork/Rfree No. atoms Protein Ligand/ion Water B-factors (2) Protein Ligand/ion Water R.m.s. deviations Bond lengths () Bond angles () C2 49.5, 4.8, 19.4 90.0, 96.7, 90.0 1.00 18.4/19.0 (36.1/41.1) 7.15 (3.26) 95.1 (84.7) 5.0 (3.7) 16.14- 1.00 (1.11- 1.00) 2396 (651) 14.6/18.0 (23.2/33.8) 133 0 3 6.07 n/a 28.17 0.015 1.790

Values in parentheses refer to the highest resolution shell.

Ivanova et al. www.pnas.org/cgi/content/short/0910080106

9 of 9