Identification of a hinge residue controlling islet amyloid polypeptide self-assembly and cytotoxicity

The islet amyloid polypeptide (IAPP) is a 37-residue peptide hormone whose deposition as amyloid fibrils in the pancreatic islets is associated with type 2 diabetes. Previous studies have suggested that residue Asn-21 plays a critical role in the in vitro self-assembly of IAPP. Herein, we studied structure–self-assembly relationships focusing on position 21 to gain detailed insights into the molecular mechanisms of IAPP self-assembly and to probe the conformational nature of the toxic assemblies associated with β-cell death. Thioflavin T (ThT) fluorescence, CD spectroscopy, and transmission EM analysis revealed that the Asn-21 amide side chain is not required for IAPP nucleation and amyloid elongation, as N21A and N21G variants assembled into prototypical fibrils. In contrast, Asn-21 substitution with the conformationally constrained and turn-inducing residue Pro accelerated IAPP self-assembly. Successive substitutions with hydrophobic residues led to the formation of ThT-negative β-sheet–rich aggregates having high surface hydrophobicity. Cell-based assays revealed no direct correlation between the in vitro amyloidogenicity of these variants and their toxicity. In contrast, leakage of anionic lipid vesicles disclosed that membrane disruption is closely associated with cytotoxicity. We observed that the N21F variant self-assembles into worm-like aggregates, causing loss of lipid membrane structural integrity and inducing β-cell apoptosis. These results indicate that specific intra- and intermolecular interactions involving Asn-21 promote IAPP primary nucleation events by modulating the conformational conversion of the oligomeric intermediates into amyloid fibrils. Our study identifies position 21 as a hinge residue that modulates IAPP amyloidogenicity and cytotoxicity.

The aggregation and tissue deposition of proteins into amyloid fibrils are the hallmark of numerous diseases, including Alzheimer's disease, type II diabetes, and various amyloidoses (1,2). In over 90% of patients afflicted with type II diabetes, amyloid deposits are observed in the extracellular space of pancreatic islets of Langherhans (3)(4)(5). The main component of islet amyloids is the peptide hormone islet amyloid polypeptide (IAPP, 3 or amylin) (6,7). The accumulation of IAPP-insoluble aggregates correlates closely with the duration and severity of the disease and with the loss of ␤-cell mass (4). The link between islet amyloid and type II diabetes initially led to the postulate that amyloid fibrils mediate ␤-cell degeneration. This hypothesis was reinforced by the early work of Lorenzo et al. (8), demonstrating the cytotoxicity of IAPP fibrils on pancreatic islet cells. However, most recent studies have indicated that oligomeric and nonfibrillar species cause cell death (9 -11). Using time-resolved analysis, it has been observed that IAPP amyloid structures are nontoxic to INS-1 cells and that the toxic species are low-order oligomers, which lack a ␤-sheet structure and hydrophobic patches (12). Recently, IAPP fibrils have been shown to be toxic to RIN-5F cells, and it has been proposed that a quaternary structure characterized by pairs of ␤-sheets joined by a dry interface represents the toxic spine (13). The opposing conclusions between these two elegant studies (12,13) are likely associated with differences in the experimental conditions (monomerization, buffer, aging time, cells, toxicity assays, etc.) and/or with the undetectable presence of toxic oligomers in the fibril preparation used by Krotee et al. (13). Thus, although a substantial number of studies have investigated the conformational nature of IAPP toxic species, the subject is still the matter of active debates.
IAPP is a 37-residue hormone that is co-expressed and cosecreted with insulin by pancreatic ␤-cells (14). The peptide exhibits, under its monomeric state, a conformational ensemble populated by disordered structures, although it diverges from an absolute random coil by the presence of transient helical conformations (15). In membrane mimetic environments, which are known to accelerate fibrillization, IAPP is mainly characterized by helical conformation (16,17). A structure characterized with three antiparallel ␤-strands has been recently reported when the peptide is trapped into negatively charged lipid nanodiscs (18). Upon self-assembly into cross-␤sheet quaternary structure, each monomer adopts a U-shaped conformation with two ␤-strands connected by a loop. According to solid-state NMR, the ␤-strands comprise residues 8 -17 and 28 -37, and the loop involves residues 18 -27 (19). In the EPR model, the two ␤-strands comprise residues 14 -19 and 31-36 (20). Conformational changes associated with the transition of IAPP from its soluble state ensembles into ordered amyloids and the molecular interactions governing this transition remain elusive. Helical conformation has been proposed to be critical for self-recognition, as oligomerization could be thermodynamically linked with helix formation within the 5-20 segment (21,22). In contrast, by using a helix-disrupting analog, it has been proposed that helical species are off-pathway and that preventing helical folding increases cytotoxicity (23). Moreover, it has been reported that helical conformations are mainly found in the monomeric state and could seed oligomer formation, although they are not mandatory for amyloid growth (24).
Elucidating the molecular determinants promoting oligomerization and the interactions initiating amyloidogenesis is critical to better define the nature of the toxic proteospecies. Studies have exploited residue-specific modifications to better define the driving forces of IAPP self-assembly (25)(26)(27)(28)(29)(30). Sequence differences between human and mouse IAPP and mutational studies have initially indicated that the segment 20 -29 dictates aggregation and toxicity (11). Nevertheless, substitutions outside the 20 -29 amyloidogenic region, including A13E, V17C, and Y37L, have been shown to alter amyloid formation (27,31). It has been shown that residue Asn-21 plays a key role in the in vitro self-assembly of IAPP, as its consecutive replacement by Leu, Ser, and Asp inhibits amyloid formation (25,30). Position 21 is particularly intriguing because it is located in the disordered loop joining the two ␤-strands in the amyloid conformation (19,20) (Fig. 1). The side chain of Asn-21 projects outward from the protofilament core; thus, it could participate in fibril packing. This position is also located at the intersection of the putative 5-20 helical segment and the 20 -29 amyloidogenic core. Moreover, Asn-21 is adjacent to Ser-20, for which the Asian mutation S20G is known to increase amyloidogenicity and toxicity (32). These observations strongly suggest that this residue plays a critical role in the conformational conversion modulating oligomerization, nucleation, and/or amyloid growth.
Accordingly, we performed a structure-assembly relationship study focusing on position 21 by tuning the side-chain physicochemical properties and the local conformational freedom of the backbone. The data indicated that the introduction of a hydrophobic residue at this position locks the peptide into highly toxic ␤-sheet-rich aggregates, whereas a Pro hastens IAPP self-assembly into nontoxic amyloid fibrils. The present study reveals that Asn-21 acts as a molecular hinge modulating IAPP amyloid formation and cytotoxicity.

Rational design of N21X mutants
To gain mechanistic insights into IAPP self-assembly and toxicity to ␤-cells, we prepared a library of analogs modified at Asn-21. We probed the contribution of the amide side chain by successively incorporating Ala and a negatively (Asp) and a positively (diaminobutyric acid (Dab)) charged residue. Dab was used instead of Lys to maintain the length of the side chain. Asn-21 was substituted with hydrophobic residues with high ␤-sheet propensity (Phe and Leu) and with Phe capable ofinteractions. Conformational modifications were introduced to favor turn (Pro, Gly, and D-enantiomer). In fact, D-amino acids are known to promote and/or stabilize turn conformation (33,34). The C-␣-methylated aminoisobutyric acid (Aib) residue, known to promote helical folding (35), was incorporated. These substitutions (Pro, Gly, Asn, and Aib) modulate the local conformational freedom of the peptide backbone. Rodent IAPP (rIAPP), which contains three Pro residues and is less prone to aggregation and is nontoxic, was used as a negative control (36).

Substitutions at position 21 modulate kinetics of self-assembly
The effect of site-specific modifications on self-assembly was initially evaluated by thioflavin T (ThT) fluorescence. ThT is a small dye whose fluorescence emission increases sharply upon its binding to the cross-␤-sheet quaternary structure (37). IAPP amyloid formation can be ascribed to a nucleated polymerization where the three distinctive phases (lag phase, elongation phase, saturation phase) are governed by different kinetics. Dynamic and transient oligomeric species are generated during the lag phase, which is the thermodynamic rate-limiting step. Upon formation of competent oligomer(s), or nuclei, the elongation phase begins, leading to the rapid growth of protofilaments and fibrils until equilibrium is reached. The sigmoidal curve obtained by measuring ThT fluorescence over time is fitted to a Boltzmann sigmoidal, and the lag time can be extracted. Under the conditions of the assay (20 mM Tris-HCl, pH 7.4, 40 M ThT, 25°C, nonbinding surface 96-well plates, 10-min interval between measurements, no agitation), a lag time of 8.6 Ϯ 1.8 h was observed for IAPP at 12.5 M (Fig. 2) and 4.2 Ϯ 1.2 h at 25 M (Fig. S1).
Introducing a hydrophobic side chain (N21F, N21L) at position 21 abolished the formation of ThT-positive assemblies ( Fig. 2A). This is in agreement with the previous work of Miranker and colleagues (25). Interestingly, the presence of an amide group on the residue 21 side chain is not a prerequisite for the formation of a ThT-positive cross-␤-sheet structure, as the substitution N21A led to a slight acceleration of self-assembly. This result is unexpected, considering that the mutant N21S has been previously shown to self-assemble into ThTnegative amorphous aggregates (25). Introducing a negative or A, primary sequence of human IAPP with Asn-21 residue in red boldface type and the amyloidogenic core in blue. IAPP has a disulfide between Cys-2 and Cys-7 and a C-␣-amidated C terminus. The two ␤-sheets inferred from the solidstate NMR amyloid fibril model are indicated with gray arrows. B, solid-state NMR model of IAPP protofilament with the Asn-21 side chain indicated (19).

Self-assembly-toxicity relationships of IAPP
a positive charge without altering the side chain length (N21D or N21Dab) inhibited the formation of ThT-positive species. These results indicate that IAPP amyloid formation tolerates the elimination of the position 21 amide group, although it does not allow the introduction of a charged or hydrophobic residue.
Considering that IAPP self-assembly occurs in the absence of the Asn-21 amide group, we investigated the role of backbone conformational entropy and side-chain chirality. Introducing the turn-inducing residue Pro led to an acceleration of amyloid assembly, with N21P displaying a lag time of 4.2 Ϯ 0.8 h and a significant increase of ThT end-point fluorescence (Fig. 2B). This observation is intriguing, as a Pro residue is recognized to inhibit amyloid assembly (36,38,39). For instance, single-point mutation I26P has been shown to transform IAPP into a potent amyloid inhibitor (28). When Asn-21 was replaced by Aib, known to favor ␣-helix, amyloid assembly was slowed down to a lag time of over 12 h, with some ThT assays at 12.5 M showing no increase of fluorescence over 20 h (Fig. S2). Increasing the conformational freedom by incorporating Gly led to kinetics somewhat similar to the N21A analog. It is worth mentioning that N21Aib, N21A, and N21G showed a somewhat high heterogeneity between ThT kinetic assays performed at 12.5 M (Fig. S2). This observation suggests that the Asn-21 amide function facilitates the IAPP nucleation step through specific intra-or intermolecular hydrogen bonds. Substituting Asn-21 by its corresponding D-amino acid hastened amyloid formation, although it led to a lower fluorescence at the saturation phase.
Similar trends were obtained with all peptides used at 25 M (Fig. S1). These results show that modifications favoring turn conformation (Pro and D-Asn) hasten IAPP self-assembly, whereas restricting the local conformation to the helical space (Aib) delays amyloid formation.

Asn-21 modulates the conformational conversion associated with self-assembly
The effect of site-specific modifications on conformational transitions was evaluated by CD spectroscopy. Peptides were incubated at room temperature under quiescent conditions (20 mM Tris-HCl, pH 7.4, 50 M IAPP), and far-UV CD spectra were recorded. As expected, immediately after solubilization of monomerized/lyophilized IAPP, a random coil CD spectrum was recorded (Fig. 3). No conformational shift was observed during the first 6 h incubation of WT IAPP, whereas a ␤-sheet secondary structure (single minimum at 220 nm) was observed upon 24 h. The presence of a hydrophobic group (Phe, Leu) confined IAPP into a ␤-sheet conformation immediately after solubilization. These ␤-sheet-rich species assembled under these conditions were also ThT-negative (Fig. S3). N21D and N21Dab remained in a random coil conformation, even after 48-h incubation. Although the N21A and N21G mutants showed a ThT-positive signal in the kinetic assays (Fig. 2), the two peptides remained in a random coil conformation after 24 h of incubation, and a ␤-sheet signal was observed only after 48 h of incubation. Moreover, under the conditions of the CD Results were analyzed using Student's t test, and statistically significant difference (between IAPP and analogs) was established at p Ͻ 0.01 (*). B-D, data are not shown for N21F and N21L because results could not be fitted to a sigmoidal curve. a.u., arbitrary units.

Self-assembly-toxicity relationships of IAPP
experiment, a low ThT fluorescence was measured for N21A and N21G (Fig. S3). Substitution of Asn-21 by an Aib abolished conformational transition. It is worth mentioning that time point measurements under quiescent conditions in a microcentrifuge tube ( Fig. 3 and Fig. S3) cannot be directly compared with the microplate-based ThT kinetics ( Fig. 2 and Figs. S1 and S2). We recently reported that, although the microplate kinetic assay is done under quiescent conditions (i.e. no agitation between each reading), the displacement of the microplate within the fluorimeter occurring during measurements is sufficient to accelerate amyloid formation (40). Thus, self-assembly of N21A, N21G, and N21Aib appears to be more dependent on agitation than the WT IAPP. The mechanical forces associated with agitation are known to enhance amyloid formation by promoting mass transport and amplifying the number of fibrils by fragmentation (secondary nucleation) (41).
Time point CD analysis of N21P and N21n mutants revealed their prompt random coil-to-␤-sheet conformational conversion ( Fig. 3), in agreement with ThT kinetics. A ThT-positive signal was measured for N21n at time 0 h (dead time of 3-4 min), confirming the prompt self-assembly of this analog (Fig.  S3). In fact, short fibrils could even be observed by atomic force microscopy immediately after the solubilization of N21n, whereas no fibrillar aggregates were visible for WT (Fig. S4). Control rIAPP remained in a random coil conformation (Fig.  S5).
The self-assembly of N21X derivatives was further characterized by measuring the formation of hydrophobic clusters using 8-anilino-1-napthalenesulfonic acid (ANS). The fluorescence intensity of ANS increases upon its binding to protein hydrophobic patches. Random coil ThT-negative WT IAPP species (0 h) did not induce any ANS fluorescence (Fig. 4). Upon 24-h incubation, a positive ANS signal was measured. This result indicates that the formation of ThT-positive ␤-sheet-rich assemblies ( Fig. 3 and Fig. S3) is associated with the emergence of hydrophobic clusters (Fig. 4). Interestingly, N21F and N21L exhibited a strong signal of ANS fluorescence immediately after their solubilization. Thus, incorporation of a hydrophobic res-idue at position 21 prompts the formation of ␤-sheet-rich ThT-negative aggregates with high surface hydrophobicity. Successive substitution of Asn-21 with Asp and Dab precluded the formation of ANS-positive assemblies, whereas the Asn-to-Ala mutation reduced ANS fluorescence. As anticipated from the kinetic assays and time point CD analysis, incorporation of Pro and the inversion of Asn chirality led to a positive ANS signal. As observed for ThT (Fig. S3), freshly dissolved N21n showed an increase of ANS fluorescence (0 h; Fig. 4). The Asnto-Gly mutation delayed time-dependent increase of ANS fluorescence, as observed for ThT fluorescence. Time point ANS fluorescence (Fig. 3) was plotted as a function of time point ThT fluorescence (Fig. S3), and we observed that the formation of ThT-positive assemblies is associated with the emergence of hydrophobic clusters for WT IAPP, N21n, and N21P (Fig. S6). The correlation between ANS and ThT signal was not present for N21F and N21L, whereas a low correlation was observed for N21G.

Asn-21 mutations modify the supramolecular architecture of assemblies
The mesoscopic architecture of the assembled proteospecies was analyzed by negative-stain transmission EM (TEM). Long, unbranched, and twisted fibrils were obtained for WT IAPP after 24 h of aging under the conditions of the time point analysis (Fig. 5). Amorphous aggregates were mainly observed for the N21F and N21L mutants, although short fibrils could be detected in some N21L samples ( Fig. 5 and Fig. S7). TEM analysis validated that the addition of a charge at position 21 inhibits fibril formation, although fibril-like assemblies could be observed for N21Dab (Fig. S8). In sharp contrast, successive replacements of Asn-21 with Pro and its D-enantiomeric counterpart led to the formation of a dense network of fibrils. Although the time point CD and ThT analysis of N21A and N21G mutants suggested the absence of amyloid assemblies after 24 h of aging ( Fig. 3 and Fig. S3), TEM analysis revealed the presence of fibrils for both peptides. Poorly defined fibrils could

Self-assembly-toxicity relationships of IAPP
be observed in some N21Aib preparations ( Fig. 5 and Fig. S9), in agreement with the low reproducibility between ThT kinetics.

Toxicity of N21X mutants does not correlate with their in vitro amyloidogenicity
The relationships between the in vitro amyloidogenicity and the cytotoxicity of N21X analogs were investigated by evaluating the viability of INS-1 cells upon treatment with freshly dissolved monomerized peptides. As reported (12,29,42), a concentration-dependent decrease of viability was observed for IAPP, with a cell viability below 20% of control at 50 M (Fig. 6). In contrast, the nonamyloidogenic rIAPP did not induce cell death, even at a concentration of 50 M. N21A, N21P, and N21n mutants, which self-assembled into amyloids, reduced the viability of INS-1 cells as the WT peptide. Incorporation of a positive or a negative charge at position 21 reduced the toxicity of IAPP, although both analogs were toxic at 50 M (Fig. 6). The hydrophobic-substituted derivatives were toxic to ␤-cells, although N21F and N21L did not form amyloids. These observations indicate that nonamyloidogenic IAPP derivatives (N21Dab, N21D, N21F, and N21L) can be toxic toward INS-1E cells. In contrast, N21Aib and N21G were nontoxic. Toxicity (Fig. 6) was plotted as a function of in vitro amyloidogenicity ( Fig. 1) to evaluate the correlation between these two parameters. No direct correlation between in vitro amyloidogenicity and cytotoxicity was observed (Fig. S10), although mutagenesis at position 21 dramatically modulates both the cytotoxicity and the self-assembly of IAPP.
It was proposed that toxic IAPP species are mainly pre-amyloid intermediates that are transiently formed during the lag phase (12). Thus, the presence and distribution of oligomers populating the lag phase were evaluated for selected mutants using

Self-assembly-toxicity relationships of IAPP
photoinduced cross-linking of unmodified proteins (PICUP) followed by SDS-PAGE analysis (43). As observed in Fig. S11, an essentially similar distribution of proteospecies, from monomers to hexamers, was obtained for WT, N21P, N21n, N21G, and rIAPP. In contrast, cross-linking of the ThT-negative and toxic N21F mutant only revealed monomers and dimers and an absence of higher oligomeric species. Considering that N21G (amyloidogenic) and rIAPP (nonamyloidogenic) are not cytotoxic (Fig. 6), the PICUP results indicate that toxicity is not necessarily associated with the in vitro aggregation propensity and that not all oligomers are toxic, as shown previously (12).

Mutagenesis of Asn-21 modulates lipid membrane disruption
The mechanism by which IAPP induces cell death is complex and is still the subject of active research (10,44). One of the most accepted upstream events is disruption of the plasma membrane structural integrity by prefibrillar proteospecies (45,46). In addition, lipid membranes are known to hasten amyloid formation and might change the pathway(s) by which IAPP self-assembles, compared with aqueous solution (45,47). Thus, we sought to evaluate how mutagenesis at position 21 affects the kinetics of amyloid formation in the presence of lipid vesicles and the ability of IAPP to disrupt membranes. First, ThT fluorescence assays were performed in the presence of large unilamellar vesicles (LUVs) composed of phosphocholine/ phosphoglycerol (DOPC/DOPG, 7:3). The presence of anionic vesicles dramatically hastened amyloid formation of WT, N21P, and N21n peptides, with a lag time under 30 min for N21n and WT and 90 min for N21P ( Fig. 7A; peptide concentration of 12.5 M). In contrast, no ThT signal was measured for the N21G and N21F mutants and rIAPP over the 4-h incubation time. Next, we examined the ability of these peptides to induce leakage of DOPC/DOPG vesicles. Calcein was encapsulated at high concentration within LUVs, leading to fluorescence self-quenching. Membrane disruption releases the dye, and the fluorescence is restored. LUVs loaded with calcein were incubated in the presence of 12.5 M peptides, and fluorescence was measured for 4 h. For WT IAPP and the N21n and N21P mutants, an increase of fluorescence was observed over time with membrane leakage of 60 -80% after 4 h (Fig. 7B). Strikingly, the nontoxic amyloidogenic N21G analog, which does not form ThT-positive species in the presence of LUVs, induced a low membrane leakage. Actually, N21G mutant shows a similar behavior in the presence of membrane vesicles as the nontoxic rIAPP (Fig. 7). Interestingly, although the N21F analog did not form ThT-positive signal in the presence of lipid vesicles, a high percentage of membrane leakage was measured. These results expose a certain association between cytotoxicity (Fig. 6) and membrane disruption (Fig. 7).

Residue 21 controls the assembly-dependent cytotoxicity of IAPP
Bearing in mind that Asn-21 mutagenesis modulates the morphology of the resulting assemblies (Fig. 5), we evaluated the toxicity of the proteospecies formed after a prolonged incu-

Self-assembly-toxicity relationships of IAPP
bation period. WT IAPP, N21F, N21P, N21n, and rIAPP were incubated at 150 M under quiescent conditions at room temperature. At different incubation times, peptide solution was applied to INS-1E cells, and cellular viability was evaluated. As anticipated, all peptides, except for the rIAPP control, were highly toxic at time 0 h (Fig. 8A). N21P and N21n assemblies incubated for 24 h were significantly less toxic to cells compared with IAPP. The proteospecies generated after 120 h of aging for WT, N21P, and N21n were nontoxic. TEM images validated that IAPP, N21n, and N21P assembled into well-defined amyloid fibrils at 150 M (Fig. 8B). These data support previous studies indicating that well-defined fibrils are poorly toxic. In sharp contrast, the Asn-to-Phe substitution locked IAPP into proteospecies, which remained highly toxic over time (Fig. 8A). As observed by TEM, N21F formed short and poorly defined wormlike fibrils when incubated at 150 M for 48 h (Fig. 8B). Interestingly, immediately upon its solubilization at 150 M, the N21F analog assembled into amorphous spherical aggregates that slowly evolved into short wormlike assemblies (Fig. S12).
IAPP-induced cytotoxicity is associated with numerous downstream cellular events, including oxidative stress, mitochondrial dysfunction, endoplasmic reticulum stress, and apoptosis (9,44,48). We evaluated the activation of caspase-3 upon cell treatment with the different proteospecies. As caspase-3 activation is an upstream event to cell death, cells were treated for 3 h with the proteospecies. All four peptides under their nonfibrillar states (0 h) induced a significant increase of caspase-3 activation (Fig. 8C). N21F short fibrillar species obtained after 48 h of aging strongly activated caspase-3, in contrast to IAPP, N21P, and N21n. Moreover, we measured the ability of the assemblies formed after 48 h of aging to induce membrane disruption using the calcein release assay. Interestingly, N21F wormlike aggregates disrupted anionic LUVs, whereas well-defined amyloid fibrils (WT, N21P, and N21n) had no significant effect on lipid membrane (Fig. 8D). We investigated whether the cytotoxicity and membrane leakage ability of N21F assemblies could be associated with the presence of remaining oligomers/monomers in solution. Accordingly, an N21F aggregation mixture (48 h; 150 M) was centrifuged at 35,000 ϫ g for 45 min, and the presence of soluble species in the supernatant was evaluated by analytical reverse-phase HPLC (RP-HPLC). As observed in Fig. S13, no N21F was detected by HPLC in the supernatant. These data indicate that N21F ThTnegative ␤-sheet-rich assemblies are highly toxic to pancreatic ␤-cells, increasing the complexity of identifying specific IAPP toxic species. Nonetheless, we cannot rule out the possibility that the N21F wormlike aggregates disassemble and release soluble species, which induce cell death.

Discussion
A complex network of specific intra-and intermolecular interactions, which give rise to off-and on-pathway intermediates, governs IAPP self-assembly from a disordered peptide into

Self-assembly-toxicity relationships of IAPP
highly ordered fibrils. Intermediate assemblies were identified as the most toxic species of the amyloidogenic cascade. Thus, it is important to pinpoint the chemical determinants governing these secondary and quaternary structural transitions. Whereas IAPP amyloid formation is somewhat tolerant to mutations, previous studies have revealed that self-assembly does not tolerate substitution at Asn-21 (25,30). Herein, we performed a structure-assembly relationship study at position 21 to delineate the molecular mechanisms of self-assembly and to better define the conformational species associated with ␤-cell death.
In contrast to expected results based on previous studies, we showed that an Asn-21 amide group is not a prerequisite for nucleation and amyloid elongation. As shown in the kinetic assays, substitution by Ala led to a shorter lag time, and the N21A mutant formed well-defined fibrils. This result indicates that hydrogen-bonding interactions involving the Asn amide side chain are not mandatory for IAPP amyloid formation. Nonetheless, N21A amyloidogenesis was significantly delayed when assembly occurred in a test tube and was followed by time point ANS, ThT, and CD measurements. A similar trend was observed with the mutant N21G. N21A and N21G mutants show high heterogeneity between ThT kinetics in microplate assays. Microplate displacement within the fluorimeter induces significant agitation, hastening amyloid formation through mass transport and fibril fragmentation (secondary nucleation). Moreover, ThT kinetic assays are performed in nonbinding surface microplates, whereas the surface of microcentrifuge tubes is nontreated. It has been reported that hydrophobic surfaces, such as nonbinding coating, hasten amyloid formation by promoting the primary nucleation rate (41). These observations suggest that removing the amide group (i.e. N21A and N21G) prolongs the lag phase (delayed nucleation) under fully quiescent conditions and in the absence of a hydrophobic surface. In contrast, the nonbinding surface of the microplate and/or agitation within the fluorimeter counteract this effect by promoting primary and/or secondary nucleation events.
Interestingly, substitution of Asn-21 with the conformationally constrained and turn-inducing Pro accelerated amyloid formation. This result is surprising, considering that the incorporation of Pro within an amyloidogenic peptide is a common strategy to inhibit amyloid formation. This substitution constitutes the first X-to-Pro mutation within the amyloidogenic 20 -29 segment promoting self-assembly. Inversing the chirality of Asn-21, a modification known to favor ␤-turn and to destabilize ␣-helix (34), dramatically hastened amyloid formation. Short ThT-positive fibrils were even observable immediately after solubilization of N21n. Previous studies have suggested that the segment 20 -29 mediates initial self-recognition and that early oligomers contain stacks of parallel ␤-sheet within the 23 FGAIL 27 region (24,49,50). However, below a critical concentration, IAPP self-recognition appears to be primarily initiated by the N-terminal region, which than propagates to the central domain (51). Regardless of the contacts initiating aggregation, a high-energy structural rearrangement within the 20 -29 segment is necessary for the conversion from oligomers to fibrils. The energy barrier induced by the FGAIL ␤-sheet oligomers is predicted to slow down amyloid formation and to be at the origin of the lag phase (50). In the NMR and EPR atomistic structural models of amyloids, Asn-21 is located at the ␤-strand-loop interface, whereas in the oligomers, this residue is located close to the FGAIL ␤-sheet. Accordingly, the N21P mutation provides sufficient torsional driving forces to overcome the FGAIL ␤-sheet intermediate free energy barrier while not affecting the initial self-recognition events. This promotes primary nucleation by facilitating the structural conversion of the oligomer into the nuclei conformation.
Aromatic stacking and hydrophobic collapse are key driving forces for aggregation. IAPP self-recognition involving the 20 -29 amyloidogenic core is mainly directed by hydrophobic interactions (52). It has been reported, using both the fulllength peptide and IAPP (20 -29), that the contributions of Phe-23 in amyloid formation are a function of its hydrophobicity and ␤-sheet propensity, whereas its ability to forminteractions is not critical (27,53). In the present study, we observed that the successive substitutions of Asn-21 with Leu and Phe led to the prompt formation of ThT-negative ␤-sheetrich aggregates with high surface hydrophobicity. Some short and poorly defined fibrils could also be observed for N21L and N21F when incubated for a long period at 50 and 150 M, respectively. The increase of hydrophobicity and/or ␤-sheet propensity of the 20 -29 segment induced by these mutations promote(s) the formation of low energy nonamyloid aggregates, most likely due to the propagation of the FGAIL parallel ␤-sheet. In sharp contrast, incorporation of a charge (i.e. Asp or Dab) inhibited amyloid assembly, probably due to electrostatic repulsion between monomers. It has been shown that the consecutive substitutions of the neighboring Asn-22 residue with Asp and Leu have a modest impact on IAPP amyloid formation (25,30). This clearly indicates that Asn-21 plays a dictating role in IAPP self-assembly and is involved in specific intra-or intermolecular interactions controlling nucleation and/or elongation events, which is not the case for Asn-22.
Whereas the causative link between islet amyloids and type II diabetes was initially described over 100 years ago (3), the conformational nature of the proteotoxic species remains ambiguous. Whereas most studies have reported that soluble oligomers populating the lag phase are causing ␤-cell death (10,12,54), some earlier and current researches have proposed that amyloid fibrils could also be cytotoxic (8,13). Considering that Asn-21 mutagenesis modulates the kinetics of self-assembly and the final supramolecular morphology, we evaluated the toxicity of freshly dissolved IAPPs (monomers/oligomers) and of assemblies generated after different aging periods. We observed a concentration-dependent ␤-cell toxicity for most N21X derivatives. Interestingly, although N21F and N21L did not self-assemble into amyloids, both mutants showed a high toxicity. In sharp contrast, N21G was nontoxic, although this mutant is prone to form amyloid fibrils. These results highlight that the relation between amyloidogenicity and cytotoxicity cannot be rationalized according to in vitro biophysical studies performed in homogeneous solution. Moreover, the nontoxic rIAPP and N21G mutant readily oligomerized, indicating that not all oligomers are cytotoxic and that specific structural and/or physicochemical properties are determining factors for Self-assembly-toxicity relationships of IAPP cellular toxicity (12,55). Interestingly, the nontoxic mutant N21G induced low leakage of anionic LUVs, comparable with the rIAPP. This suggests the existence of a correlation between cytotoxicity and in vitro membrane disruption. However, this observation should be taken with caution, as it has been reported that there is no direct correlation between cell toxicity and the ability to disrupt model membranes (56). Besides, it has been shown using the N-terminal 1-19 fragment that membrane disruption can occur independently of amyloid formation (57) and that the capacity of hIAPP (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19) to perturb membrane is associated with a specific transmembrane orientation of the peptide (58). Thus, differences between the membraneinduced conformation of the WT and the N21G variant can be related to the observed divergence effect on lipid membrane integrity.
Amyloid fibrils assembled from WT IAPP, N21P, and N21n were poorly toxic, whereas the N21F analog formed toxic wormlike aggregates, which induce disruption of anionic lipid membranes. N21F cytotoxic assemblies are ThT-negative, display a high ␤-sheet content, and have solvent-exposed hydrophobic patches. This is in agreement with previous studies stipulating that the toxicity of protein aggregates is associated with their surface hydrophobicity (59,60). However, the structural and physico-chemical properties of N21F aggregates diverge substantially from those of the toxic IAPP intermediates reported by Abedini et al. (12), which lack extensive ␤-sheet structure and do not have persistent hydrophobic patches.
Thus, it appears that multiple IAPP quaternary species could contribute to ␤-cell death, perhaps by using different upstream mechanisms. Consequently, it remains challenging to determine the specific conformational features of cytotoxic prefibrillar and/or amyloid assemblies.
Overall, the present work indicates that specific intra-and intermolecular interactions involving Asn-21 are governing nuclei formation and protofilament elongation. Asn-21 could promote IAPP primary nucleation by facilitating the conformational conversion of the FGAIL ␤-sheet intermediate into poorly toxic fibrillar quaternary structure. This study identifies position 21 as a hinge residue that controls both the amyloidogenicity and the cytotoxicity of IAPP, ultimately supporting the rational development of therapeutic strategies to arrest aggregation and IAPP-induced pancreatic ␤-cell loss.

Peptide synthesis and purification
Peptides were synthesized on solid support using Fmoc (fluorenylmethyloxycarbonyl chloride) chemistry. Oxazolidine dipeptide derivatives were used to facilitate synthesis (61). Peptides were cleaved from the Rink Amide AM resin with a mixture of TFA, ethanedithiol, phenol, and water. Crude peptides were precipitated with ethyl ether, solubilized in water, and lyophilized. Peptides were dissolved in 100% acetic acid and diluted to 35% to be purified by RP-HPLC on a C 18 column. Fractions were analyzed by TOF MS using a LC/electrospray ionization-TOF (Table S1 and Fig. S14). Peptides were cyclized with 100% DMSO overnight and then diluted to 20% to be puri-fied by RP-HPLC a second time. Fractions with purity higher than 95% were pooled and lyophilized (Figs. S15-S17).

IAPP monomerization and sample preparation
Aliquots of monomerized IAPP were prepared by dissolving the lyophilized and pure peptide in 100% hexafluoro-2-propanol to a concentration of 1 mg/ml. The solution was filtered through a 0.22-m hydrophilic polyvinylidene difluoride filter and sonicated for 20 min before being lyophilized. The resulting peptide powder was solubilized for a second time in hexafluoro-2-propanol to a concentration of 1 mg/ml and sonicated for 20 min, and the solution was aliquoted and lyophilized again. Monomerized IAPP samples were kept dried at Ϫ80°C until used, but not for longer than 4 weeks.

Kinetics of amyloid formation by ThT fluorescence
Solutions were prepared by dissolving the lyophilized peptide at a concentration of 50 M in 20 mM Tris-HCl, pH 7.4. Assays were performed at 25°C without stirring in sealed blackwall clear-bottom 96-well nonbinding surface plates with 100 l/well. Final peptide concentrations were 12.5 and 25 M, and ThT concentration was 40 M. LUVs were added (or not) at a final concentration of 500 M. Fluorescence, excitation at 440 nm and emission at 485 nm, was measured every 10 min. Data obtained from triplicate wells were averaged, corrected by subtracting the corresponding control reaction, and plotted as fluorescence versus time. Data of time dependence of ThT fluorescence were fitted to a sigmoidal Boltzmann model, where T 50 is the time required to reach half of the fluorescence intensity, k is the apparent first-order constant, and Y max and Y 0 are, respectively, the maximum and initial fluorescence values.
The lag time is described as T 50 Ϫ 2/k. Data (lag time and final ThT) of at least four different lots of peptides were averaged and were expressed as the mean Ϯ S.D. Evaluation of the results was made using Student's t test, and statistically significant difference (between WT and mutants) was established at p Ͻ 0.01.

CD spectroscopy
Lyophilized aliquots of peptide were dissolved in 20 mM Tris-HCl, pH 7.4, at 50 M and were incubated at room temperature without agitation. Samples were incorporated into a 1-mm path length quartz cell. Far-UV CD spectra were recorded from 190 to 260 nm using a Jasco J-815 CD spectropolarimeter at 25°C. The wavelength step was set at 0.5 nm with an average time of 10 s/scan at each wavelength step. Each collected spectrum was background-subtracted with peptide-free buffer. The raw data were converted to mean residue ellipticity.

Time point fluorescence spectroscopy (ThT and ANS)
Solutions were prepared by dissolving the peptides at a concentration of 50 M in 20 mM Tris-HCl, pH 7.4, and were incubated at room temperature without agitation. At the indicated time, peptide was diluted to a final concentration of 25 M in Self-assembly-toxicity relationships of IAPP the presence of a final concentration of 40 M ThT or 50 M ANS. Fluorescence was measured in ultra-micro 10-mm length cells using a PTI QuantaMaster spectrofluorometer. For ThT, excitation wavelength was set at 440 nm, and the emission spectra from 450 to 550 nm was recorded. For ANS, the excitation wavelength was set at 355 nm, and the emission scan was recorded from 385 to 585 nm. For each experiment, control reactions (without IAPP) were carried out. Data obtained from at least three experiments were averaged.

Transmission EM
Peptide samples were solubilized in a 20 mM Tris-HCl buffer, pH 7.4, to a final concentration of 50 or 150 M (for timeresolved toxicity assays) and incubated at room temperature under fully quiescent conditions. Samples were diluted to 10 M before being applied on glow-discharged carbon films on 300-mesh copper grids. Peptides were adsorbed and were negatively stained with 1.5% uranyl formate for 45 s. Images were recorded using an FEI Tecnai 12 BioTwin microscope operating at 120 kV and equipped with an AMT XR80C CCD camera system.

Atomic force microscopy
Peptides were solubilized in a 20 mM Tris-HCl buffer, pH 7.4, to a concentration of 50 M and incubated at room temperature without agitation. After the indicated time of incubation, the peptide was diluted in 1% acetic acid and immediately applied to freshly cleaved mica. The mica was washed twice with deionized water and air-dried for 24 h. Samples were analyzed using a Veeco/Bruker Multimode atomic force microscope using tapping mode with a silicon tip (2-12-nm tip radius, 0.4 newton/m force constant) on a nitride lever. Images were taken at 0.5 Hz and 1024 scans/min.

Photoinduced cross-linking
Peptides were incubated at 50 M in 20 mM Tris-HCl, pH 7.4, at 25°C for the indicated periods. Cross-linking solution was added to a final concentration of 70 M Tris-(bipyridyl)Ru(II) and 1.4 mM ammonium persulfate. Aliquots were illuminated with a 150-watt incandescent bulb for 10 s. Reaction was quenched by the addition of 1 M DTT. Prefibrillar species were separated by SDS-PAGE using a 15% Tris-Tricine gel and visualized by silver staining.

Cell viability assay
Rat INS-1 cells were cultured in black-wall clear-bottom 96-well plates (tissue culture-treated) at a density of 20,000 cells/well (100 l/well) in RPMI 1640 medium supplemented with 10% FBS, 100 units/ml penicillin, 100 mg/ml streptomycin, 10 mM HEPES, 1 mM sodium pyruvate, 50 mM ␤-mercaptoethanol. After a 48-h incubation at 37°C in a 5% CO 2 , cells were treated by the direct addition of 50 l of peptide solutions (20 mM Tris-HCl, pH 7.4) at 3ϫ final concentration. Cells were incubated another 24 h, and cellular viability was measured using the resazurin reduction assay. Cell viability (in percentage) was calculated from the ratio of the fluorescence of the treated sample to the control (vehicle-treated). Data of at least four assays (with different peptide lots) were averaged and were expressed as the mean Ϯ S.D. Statistical analysis was performed with Prism version 6.0 software using Student's t test, and statistically significant difference (between WT and mutants) was established at p Ͻ 0.01.

Caspase-3 activity assay
Rat INS-1 ␤-cells were cultured in 12-well plates (tissue culture-treated) at a density of 400,000 cells/well in supplemented RPMI 1640 medium. After a 48-h incubation, cells were treated by the direct addition of peptide solution (50 M) for 3 h. Cells were lysed on ice for 30 min followed by a centrifugation at 16,000 ϫ g for 20 min. Protein extracts in supernatant were measured by a caspase-3 colorimetric assay (Sigma-Aldrich). Commercial caspase-3 enzyme provided with the kit was used as a positive control. Caspase-3 activity after a 2-h reaction was evaluated at 37°C by measuring absorbance at 405 nm.

Large unilamellar vesicle preparation and membrane leakage
DOPC and DOPG lipids were solubilized in 100% chloroform at a ratio of 7:3, and solvent was evaporated. The lipid film was hydrated in 20 mM Tris-HCl, pH 7.4, buffer containing 70 mM calcein, or not, for 30 min. Solubility of calcein was increased with the dropwise addition of NaOH. The solution was freeze-thawed five times, and lipids were extruded through a 0.1-m nucleopore membrane. Nonencapsulated calcein was removed using a 10-ml Sephadex G-25 column. Lipid concentration was determined using an P i detection colorimetric assay. The size and homogeneity of LUVs were analyzed by dynamic light scattering, and an average diameter of 100 nm was obtained. For membrane leakage, 50 M peptide was used, and concentration of LUVs was fixed at 500 M. The excitation wavelength was set at 495 nm, and the emission scan was recorded from 500 to 540 nm in an ultra-micro cell. Fluorescence was measured over incubation time. The control used to determine 100% leakage (F max ) was calcein-LUVs with 0.2% Triton X-100. Dye leakage was reported using the following equation.
% of leakage ϭ ͑F Ϫ F baseline ͒/͑F max Ϫ F baseline ͒ (Eq. 2) Author contributions-E. G., P. T. N., and X. Z. conducted the experiments and analyzed the results. E. G., P. T. N., and S. B. wrote the paper and prepared the figures. S. B. supervised the project.