Structural and biological mimicry of protein surface recognition by α/β-peptide foldamers

We selected a recently reported gp41 CHR-derived α-peptide, 3 (27), as the starting point for α→β modification (Fig. 1). α-Peptide 3, also known as T-2635, is 50% mutated as compared with the wild-type gp41 CHR domain and contains a combination of Xxx→Ala substitutions and engineered i→i+4 salt bridges that enhance α-helical propensity (27). α-Peptide 3 represents one of the most successful examples reported to date of improving the antiviral efficacy of gp41 CHR α-peptides via modification of the α-amino acid sequence. We began with the side chain sequence optimized in 3 and explored the orthogonal variable of backbone composition in the form of α→β residue substitution [a method we term “sequence-based design” (24–26)]. In α/β-peptide 4, a subset of the α-residues in 3 has been replaced by corresponding β³-residues (Fig. 1A). Thus, α/β-peptide 4 has the sequence of side chains found in 3 displayed on an unnatural backbone. The β³-residues of 4 are incorporated in an ααβαααβ pattern, which generates a stripe of β-residues that runs along one side of the helix (24). Our design places the β-stripe in 4 distal along the helix circumference to the molecular surface that packs against the gp41 NHR domain trimer in the six-helix bundle.

We used a previously reported competition fluorescence polarization (FP) assay (15) to compare 3 and 4. The assay measures displacement of a fluorescently labeled CHR α-peptide from an engineered five-helix bundle protein, gp41–5, which contains three NHR segments and two CHR segments. Affinity for the gp41–5 protein construct is expected to correlate with the ability of CHR-mimetic agents to bind to the gp41 prehairpin intermediate formed just before HIV-cell fusion (15). As expected, α-peptide 3 binds very tightly to gp41–5 (K_i < 0.2 nM; Table 1). The analogous α/β-peptide 4, however, displays only weak affinity for gp41–5, >10,000-fold lower than that of 3. The modest potency of α/β-peptide 4 in this protein-based assay is similar to that displayed by many small molecules and peptidomimetics in comparable experiments (11–17).

In an effort to understand the dramatic differences in binding between 3 and 4, we prepared and characterized chimeric α/β-peptides 5 and 6. Both 5 and 6 contain a pure α segment at the N terminus and an α/β segment at the C terminus (28); these oligomers are chimeras of α-peptide 3 and α/β-peptide 4. α/β-Peptide 5 displays very high affinity for gp41–5, indistinguishable from that of α-peptide 3; however, extending the α/β segment toward the N terminus (as in 6) causes a significant loss of affinity. The sensitivity of the N-terminal segment to α→β³ modification is consistent with data showing that side chains in this region, especially those corresponding to Trp₃, Trp₆, and Ile₁₀ in 3, play a crucial role in CHR binding to the NHR trimer (29).

α/β-Peptides 5 and 6 represent an improvement in gp41 mimicry relative to 4 (vide supra), but it would be desirable to place β-residues throughout an α/β-peptide sequence to maximize resistance to proteolysis (30–32). Each α→β³ replacement, however, adds a flexible bond to the backbone, which should increase the conformational entropy penalty associated with helix formation (26, 33, 34). The greater conformational entropy of the unfolded state of 4 relative to 3, arising from eleven α→β³ replacements, may account for the large difference in binding affinity for gp41–5 between these two oligomers. β-Residues provide an avenue for conformational preorganization that is made uniquely possible by their chemical structure. Incorporation of cyclic β-residues (e.g., ACPC and APC; Fig. 1A) can constrain the C_α-C_β backbone torsion and thereby enhance folding propensity without disrupting backbone amide hydrogen bonding (35–37).

We probed the impact of conformational preorganization in the context of gp41 mimicry by replacing a subset of β³-residues with cyclic analogues. The first comparison involved α/β-peptide 7, the analogue of 6, in which the three β³-hAla residues are replaced by ACPC (Fig. 1A). Both β³-hAla and ACPC are nonpolar, and this similarity was expected to maintain the physical properties that emerge from side chain sequence. The >30-fold higher affinity for gp41–5 displayed by 7 relative to 6 supports our hypothesis that residue-based rigidification is a useful complement to sequence-based design for development of peptide-mimetic foldamers. Replacement of two β³-hArg residues in oligomer 7 with APC, a heterocyclic analogue of ACPC, leads to α/β-peptide 8, which showed a very high affinity for gp41–5. APC₃₆ of 8 is in a region that does not engage the NHR region contained in gp41–5; this observation may explain the similar K_i values of 7 and 8. Additional evidence of the favorable contribution of cyclic β-residues comes from comparison of oligomers 4, 9, and 10, each of which has β-residues throughout the sequence. α/β-Peptide 9 was generated from 4 by four β³-hAla→ACPC replacements, which leads to a >45-fold improvement in K_i. Replacement of the three β³-hArg residues of 9 with APC, to generate 10, improves K_i by a further ∼10-fold. Relative to completely flexible α/β-peptide 4, rigidified analogue 10 (K_i = 9 nM) shows ∼380-fold enhanced binding to gp41–5.

We investigated the interactions of CHR α-peptide 3 and α/β-peptide analogs 4, 5, 8, and 10 with gp41 NHR peptide 1 by circular dichroism (CD) spectroscopy. When mixed with gp41 CHR α-peptides, NHR-peptide 1 forms a six-helix bundle that is thought to represent the postfusion state adopted by gp41 in the course of viral entry (4). α-Peptide 3 showed significant helical content at 20 μM in PBS, consistent with previously published data (Fig. 2A) (27). α/β-Peptide 4 showed little helicity under similar conditions; however, analogue 10, with seven β³→cyclic-β substitutions, showed an intense CD minimum, consistent with a well-folded α/β-peptide helix (24). We compared the observed CD spectrum for each 1:1 mixture of NHR+CHR peptide (Fig. 2B, solid lines) to that calculated by averaging spectra for the corresponding individual oligomers (Fig. 2B, dashed lines). α/β-Peptides 5, 8, and 10, which displayed high affinity for gp41–5 in the competition FP assay, each showed a significant degree of induced helicity when mixed with NHR α-peptide 1, consistent with six-helix bundle formation. By contrast, α/β-peptide 4, which has only modest affinity for gp41–5, showed essentially no interaction with 1. The magnitude of the CD signatures among the well-folded mixtures is similar, but the ratio of intensities at 208 and 222 nm changes as a function of β-residue content. This trend is consistent with our previous studies on helical oligomers containing mixed α/β backbones (24, 26). The complexes formed by 1+3, 1+5, 1+8, and 1+10 each showed highly cooperative thermal transitions (Fig. 2C). The trend in apparent T_m values (T_m,app) correlates with differences in affinity among 3, 5, 8, and 10 for gp41–5 in the competition FP assay (Table 1); that is, stronger binding to gp41–5 correlates with more stable assembly with NHR peptide 1.

We used X-ray crystallography to compare the heteromeric six-helix bundles formed by NHR α-peptide 1 with CHR α-peptide 3, chimeric CHR α/β-peptide 8, or CHR α/β-peptide 10 (Fig. 3, Table S1, and Fig. S1). Although the mutations to the native CHR sequence that lead to α-peptide 3 were not intended to modify the nature of its binding interactions with the gp41 NHR domain (27), we sought direct evidence that the six-helix bundle structure was unchanged relative to that formed by 1 and the native CHR sequence (4). We obtained a cocrystal of α-peptides 1 and 3 and solved the structure to 2.0 Å resolution. The resulting six-helix bundle (Fig. 3A) is essentially identical to that previously reported for 1+2 (4), which contains the native CHR sequence; the root mean square deviation (rmsd) is 0.73 Å for C_α atoms.

We obtained a crystal of the 1+10 complex and solved the structure to 2.8 Å resolution. α-Peptide 1 and α/β-peptide 10 combine to form a six-helix bundle that is similar to that formed by 1+3 (Fig. 3A). A crystal containing only α/β-peptide 10 was obtained as well. The structure of 10 alone (Fig. S2), solved to 2.1 Å resolution, revealed a parallel trimeric helix bundle with a hydrophobic core comprising the residues that engage the gp41 NHR trimer 1+10. The self-assembly of α/β-peptide 10 in the crystalline state parallels the behavior previously observed for prototype α-peptide 3, which was shown to self-assemble in solution (27).

The core NHR trimers in the structures of 1+10 and 1+3 are highly homologous (0.65 Å C_α rmsd for NHR residues 3–30). When the two bundles are aligned via the NHR trimer (Fig. 3B), the CHR helices track very closely in the C-terminal segment (0.84 Å C_α rmsd for residues 16–33), but diverge near the N terminus (4.2 Å C_α rmsd for residues 2–15). This divergence reflects a greater superhelical twist in α-peptide 3 relative to α/β-peptide 10. The divergent portion of the helix formed by 10 contains the two Trp residues that, in CHR α-peptides, are essential for stable six-helix bundle formation (Fig. 3C) (29). In the structure of 1+10, the side chains of Trp₃ and Trp₅ were not resolved in electron density, suggesting a high degree of disorder. In addition, significant disorder was observed in the side chains of NHR residues Lys₂₉ and Trp₂₆, which pack around CHR Trp₅ in the 1+3 complex.

Given the well-established role of the gp41 CHR domain Trp-Trp-Ile motif in six-helix bundle formation (29), the observation that the N-terminal segment of α/β-peptide 10 does not engage the NHR binding pocket in the crystal structure of the 1+10 complex is intriguing. Removal of the first 10 residues of α/β-peptide 10 leads to oligomer 11, in which the Trp-Trp-Ile motif is not present. α/β-Peptide 11 showed no measurable affinity for gp41–5 (K_i > 10 μM), indicating that the N-terminal segment of 10 is essential for high-affinity binding to gp41–5 in solution.

Motivated by the differences between the CHR domain, N-terminal segments in the 1+3 complex and the 1+10 complex, we investigated the structure of NHR peptide 1 in complex with CHR α/β-peptide 8, a chimera of α-peptide 3 and α/β-peptide 10. We crystallized the 1+8 complex and solved the structure to 2.8 Å resolution. Relative to α/β-peptide 10, chimeric α/β-peptide 8 tracks much more closely with the CHR helix (3) in the all α-peptide six-helix bundle formed by 1+3 (1.4 Å C_α rmsd for residues 2–33; Fig. 3 A and B). The side chains of the Trp-Trp-Ile motif in the N-terminal segment of 8 show the expected packing into the binding pocket on the NHR core trimer (Fig. 3C). Based on this result and the behavior of truncated α/β-peptide 11, we suspect that the lack of direct contact between the N-terminal portion of 10 and the NHR trimer in the 1+10 complex is an artifact of crystal packing.

We performed two sets of experiments to evaluate the activities of α-peptide 3 and α/β-peptides 4, 5, and 10 in a biological context. We first compared the oligomers in a cell-cell fusion assay based on expression of the env gene of the HIV-1 clone HxB2, an assay that is commonly used to model gp41-mediated HIV-cell fusion (38). The cell-cell fusion assay results (Table 1) showed that α/β-peptides 5 and 10 have IC₅₀ values indistinguishable from that of α-peptide 3, whereas α/β-peptide 4 is much less effective. We then evaluated 3, 4, 5, and 10 for the ability to prevent HIV infection of the cell line TZM-bl (39). These studies used one T-cell-line-adapted strain and three primary isolates of HIV; two of the strains are X4-tropic, and the other two are R5-tropic. The infectivity assay results (Table 1 and Fig. S3) show similar biological potencies among 3, 5, and 10 for HIV-1 strains that use different coreceptors; this finding suggests the blocking of a necessary, shared step in entry through peptide interactions with conserved regions of gp41. It may be noted that there is imperfect correlation between K_i for binding to gp41–5 and IC₅₀ values in cell-based assays among the compounds reported here. For example, the affinity of 10 for gp41–5 was >45-fold lower than that of 5, yet IC₅₀ values for 10 were sometimes lower than for 5. There are several possible reasons for this discrepancy. Sequence differences between the CHR and NHR domains found in gp41–5 and those found in the viruses tested may lead to better correlation between gp41–5 binding affinity and antiviral activity against some strains relative to others. In addition, it has previously been suggested that the association rates for CHR peptides binding to gp41 are a better predictor of relative antiviral potencies than are equilibrium binding affinities (40). The rigidified backbone in 10 may alter its association rate with gp41 relative to that of 5. Sensitivity to gp41-derived fusion inhibitors may be affected by many factors that differ among strains of virus, including the amount of Env incorporated into the virion, the strength of Env interactions with CD4 and with coreceptors, the kinetics and energetics of the fusion process, as well as amino acid variation in the binding site for inhibitory peptides. Overall, the antiviral assays results support our hypothesis that CHR-derived α/β-peptides effectively mimic gp41 in a complex biological milieu.

An important motivation for the development of foldamer antagonists of protein–protein interactions is the prospect of diminishing sensitivity to proteolytic degradation. Rapid destruction in vivo represents a significant drawback to the clinical use of α-peptide drugs. We compared the susceptibilities of α-peptide 3 and α/β-peptides 4 and 10 toward degradation by proteinase K, a promiscuous serine protease. Under the assay conditions, α-peptide 3 was completely degraded within minutes; mass spectrometry revealed hydrolysis of at least 10 different amide bonds in the sequence (Fig. S4). α/β-Peptide 4, with exclusively α→β³ substitution, showed 20-fold improvement in stability relative to prototype α-peptide 3. Rigidified α/β-peptide 10 showed an even greater improvement in stability over α-peptide 3 (280-fold). The greater stability of α/β-peptide 10 relative to α/β-peptide 4 likely results from the greater helical propensity of 10, as detected by CD. The small number of proteolysis products observed for α/β-peptide 10 supports previous observations that β-residues in mixed α/β backbones tend to protect neighboring amides from proteolytic cleavage (25, 31).

Detailed protocols for the synthesis and purification of 1–11 and CHR peptide modified at the N terminus with 5-carboxyfluoroscein (Flu-C38) can be found in the SI Materials and Methods.

Expression and purification of gp41–5 were carried out as previously described (15). The plasmid for gp41–5 was provided by Professor Stephen Harrison (Harvard University, Cambridge, MA). The full sequence of the construct used in this study, which differs slightly in the loop regions from the published sequence, is below.

MSGIVQQQNNLLRAIEAQQHLLQLTVWGIKQLQARILSGGSGGWMEWDREINNYTSLIHSLIEESQNQQEKNEQELLGGSGGSGIVQQQNNLLRAIEAQQHLLQLTVWGIKQLQARILSGGSGGWMEWDREINNYTSLIHSLIEESQNQQEKNEQELLGGSGGSGIVQQQNNLLRAIEAQQHLLQLTVWGIKQLQARIL

The general experimental considerations for the FP assays, modified somewhat from the reported gp41–5 FP assay protocol, (15) were based on published methods (25). Briefly, the binding affinity of Flu-C38 for gp41–5 was measured by titrating gp41–5 against a 0.2 nM solution of labeled peptide. The K_d of the tracer was determined to be 0.4 ± 0.1 nM. Competition fluorescence polarization assays were carried out by 1:100 dilutions of inhibitor stocks in DMSO into a solution containing 2 nM gp41–5 and 1 nM Flu-C38.

Crystals were grown by hanging drop vapor diffusion. Diffraction data for the 1+3 and 1+8 complexes were collected on a Bruker X8 Proteum Diffractometer using Cu K_αradiation. Diffraction data for the crystals of 10 and the 1+10 complex were collected at the Life Sciences Collaborative Access Team beamline 21-ID-G at the Advanced Photon Source, Argonne National Laboratory. All structures were solved by molecular replacement. Detailed protocols for crystallization, data collection, data processing, structure solution and refinement can be found in the SI Materials and Methods.

Circular dichroism measurements were carried out on an Aviv 202SF Circular Dichroism Spectrophotometer. Spectra were recorded in a 2-mm cell with a step size of 1 nm and an averaging time of 5 s. Thermal melts were carried out in a 1-mm cell, monitoring at 222 nm, with 5 °C increments and an equilibration time of 10 min between each temperature change (a 20-min equilibration time gave similar results). Thermal unfolding data were fit to a simple two-state folding model using GraphPad Prism to obtain the reported T_m,app values.

A cell-to-cell-fusion assay based on the envelope glycoprotein of the HIV-1 clone HXB2 expressed in CHO cells and with U373-MAGI cells as targets was carried out as previously described (38). All of the α/β peptides showed no cytotoxicity at 5 μM, as judged by measuring the basal level of β-galactosidase expression in the U373-MAGI target cells. Inhibition of HIV-1 infectivity was measured on TZM-bl (JC53BL) cells, which express CD4, CXCR4, CCR5, and the luciferase gene under the control of HIV-1 long terminal repeat (LTR) (39). Viral stocks produced in PBMC of four HIV-1 strains were used: NL4–3, a clone derived from the X4-tropic T cell line-adapted isolate IIIB of clade B; HC4, an X4 primary isolate of clade B (47); and the two R5 primary isolates, CC 1/85 (clade B) (48) and DJ258 (clade A) (49). Briefly, TZM-bl cells were seeded the day before inoculation at a density of 10⁵ cells per mL, 100 μL per well. Serially diluted peptide in 50 μL (or medium alone as a control) was added to each well. Then the virus, 40 TCID₅₀ in 50 μL, or medium only as a background control, was added to each well. On the third day, the wells were inspected by light microscopy. Wells with and without peptide were compared for cell confluency and morphology. No signs of toxicity were discerned at the highest concentrations of peptide used. The infectivity was then quantified in relative light units with the Bright-Glo Luciferase Assay System (Promega) according to the manufacturer's instructions. The experiment was performed three times. The signal of test wells was normalized to that of control wells without inhibitor after background subtraction from both. The percent inhibition of infectivity was expressed as a function of the log₁₀ concentration of inhibitor in nM. A four-parameter sigmoid function was fitted to the data in Prism (GraphPad). The R² values for the fits were 0.95–1.0 for NL4–3; 0.98–1.0 for HC4; 0.95–0.98 for CC 1/85; and 0.92–0.98 for DJ258.

Assays were carried out as previously described (25). A detailed protocol can be found in SI Materials and Methods.