Structure-Based Design of Non-Natural Amino Acid Inhibitors of Amyloid Fibrillation

Many globular and natively disordered proteins can convert into amyloid fibers. These fibers are associated with numerous pathologies1 as well as with normal cellular functions2,3, and frequently form during protein denaturation4,5. Inhibitors of pathological amyloid fibers could serve as leads for therapeutics, provided the inhibitors were specific enough to avoid interfering with normal processes. Here we show that computer-aided, structure-based design can yield highly specific peptide inhibitors of amyloid formation. Using known atomic structures of segments of amyloid fibers as templates, we have designed and characterized an all D-amino acid inhibitor of fibrillation of the tau protein found in Alzheimer’s disease, and a non-natural L-amino acid inhibitor of an amyloid fiber that enhances sexual transmission of HIV. Our results indicate that peptides from structure-based designs can disrupt the fibrillation of full-length proteins, including those like tau that lack fully ordered native structures. Correspondence to: David Eisenberg, Box 951570, UCLA, Los Angeles CA 90095-1570, 310—825-3754, [email protected], Fax: 310-206-3914. *These authors contributed equally to this work. Supplementary Information is linked to the online version of the paper at www.nature.com/nature. Author Contributions S.A.S., J.K., D.B., J.M. and D.E. designed the project. J.K. and S.A.S. created design protocol. J.K. designed D-peptides. L.J. expanded design methodology and designed non-natural amino acid peptides. S.A.S, H.W.C., and A.Z. performed fluorescence experiments and electron microscopy. A.Z. determined the structure of GGVLVN. O.Z. performed HIV infectivity experiments. J.T.S. performed kinetic data analysis. S.A.S performed NMR experiments. S.A.S, J.K., and D.E. wrote the manuscript coordinating contributions by other authors. Atomic coordinates and structure factors for the reported GGVLVN structure have been deposited in the Protein Data Bank with accession code 3PPD. Reprints and permissions information is available at www.nature.com/reprints. The authors declare no competing financial interests. NIH Public Access Author Manuscript Nature. Author manuscript; available in PMC 2014 June 27. Published in final edited form as: Nature. ; 475(7354): 96–100. doi:10.1038/nature10154. N IH -P A A uhor M anscript N IH -P A A uhor M anscript N IH -P A A uhor M anscript The finding that dozens of devastating pathologies, including Alzheimer’s disease, are associated with amyloid fibers has stimulated research on fiber inhibition. One approach employs the self-associating property of proteins that form fibers to poison fibrillation with short peptide segments6–11. A second approach is based on screening for molecules that can disrupt fiber formation12,13. Here we take a third approach to fiber inhibition: structurebased design of non-natural peptides targeted to block the ends of fibers. With advanced sampling techniques and minimizing an appropriate energy function, we computationally identify novel candidate inhibitors from a vast peptide space that interact favorably with our template structure. This approach has become possible following the determination of several dozen fiber-like atomic structures of segments from amyloid-forming proteins14–16. These structures reveal a common motif termed a steric zipper, in which a pair of β-sheets is held together by the interdigitation of their side-chains14. Using the steric-zipper structures formed by segments of two pathological proteins as templates, here we design inhibitors that cap fiber ends. As we show, the inhibitors greatly slow fibrillation of their parent proteins, offering a route to designed chemical interventions, and also supporting the hypothesis that steric zippers are the principal structural element of these fibers. One of the two fiber-like steric zippers that we have chosen as a target for inhibitor design is the hexapeptide 306VQIVYK311 from tau, a protein that forms intracellular amyloid fibers in Alzheimer’s disease17. This segment has been shown to be important for fibrillation of the full-length protein and itself forms fibers with biophysical properties similar to full-length tau fibers15,18,19. Our second template for inhibitor design, identified by the 3D Profile algorithm20,21, is the steric-zipper structure of the peptide segment GGVLVN from the amyloid fiber formed by 248PAP286, a proteolytic fragment of prostatic acid phosphatase (PAP), a protein abundant in semen. 248PAP286 fibers (also termed SEVI, or Semen derived Enhancer of Virus Infection) enhance HIV infection by orders of magnitude in cell culture studies, while the monomeric peptide is inactive22. Our computational approach to designing non-natural peptides that inhibit fibrillation is summarized in Fig. 1 for the VQIVYK segment of tau; the same general strategy is used for the GGVLVN segment of 248PAP286. In both systems, we design a tight interface between the inhibiting peptide and the end of the steric zipper to block additional segments from joining the fiber. By sampling Lor Damino acids, or commercially available non-natural amino acids, we can design candidate inhibitors with side chains that maximize hydrogen bonding and apolar interactions across the interface. We hypothesize that the steric-zipper structures of the VQIVYK and GGVLVN segments represent the spines of the fibers formed by their parent proteins. Supporting our hypothesis are our results that D-amino acid inhibitors designed on the VQIVYK steric zipper template inhibit fiber formation not only of the VQIVYK segment, but also of two tau constructs, K12 and K1923,24 (Fig. 2a). Similarly, the peptide composed of non-natural amino acids designed on the GGVLVN template inhibits the fibrillation of 248PAP286 and greatly inhibits the HIV infectivity of human cells in culture. Sievers et al. Page 2 Nature. Author manuscript; available in PMC 2014 June 27. N IH -P A A uhor M anscript N IH -P A A uhor M anscript N IH -P A A uhor M anscript To design a D-amino acid hexapeptide sequence that interacts favorably with the VQIVYK steric zipper15, and also prevents further addition of tau molecules to the fiber, we used the RosettaDesign software25. This led to the identification of four D-amino acid peptides: DTLKIVW, D-TWKLVL, D-DYYFEF, and D-YVIIER, in which Dsignifies that all αcarbon atoms are in the D-configuration (Fig. 2b,c, Supplementary Figs. 1, 2 and Supplementary Table 1). In the D-TLKIVW design model (Fig. 2b, c and Supplementary Fig. 1), the inhibitor packs tightly across the top of the VQIYVK steric-zipper structure, maintaining all main chain hydrogen bonds. The side-chain hydrogen bonding between layers of stacked Gln307 residues is replaced in the designed interface by an interaction with D-Lys3. Several hydrophobic interactions between D-TLKIVW and the two VQIVYK βstrands contribute to the favorable binding energy (Supplementary Table 1). In the design, the D-peptide blocks the addition of another layer of VQIVYK, both above the D-peptide and across on the mating β-sheet (Supplementary Fig. 3). D-Leu2 of the designed inhibitor prevents the addition of a VQIVYK molecule above it through a steric clash with Ile308 of VQIVYK and on the mating sheet through a clash with Val306 and Ile308 (Supplementary Fig. 3). These steric clashes involving D-Leu2 are intended to block fiber growth. We used fluorescence spectroscopy and electron microscopy to assess whether the designed D-peptides inhibit the fibrillation of the tau segment VQIVYK, and the K12 and K19 tau constructs. Among our designed inhibitors, D-TLKIVW is the most effective (Supplementary Fig. 4). Electron microscopy verified that incubation with D-TLKIVW prevents K19 fibrillation, which would otherwise have occurred within the elapsed time (Fig. 1 upper right). D-TLKIVW delays fiber formation even at sub-equimolar concentration relative to VQIVYK, K12, and K19 (Supplementary Fig. 5). Five-fold molar excess of DTLKIVW delays K12 fibrillation for more than two weeks in some experimental replicates (Supplementary Fig. 5c, d). In ten-fold molar excess, D-TLKIVW prevents the fibrillation of K12 in the presence of preformed K12 fiber seeds, suggesting that the peptide interacts with fibers (Fig. 2d). Also, kinetic analysis shows that the fiber elongation rate decreases in the presence of increasing concentrations of inhibitor peptide (Supplementary Fig. 6). The large increase in lag time in un-seeded reactions may be due to interactions with small aggregates on pathway to fiber formation. To investigate the specificity of the designed inhibitor, we tested scrambled sequence variants of D-TLKIVW that have poor calculated energy scores and unfavorable packing (Supplementary Table 1). The scrambled peptides D-TIKWVL, D-TIWKVL, and DLKTWIV exhibit little inhibitory effect at an equimolar ratio with VQIVYK, K12, and K19 (Fig. 2e and Supplementary Fig 7), showing that the inhibition is sequence-specific. Also, the diastereomer, L-TLKIVW, is less effective than D-TLKIVW (Supplementary Fig 8). As a further test of the specificity of our design, we confirmed that D-TLKIVW is unable to block the fibrillation of amyloid beta, also associated with Alzheimer’s disease. This suggests that the D-peptide inhibitor is not general for amyloid systems, but is specific for the VQIVYK interface in tau (Supplementary Fig. 9). Such specificity for designed inhibitors is essential if they are not to interfere with proteins that natively function in an amyloid state3. Sievers et al. Page 3 Nature. Author manuscript; available in PMC 2014 June 27. N IH -P A A uhor M anscript N IH -P A A uhor M anscript N IH -P A A uhor M anscript To confirm that the designed D-peptide inhibits in accord with the design model (Fig. 2b,c and Supplementary Fig. 1), we performed several additional tests. First, we visualized the position of the inhibitor D-TLKIVW relative to fibers of the tau construct K19 using electron microscopy. We covalently linked maleimido-nanogold particles to both the inhibitor and, separately, to a scrambled hexapeptide, D-LKTWIV. We used a blind counting assay and found that D-TLKIVW shows a significant binding preference for the end of fibers compared to nanogold alone, in contrast to the scrambled control peptide, DLKTWIV (Fig. 3a and Supplementary Fig. 10). As a further test of the model, we used nuclear magnetic resonance (NMR) to characterize the binding affinity of D-TLKIVW for tau fibers. The 1H NMR spectra for D-TLKIVW were collected in the presence of increasing concentrations of VQIVYK or K19 fibers. Because neither K19 nor VQIVYK contains tryptophan, we were able to monitor the 1H resonance of the indole proton of the tryptophan in our inhibitor. When bound to a fiber, the inhibitor, D-TLKIVW, is removed from the soluble phase and the 1H resonance is diminished (Fig. 3b and Supplemental Fig. 11)26. As a control, both D-TLKIVW and the non-inhibiting peptide D-LKTWIV were included in the same binding reaction mixture. As shown in Fig. 3b, the D-TLKIVW indole resonance is reduced greatly, whereas the DLKTWIV indole resonance is only slightly affected. Spectra of the two peptides are shown in Supplementary Fig. 12. By monitoring the D-TLKIVW indole resonance over a range of VQIVYK fiber concentrations, we estimate the apparent Kd value of the interaction between D-TLKIVW and VQIVYK fibers to be ~2 μM (Supplementary Fig. 11a and Methods). This value corresponds to a standard free binding energy of ~7.4 kcal/mol, with ~2.5 kcal/mol from apolar interactions, and ~4.9 kcal/mol from six hydrogen bonds (see Methods). Repeating the NMR binding experiment with K19 fibers yields a similar trend (Supplementary Fig 11b). To determine whether D-TLKIVW has affinity for soluble VQIVYK, we measured 1H NMR spectra of D-TLKIVW and D-LKTWIV in the presence of increasing amounts of soluble VQIVYK. Only a slight change in the chemical shifts of the indole proton peaks of D-TLKIVW and D-LKTWIV is observed, even at 70-fold molar excess of VQIVYK (Supplementary Fig. 13). This, together with the ability of the peptide to prevent seeded fibrillation, suggests that D-TLKIVW does not interact with monomers, but rather with a structured, fiber-like species. For another test of our design model, we replaced the D-Leu residue with D-Ala in DTLKIVW. Our structural model suggests that D-Leu2 of D-TLKIVW is important for preventing tau fibrillation because of its favorable interaction with the Ile residue of the VQIVYK molecule below and with Ile and the first Val of VQIVYK across the steric zipper (Fig. 2b,c and Supplementary Fig. 1). The D-Ala replacement eliminates these interactions and, further, removes a steric clash that would occur were another VQIVYK molecule placed across from the inhibitor (Supplementary Fig. 3 and Supplementary Table 1). When the D-Ala variant is incubated with VQIVYK and the tau constructs, it shows no inhibitory effect on fibrillation (Fig. 2f and Supplementary Fig. 14). This confirms that D-Leu2 is critical for the efficacy of D-TLKIVW, consistent with our model. In summary, while our electron microscopy, NMR, and D-Ala replacement results support a model in which the designed D-TLKIVW peptide binds to the ends of tau fibers, they do not Sievers et al. Page 4 Nature. Author manuscript; available in PMC 2014 June 27. N IH -P A A uhor M anscript N IH -P A A uhor M anscript N IH -P A A uhor M anscript constitute proof that the inhibitor binds exactly as anticipated in the design (Supplementary Fig. 15). To expand on our design methodology, we computationally designed an inhibitor of 248PAP286 fibrillation containing non-natural L-amino acids (Fig. 4b, Supplementary Fig. 16) using the GGVLVN structure as a template (Fig. 4a and Supplementary Table 2). This peptide, W-H-K-chAla-W-hydroxyTic (WW61), contains an alanine derivative, βcyclohexyl-L-alanine (chAla) and a tyrosine/proline derivative, 7-hydroxy-(S)-1,2,3,4tetrahydroisoquinoline-3-carboxylic acid (hydroxyTic), both of which increase contact area with the GGVLVN template. The non-natural chAla forms hydrophobic interactions with the leucine residue in the steric zipper interface, and hydroxyTic supports the favorable placement of chAla through hydrophobic packing (Fig. 4b and Supplementary Fig. 16b). Moreover, we hypothesize that the bulky side-chains and steric constraints of hydroxyTic provide hindrance to further fibril growth. This designed peptide, WW61, effectively delays both seeded and unseeded fibrillation of 248PAP286 in vitro (Fig. 4c, Supplementary Figs. 17,18). In the presence of twofold molar excess of this inhibitor, seeded fibrillation is efficiently blocked for more than two days (Fig. 4c). Further, we see that increasing the concentration of this inhibitor extends the fibrillation lag time (Fig. 19). These inhibition assay results were further confirmed by electron microscopy (Supplementary Fig 20). As a control for specificity, we tested the effect of GIHKQK, from the N-terminus of 248PAP286, and PYKLWN, a peptide with the same charge as WW61. Neither peptide affected fibrillation kinetics, indicating that the inhibitory activity of the designed peptide is sequence specific (Supplementary Fig. 21). Because 248PAP286 fibers (SEVI) have been shown to enhance HIV infection22, we tested whether WW61 is able to prevent this enhancement using a functional assay. In this experiment, we treated HIV particles with 248PAP286 solutions that had been agitated for 20 hours (to allow fiber formation) in the presence or absence of WW61, and infected TZM-bl indicator cells. As previously observed, SEVI efficiently enhanced HIV infection22. However, 248PAP286 incubated with the designed inhibitor prevented HIV infection (Fig. 4d). We performed several control experiments to verify that the lack of infectivity observed in the assay is indeed due to the inhibition of SEVI formation. First, we confirmed that in the absence of SEVI, the designed inhibitor WW61 does not affect HIV infectivity (Supplementary Fig. 22a). We also found that the control peptides GIHKQK and PYKLWN, which do not inhibit 248PAP286 fibrillation, fail to decrease HIV infectivity (Supplementary Fig 22b). Additionally, we observed that WW61 has no inhibitory effect on poly-lysine mediated HIV infectivity27, further ruling out a non-specific electrostatic interaction mechanism (Supplementary Fig. 22a). Together, these results demonstrate that a peptide capable of preventing 248PAP286 fibrillation also inhibits the generation of virus-enhancing material. Structure-based design of inhibitors of amyloid fibrillation has been challenging in the absence of detailed information about the atomic-level interactions that form the fiber spine. Sievers et al. Page 5 Nature. Author manuscript; available in PMC 2014 June 27. N IH -P A A uhor M anscript N IH -P A A uhor M anscript N IH -P A A uhor M anscript To date, one of the most successful structure-based approaches to preventing fibrillation has been to stabilize the native tetrameric structure of transthyretin28. While that approach is well-suited to prevent fibrillation of proteins with known native structures, other proteins involved in amyloid-related diseases, such as tau, amyloid beta, and 248PAP286 lack fully ordered native structures29. Our structure-based approach designs inhibitors independent of native structure. Instead, the templates are atomic-level structures of short fiber-forming segments14,15. Using these fiber-like templates, and adopting computational methods successful in designing novel proteins and protein-protein interfaces25,30, we have created specific inhibitors of proteins that normally fibrillate. These results support the hypothesis that the steric zipper is a principal feature of tau-related and SEVI fibers, and suggest that with current computational methods and steric-zipper structures, we have the tools to design specific inhibitors to prevent the formation of other amyloid fibers.