Structural Insight into IAPP‐Derived Amyloid Inhibitors and Their Mechanism of Action

Abstract Designed peptides derived from the islet amyloid polypeptide (IAPP) cross‐amyloid interaction surface with Aβ (termed interaction surface mimics or ISMs) have been shown to be highly potent inhibitors of Aβ amyloid self‐assembly. However, the molecular mechanism of their function is not well understood. Using solution‐state and solid‐state NMR spectroscopy in combination with ensemble‐averaged dynamics simulations and other biophysical methods including TEM, fluorescence spectroscopy and microscopy, and DLS, we characterize ISM structural preferences and interactions. We find that the ISM peptide R3‐GI is highly dynamic, can adopt a β‐like structure, and oligomerizes into colloid‐like assemblies in a process that is reminiscent of liquid–liquid phase separation (LLPS). Our results suggest that such assemblies yield multivalent surfaces for interactions with Aβ40. Sequestration of substrates into these colloid‐like structures provides a mechanistic basis for ISM function and the design of novel potent anti‐amyloid molecules.

It has been recognized recently that liquid-liquid phase separation (LLPS) plays an important role for self-organization of membrane-less cellular organelles. [4] In particular, proteins containing low-complexity sequences can form protein-rich droplets. [5] Phase separation is the driving force for the formation of membrane-less cellular organelles such as nucleoli, stress granules,P -bodies,a nd other cellular compartments. [6] Similar to stress granules,hydrophobic small molecules undergo LLPS,a dopt colloidal structures in aqueous environment, [7] and recruit amyloidogenic proteins into their core in which amyloids adopt an altered structure that prevents amyloid neurotoxicity. [8] Thee levated local concentration facilitates interactions with the amyloid.
Herein, we show that R3-GI is highly dynamic,can adopt a b-like structure,a nd oligomerizes into colloid-like assemblies.O ur results suggest that formation of such ISM assemblies provides am ultivalent surface for interactions with Ab40, resulting in its sequestration into off-pathway nontoxic aggregates.T he suggested mechanism provides apossible mechanistic scenario for the potent amyloid inhibitor function of ISMs.

Results
Our studies focused on the two ISMs R3-GI and K3-L3-K3-GI for the following reasons ( Figure 1A and Table S1 in the Supporting Information): 1) R3-GI was used in solutionstate NMR studies as the intrinsically low solubility of all ISMs with hydrophobic linkers makes them unsuitable for solution-state NMR spectroscopy.2 )K3-L3-K3-GI was used in MAS solid-state NMR studies as it is areasonably soluble and functional analogue ( Figure S1) of the sparingly soluble but highly potent L3-GI (LLL in the linker), which shows the largest effects in terms of substrate interaction. [1b] At hird ISM, the non-inhibitor G3-GI containing the flexible GGG tripeptide as linker,was used as ac ontrol peptide.

ISM Self-Association and Exchange between Monomeric and Oligomeric States
Most of the ISMs self-associate with apparent binding affinities (app. K d )i nt he low-to mid-nanomolar concentration range. [1b] In the case of R3-GI, an app. K d value of 77 nm was determined by titrating synthetic N-terminal fluorescently labeled peptide (Fluos-R3-GI) with R3-GI ( Figure 1A). Theh igh oligomerization propensity of R3-GI was further confirmed by concentration-dependent CD studies ( Figure S2 D), chemical cross-linking ( Figure 1B), and dynamic light scattering (DLS;F igure 1C). In addition, differential interference contrast (DIC) microscopy,f luorescence microscopy,a nd transmission electron microscopy showed that R3-GI aggregation resulted in granule-like, high-molecular-weight structures ( Figure 1D-F). As expected, R3-GI assemblies within these structures have as ignificantly retarded translational diffusion coefficient in comparison to particles in isotropic solution ( Figure 1G). Next, we recorded solution-state NMR spectra of R3-GI at ac oncentration of 1.5 mm ( Figure S2 A). At this concentration, peptide self-association should result in aggregates that are too large to be observable by solution-state NMR spectroscopy.U nexpectedly,h igh-intensity and high-resolution spectra were obtained that are characteristic for monomeric, random-coil peptides.T he 13 Cc hemical shifts,w hich are sensitive to secondary structure, [9] yielded no indication for formation of a-helical or b-sheet secondary structure elements ( Figure S2 C).
In order to resolve the apparent discrepancy between the NMR findings and the results of the other biophysical studies, we performed fluorescence recovery after photobleaching (FRAP) experiments ( Figure 1H). We found that the fluorescence of the granules recovered within seconds,indicating that peptides exchange between the peptide-dense phase and bulk solution. In addition, we conducted saturation transfer difference (STD) experiments for R3-GI ( Figure 1I). We observed very intense signals in this experiment, suggesting that R3-GI undergoes chemical exchange between am onomeric random-coil-like conformation and an aggregated state that is too large to be observable by solution-state NMR analysis.I nc ontrast, an on-aggregating monomeric peptide used as ac ontrol yielded no STD signals.N otably,s imilar behavior has been observed previously for the Alzheimers disease Ab peptide,w hich has been shown to exchange between as oluble and an aggregated state. [10] These findings are in agreement with DOSY experiments ( Figure S2 E, F). We observed as maller apparent diffusion coefficient that is consistent with R3-GI undergoing at ransition between am onomeric and an aggregated state.N ext, we quantified the amount of R3-GI in the aggregated state versus solution. Fort his purpose,w ed etermined the intensities of the fluorescent granules with respect to background ( Figure S3). We found apartitioning coefficient of R3-GI on the order of 2.9, which corresponds to an approximately threefold higher concentration in aggregates than in free solution.

NMR Structural Characterization of R3-GI
Subsequently,w erecorded NOESY spectra to assign the resonances of R3-GI ( Figure S2 A) and to determine the structure of the peptide.N -Methylation increases the population of a cis peptide conformer,and has been suggested to induce at urn structure similar to ap roline. [11] As expected, three sets of resonances are observed in the N-methyl region (residues N15-L20). We estimated the populations of the three conformers G17(trans)-I19(trans), G17(cis)-I19(trans), and G17(trans)-I19(cis)tobeonthe order of 64 %, 32 %, and 4% ( Figure S4). TheG 17(cis)-I19(cis)c onformer is not sufficiently populated to be observable by NMR spectroscopy.Furthermore,wefound different sets of resonances at the N-terminal half of the peptide (residues F8-H11;F igure S5), suggesting that N-methylation assists in turn formation of the monomeric peptide.
TheSTD NMR and FRAP experiments demonstrate that R3-GI exchanges between am onomeric and an oligomeric form. Thee xperimental NOEs are thus transfer-NOEs [12] containing contributions from the monomeric and the oligomeric state of the peptide.I nf act, the observed NOEs are very intense,u nderlining the exchange contribution to the NOEs.F igure 2A summarizes the experimental long-range 1 H, 1 HN OE connectivities for R3-GI. Theo bserved contacts are indicative for as tructure containing al oop.W ei nvestigated further the salt, temperature,and pH dependence for loop formation ( Figures S6 and S7). Whereas the salt concentration did not have as ignificant impact on the intensity of the long-range cross-peaks in R3-GI, we found that conditions of low pH significantly increased the intensity of the long-range cross-peaks.S imilarly,w ef ound that low temperatures increase the fraction of peptides adopting the turn-like structure ( Figure S7). Interestingly,t he (N7-I19) [2] cross-peak intensity seems to correlate with the pK a value of the histidine imidazole ring ( Figure S8). We speculate that al ower pH and protonation of the histidine side chain is beneficial for loop formation in the aggregated state.A tt he same time,low pH has no influence on the population of the two conformers observed in the N-terminal half of the peptide ( Figure S4). We observed long-range NOEs for both conformer 1( G17(trans)-I19(trans)) and conformer 2( G17(cis)-I19(trans); Figure 2A). By contrast, the non-inhibitor peptide G3-GI shows only weak long-range NOEs if any,s uggesting that the loop-like structure is not adopted for G3-GI (Figure S9). These results are in good agreement with previous results and support the hypothesis underlying the design of the ISMs. [1b] Conformational ensembles representing the R3-GI conformers 1a nd 2w ere generated by metadynamic metainference [13] using 221 and 35 inter-residue distance restraints for the first and the second conformer, respectively (Table S4 and  Table S5). Metadynamic metainference represents an extension of the inferential structure determination approach introduced by Nilges and co-workers for heterogenous systems. [14] Using this method, an optimal coupling of simulations and equilibrium experiments allows one to determine the overall ensembles of structures that are compatible with the experimental data, in this case with the NOE-derived distances.The calculated ensembles for the two conformers are highly heterogeneous.I nfact, aclose inspec-tion of the ensembles reveals significant differences.T he G17(trans)-I19(trans)e nsemble is characterized by an equilibrium between two populations.T he first conformer is lacking any secondary structure and features alarge radius of gyration (ca. 1.3 nm), while the second conformer is characterized by aloop forming a b-like structure involving residues N7-V10 to S21. Thef ree energy for members of the two different populations is rather similar, suggesting that conformers of the two populations may interconvert on af ast timescale (microseconds or less). By contrast, the ensemble for the G17(cis)-I19(trans)c onformer does not show any indication for al oop-like structure and is overall more compact with an average radius of gyration of 0.9 nm, reflecting the observed NOE between N7 and I19. The conformational ensembles suggest that the peptide is overall disordered in solution with some preference for a b-like structure,i np articular for the G17(trans)-I19(trans)c onformer.
TheN OE intensities cannot easily be disentangled into contributions originating from the monomeric and the oligomeric state of the peptide.I no rder to probe peptidepeptide contacts in the oligomer, we prepared amixed sample that contained 50 %u nlabeled (R3-GI) and 50 %l abeled peptide (R3-GI*; labeling scheme depicted in Figure 3). In the experiment, am agnetization filter element was applied during the first evolution period t 1 to remove magnetization of protons that are directly bound to 13 Cn uclei, following ad ouble half-filter approach. [15] After the NOESY mixing

Angewandte Chemie
Research Articles 5774 www.angewandte.org time, 13 C-bound protons were selected for detection. We found anumber of sequential connectivities between labeled and non-labeled residues within one peptide (e.g., H11b-V10a). These cross-peaks are detected either above or below the diagonal, indicating that filtering of magnetization works in the intended way.I na ddition to these intramolecular sequential connectivities,w eo bserved many correlations, which yielded asymmetric cross-peak both below and above the diagonal (A6b-F8b,A 6 b-L9b,F 8 b-L9b,L 9 b-L9d). These correlations are due to intermolecular connectivities as they involve potentially labeled amino acids in both evolution periods.
In order to gain more structural insight, we performed MARTINI coarse-grain MD simulations (Figure 4). [16] 54 monomers were solvated in ac ubic box of 22.7 nm lateral length. Thes tructure of the individual monomers was described by employing an elastic network centered around arepresentative G17(trans)-I19(trans) b-turn-like conformer. Interestingly,w hile the molecules self-assembled quickly on the timescale of the simulation (after 10 ms, the average oligomer size was stable between 24 and 30 monomers), the oligomers were overall very dynamic,s howing ab road distribution of sizes and as ignificant fraction of free monomers ( Figure 4A,B). Furthermore,m onomers can exchange among oligomers on the microsecond timescale in agreement with STD experiments.T he monomer-monomer interface is well defined and characterized by few intermolecular interactions ( Figure 4C). Arginine side chains are solvent-exposed ( Figure 4D). Intermolecular interactions involve almost only amino acids 7-10. This is in agreement with the experimental intermolecular NOE contacts ( Figure 3C), suggesting that the dynamic oligomer may represent well the macroscopic behavior of R3-GI at as mall scale.A sacontrol, five additional simulations were carried out to investigate the role of the linker sequence ( Figure S11). First, the structure of the R3-GI monomer was relaxed by removing the elastic network around the b-like structure,w hich resulted in am ore disordered peptide.I nterestingly,t he overall behavior of the system is robust with respect to this property.T he resulting oligomer has similar dynamics and as imilar intermolecular interface,w hile the average oligomer size is slightly decreased. Further control simulations with and without an elastic network were performed for G3-GI and for the peptide (SG) 10 Sw ith as upposedly totally flexible linker.F or G3-GI and (SG) 10 S, the oligomer dynamics disappears.Here, monomers self-assemble rapidly into 54-mers.I nterestingly, differences can be observed for intermonomer interactions. While G3-GI retains specific intermolecular interactions very similar to R3-GI, the specific contacts are lost for the (SG) 10 S peptide.The simulations indicate that both R3-GI and G3-GI can form oligomers with robust intermolecular interactions. Thet hree arginine residues in the loop of R3-GI induce a  13 Ccorrelation spectrum of R3-GI*. C) 1 H(w 1 ), 1 H(w 3 )correlation spectrum extracted from the 3D NOESY experiment. During w 1 ,p rotons were selected that are directly bonded to 12 C, whereas 13 C-bound protons were filtered during w 3 .I naddition to trivial sequential connectivities (e.g.,H11b-V10a)t hat appear only on one side of the diagonal, anumber of symmetric cross-peaks (in red) were observed that are due to intermolecular interactions.
b-like structure that may determine the specific dynamic properties of the oligomers,w hich are essential for its inhibitory function.

Substrate Interactions
To investigate substrate interactions,weturned to the ISM K3-L3-K3-GI. Upon titration with Ab40, hetero-complexes precipitated quickly out of solution ( Figure 5A,t op). As ac ontrol, monomeric Ab40 was incubated with the noninhibitor G3-GI, and no effects on Ab40 solubility were observed ( Figure 5A,b ottom). At the same time,t he chemical shifts of Ab40 upon titration of the ISM K3-L3-K3-GI are not affected ( Figure 5B). At atenfold molar excess of the ISM peptide K3-L3-K3-GI with respect to Ab40, the 1D-1 Hs olution-state NMR spectrum is almost empty (Figure 5A,i nset), indicating that the two peptides co-precipitated. By contrast, high intensities were observed for the control sample Ab40·G3-GI ( Figure 5A,inset). Furthermore, fluorescence microscopy showed that K3-L3-K3-GI or R3-GI co-localize with Ab40 in the aggregates.B yc ontrast, no colocalization with Ab40 was observed in the case of G3-GI ( Figure 5C). TheK 3-L3-K3-GI-inducedA b40 aggregates were analyzed by TEM and solid-state NMR spectroscopy.T EM indicates the presence of mainly amorphous aggregates while short Ab40 protofibril-like assemblies were also observed ( Figure 6A,B). At first sight, the ISM-induced aggregates appear heterogeneous.H owever,s olid-state NMR experiments yielded high-resolution spectra, indicating that the ISM-Ab40 co-assemblies are homogeneously structured (Figure 6C). In fact, the spectral resolution obtained for these aggregates is very similar to the resolution achieved for Ab40 fibrils that were obtained after several rounds of seeding. [17] We performed chemical shift assignments to identify the residues of Ab40 that are part of the core of the ISM-induced aggregates (Table S6). Based on 13 Ca and 13 Cb NMR chemical shifts,wepredicted the b-strand secondary structure elements for K3-L3-K3-GI-induced Ab40 aggregates (Figure 6D). We found that the same residues as in an Ab amyloid fibril are immobilized and involved in the b-sheet core. [17] Furthermore,ac omparison of the NMR secondary chemical shifts for the two preparations shows ah igh degree of similarity ( Figure 6E), indicating that the fold of the two aggregates is rather related. In addition, we observed aTEDOR cross-peak involving the carboxyl group of residue Asp-23 and the e-amino group of Lys-28 ( Figure 6F), suggesting the formation of as alt bridge between the two residues.T his interaction is ac haracteristic feature of all Ab40 fibril structures determined thus far, [18] and confirms that also K3-L3-K3-GI-induced Ab40 aggregates adopt a b-arch-like fold upon interaction with as ubstrate amyloid in the solid state.Even though the morphology of the K3-L3-K3-GI-induced Ab40 aggregates is rather different from that of Forreference, the correlation spectrum obtained for Ab40 fibrils is shown in black. To prepare the Ab40 fibril sample, monomeric Ab40 was grown using 5% seeds, following the protocol described by Lopez and co-workers. [17] (D) Ca-Cb chemical shift differences for K3-L3-K3-GIinduced Ab40 aggregates (top). TALOS + [19] -predicted secondary structure propensity( bottom).E )Correlation of the NMR chemical shifts observed for K3-L3-K3-GI-induced Ab aggregates and Ab fibrils. On top and bottom, correlations for the absolute and secondary Ca chemical shifts are shown. Secondary Ca chemicals hifts indicate differences from random-coil chemical shifts. The correlation coefficientiso nthe order of R = 0.954 and R = 0.680, respectively.The correlation coefficient is very high, indicatingthat the conformation in the two different preparationsi s surprisinglysimilar.F)2D 13 C, 15 NTEDOR MAS solid-state NMR spectra for Ab40 fibrils (top) and K3-L3-K3-GI-induced Ab40 aggregates(bottom). Only the spectral region containing amino side chain nitrogen chemical shifts is shown. Forb oth samples, along-range correlation peak between Ne of Lys-28 and the carboxylic carbon atom of Asp-23 is observed,i ndicatingthat asalt bridge is formed in the K3-L3-K3-GI-induced Ab40 aggregates. The relative intensity of the long-range NH 3 -COO À cross peak appears to be larger in the K3-L3-K3-GI/Ab40 sample, indicating that this structure is presumably more compact.

Angewandte Chemie
Research Articles 5778 www.angewandte.org Ab40 amyloid fibrils,weconclude that both complexes adopt as imilar bsheet/turn/b-sheet molecular architecture.H owever,amore detailed structural analysis is necessary to characterize the exact structural features of ISM-induced Ab aggregates.

Discussion
We have found that the ISM inhibitor R3-GI can adopt a b-like fold in solution. At the same time,w e observed supramolecular,granule-like structures by DIC and fluorescence microscopy,which suggests that R3-GI may undergo LLPS.S TD NMR and FRAP experiments implied that the peptide exchanges between am onomeric and ah igh-molecular-weight soluble aggregated state.M AS solidstate NMR experiments showed that the b-arch-like architecture of an Ab40 amyloid fibril with the characteristic salt bridge between Asp-23 and Lys-28 is preserved in the solid state in the amyloid-inhibitor complex. R3-GI peptides thus form highly dynamic assemblies that provide asuitable surface for sequestration of Ab40.
Conventional inhibitors,f ollowing the classical key-lock or induced-fit principle,t arget specific structural motifs,f or example,adeep hydrophobic binding pocket. [20] To be potent, these inhibitors have to be very specific with ah igh binding affinity.W hent he binding specificity is reduced, ligands can exploit multivalent interactions to yield increased avidity. Bacterial toxin inhibitors,f or example,a re based on multivalent scaffolds. [21] Similarly,m ost protein-carbohydrate interactions are multivalent to compensate for their low affinities.M ultivalency has been employed to trigger signal transduction by inducing receptor clustering, [22] and to design amyloid Ab inhibitors [23] where small-molecule inhibitors have been covalently coupled to chaperones to increase their steric bulk. We suggest that R3-GI and related ISMs exploit multivalencybyself-association. Thecorrect spatial arrangement of side chains allows them to efficiently capture Ab40 and direct it into off-pathway non-toxic aggregates. [24] Figure 7s chematically illustrates different amyloid selfassembly inhibition mechanisms involving classical single-site binding,t ogether with intervention strategies that employ colloid formation or the here suggested ISM-induced LLPSlike process.S ingle-site binding events are the classical paradigm for ad rug-enzyme complex. Because of the low affinity and the lack of deep binding pockets,t his class of inhibitors is not very effective for amyloids (Figure 7, top). Nevertheless,ithas been shown that multi-ligand interactions of tramisprosate with monomeric Ab42 can prevent amyloid oligomer formation. [25] Hydrophobic small molecules can form colloids that resemble protein liquid droplets in size. [7] These small-molecule colloids are,h owever, unspecific inhibitors,a st hey interact promiscuously with hydrophobic regions of aprotein and eventually induce protein unfolding. We have shown previously that NSAIDs (non-steroidal antiinflammatory drugs) such as sulindac sulfide can bind into hydrophobic cavities of amyloid fibrils,a nd stabilize aggregates. [8,26] TheNSAID accelerates Ab peptide aggregation by recruiting Ab peptides into its colloidal core (Figure 7, center).
Our data suggest that ISMs may act via as imilar mechanism:I SMs are active at very low concentrations and make use of multivalent interactions,which may dramatically increase avidity (Figure 7, bottom). Multivalent interactions between an inhibitor molecule and an amyloid fibril have been exploited by Wall and co-workers, [27] who designed an ahelical peptide with al ysine basic side chain to interact with negatively charged residues exposed on the surface of amyloid structures.T hese inhibitors are used as imaging reagents and stabilize af ibrillar fold. In contrast, ISMs prevent amyloid formation and redirect Ab40 into an offpathway aggregate that lacks cellular toxicity.I mportantly, R3-GI self-assembly is likely required but not sufficient for its amyloid inhibitor function. In fact, macroscopic particles are observed in solution for both the non-inhibitor G3-GI and the inhibitor R3-GI in DIC experiments.H owever,o nly R3-GI yields long-range contacts in NOESY experiments,s upporting the idea that the specific loop-containing structure of R3-GI identified here is essential for its potent inhibitory function. [1b] Notably,A b40 in solution exhibits ac onformational preference for b-arch-like structures as well. [28] Thus, R3-GI and related ISMs may exert their inhibitory function Figure 7. Differentm echanistic pathways of ligand binding to amyloidogenic proteins. Top: mono-valent ligands (e.g.,small molecules) bind to aspecific site in ap articular peptide strand. Middle:incolloids, the local small-molecule concentration is increased, facilitating ligand binding. Amyloid peptides are recruited into the colloid. Bottom:the IAPP-derived ISM inhibitor R3-GI is suggested to self-assemble and recruit Ab40 into non-toxic aggregates. Circles and squares represent hydrophilic and hydrophobic side chains, respectively. The inner core of the ISM is shown in light blue to indicate its hydrophobic environment.
by providing as tructural template to which specific,a myloidogenic Ab40 conformers can adhere.

Conclusion
Theh ighly potent inhibitor function of the ISMs renders them well suited templates for the development of antiamyloid drugs. [1b, 24b] Our findings provide am olecular basis for understanding their function and should thus assist in the design of novel potent anti-amyloid drugs.I na ddition, they may contribute to elucidating the mechanism of previously reported self-assembling peptide inhibitors designed to mimic surfaces involved in self-or cross-amyloid interactions,f or which the mode of action is thus far not understood. [29]