Cyclic Modular β-Sheets

Abstract

The development of peptide β-hairpins is problematic, because folding depends on the amino acid sequence and changes to the sequence can significantly decrease folding. Robust β-hairpins that can tolerate such changes are attractive tools for studying interactions involving protein β-sheets and developing inhibitors of these interactions. This paper introduces a new class of peptide models of protein β-sheets that addresses the problem of separating folding from sequence. These model β-sheets are macrocyclic peptides that fold in water to present a pentapeptide β-strand along one edge; the other edge contains the tripeptide β-strand mimic Hao [JACS 2000, 122, 7654] and two additional amino acids. The pentapeptide and Hao-containing peptide strands are connected by two δ-linked ornithine (^δOrn) turns [JACS 2003, 125, 876]. Each ^δOrn turn contains a free α-amino group that permits the linking of individual modules to form divalent β-sheets. These “cyclic modular β-sheets” are synthesized by standard solid-phase peptide synthesis of a linear precursor followed by solution-phase cyclization. Eight cyclic modular β-sheets 1a–1h containing sequences based on β-amyloid and macrophage inflammatory protein 2 were synthesized and characterized by ¹H NMR. Linked cyclic modular β-sheet 2, which contains two modules of 1b, was also synthesized and characterized. ¹H NMR studies show downfield α-proton chemical shifts, ^δOrn δ-proton magnetic anisotropy, and NOE crosspeaks that establish all compounds but 1c and 1g to be moderately or well folded into a conformation that resembles a β-sheet. Pulsed-field gradient NMR diffusion experiments show little or no self-association at low (≤ 2 mM) concentrations. Changes to the residues in the Hao-containing strands of 1c and 1g improve folding and show that folding of the structures can be enhanced without altering the sequence of the pentapeptide strand. Well-folded cyclic modular β-sheets 1a, 1b, and 1f each have a phenylalanine directly across from Hao, suggesting that cyclic modular β-sheets containing aromatic residues across from Hao are better folded.

Introduction

Synthetic compounds that imitate the hydrogen-bonding edges and side-chain faces of protein β-sheets are valuable tools for studying and regulating protein–protein and protein–DNA interactions.¹ Peptides that fold into β-hairpins are especially suited for such studies, because these β-sheet models permit the use of actual amino acid sequences from protein β-sheets. Several reports of β-hairpins that bind DNA or proteins have been published in recent years.² Waters and coworkers developed a divalent β-hairpin that binds single-stranded DNA with an affinity similar to that of proteins.^2c Wells and coworkers and Robinson and coworkers developed β-hairpins that bind the Fc domain of human immunoglobulin G and inhibit the binding of immunoglobulin G to protein A.^2a,d Robinson and coworkers also developed a β-hairpin inhibitor of the p53–HDM2 protein–protein interaction.^2b

The development of peptide β-hairpins has historically been difficult, but groundbreaking work in the 1990s led to reports of peptide β-hairpins that were either isolated from protein sequences or designed de novo.^3,4 Subsequent studies have revealed the importance of cross-strand hydrophobic and aromatic–aromatic side chain interactions and of good β-turn sequences for the folding of β-hairpins.^5,6 In spite of this progress, the development of β-hairpins remains problematic, because β-hairpin folding is intimately linked to the amino acid sequence of the peptide strands and because well-folded structures often form intractable aggregates.^5b Only certain sequences give folded structures, and changes in the sequence of a folded β-hairpin can disrupt folding by disrupting key side-chain–side-chain interactions. In studies with well-folded β-hairpins, Gellman and coworkers carried out carefully selected mutations to show the dependence of β-hairpin folding on interstrand hydrophobic side-chain–side-chain interactions.^6b,c Cochran and coworkers^6a and Serrano and coworkers^4c have reported similar sequence dependences. To probe the effects of mutations within a β-hairpin on intermolecular interactions, it is important to be able to change the sequence of the β-strands while maintaining a folded structure and avoiding aggregation. Robust models of protein β-sheets, which tolerate a variety of sequences, are needed.

Robust models of protein β-sheets can be created by combining peptides with peptidomimetic templates that nucleate and reinforce a β-sheet structure.^5a One class of templates is the turn mimics, which nucleate parallel or antiparallel β-sheet structures in attached peptide chains. Examples of turn mimics include structures such as the dibenzofuran-based amino acid developed by Kelly and coworkers,⁷ the β-peptide reverse-turn mimic developed by Gellman and coworkers,⁸ and the oligourea “molecular scaffolds” developed by Nowick and coworkers.⁹

Another class of templates is the β-strand mimics, which duplicate the pattern of hydrogen-bond donors and acceptors of one edge of a peptide β-strand. These templates can serve as frameworks to nucleate and reinforce the β-strand conformation of attached polypeptide strands. A pioneering example of such a template is the 2,8-diaminoepindolidione β-strand mimic developed by Kemp and coworkers in the late 1980s.¹⁰ More recently developed examples of β-strand mimics, such as the Hao amino acid, invented by Nowick and coworkers,¹¹ and the @-unit, invented by Bartlett and coworkers,¹² are easily incorporated into peptides and nucleate the formation of well-folded β-hairpin-like structures in organic solvents. Other templates such as the hydrogen-bonded duplex developed by Gong and coworkers¹³ and the pyrrolinone-based peptidomimetics developed by Hirschmann and coworkers¹⁴ were not designed for use in β-hairpins but do mimic certain aspects of β-sheets.

Although the existing β-sheet templates have allowed the creation of many novel β-sheet structures, a robust model of protein β-sheets that folds in water is still lacking. Most of the existing templates require either organic solvents or sequence-dependent interactions with neighboring amino acids to nucleate a β-sheet structure. There is considerable room for improvement in developing β-sheet structures that fold in aqueous solution and tolerate a greater variety of different amino acid sequences.

In this paper, we address these issues with a new class of cyclic models of protein β-sheets, 1. These compounds contain two of our previously reported peptidomimetic templates and tolerate a variety of different amino acid sequences. Many of these compounds fold into β-sheets and do not aggregate or self-associate significantly at low-millimolar concentrations. These individual β-sheets are readily linked to form multivalent structures with more than one β-sheet domain.

These “cyclic modular β-sheets” are 42-membered rings that contain a pentapeptide in the “upper” strand, the amino acid Hao¹¹ and two α-amino acids in the “lower” strand, and two δ-linked ornithine β-turn mimics.¹⁵ Hao is a relatively rigid tripeptide β-strand mimic that serves as a template for the folding of the upper strand and blocks the lower strand to minimize edge-to-edge aggregation.^11,16 The two α-amino acids in the lower strand permit tuning the folding and solubility of the cyclic modular β-sheets without changing the upper strand. The two δ-linked ornithine (^δOrn) residues form hairpin turns and have free α-amino groups that serve as sites for linking the cyclic modular β-sheets.^15,17 We developed a cyclic β-sheet model, because cyclization is known to confer conformational stability to synthetic peptide β-hairpins^6e,18 and to natural peptide β-hairpins such as gramicidin S.^19,20 This paper describes the synthesis and NMR structural studies of eight cyclic modular β-sheets and one linked cyclic modular β-sheet.

Results

Design

Pentapeptide sequences based on β-amyloid (Aβ)²¹ and macrophage inflammatory protein 2 (MIP-2)²² were incorporated into the upper strands of cyclic modular β-sheets 1a–1h (Chart 1). The upper strands of 1a and 1b each present opposite edges of a hydrophobic sequence that plays a key role in the fibrilization of Aβ: 1a contains Aβ_16–20 (KLVFF) and 1b contains Aβ_17–21 (LVFFA).²³ Selection of the two residues for the lower strands of 1a and 1b was guided by observations of cross-strand β-sheet pairings in various proteins.^24,25 The initial selection of Leu–Val for positions 6 and 7 of 1b resulted in poor solubility in water; the sequence was subsequently changed to Leu–Lys to improve the solubility.

Chart 1.

The upper strands of 1c and 1d both contain a hydrophobic sequence from an alternate region of Aβ, Aβ_30–34 (AIIGL). The Leu–Lys sequence that was suitable for the lower strand of 1b was also included in the lower strand of 1c, but was replaced with Tyr–Lys in 1d to improve folding. The upper strands of 1e and 1f contain the pentapeptide sequences AIIAL and AIIFL, respectively, which are modified versions of the Aβ_30–34 sequence. Cyclic modular β-sheets 1e and 1f also contain Leu–Lys in the lower strands.

The upper strands of cyclic modular β-sheets 1g and 1h contain a relatively hydrophilic sequence from the β-sheet dimerization interface of MIP-2 (SLSVT). The lower strand of 1g contains Ala–Thr, which in MIP-2 are the cross-strand neighbors of the first two residues of the pentapeptide sequence. The Ala–Thr sequence in the lower strand of 1g was replaced with Tyr–Thr in 1h to improve folding.

We prepared linked cyclic modular β-sheet 2 to demonstrate the potential of this system to generate multivalent β-sheet structures. Linked cyclic modular β-sheet 2 contains two modules derived from 1b. The two ^δOrn β-turn mimics provide two possible linkage sites on each cyclic modular β-sheet. The two modules of linked cyclic modular β-sheet 2 are connected at the “left” ^δOrn (^δOrn1) by an N,N′-ethylenediaminediacetic acid linker (HO₂CCH₂NHCH₂CH₂NHCH₂CO₂H). An initial version of this linked cyclic modular β-sheet, which contained a 3,6-dioxaoctanedioic acid linker (HO₂CCH₂OCH₂CH₂OCH₂CO₂H), exhibited poor solubility in water. The N,N′-ethylenediaminediacetic acid linker was then chosen to solve this problem.

Synthesis.²⁶

The cyclic modular β-sheets are conveniently prepared by standard Fmoc solid-phase synthesis of the corresponding linear peptides on 2-chlorotrityl chloride resin, followed by solution-phase cyclization.²⁷ The initial step of the synthesis involves loading commercially available Boc–Orn(Fmoc)–OH onto 2-chlorotrityl chloride resin (Scheme 1). Amino acids are added to the growing peptide by removing the terminal Fmoc group with piperidine and then adding a solution of the Fmoc-protected amino acid and the coupling reagents HCTU²⁸ and N,N-diisopropylethylamine (DIEA). Most coupling reactions are completed within 30 minutes. The coupling of Fmoc*–Hao–OH^11a,29 is sluggish and is incomplete even after several hours when HCTU and DIEA are used as the coupling reagents. The reaction proceeds cleanly in 4–6 hours, however, when 2,4,6-collidine is used in place of DIEA.³⁰

Scheme 1.

After the coupling of the final amino acid, the terminal Fmoc group is removed and the linear protected peptide is cleaved from the resin by treatment with 50% acetic acid and 10% methanol in methylene chloride. Any residual acetic acid is removed by repeated rotary evaporation with hexanes, which forms an azeotrope with acetic acid. This important step eliminates the potential of capping the linear protected peptide with acetic acid instead of cyclization. Analytical HPLC typically shows that the linear protected peptide is relatively clean; the HPLC trace of the linear protected peptide intermediate in the synthesis of 1g is representative (Figure 1a).

Figure 1.

Analytical HPLC traces of (a) linear protected peptide intermediate in the synthesis of 1g, (b) unpurified cyclic modular β-sheet 1g, and (c) purified cyclic modular β-sheet 1g. Conditions: Alltech Alltima Rocket C18 column (53 × 4.5 mm), gradient of 10–90% CH₃CN in H₂O with 0.1% TFA.

The cyclization reaction occurs cleanly with HCTU and DIEA in dilute (ca. 0.5 mM) DMF solution. Since the C-terminus of the protected linear peptide contains an α-amino acid carbamate (BocNH–CHR–COOH), rather than an α-amino acid amide (RCONH–CHR–COOH), epimerization during this reaction is not expected. Treatment of the cyclic intermediate with trifluoroacetic acid (TFA) removes the acid-labile protecting groups (Figure 1b), and purification of the deprotected peptide by reverse-phase HPLC affords cyclic modular β-sheet 1 as the TFA salt (Figure 1c). A 15–25% yield, based on the loading of the resin with Boc-Orn(Fmoc)-OH, is typically obtained; a 0.1 millimole scale synthesis typically affords 20–30 milligrams of the purified cyclic modular β-sheet.

Linked cyclic modular β-sheets are prepared by connecting individual modules at one of the ^δOrn (^δOrn1 or ^δOrn2) α-amino groups with a dicarboxylic acid linker. The two ^δOrn residues in each cyclic modular β-sheet must be differentiated, and this differentiation can be accomplished with the Cbz protecting group. Commercially available Cbz–Orn(Fmoc)–OH is used instead of Boc–Orn(Fmoc)–OH at the turn position that serves as the linkage site. The Cbz protecting group is removed by catalytic hydrogenation, which leaves the Boc protecting group on the other ^δOrn α-amino group intact.

The synthesis of linked cyclic modular β-sheet 2 begins with the synthesis of fully protected cyclic peptide 3 (Scheme 2). Catalytic hydrogenation of 3 with palladium on carbon selectively removes the Cbz protecting group. The resulting Boc-protected intermediate 4 is purified by reverse-phase HPLC (Figure 2a). Coupling of 4 to the Boc-protected N,N′-ethylenediaminediacetic acid linker³¹ in a 1.0:0.4 molar ratio, with HCTU and DIEA, yields linked intermediate 5 as a relatively clean product. Treatment of 5 with TFA removes the acid-labile protecting groups (Figure 2b). Purification by reverse-phase HPLC affords linked cyclic modular β-sheet 2 as the TFA salt. The synthesis of 2 from 2-chlorotrityl chloride resin and Cbz–Orn(Fmoc)–OH proceeded in 9% overall yield; a 0.2 millimole scale synthesis afforded 24 mg of 2.

Scheme 2.

Figure 2.

Analytical HPLC traces of (a) purified intermediate 4, (b) unpurified linked cyclic modular β-sheet 2, and (c) purified linked cyclic modular β-sheet 2. Conditions: Alltech Alltima Rocket C18 column (53 × 4.5 mm), gradient of 10–90% CH₃CN in H₂O with 0.1% TFA.

¹H NMR Structural Studies

¹H NMR studies (1D, 2D TOCSY, and 2D ROESY) of cyclic modular β-sheets 1 and linked cyclic modular β-sheet 2 in aqueous solutions (D₂O and H₂O/D₂O mixtures) establish the β-sheet folding and reveal the extent of folding.³² Analysis of the α-proton chemical shifts, ^δOrn δ-proton magnetic anisotropy, and NOE crosspeaks establishes well-folded β-sheet structures for 1a, 1b, 1f, and 2, moderately folded β-sheet structures for 1d, 1e, and 1h, and poorly folded β-sheet structures for 1c and 1g.

The ¹H NMR spectra of cyclic modular β-sheets 1a–1h show peaks that are well dispersed and sharp (linewidths ca. 2–4 Hz); the ¹H NMR spectrum of linked cyclic modular β-sheet 2 shows peaks that are also well dispersed but slightly broader (linewidth ca. 5 Hz).³³ The spectrum of 1a in D₂O, shown in Figure 3, is representative of the spectra of 1b–1h. Peaks corresponding to a minor (4%) species are present in the spectrum of 1a, suggesting the occurrence of an alternate conformation. Minor peaks also occur in the spectra of 1b–1h and 2. The minor peaks are most intense in 1d and 1f (10%). ROESY experiments on 1f at 50 °C show chemical exchange crosspeaks (EXSY) that confirm that the minor peaks originate from slow exchange with an alternate conformation.³⁴

Figure 3.

600 MHz ¹H NMR spectrum of a 2 mM solution of 1a in D₂O with 50 mM CD₃CO₂D and 50 mM CD₃CO₂Na at 6 °C.³³ Peaks marked with an asterisk (*) likely correspond to an alternate conformation.

α-Proton Chemical Shifts

The chemical shifts of the α-protons of peptides in β-sheets are generally downfield of those in unstructured “random coil” environments.³⁵ Linear control peptides 6a–6g serve as references for the random coil conformations of the cyclic modular β-sheets (Chart 2).³⁶ Each of these linear control peptides contains two ^δOrn residues and one of the six different pentapeptide sequences of 1a–1h. ROESY studies of the control peptides show no long-range NOE crosspeaks, suggesting that these peptides are largely unstructured. Table 1 lists the α-proton chemical shifts of cyclic modular β-sheets 1a–1h, linked cyclic modular β-sheet 2, and control peptides 6a–6g.

Chart 2.

Table 1.

α-Proton Chemical Shifts (ppm) of the Cyclic Modular β-Sheets and Linear Controls.^a

	Position
	1	2	3	4	5	6	7
1a	4.67	4.32	4.26	4.78	4.60	4.16	4.60
1b	4.65	4.41	4.98	4.60	4.37	4.83	4.54
1c	4.42	4.14	4.09	3.94	4.32	4.43	4.25
1d	4.47	4.18	4.27	3.93	4.36	4.63	4.27
1e	4.45	4.11	4.28	4.35	4.12	4.43	4.29
1f	4.55	4.08	4.38	4.74	4.42	4.38	4.49
1g	4.53	4.38	4.53	4.26	4.26	4.58	4.26
1h	4.59	4.06	4.64	4.40	4.27	4.70	4.22
2	4.61	4.47	5.04	4.61	4.42	4.92	4.58
6a	4.33	4.35	4.06	4.59	4.40	—	—
6b	4.41	4.04	4.57	4.52	4.07	—	—
6c	4.39	4.13	4.16	3.93	4.24	—	—
6e	4.38	4.12	4.13	4.31	4.21	—	—
6f	4.38	4.06	4.14	4.62	4.21	—	—
6g	4.54	4.41	4.47	4.22	4.19	—	—

^a500 or 600 MHz data were collected on 2 mM solutions in D₂O at 6 °C. Data for 1a–1f and 6a–6f were collected in a buffer of 50 mM CD₃COOD and 50 mM CD₃COONa; data for 1g, 1h, 2, and 6g were collected in pure D₂O.³²

Comparison of the chemical shifts of the α-protons of peptides 1 and 2 to those of linear controls 6 reflects the folding of the cyclic modular β-sheets. Figure 4 illustrates the ΔδH_α values for the pentapeptide strands of 1a–1h and 2, where ΔδH_α corresponds to the difference between the chemical shifts of the α-protons of the cyclic modular β-sheets and the linear controls. Positive ΔδH_α values provide evidence that the cyclic modular β-sheets are folded. The ΔδH_α values of four or five of the α-protons in the upper strands of 1a, 1b, 1d, 1f, and 1h are positive, suggesting at least partial β-sheet folding. The ΔδH_α values are largest for 1a, 1b, and 1f, suggesting that these cyclic modular β-sheets are the best folded in the series. Linked cyclic modular β-sheet 2, which contains two modules derived from 1b, shows ΔδH_α values that are only slightly different than those of 1b, suggesting that the folding of each module of 2 is similar to that of 1b.

Figure 4.

ΔδH_α [δH_α(cyclic modular β-sheet)-δH_α(linear control)] values for the upper strands of the cyclic modular β-sheets.

The ΔδH_α values are smaller for 1d and 1h, suggesting that these cyclic modular β-sheets are not as strongly folded as 1a, 1b, and 1f. A strongly negative (−0.35 ppm) ΔδH_α value for residue 2 of 1h may result from the magnetic anisotropy of the tyrosine aromatic side chain at position 6, the cross-strand neighbor of residue 2. The ΔδH_α values of at least two β-protons of 1c, 1e, and 1g are essentially zero or slightly negative, and most of the other ΔδH_α values are only slightly positive. These data suggest that 1c, 1e, and 1g are poorly folded.

^δOrn δ-Proton Magnetic Anisotropy

The ¹H NMR spectra of 1a–1f and 1h show magnetic anisotropy between the diastereotopic pro-R δ-proton (H_δR) and pro-S δ-proton (H_δS) resonances of both ^δOrn residues (Figure 5 and Figure 6). The magnitude of the magnetic anisotropy (0.17–0.62 ppm) suggests the ^δOrn residues to be moderately or well folded and provides a measure of the folded populations. In contrast, the δ-protons of control peptides 6a–6f show ≤0.01 ppm magnetic anisotropy, and those of 6g show 0.01 and 0.05 ppm magnetic anisotropy (Figure 6). Previous and ongoing studies in our laboratories strongly suggest that roughly 0.6 ppm is the maximum magnetic anisotropy of ^δOrn δ-protons in aqueous solution when the ^δOrn has an unacylated α-amino group.³⁷ These studies involve three different systems: a cyclic β-sheet peptide related to those studied by Gellman,¹⁵ cyclic modular β-sheets 1a–1h and other cyclic modular β-sheets in this family, and a family of larger macrocyclic model β-sheets.³⁸

Figure 5.

Magnetic anisotropy between the diastereotopic δ-proton resonances of each ^δOrn residue in the cyclic modular β-sheets.

Figure 6.

Expansions of the 600 MHz ¹H NMR spectra of 2 mM solutions of 6b, 1e, and 1f in D₂O at 6 °C showing the ^δOrn δ-proton resonances.³³ Assignments of the diastereotopic δ-proton resonances of 1e are tentative and are based on analogy with 1f.

The ¹H-NMR spectra of cyclic modular β-sheets 1a, 1b, and 1f show large ^δOrn δ-proton magnetic anisotropy (0.48–0.62 ppm), demonstrating these turns to be well folded (Figure 5). The spectrum of linked cyclic modular β-sheet 2 shows 0.72 ppm δ-proton magnetic anisotropy for ^δOrn1 and 0.52 ppm δ-proton magnetic anisotropy for ^δOrn2, demonstrating that the turns in 2 are also well folded. The exceptionally large value for linked ^δOrn1 in 2 (0.72 ppm) likely arises from the magnetic anisotropy of the carbonyl group of the amide linker that is attached to ^δOrn1. The spectra of 1c, 1d, 1e, and 1h show smaller ^δOrn δ-proton magnetic anisotropy (0.17–0.35 ppm), demonstrating these turns to be moderately folded. The spectrum of 1g shows essentially no δ-proton anisotropy in ^δOrn1 (≤ 0.01 ppm), demonstrating this turn to be unfolded, and 0.13 ppm anisotropy in ^δOrn2, demonstrating this turn to be slightly folded.

The ¹H NMR spectra of the cyclic modular β-sheets with well-folded ^δOrn turns show characteristic coupling patterns of the ^δOrn δ-proton resonances that are consistent with the model that our research group has previously proposed for the ^δOrn β-turn mimic (Figure 7).¹⁵ The ¹H NMR spectrum of cyclic modular β-sheet 1f clearly illustrates these coupling patterns (Figure 6). Each ^δOrn of 1f gives two diastereotopic δ-proton resonances that are separated by ca. 0.5 ppm. The downfield resonance (H_δS) appears as a broad triplet, with a coupling constant of ca. 13 Hz, while the upfield resonance (H_δR) appears as a broad doublet with a coupling constant of 13 or 14 Hz. The triplet of H_δS results from two strong couplings—geminal coupling with H_δR and vicinal coupling with H_γR, which is in an anti orientation. Coupling with H_γS, which is in a gauche orientation, is sufficiently weak that it only contributes to broadening of the triplet. The doublet of H_δR results from one strong coupling—geminal coupling with the H_δS. Coupling with the γ-protons, which are both gauche, is sufficiently weak that it only contributes to broadening or slight further splitting of the doublet.

Figure 7.

Model of a ^δOrn β-turn mimic (Ac–^δOrn–NHMe, global minimum: MacroModel v7.0; AMBER* force field; GB/SA H₂O solvent model).¹⁵ An arrow indicates the NOE between the H_δS and H_α.

In contrast to the differentiated ^δOrn δ-proton coupling patterns of the well-folded ^δOrn turns, moderately and poorly folded ^δOrn turns give coupling patterns that suggest an ensemble of different conformations. The spectrum of 1e illustrates these coupling patterns and shows a doublet of triplets for each of the diastereotopic H_δ resonances of each ^δOrn residue (Figure 6). These coupling patterns show each δ-proton to have a strong (ca. 13 Hz) coupling with the geminal H_δ and a weaker (ca. 6 Hz) coupling with each vicinal H_γ.

NOE Crosspeaks

NOEs between peptide strands are a hallmark of β-sheets and reflect proximity between residues that are otherwise remote in the unfolded peptide. The ROESY spectra of 1a at 500 and 600 MHz show NOE crosspeaks that demonstrate this cyclic modular β-sheet to be folded into a β-sheet-like conformation (Figure 8). Most importantly, these spectra show strong long-range NOEs between the α-protons of the leucine at position 2 and the valine at position 6 (2_α–6_α) and between the α-proton of the phenylalanine at position 4 and aromatic proton 6 of Hao (4_α–Hao₆). These NOEs are associated with the characteristic short interstrand H_α–H_α contacts in the non-hydrogen-bonded pairs within β-sheets and are illustrated in Figure 9. The ROESY spectra of 1a also show strong NOEs between the α-proton of each ^δOrn and the corresponding pro-S δ-proton (^δOrn_α–δS), demonstrating the folding of the ^δOrn turns (Figure 8). These NOEs agree with the model of a ^δOrn β-turn mimic (Figure 7), in which only the H_δS is near the H_α, and are also illustrated in Figure 9.

Figure 8.

Selected expansions of the ROESY spectra of 1a: (a) 2 mM in D₂O with 50 mM CD₃CO₂D and 50 mM CD₃CO₂Na at 500 MHz and 6 °C with a 200-ms spin lock time. (b) 2 mM in 9:1 H₂O/D₂O with 50 mM CD₃CO₂D and 50 mM CD₃CO₂Na at 600 MHz and 6 °C with a 200-ms spin lock time and gradient water suppression.

Figure 9.

Key NOEs shown in the ROESY spectra of cyclic modular β-sheets 1a, 1b, 1d, 1e, 1f, and 1h in D₂O.

The 600 MHz ROESY spectra of 1b, 1f, and 1h also show strong 2_α–6_α, 4_α–Hao₆, and ^δOrn_α–δS NOEs, suggesting these cyclic modular β-sheets to also be folded into β-sheet-like conformations. The spectrum of linked compound 2 also shows these NOEs, suggesting both modules to be folded. The spectra of 1d and 1e show strong 2_α – 6_α and ^δOrn_α–δS NOEs but relatively weak 4_α – Hao₆ NOEs, suggesting that these cyclic modular β-sheets are not as strongly folded as 1a, 1b, 1f, and 1h. The spectrum of 1c does not show a 4_α–Hao₆ NOE and only shows weak 2_α–6_α and ^δOrn_α–δS NOEs. The spectrum of 1g does not show 2_α – 6_α or ^δOrn1_α–δS NOEs and only shows weak 4_α–Hao₆ and ^δOrn2_α–δS NOEs. The absence of key NOEs in each of these compounds suggests that 1c and 1g are poorly folded. Table 2 summarizes these key NOEs for cyclic modular β-sheets 1a–1h and 2.

Table 2.

^δOrn Turn and Interstrand NOEs in the Cyclic Modular β-Sheets.^a

	2α-6α	4α-Hao6	δOrn1α-δS	δOrn1α-δRb	δOrn2α-δS	δOrn2α-δRb
1a	strong	strong	strong	—	strong	—
1b	strong	strong	strong	—	strong	—
1c	weak	—	weak	weak	weak	weak
1d	strong	weak	strong	—	strong	—
1e	strong	weak	strong	weak	strong	weak
1f	strong	strong	strong	weak	strong	weak
1g	—	weak	—	—	weak	—
1h	strong	strong	strong	—	strong	—
2	strong	strong	strong	—	strong	—

^aA dash (—) indicates that no NOE crosspeak was detected.

^bNOEs suggesting alternate turn conformations.

800 MHz NOESY spectra of 1a and 1b show interstrand NH–NH and H_α–NH NOEs that further demonstrate a β-sheet structure for these compounds. Figure 10 summarizes these data. These NOEs were largely not seen in the 500 or 600 MHz ROESY spectra of 1a, 1b, or any of the other cyclic modular β-sheets. The weakness or absence of these NOE crosspeaks may reflect the lower sensitivity of the 500 or 600 MHz ROESY experiments and the longer distances associated with NH–NH and H_α–NH NOEs in β-sheets (ca. 3.2 or 3.3 Å vs. ca. 2.3 Å for interstrand H_α–H_α³⁹).

Figure 10.

Interstrand NH–NH and H_α–NH NOEs in the 800 MHz NOESY spectra of 1a and/or 1b in 9:1 H₂O/D₂O. The hydrazide NH groups do not show up under these conditions and do not give NOEs.

Several of the cyclic modular β-sheets show NOEs that suggest a rapid equilibrium with alternate conformations. For example, a weak 4_α–Hao₄ NOE occurs in the 600 MHz ROESY spectrum of 1a, suggesting an alternate rotamer involving Hao (Figure 8b). Weak ^δOrn_α–δR crosspeaks occur in the 600 MHz ROESY spectra of 1c, 1e, and 1f, suggesting alternate turn conformations (Table 2). A few weak NOE crosspeaks occur in the 800 MHz NOESY spectrum of 1a, also suggesting alternate conformations.

While the ^δOrn turn and interstrand NOE crosspeaks demonstrate folding of the cyclic modular β-sheets, intrastrand NOE crosspeaks are consistent with extended (β-strand) conformations of the peptide strands. The 600 MHz ROESY spectrum of 1a in 9:1 H₂O/D₂O shows strong interresidue H_α–NH(i,i + 1) NOE crosspeaks and weak intraresidue H_α–NH(i,i) NOE crosspeaks (Figure 8b). It should be noted that this pattern of strong interresidue NOEs and weak intraresidue NOEs does not rigorously distinguish β-sheet structures from non-β-sheet structures. This pattern occurs in all of the cyclic modular β-sheets and in control peptides 6.

PFG NMR Diffusion Studies

Pulsed-field gradient (PFG) NMR diffusion studies provide insight into the molecular weights and association states of molecules in the same sample and under the same conditions as NMR structural studies.⁴⁰ PFG NMR diffusion studies are based on the attenuation of the NMR signal that occurs when opposing z-axis magnetic field gradient pulses are applied with an intervening delay. This attenuation results from molecular diffusion and permits the calculation of the diffusion coefficient (D). PFG NMR diffusion studies of cyclic modular β-sheets 1a–1h show little or no self-association at the 2 mM concentration used for structural studies.⁴¹

The diffusion coefficients of cyclic modular β-sheets 1a–1h and 2 were measured at 2.0 mM (Table 3), and that of 1a was also measured at 0.1, 0.33, 1.0, 3.3, and 10 mM to assess the effects of concentration (Figure 11). The diffusion coefficient of 1a shows little or no decrease from 0.1 mM to 1.0 mM (from 11.9 (±1.1) × 10⁻⁷ cm²/s to 11.2 (± 0.4) × 10⁻⁷ cm²/s). It decreases slightly at 2.0 mM (10.7 (± 0.3) × 10⁻⁷ cm²/s), more at 3.3 mM (10.0 (± 0.1) × 10⁻⁷ cm²/s), and substantially at 10 mM (7.7 (± 0.2) × 10⁻⁷ cm²/s). These changes suggest significant self-association at 10 mM, slight self-association at 2.0 mM, and no significant self-association at lower concentrations. Consistent with these trends, the ¹H NMR peak widths of 1a do not increase significantly from 0.1 mM to 2.0 mM but increase significantly from 2.0 mM to 10 mM. The ¹H NMR chemical shifts do not change significantly from 0.1 mM to 10 mM, suggesting that folding does not depend on self-association.

Table 3.

Diffusion Coefficients of the Cyclic Modular β-Sheets.^a

	MW	D (cm2/s)
gramicidin S	1140	12.9 (± 0.5) × 10−7
1a	1327	10.7 (± 0.3) × 10−7
1b	1283	11.9 (± 0.7) × 10−7
1c	1172	12.9 (± 0.8) × 10−7
1d	1222	12.0 (± 0.2) × 10−7
1e	1186	12.2 (± 0.3) × 10−7
1f	1263	11.4 (± 0.3) × 10−7
1g	1123	12.8 (± 0.3) × 10−7
1h	1214	12.4 (± 0.5) × 10−7
2	2705	7.2 (± 0.1) × 10−7

^aThe diffusion coefficients were measured by 800 MHz PFG NMR diffusion experiments at 6 °C in D₂O.⁴¹ The concentrations of 1a–1h and 2 were 2 mM; the concentration of gramicidin S was 1.1 mM. Data for 1a–1f were collected in a buffer of 50 mM CD₃COOD and 50 mM CD₃COONa; data for gramicidin S, 1g, 1h, and 2 were collected in pure D₂O. Each value represents an average measured for multiple resonances. Uncertainties (± values) are the standard deviations of these measurements.

Figure 11.

Diffusion coefficient of 1a as a function of concentration. The diffusion coefficient was measured by 800 MHz PFG NMR diffusion experiments at 6 °C in D₂O in a buffer of 50 mM CD₃COOD and 50 mM CD₃COONa.⁴¹ Each value represents an average measured for three resonances. Error bars are the standard deviations of these measurements.

The diffusion coefficients of 1a–1h at 2.0 mM range from 10.7 (± 0.3) × 10⁻⁷ cm²/s to 12.9 (± 0.8) × 10⁻⁷ cm²/s and, as expected, generally increase with decreasing molecular weight (Table 3). These values are comparable to that of gramicidin S (12.9 (± 0.5) × 10⁻⁷ cm²/s), a monomeric cyclic peptide with a molecular weight close to that of the cyclic modular β-sheets, which was used as a reference. The similarities of the diffusion coefficients of 1b–1h to those of gramicidin S and 1a suggests little or no self-association of these compounds at 2 mM.

The diffusion coefficient of linked cyclic modular β-sheet 2 is 7.2 (± 0.1) × 10⁻⁷ cm²/s at 2.0 mM (Table 3). This value is substantially lower than that of the monovalent homologue 1b, as expected with the larger size of 2.⁴²

Discussion

Collectively, the ¹H NMR structural studies show that 1a, 1b, 1f, and 2 adopt well-folded β-sheet conformations, 1d, 1e, and 1h adopt moderately folded β-sheet conformations, and 1c and 1g adopt poorly folded β-sheet conformations (Table 4). The well-folded cyclic modular β-sheets show large ΔδH_α values for four or five α-protons in the pentapeptide strand, ^δOrn δ-proton magnetic anisotropy of at least 0.45 ppm, and strong key NOE crosspeaks (Figure 9). The moderately folded cyclic modular β-sheets also demonstrate a β-sheet structure, but have smaller ΔδH_α values, more moderate ^δOrn δ-proton magnetic anisotropy, and weaker NOE crosspeaks, which suggests that the population of β-sheet structure is lower. The poorly folded cyclic modular β-sheets show two or more negative or essentially zero ΔδH_α values, moderate or small ^δOrn δ-proton magnetic anisotropy, and only three or fewer of the four key NOEs in Figure 9. Although portions of the poorly folded cyclic modular β-sheets may be folded, the population of β-sheet structure is low and the folding is incomplete.

Table 4.

Folding of cyclic modular β-sheets 1.

1a	cyclo(δOrn-K-L-V-F-F-Hao-V-E-δOrn)	well folded
1b	cyclo(δOrn-L-V-F-F-A-Hao-L-K-δOrn)	well folded
1c	cyclo(δOrn-A-I-I-G-L-Hao-L-K-δOrn)	poorly folded
1d	cyclo(δOrn-A-I-I-G-L-Hao-Y-K-δOrn)	moderately folded
1e	cyclo(δOrn-A-I-I-A-L-Hao-L-K-δOrn)	moderately folded
1f	cyclo(δOrn-A-I-I-F-L-Hao-L-K-δOrn)	well folded
1g	cyclo(δOrn-S-L-S-V-T-Hao-A-T-δOrn)	poorly folded
1h	cyclo(δOrn-S-L-S-V-T-Hao-Y-T-δOrn)	moderately folded

The spectra of divalent cyclic modular β-sheet 2, which contains two modules derived from 1b, show that the folding of the modules is comparable to that of 1b. This result is significant, because it shows that acylation of a ^δOrn turn residue does not disrupt the folding of a cyclic modular β-sheet. This finding is similar to that of Waters and Cooper, who showed that modification of the side chain of an asparagine turn residue caused little change in the folding of a β-hairpin peptide.⁴³ We have now prepared several other linked cyclic modular β-sheets that are linked at either of the two ^δOrn residues. In all cases the folding of the linked compounds is comparable to that of the monovalent homologues.

Varying the two α-amino acids in the lower strand can enhance the folding of a cyclic modular β-sheet without altering the sequence of the upper pentapeptide strand. The development of homologues of 1c and 1g with enhanced folding demonstrates how tuning the lower strand can improve β-sheet folding. Cyclic modular β-sheet 1g contains sequences from MIP-2 in the upper and lower strands. After ¹H NMR studies of 1g showed poor folding, we attempted to increase the folding by incorporating residues into the lower strand that would allow for greater cross-strand hydrophobic interactions. Our initial change to 1g of replacing the alanine of position 6 with valine resulted in slightly improved β-sheet folding. Our second change of incorporating tyrosine into position 6 resulted in greater improvement in folding and gave moderately folded cyclic modular β-sheet 1h. Insight from the design of 1h led to improvements to poorly folded 1c. Replacement of the leucine at position 6 of 1c with tyrosine resulted in moderately folded 1d.

Tuning of the lower strand was also necessary to prepare a cyclic modular β-sheet related to the huntingtin protein associated with Huntington’s disease and containing five glutamine residues in the upper strand.^44,45 We initially selected two arginine residues for the lower strand to enhance solubility. ¹H NMR studies on this compound showed essentially no ^δOrn δ-proton magnetic anisotropy (≤0.01 ppm) and no detectable NOE crosspeaks, thus indicating the compound to be completely without any β-sheet-like structure. Attempts at improving the folding by replacing the arginines in the lower strand with hydrophobic residues eventually resulted in an analogue that is moderately folded.

While some of our initial cyclic modular β-sheet designs have resulted in poorly folded structures, some others resulted in structures with poor water solubility. This is a potential problem for cyclic modular β-sheets that contain hydrophobic pentapeptide sequences, such as those from β-amyloid. This problem is illustrated by the homologue of 1b that we first prepared, which contained Val–Leu in the lower strand. The compound had poor solubility (<0.5 mM) in water, and no studies of the structure of this compound were attempted. Replacement of the valine in the lower strand with lysine substantially increased the solubility.

¹H NMR studies of 1a, 1b, and 1f suggest that pentapeptide sequences that allow cross-strand aromatic–aromatic interactions involving the Hao aromatic ring result in the best-folded cyclic modular β-sheets.^46,47 The three monovalent cyclic modular β-sheets in this paper that are well folded each contain at least one phenylalanine residue across from Hao. The phenylalanine side chains at position 4 of the well-folded cyclic modular β-sheets 1a, 1b, and 1f can participate in aromatic–aromatic interactions with the Hao aromatic ring, and the spectra of these compounds suggest that such interactions do occur. The spectra of 1a show ortho-proton resonances of the side chain of the phenylalanine at position 4 that are strongly upfield by ca. 0.5 ppm of those in control 6a. The spectra of 1b and 1f also show ca. 0.5 ppm upfield shifting of the ortho-proton resonances of the phenylalanine at position 4 relative to the respective controls. ROESY studies of 1a, 1b, and 1f show NOE crosspeaks between the phenylalanine aromatic rings and the Hao aromatic ring, further suggesting cross-strand interactions.⁴⁸

Cyclic modular β-sheets 1c, 1e, and 1f demonstrate the positive effect that a phenylalanine at position 4 can have on folding. These compounds are identical except for the residues at position 4: 1c contains glycine, 1e contains alanine, and 1f contains phenylalanine. Compound 1c is poorly folded, 1e is moderately folded, and 1f is well folded.

Conclusions

Combination of the Hao amino acid β-strand mimic with two ^δOrn β-turn mimics and α-amino acids results in cyclic modular β-sheets that present pentapeptide β-strands along one edge. The linear peptide precursors are quickly and easily synthesized by standard Fmoc solid-phase chemistry, cyclization is clean and efficient, and the resulting cyclic modular β-sheets are easily purified. The α-amino groups of the ^δOrn residues provide attractive sites for linking individual cyclic modular β-sheets. Several cyclic modular β-sheets have been prepared that show clear evidence (α-proton chemical shifts, ^δOrn δ-proton magnetic anisotropy, and NOE crosspeaks) of being folded into a β-sheet-like conformation. A linked cyclic modular β-sheet also shows evidence of being folded. PFG NMR diffusion studies demonstrate that little or no self-association occurs in the cyclic modular β-sheets at low millimolar (≤ 2mM) concentrations.

The potential for improving the folding of cyclic modular β-sheets by tuning the lower strand is a powerful feature that expands the number of sequences that can be placed in the top strand. The modularity of the cyclic modular β-sheets is another powerful feature. Monovalent cyclic modular β-sheets can be thought of as discrete building blocks for the synthesis of divalent structures with two separate β-sheet domains. Such divalent structures have the potential to be ligands for biomolecular assemblies that present multiple binding sites (e.g., β-amyloid or huntingtin aggregates). Preparation of multivalent cyclic modular β-sheets with more than two β-sheet domains should also be possible. Both monovalent and multivalent cyclic modular β-sheets are potential tools with which to mimic β-sheet interactions involving proteins and to inhibit those interactions.

Supplementary Material

si20061214_020. Supporting Information Available.

Experimental procedures and NMR spectra, mass spectra, and HPLC traces for the cyclic modular β-sheets; full citation for Reference ⁴⁵. This material is available free of charge via the Internet at http://pubs.acs.org.