Design and characterization of swapped-domain constructs of HIV-1 glycoprotein-41 as receptors for drug discovery

Abstract

Four new swapped-domain constructs of the ectodomain of human immunodeficiency virus type 1 glycoprotein-41 (gp41) were prepared. The gp41 ectodomain consists of 50-residue N-heptad repeat (NHR), 36-residue disulfide-bonded loop and 39-residue C-heptad repeat (CHR). It folds into a hairpin structure that forms a trimer along the NHR axis. The swapped-domain proteins feature CHR domains of length 39, 28 or 21 residues preceding a 4-residue loop and a 49- or 50-residue NHR domain. The effect of CHR truncation was to expose increasing lengths of the NHR groove, including the conserved hydrophobic pocket, an important drug target. A novel method for preparing proteins with extended exposed hydrophobic surfaces was demonstrated. Biophysical measurements, including analytical ultracentrifugation and ligand-detected Water-Ligand Observed via Gradient Spectroscopy and ¹H–¹⁵N-HSQC NMR experiments, were used to confirm that the proteins formed stable trimers in solution with exposed binding surfaces. These proteins could play an important role as receptors in structure-based drug discovery.

Introduction

The human immunodeficiency virus (HIV) fusion protein, glycoprotein-41 (gp41) is an important target for drugs against viral fusion. Currently, the only FDA-approved fusion inhibitor is the peptide enfuvirtide (T20), derived from the C-domain of gp41. There are no small-molecule drugs that inhibit fusion. gp41 is vulnerable to inhibition during the large conformational transition that follows receptor binding by gp120 (Fig. 1). Recent crystal and cryoelectron microscopy structures of a prefusion form of the Env trimer revealed a gp41 conformation in which three N-heptad repeat (NHR) helices extended for 30 Å along the trimer axis, and the C-heptad repeat (CHR) helices were wrapped around the base of the trimer (Julien et al., 2013; Lyumkis et al., 2013). Both the NHR and CHR regions made extensive contacts with gp120 residues. Once gp120 disassociates, gp41 refolds, eventually forming a trimer of hairpins or six helix bundle structure in which residues 542–591 of the NHR form a 77 Å coiled coil homotrimer, and the CHR helical domains fold back into the grooves of the NHR (Chan et al., 1997; Weissenhorn et al., 1997; Caffrey et al., 1998). gp41 N- and C-peptides are effective fusion inhibitors, supporting the existence of a long-lived pre-hairpin intermediate with exposed NHR and CHR domains.

Fig. 1.

A model for the conformational changes of gp41 from its prefusion complex with gp120 to a final trimer of hairpins structure with fusion and transmembrane peptides coalescing to form a fusion pore. The existence of pre-hairpin intermediates (PHIs) with exposed NHR and CHR domains is supported by biochemical data on composite peptide inhibitors.

Small-molecule drug discovery has focused on a highly conserved hydrophobic pocket (HP) formed by two NHR peptides of the homotrimer between residues 565 and 581 (Fig. 2) (Chan et al., 1998). The HP binding domain (HPbd) includes the ‘WWI’ motif of CHR residues 628–635, which make important contacts in this pocket. We (Zhou et al., 2011, 2014; Whitby et al., 2012) and others (Debnath et al., 1999; Katritzky et al., 2009; Wang et al., 2010; He et al., 2013) have developed low-molecular-weight inhibitors targeting the HP, and we have found a correlation between pocket binding affinity and ability to inhibit fusion (Zhou et al., 2014). However, optimization of the inhibitors has been constrained by the lack of structural details defining the gp41—inhibitor interaction. In addition, our peptide and small-molecule studies have revealed a critical role for a region extending beyond the HP between residues 582 and 591 (Gochin and Cai, 2009; Chu and Gochin, 2013). The C-peptide segment apposing this upstream domain has been found to contribute to six helix bundle stability and antiviral activity (He et al., 2008), and we have determined this region of the NHR to contain a druggable pocket (Chu and Gochin, 2013).

Fig. 2.

Design of swapped-domain constructs of gp41. Hairpins are reversed, compared with the gp41 ectodomain structure shown at left, with the N-terminus at the beginning of the CHR and C-terminus at the end of the NHR, and the two domains connected by a short loop. The reverse hairpin permits exposure of the predicted CT pocket and HP on the NHR, by successive truncation of the CHR. Note that only one monomer of the expected trimer is shown for clarity; the trimer axis is formed by three NHR domains forming a coiled coil, and the binding pockets are created by trimerization.

Here we have prepared and evaluated swapped-domain constructs of gp41 in which the CHR precedes the NHR in sequence, separated by a short loop (Fig. 2). Stewart et al. (2007, 2010) demonstrated this design as a way to expose the HP for binding by small molecules. A short CHR abutting a longer NHR ensured pocket exposure. Stewart et al. conducted an NMR study of a small-molecule bound in the HP of Protein-1, which formed a reverse hairpin trimer consisting of CHR residues 639–664 and NHR residues 542–584. However, Protein-1 was difficult to make, since enzymatic cleavage of an N-terminal His-tag covering the HP failed, and the authors resorted to ammonium sulfate precipitation for crude extraction. A persistent yellow color in the samples, possibly due to endotoxin or lipid association, was removed by high-speed ultracentrifugation. In our hands, similar constructs were poorly overexpressed, appeared in the insoluble fraction and aggregated readily, and we were not able to remove the yellow color. As part of our goal to evaluate small-molecule binding to gp41, we have redesigned the reverse hairpin sequences and the methods used to prepare them, and have obtained several constructs with high levels of overexpression, ease of purification and promising properties. We have extended the NHR to residue 591 and made constructs with three different CHR lengths, and variations of the wild-type sequence. We demonstrate that the receptors display the desired binding pockets between residues 565 and 591 and are sensitive to peptide and small-molecule binding.

Materials and methods

Reagents

All reagents were used without further purification. The fragment library was purchased from Chembridge, ∼1 mg of each compound, dissolved in DMSO to obtain 200 mM solutions (Chu and Gochin, 2013). Fragments P1C02 and P5C04 were repurchased in larger quantities for further testing.

Construction of plasmids

C39(L4)N50

We began by sub-cloning the gp41-portion of the sequence of Protein-1 (Stewart et al., 2007) between Nde1 and BamH1 of pET-21a. A stop was inserted in front of the pET-21a C-terminal (CT) 6-his tag, and an N-terminal 6-his tag followed by TEV protease recognition sequence inserted by QuikChange mutagenesis (Agilent). This construct was then extended both on the NHR and CHR, and the loop between domains changed from DGD to ‘L4’ = SGGG in order to obtain a construct with the gp41 wild-type CHR and NHR sequences in the arrangement C39(L4)N49. Additionally, we performed mutagenesis to alter the wt-CHR sequence at 10 positions. A final modification of the construct entailed appending the sequence QKR on to the C-terminus (Q being the final wild-type NHR residue). The resultant construct contained a modified CHR, but maintained the full wild-type NHR residues 542–591 (with residues KR appended at the C-terminus) and was named C39(L4)N50 (Fig. 3).

Fig. 3.

Sequences of the swapped-domain construct precursors synthesized in E. coli. The sequence of the final product begins immediately after the arrow indicating the cleavage site. CHR and NHR domains are enclosed in gray and black boxes, respectively. Residues modified from the wild-type HXB2 sequence are underlined. Glutamic acid and arginine residues that could form i, i + 4 salt bridges on the outside of the putative six helix bundle are shown in italics.

C21(L4)N49

A number of variants of this construct were designed before the construct resulting in the final protein was obtained. The constructs were made by starting with C39(L4)N49, followed by removal of the TEV protease site between N-terminal 6-his tag and beginning of the CHR, and insertion of a new TEV protease site in the middle of the CHR, immediately preceding L645. Several variants of known TEV protease recognition sites were inserted, as well as different composition and length spacer sequences flanking the recognition site to the N- and C-terminal sides. We refer to the cleavage site together with flanking sequences as a ‘loop-out’, since they ideally would form a protease-accessible sequence looping out and interrupting the CHR helix. Finally, we also generated a variant of C21(L4)N49 which contained the acid-labile sequence Asp-Pro inserted at the TEV recognition site. This construct eventually became the final C21(L4)N49. From this construct, C21(L4)N50 was made by appending the residues QKR, by mutagenesis.

C28(L4)N50

This construct was obtained from C21(L4)N50 by insertion of additional CHR residues. A number of variants of the Asp-Pro acid-labile loop-out sequence were constructed.

GB1-C-peptide fusions

We designed two fusion proteins composed of a peptide from the HIV type 1 (HIV-1) gp41 CHR domain and GB1 (domain B1 of streptococcal protein G). They contained an N-terminal 6His-tag followed by the GB1 domain, a linker of residues GGSGGS and gp41 residues N616 through either I635 or I642, designated as GB1ⁱ⁶³⁵ and GB1ⁱ⁶⁴², respectively. Several alterations were made in the CHR sequence from wild type, including the use of the same CHR modifications as were used for C39(L4)N50.

Protein expression and purification

C39(L4)N50

The precursor protein was overexpressed in Escherichia coli BL21(DE3) Gold (Agilent Technology) in M9 minimal medium (Cai et al., 1998). For ¹⁵N-labeled protein, 1 g/l of ¹⁵N-ammonium chloride (CIL, Massachusetts, MA) was used, and to improve yields, the M9 contained 6 g/l dextrose. Protein expression was induced with 0.5 mM IPTG at OD₆₀₀ ∼1.2, and continued for 16–18 h at 12°C. The cells were harvested at OD₆₀₀ ∼3.7 and spun down for 30 min at 5000 g, 4°C. The cells were lysed for 30 min under gentle rocking at room temperature using 70 ml/l culture of BPER lysis reagent (Thermo Scientific) with pH adjusted to 8.0, and spun down for 1 h at 40 000 g, 4°C. The supernatant was filtered through 0.22 μm syringe filters and passed over 2 × 5 ml Nickel HisTrap-HP columns (GE Healthcare) in series, equilibrated in 50 mM Tris pH 7.5, 200 mM NaCl, 20 mM imidazole, 0.02% NaN₃. A linear imidazole gradient (20−500 mM) was applied with the protein eluting at ca. 280 mM imidazole. After overnight dialysis at room temperature into TEV Protease digestion buffer (50 mM Tris, 150 mM NaCl, 5 mM β-mercaptoethanol, 0.02% NaN₃), the protein was digested with ca. 1:15 mg/mg of His-tagged TEV Protease overnight at room temperature. The digested protein was passed over a 5 ml Nickel HisTrap-HP column equilibrated in 50 mM Tris pH 8.0, 200 mM NaCl, 20 mM imidazole and 0.02% NaN₃ and the flow-through collected and dialyzed into 10 mM MES pH 5.5, 50 mM arginine base, 50 mM glutamic acid, 0.02% NaN₃. The protein was then concentrated using an Amicon Ultra 15 ml 3K MWCO filtration unit (Millipore) and purified by gel filtration on HiLoad 26/60 Superdex 75 pg (GE Healthcare) equilibrated in the same buffer.

C28(L4)N50, C′21(L4)N50, C21(L4)N49

The precursor proteins to the HP-exposing constructs were overexpressed and harvested, lysed and spun down as described above, with the modification that the pH of the BPER lysis buffer was adjusted to 8.5. The filtered supernatant was applied to 2× 5 ml Nickel HisTrap-HP (GE Healthcare), equilibrated in 50 mM Tris, pH 8.0, 200 mM NaCl, 20 mM imidazole, 5% glycerol, 0.02% NaN₃. After washing with 2% Triton X-100 (Sigma-Aldrich) and then extensively with buffer, the protein was eluted in an imidazole gradient (20–500 mM). The peak fraction was dialyzed overnight at room temperature in 50 mM formic Acid pH 3.0. After a change in dialysis buffer, the sample was fully dried down using a Hei-VAP Precision Rotovap (Heidolph) with a bath temperature of 30°C. The protein was then taken up in 8 M urea, 50 mM Tris pH 8.0, 200 mM NaCl, 20 mM imidazole binding buffer, and the Nickel column step, dialysis, and drying was repeated so as to remove any residual endotoxin, as indicated by the powder-white appearance of the final dried sample. This purified precursor protein was then taken up in 70% formic acid and warmed to 55°C for 36–42 h to hydrolyze the peptide bond between the aspartic acid and proline residues. After cleavage the sample was dried completely by rotary evaporation, and the resultant white solid taken up in 8 M urea, 50 mM Tris pH 8.0, 200 mM NaCl, 20 mM imidazole. The sample was applied to a 5 ml Nickel HisTrap-HP column and the flow-through collected. Due to a minor secondary cleavage between D664 and S665, final gel filtration purification was performed on Superdex 75 (GE Healthcare) in 8 M urea, 50 mM Tris pH 8.0, 200 mM NaCl. The desired purified proteins, with HPbd removed, were concentrated by centrifugation in Amicon Ultra 4 ml 3K-MWCO and refolded by ca. 1:80 drop dilution at room temperature from denaturant conditions (either 4 M urea or 4 M GndHCl), into 10 mM MES pH 5.5, 50 mM arginine base, 50 mM glutamic acid, 0.02% NaN₃ buffer and concentrated. Proteins remained soluble under these conditions at room temperature indefinitely, however, precipitation occurred upon refrigeration. The sample molecular weight was confirmed by Maldi-MS or ESI-MS (Supplementary data). Both unlabeled and ¹⁵N-labeled proteins were prepared using the same method.

GB1ⁱ⁶³⁵ and GB1ⁱ⁶⁴²

Proteins were expressed in E. coli BL21(DE3) in M9 minimal medium containing ¹⁵N-NH₄Cl (Cambridge Isotope labs, Inc.). Protein expression was induced at OD₆₀₀ ∼1.0 with 0.5 mM IPTG, and continued for 12 h at 12°C until OD₆₀₀ ∼3.5. Protein was extracted as before, and purified on 2 × 5 ml Nickel HisTrap-HP, followed by purification by gel filtration on Superdex 75 equilibrated in 50 mM Tris, pH 8.0, 200 mM NaCl, 4% glycerol, 0.02% NaN₃. The proteins were soluble indefinitely in this buffer at 4°C. A ¹³C–¹⁵N-double labeled sample of GB1ⁱ⁶³⁵ was obtained using 1 g/l ¹³C₆-d glucose (Aldrich). Unlabeled GB1ⁱ⁶³⁵ was also prepared using an identical method with unlabeled NH₄Cl and glucose.

Fluorescence binding experiments

HP binding was detected using 7.2 µM of bipyridated NHR peptide Fe(env2.0)₃ and 15 nM C18-e2.0-FL in Tris-acetate buffer at pH 7.0, described in detail in a previous publication (Gochin, 2012). The fragment library was screened in duplicate at a concentration of 1 mM. Twenty micromolar of bithionol was used as a positive control. Hits were further evaluated in dose–response measurements by serial dilution over the range of 1 mM to 1 µM.

LC–MS analysis

Samples were analyzed on an Agilent 1100 LC coupled with a Finnigan LCQ Duo MS system (Thermo Quest). An Ultra 120 5 µm C18Q column (50 × 2 mm i.d.) (Peeke Scientific, Redwood City, CA) was used for LC, with mobile phase A: water with 0.1% formic acid, and mobile phase B: acetonitrile with 0.1% formic acid. The gradient was as follows: 0–5 min 5% B, 5–25 min 5–100% B, 25–30 min 100% B, 30–35 min 100–5% B. The flow rate was 0.2 ml/min. Positive ions were collected over a mass range of 800–2000 Da.

NMR experiments

A GB1ⁱ⁶³⁵ HNCACB experiment (Wittekind and Mueller, 1993) was recorded at 25°C on an NMR sample containing 1.3 mM ¹³C/¹⁵N-labeled protein in 10 mM MES, pH 5.5 with 50 mM glutamic acid, 50 mM arginine and 0.02% azide using Bruker AVANCE 800 Spectrometer (Bruker Biospin, Billerica, MA) equipped with a 5 mm triple-resonance gradient probe. The data were processed with nmrPipe (Delaglio et al., 1995) and assignments were made using the program CARA (Keller, 2004). The GB1ⁱ⁶³⁵ titration experiments were performed by acquiring the ¹H–¹⁵N-TROSY spectra (Pervushin et al., 1997) at 25°C on a Bruker AVANCE 400 Spectrometer equipped with a 5 mm triple-resonance gradient probe. The data were processed with nmrPipe. The NMR samples contained 100 µM ¹⁵N-labeled GB1ⁱ⁶³⁵ and concentrations of unlabeled C28(L4)N50 of 0, 25, 50, 100, 200, 400 and 1350 µM in 10 mM MES, pH 5.5 with 50 mM glutamic acid, 50 mM arginine, 0.02% NaN₃. For GB1ⁱ⁶⁴², ¹H–¹⁵N-TROSY spectra were acquired at 400 MHz, 25°C containing 250 µM GB1ⁱ⁶⁴² alone, and with an equimolar amount of unlabeled C21(L4)N50.

The combined ¹⁵N and ¹H chemical shift perturbations were calculated as (Pellecchia et al., 1999):

(1)

A global fit of the data to dissociation constant K_D was obtained using the following equation:

(2)

where [R] is the concentration of free C28(L4)N50 and f_b is the fraction of bound GB1ⁱ⁶³⁵. is the maximum chemical shift difference obtained with 100% bound GB1ⁱ⁶³⁵ and is a parameter calculated from the fitting. [R] was obtained by solving the quadratic equation [R]² + (L_t – R_t + K_D) [R] – K_DR_t = 0. L_t and R_t are the total added concentrations of GB1ⁱ⁶³⁵ and C28(L4)N50, respectively. Fits were performed for each non-overlapped resonance, as well as globally over all resonances to obtain a single K_D value.

Water-Ligand Observed via Gradient Spectroscopy (WaterLOGSY) experiments were carried out with 128 scans at 400 MHz in Tris-acetate buffer at pH 7, with 300 µM fragments and 10 µM protein, using a 3 ms Gaussian pulse to selectively invert the water resonance. Excitation sculpting with gradients (Hwang and Shaka, 1995) was used for water suppression in 1D and WaterLOGSY spectra. Additional fragment screening was conducted with WaterLOGSY experiments using the peptide pair Fe(env3.0)₃ (10 µM)/C18-e3.0 (30 µM), as described in detail in Chu and Gochin (2013).

Sedimentation equilibrium analytical ultracentrifugation

Sedimentation equilibrium studies were performed with the constructs C39(L4)N50 at concentrations of 24 and 12 µM, C28(L4)N50 at 48 and 12 µM and C21(L4)N50 at 48, 24 and 12 µM in MES buffer, pH 5.5, containing 50 mM arginine base and 50 mM glutamic acid, at speeds of 25 000, 30 000, 35 000, 40 000 and 45 000 rpm in a Beckman XL-I ultracentrifuge at 25°C. Data from the five speeds for each sample were fit globally to single-species and/or monomer–trimer two-species models of equilibrium sedimentation by a non-linear least squares method using Igor Pro.

Results and discussion

Design of swapped-domain gp41 proteins

To create a gp41 model system for drug discovery, we prepared swapped-domain constructs which reverse the wild-type order of NHR and CHR domains, so that the CHR precedes the NHR with a short connecting loop. Figure 2 shows the topology of the reverse hairpins in the context of the full-length gp41 ectodomain. They present elements of gp41 in a manner similar to previously described core structures N47(L6)C39, N36(L6)C34 and N34(L6)C28 (Lu and Kim, 1997; Lu et al., 1999; Shu et al., 2000; Sackett et al., 2009), but have the constraining loop at the opposite end. The constructs are named for the lengths of the gp41 CHR and NHR domains, and the intervening four-residue SGGG loop. C39(L4)N50, C28(L4)N50 and C21(L4)N50 all contain a full-length NHR domain, and varied length CHR domains exposing the hydrophobic groove of the (trimeric) NHR to different extents. Included in this groove is the HP and extended CT segment. C28(L4)N50 and C21(L4)N50 were both designed to have an exposed HP in the folded trimer, while the long CHR in C39(L4)N50 contains the HPbd occluding the HP, but is expected to have an exposed CT. By comparison, Protein-1 designed by Stewart et al. (2007) had the form C26(L4)N42, and exposed the HP. The HP is not exposed in core structures N47(L6)C39, N36(L6)C34 and N34(L6)C28.

Construct design and protein expression

Plasmids were constructed with DNA sequences corresponding to the precursor proteins shown in Fig. 3. In the sequences shown, there are several helix stabilizing salt-bridge and alanine substitutions (Dwyer et al., 2007; Oishi et al., 2008) on the outer solvent-exposed face of the helical CHR. All N50-containing constructs also contain two additional CT amino acids KR, which are not part of the HIV sequence but were intended to test the effect on solubility. An additional construct C21(L4)N49 lacks the CT (Q)KR.

Precursor proteins were expressed at high yield (∼70 mg/l of culture) in E. coli with an N-terminal His tag, which was used for purification prior to removal by enzymatic or chemical cleavage. C39(L4)N50 was successfully produced using TEV protease to remove the His tag. To prepare proteins with an exposed HP in high yield in minimal media, we designed precursor proteins containing the HPbd, which covered the HP during overexpression (Fig. 3). A loop-out immediately after the HPbd was included with a TEV protease recognition sequence. Numerous variants of the loop-out sequence were screened for maximum overexpression yield and solubility, and in attempts to achieve TEV protease cleavage. The benefit of such a design was manifest in the substantially increased yield and solubility when compared with our early attempts with constructs in which the HP was exposed during overexpression. However, we were unsuccessful in our in vitro and in vivo (Kapust and Waugh, 2000) attempts to obtain digestion of the precursor protein with TEV protease. Therefore, we introduced the acid-labile Asp-Pro sequence motif into the location of the previous TEV protease recognition site, which allowed the additional HPbd segment to be successfully removed using formic acid cleavage (Crimmins et al., 2005). This left an N-terminal proline in the final constructs. The cleaved N-terminal segment was removed by a second Ni affinity column, and the desired inverted hairpin proteins were subsequently further purified by size exclusion chromatography. The proteins folded spontaneously by drop dilution from urea or guanidine into pH 5.5 buffer. Protein purity was confirmed by SDS-PAGE and Mass Spectrometry (Maldi, ESI) (Supplementary Figs. S1 and S2 and Tables SI and SII). We obtained 20–30 mg of purified protein per liter M9.

Analytical ultracentrifugation studies

The sedimentation equilibrium experiments on C39(L4)N50, C28(L4)N50 and C21(L4)N50 clearly indicated that all three proteins were trimeric in solution over the concentration range of 12–48 µM. Figure 4 shows the global fitting of the data to a single-species model for C39(L4)N50, C28(L4)N50 and C21(L4)N50. A molecular weight equal to 32.9 kDa was obtained from a global fit of the C39(L4)N50 data at 24 and 12 µM and at five speeds, consistent with a well-formed trimer (expected MW 34 307 Da, ¹⁵N-labeled protein). Similarly, a global fit calculated molecular weight from the C28(L4)N50 samples was 27.4 kDa at 48 and 12 µM at five speeds (expected MW 30 036 Da, ¹⁵N-labeled protein). A global fit of C21(L4)N50 data at 48, 24 and 12 µM at five speeds gave molecular weight 27.7 kDa (expected MW 27 370 Da, ¹⁵N-labeled protein). By employing a monomer–trimer two-species fitting, we also calculated the oligomerization affinities of the two shorter constructs. The equilibrium association constant (K_a) for C28(L4)N50 was 5.0 × 10¹⁰ M⁻² in the same conditions as above, and for C21(L4)N50 was 3.2 × 10¹⁰ M⁻² at 48 and 12 µM at four speeds (Supplementary Fig. S3). Their K_a's are close to those reported for the structurally similar ectodomain of SIV gp41 (K_a = 1.5 × 10¹¹ M⁻² and 3.1 × 10¹¹ M⁻²) (Wingfield et al., 1997; Caffrey et al., 1999). Thus, the four-residue loop size and sequence successfully accommodated the required hairpin turn for trimerization.

Fig. 4.

Sedimentation equilibrium analytical ultracentrifugation experiment on (A) 24 (left) and 12 (right) μM C39(L4)N50, (B) 48 (left) and 12 (right) μM C28(L4)N50, and (C) 48 (left), 24 (middle) and 12 (right) μM C21(L4)N50 at 25°C in MES, pH 5.5 containing 50 mM arginine base and 50 mM glutamate. Residuals between calculated and observed data are shown above each plot. Rotor speeds were 25 000, 30 000, 35 000, 40 000 and 45 000 rpm.

Interaction of HPbd peptides with reverse hairpins

C28(L4)N50 and C21(L4)N50 were examined for their ability to interact with peptides containing the HPbd. HPbd peptides were prepared as fusion proteins consisting of His-tagged GB1 (immunoglobulin-binding domain B1 of streptococcal protein G), a short spacer and an extended gp41 HPbd peptide, beginning at residue 616. The globular GB1, in addition to facilitating isotope labeling of the peptides, was used to assist with solubility as an ideal fusion partner. The spacer sequence GGSGGS was placed to prevent any steric interference from GB1 on the HP–HPbd interaction. The length of the gp41 peptide component was adjusted to match the start of the CHR of each construct. GB1ⁱ⁶³⁵ is terminated at gp41 residue I635, immediately preceding the beginning of C28(L4)N50 at residue E637. GB1ⁱ⁶⁴² was matched with C21(L4)N50 which starts at residue S644. The sequences are shown in Figs. 5 and 6, and included salt-bridge modifications and a W623A mutation.

Fig. 5.

Interaction of inverted hairpin C28(L4)N50 with HPbd peptide GB1ⁱ⁶³⁵. (A) Sequence of GB1ⁱ⁶³⁵ (see text). GB1 stands for the 56-residue immunoglobulin-binding domain B1 of streptococcal protein G. (B) Schematic representation of the expected complex of GB1ⁱ⁶³⁵ with C28(L4)N50. (C) Overlay of a section of the ¹H, ¹⁵N-TROSY-HSQC spectra at 400 MHz of 100 µM ¹⁵N-labeled GB1ⁱ⁶³⁵ in the presence of 0, 50, 100, 200, 400 and 1350 µM ¹⁴N-labeled C28(L4)N50. Overlays are shown in various shades of gray, and shifted peaks are labeled. Resonances with no shift as a function of hairpin concentration appear as very dark peaks. (D) Selected peak shifts as a function of C28(L4)N50 concentration. Lines drawn through the data are the best fit to equation (2) for each resonance. A color version of this figure is available as Supplementary data at PEDS Online.

Fig. 6.

Overlay of central region of the ¹H–¹⁵H HSQC of 250 µM ¹⁵N-labeled GB1ⁱ⁶⁴² alone (gray) and in the presence (black) of an equimolar amount of C21(L4)N50. Assignments were transferred from the spectrum of GB1ⁱ⁶³⁵ and the resonances of 636-AEYTSRI-642 are unassigned. Assigned peaks which disappear in the presence of C21(L4)N50 are labeled in bolded italics. A color version of this figure is available as Supplementary data at PEDS Online.

The GB1ⁱ⁶³⁵ backbone resonance assignments were carried out via the HNCACB experiment using ¹³C–¹⁵N-labeled protein (Supplementary Table SIII). Seven ¹H, ¹⁵N-TROSY-HSQC spectra of 100 µM ¹⁵N-labeled GB1ⁱ⁶³⁵ were measured with increasing concentrations of unlabeled C28(L4)N50 varying from 0 to 1350 µM. Only resonances of the HPbd peptide region from residue 626 on were shifted with addition of C28(L4)N50; resonances of GB1, the linker region and first 10 CHR resonances remained unchanged (Fig. 5C). A global fit to the shift data gave K_D = 267 ± 25 µM (Fig. 5D). From the observed chemical shift differences, we estimated a lower limit of 460 s⁻¹ for the off-rate and 1.8 × 10⁶ M⁻¹ s⁻¹ for the on-rate. The low affinity interaction is in line with expectations for an unconstrained short gp41 peptide, which we and others (Cole and Garsky, 2001) have found to have no discernable binding in fluorescence studies using micromolar amounts of material. The study confirms accessibility of the pocket at the high concentrations required for NMR.

A 1:1 mixture of GB1ⁱ⁶⁴² and C21(L4)N50 resulted in broadening and shifting of resonances 622-KANH-625 and disappearance of all resonances from 626 to 642 of the gp41 HPbd segment (Fig. 6). This indicates that the complex is in intermediate exchange. The chemical shift difference observed for K622 ¹⁵NH in free GB1ⁱ⁶⁴² vs. 1:1 complex suggested an upper limit of 25 s⁻¹ for the off-rate. Assuming the same on-rate as for GB1ⁱ⁶³⁵ gave a ceiling for the K_D of ∼14 µM. This is in line with a previously observed K_D = 1.2 µM for the CHR peptide 617–642 binding to NHR trimer 560–591 (Gochin and Cai, 2009). There were some differences in the CHR sequence used previously, due to alternative helix stabilizing modifications and retention of W623. Changes that affect both electrostatics and helicity are likely to affect the binding affinity (Gochin and Cai, 2009). There was no evidence that the GB1 domain hindered or degraded the association of the gp41 CHR peptide with the NHR receptor, and clear evidence that C21(L4)N50 presented an integral binding domain for the extended HPbd.

Small-molecule–inverted hairpin interactions

To evaluate whether the protein constructs were presenting the expected binding sites for small-molecule detection and evaluation, we used fragments from a library that we have screened to detect both HP and CT binders. Figure 7 shows the fluorescence and NMR assays that were used for screening the 500-member library. These assays use bipyridylated (bpy) NHR peptides that spontaneously form trimers upon addition of metal ions, and have been described previously (Cai and Gochin, 2007; Gochin, 2012; Chu and Gochin, 2013). Briefly, HP binding was detected using HP-presenting trimer Fe(env2.0)₃ (HXB2 residues 560–584) and fluorescein labeled probe peptide C18-e2.0-FL (HXB2 residues 626–642) (Fig. 7A). The Fe(bpy)₃ group quenched the fluorescence of bound probe peptide, and displacement by compound binding caused a concomitant increase in fluorescence. The data in Fig. 7B show four fragments in the library that were detected as HP binders in a single point screening assay. Subsequent dose–response titrations confirmed the activity and established inhibition constants of 11, 60 and 140 µM for P5C04, P6G07 and P6F08, respectively (Supplementary Fig. S4). CT binding was established in an NMR experiment using a longer NHR trimer Fe(env3.0)₃ (HXB2 residues 560–591) and using a C-peptide C18-e3.0 (HXB2 residues 617–642) covering the HP (Fig. 7C). An example of a fragment P1C02 detected by this assay using WaterLOGSY NMR (Dalvit et al., 2001) is shown in Fig. 7D. In the WaterLOGSY experiment, saturation of the water resonance leads to magnetization transfer from bulk water to water molecules bound to protein near the ligand binding site and then to the ligand, resulting in resonance inversion. K_D was estimated to be 500 µM (Chu and Gochin, 2013). Structures of the fragments are provided in the Supplementary Fig. S4.

Fig. 7.

NMR and fluorescence assays that were used to detect fragments binding in the HP (A and B) and CT (C and D). The peptide assay components are shown in A and C and results for select compounds in B and D. Metal-stabilized NHR peptide trimers Fe(env2.0)₃ and Fe(env3.0)₃ were used as receptors to detect HP and CT binding, respectively, using probe peptides C18-e2.0-FL and C18-e3.0 (see text). Fluorescence assays were conducted using 7.2 µM receptor, 30 nM probe peptide and 1 mM inhibitors. NMR assays were conducted using 10 µM receptor, 30 µM probe peptide and 300 µM inhibitor. The position of the fluorescent label is shown as a star. A color version of this figure is available as Supplementary data at PEDS Online.

The reversed hairpin proteins were tested for small-molecule binding using HP binding fragments P5C04 and P6F08, and CT binding fragment P1C02, at a concentration of 300 µM. WaterLOGSY NMR experiments were conducted on fragments alone and in combinations, in the presence and absence of 10 µM C39(L4)N50 or 10 µM C21(L4)N49. Binding of each of the fragments to C21(L4)N49 was readily apparent, both alone and in combination (Fig. 8), while only CT binding ligand P1C02 interacted with C39(L4)N50, as expected from the design, and no peak inversion of P5C04 occurred (Supplementary Fig. S5). A combination of a CT-binding ligand with an HP-binding ligand yielded simultaneous inversion of both in the presence of C21(L4)N49, as did a combination of two HP binding ligands. Thus, the method did not discriminate between two ligands binding in adjacent pockets or competing for the same pocket (Sledz et al., 2012). In the absence of protein, no resonance inversion occurred, with almost complete cancelation of signal (Fig. 8C). This is an important negative control, since compounds that form slow tumbling molecular aggregates can yield false-positive results in the WaterLOGSY experiment. In addition, we measured three fragments that tested negative in the binding assays of Fig. 7, and did not observe resonance inversion in the WaterLOGSY experiment. Large shifts ± 0.4 ppm of P5C04 aromatic resonances occurred upon C21(L4)N49 binding, conditions in which <1% is bound. This implies very large chemical shifts for the complex of P5C04 in the HP. Overall, the data show that the swapped-domain constructs form trimers with selective exposure of HP and CT grooves, as designed.

Fig. 8.

Aromatic region of the 400 MHz ¹H NMR spectra of fragments P1C02, P5C04 and P6F08, each at a concentration of 300 µM in Tris-acetate buffer, pH 7.0, 5% DMSO. (A) 1D NMR spectrum of P1C02 and P5C04, (B) 1D NMR spectrum of P1C02 and P5C04 with 10 µM C21(L4)N49, (C) WaterLOGSY spectrum of P1C02 and P5C04, (D) WaterLOGSY spectrum of P1C02 and P5C04 with 10 µM C21(L4)N49 and (E) WaterLOGSY spectrum of P5C04 and P6F08 with 10 µM C21(L4)N49.

Conclusions

In this study, we have prepared constructs of the gp41 ectodomain in which the CHR and NHR domains are swapped in sequence. By changing the length of the CHR domain, we were able to expose different subdomains within the NHR groove, including the highly conserved HP, a target for small molecules, D-peptides (Eckert et al., 1999; Welch et al., 2010) and antibodies (Miller et al., 2005; Luftig et al., 2006; Sabin et al., 2010). We demonstrated a novel method for preparing constructs with an exposed HP in high yield and with ease of purification, using a cleavable cover peptide to shield the hydrophobic residues during overexpression in E. coli. By varying the length of the CHR, we have demonstrated selective exposure of grooves of the NHR. Analytical ultracentrifugation confirmed that the constructs trimerize, and NMR experiments with both small molecules and peptides confirmed that they exhibit binding sites for C-peptides and small-molecule fusion inhibitors targeting the NHR. The constructs are stable at pH 5.5 indefinitely, but we noticed a gradual loss of activity at pH 7 after a day. We attempted to increase solubility by adding residues Lys-Arg at the C-terminus, but this had little effect on overall solubility. Based on our data and previous observations on Protein-1 (Stewart et al., 2010), we surmise that the swapped-domain constructs form reverse hairpin trimers, shown (in monomeric form) in Fig. 2. Our reverse hairpin proteins displayed improved solution behavior over Protein-1, which was not a discrete trimer by analytical ultracentrifugation. We attribute the difference to stabilization of the trimer by the extended NHR CT domain in C28(L4)N50 and C21(L4)N50.

We demonstrated that the constructs’ HP appeared intact, as we could determine interactions with HPbd peptides of lengths matching the exposed pocket. A six residue linker GGSGGS connecting the GB1 domain prevented any interference of GB1 in HPbd peptide binding. HSQC experiments yielded a K_D of 267 µM for HPbd peptide terminated immediately after the WWI motif at residue 635; affinity improved by more than an order of magnitude with one additional heptad repeat, in line with previous studies (Mo et al., 2004; Gochin and Cai, 2009). In WaterLOGSY ligand-detected NMR experiments, we readily identified binders in a K_D range of 11–500 µM in the presence of reverse hairpin protein at 1/30th the concentration.

The biochemical and biophysical data presented here provide an avenue for examining details of the interaction surfaces presented by NHR and the molecules that interact with them. Crystallographic and NMR studies are planned to study these complexes. Structural studies of ligands binding in the HP and adjacent sites along the NHR could assist in discovery of potent non-peptide inhibitors of fusion. Constructs presenting exposed NHR grooves will be the subject of antiviral studies in a future study, since such molecules have the potential to be active against strains resistant to C-peptide inhibitors.

Supplementary data

Supplementary data are available at PEDS online.

Funding

We gratefully acknowledge the financial support of the National Institutes of Health (GM087998 to M.G.). Molecular graphic images and 2D NMR spectra were produced using the UCSF Chimera and Sparky packages from the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco (supported by NIH P41 RR-01081). SQZ was supported by NIH grant AI097051 awarded to Dr William F. DeGrado at UCSF, and the MRSEC program of NSF (DMR-1120901) through a grant to the LRSM at the University of Pennsylvania.