Calixarene-mediated assembly of a small antifungal protein

Synthetic macrocycles such as calixarenes and cucurbiturils are increasingly applied as mediators of protein assembly and crystallization. The macrocycle can facilitate assembly by providing a surface on which two or more proteins bind simultaneously. This work explores the capacity of the sulfonato-calix[n]arene (sclx_n) series to effect crystallization of PAF, a small, cationic antifungal protein. Co-crystallization with sclx₄, sclx₆ or sclx₈ led to high-resolution crystal structures. In the absence of sclx_n, diffraction-quality crystals of PAF were not obtained. Interestingly, all three sclx_n were bound to a similar patch on PAF. The largest and most flexible variant, sclx₈, yielded a dimer of PAF. Complex formation was evident in solution via NMR and ITC experiments, showing more pronounced effects with increasing macrocycle size. In agreement with the crystal structure, the ITC data suggested that sclx₈ acts as a bidentate ligand. The contributions of calixarene size/conformation to protein recognition and assembly are discussed. Finally, it is suggested that the conserved binding site for anionic calixarenes implicates this region of PAF in membrane binding, which is a prerequisite for antifungal activity.

1. Introduction

There is growing interest in the use of synthetic macrocycles as mediators of protein assembly (van Dun et al., 2017). The special case of protein crystallization (McPherson et al., 2011) has benefitted from `molecular glues' such as calixarenes and cucurbiturils that promote crystal packing (Guagnini et al., 2018; Rennie et al., 2018). The sulfonato-calix[n]arenes (sclx<sub>n</sub>, Fig. 1) are highly water-soluble, anionic macrocycles with diverse applications in the biosciences (Baldini et al., 2017; Giuliani et al., 2015; Guo & Liu, 2014). The hydrophobic core and the anionic rim of the calixarene can facilitate protein recognition, in particular, via the entrapment of arginine or lysine side chains (McGovern et al., 2012, 2014, 2015; Wang et al., 2016; Mallon et al., 2016; Rennie et al., 2017, 2018; Doolan et al., 2018; Alex et al., 2018). Consequently, sclx<sub>4</sub> and related compounds readily co-crystallize with the highly cationic cytochrome c and lysozyme (Alex et al., 2018; Doolan et al., 2018; McGovern et al., 2012, 2014, 2015). With increasing calixarene size there tends to be more pronounced effects; for example, phosphonato-calix[6]arene (pclx<sub>6</sub>) has an approximately tenfold increase in affinity (with respect to sclx<sub>4</sub>) and prompts dimerization of cytochrome c in solution (Rennie et al., 2017). Sulfonato-calix[8]arene (sclx<sub>8</sub>) on the other hand induces a tetramer of cytochrome c (Rennie et al., 2018). Furthermore, while calix[4]arene is locked in a bowl conformation, the larger calixarenes are flexible and adopt various conformations (Fig. 1) (Atwood et al., 1992; Dalgarno et al., 2003; Gutsche & Bauer, 1985; Liu et al., 2009; Perret et al., 2006; Rennie et al., 2017, 2018; Smith et al., 2006). Accordingly, sclx<sub>8</sub> can bind to cytochrome c either via an extended `pleated loop' or a collapsed `double cone' conformation, as shown using X-ray crystallography (Rennie et al., 2018).

Figure 1 Sulfonato-calix[n]arenes. (a) Molecular structures and (b) cone (sclx4), double partial-cone (sclx6) and double cone (sclx8) conformations.

We were motivated to characterize the sclx_n series with a single protein and thus investigate systematically how the calixarene size and flexibility influence protein recognition and assembly. Furthermore, we were interested in studying a protein for which a crystal structure was not available. Acknowledging the tendency of sclx_n to complex cationic proteins we chose the Penicillium antifungal protein (PAF) (Marx et al., 1995, 2008) as a test case. PAF is a small (∼6.2 kDa, 55 residues) lysine-rich protein (13 × Lys, pI ≃ 9) and a potent agent against Aspergillus species and dermatophytes (Binder et al., 2010; Leiter et al., 2005; Palicz et al., 2016). The NMR structure is a twisted β-barrel composed of five antiparallel β-strands and stabilized by three disulfide bridges (Batta et al., 2009; Fizil et al., 2015, 2018). Lys30, Phe31, Lys34, Lys35 and Lys38 (loop 3) belong to a conserved region of PAF that is important for antifungal activity (Batta et al., 2009; Sonderegger et al., 2016; Garrigues et al., 2017). Similar to defensins, the mechanism of antifungal action is postulated to require interaction with anionic components on the cell membrane (Binder et al., 2010; Garrigues et al., 2017; Silva et al., 2014). Recent X-ray crystal structures have revealed how defensin–phospholipid binding leads to oligomerization, suggesting a mechanism for membrane permeation (Poon et al., 2014; Kvansakul et al., 2016; Cools et al., 2017; Järvå et al., 2018). These observations provided further motivation to characterize PAF binding with anionic receptors.

Here, we report three PAF–sclx_n crystal structures, demonstrating the fitness of calixarenes as crystallization agents. Interestingly, all three calixarenes were bound to PAF, mainly at the conserved loop 3. A similar interaction site was determined by NMR studies; these results suggest that loop 3 is favoured for recognition by anionic receptors. The largest calixarene sclx₈ mediated a PAF dimer that was observed both crystallographically and in solution. The thermodynamics of PAF–sclx_n interactions were characterized by isothermal titration calorimetry, providing further evidence of PAF dimerization via sclx₈. The results are discussed in the context of protein assembly and membrane binding. Finally, insights into protein complexation by flexible calixarenes are provided, including the role of PEG fragments at the protein–calixarene interface.

2. Experimental

2.1. Materials

PAF was produced as described (Batta et al., 2009; Sonderegger et al., 2016). The calixarenes were purchased from TCI Chemicals. Stock solutions of sclx₄, sclx₆ and sclx₈ were prepared in water and the pH was adjusted to 6.0.

2.2. Crystallization trials

Co-crystallization experiments were performed by the hanging-drop vapour-diffusion method at 20°C. The reservoir solution was 20–30% PEG 3350 and 50 mM sodium acetate, pH 5.6. A range of protein (0.7–7.0 mM PAF) and ligand (5–40 mM sclx₄) concentrations were tested for PAF–sclx₄ co-crystallization. Drops were prepared by combining sequentially 1 µl each of reservoir solution, protein and sclx₄. Crystals grew at 7 mM PAF and 40 mM sclx₄. In the case of PAF–sclx₆ and PAF–sclx₈, the protein–ligand solutions were premixed before combining with the reservoir solution. Co-crystals were obtained with 10 mM sclx₆ and 40 mM sclx₈. Crystals grew in 4–5 days (sclx₄), 2–3 weeks (sclx₆) or 6–8 weeks (sclx₈).

The crystallization of ligand-free PAF (7 mM) was performed with an Oryx 8 Robot (Douglas Instruments) and a sparse matrix screen (JCSG++, Jena Bioscience). Spherulites were obtained in C6 (40% PEG 300, 100 mM potassium phosphate citrate pH 4.2) and needles grew in D7 (40% PEG 400, 100 mM Tris–HCl pH 8.5, 200 mM lithium sulfate). Manual crystallization trials under these conditions did not yield suitable crystals.

2.3. X-ray data collection

Crystals were cryo-protected in reservoir solution supplemented with 20% glycerol and cryo-cooled in liquid nitrogen. Diffraction data were collected at the SOLEIL synchrotron (France) to 1.30, 1.45 and 1.50 Å for PAF–sclx₄, PAF–sclx₆ and PAF–sclx₈, respectively. Datasets were collected using φ scans of 0.1° over 200° (PAF–sclx₄), 180° (PAF–sclx₆) and 110° (PAF–sclx₈) using an EIGER X 9M detector. In the case of pure PAF, a dataset extending to 3.0 Å was collected for the spherulites (condition C6), but was difficult to index/integrate in both XDS and iMOSFLM. The needle-like crystals (condition D7) did not diffract.

2.4. Structure determination

The observed reflections for PAF–sclx₄ were processed with XDS (Kabsch, 2010), whereas iMOSFLM (Battye et al., 2011) was used for the PAF–sclx₆ and PAF–sclx₈ datasets. In all cases, the data were scaled using POINTLESS (Evans, 2011) and AIMLESS (Evans & Murshudov, 2013). Xtriage (PHENIX, Adams et al., 2010) suggested pseudo-merohedral twinning for the PAF–sclx₄ data with twin law −h, −k, −h −l, and estimated twin fractions of 0.025 (Britton analyses), 0.066 (H-test) and 0.022 (maximum-likelihood method). The structure was determined by molecular replacement in PHASER (McCoy et al., 2007) by using the NMR structure (PDB reference 2mhv, conformer 1; Fizil et al., 2015) as the search model. A satisfactory solution (LLG, 134; TFZ, 7.4) was obtained with a search model in which residues 1–2, 17–24 and 47–49 were deleted and all six cysteines were replaced by alanine. The coordinates and restraints for sclx₄ (ligand ID T3Y) were added in COOT. Twin refinement did not result in any significant improvement in the electron density. No twinning was indicated for the PAF–sclx₆ or PAF–sclx₈ data. The structures were solved by molecular replacement using the structure of PAF–sclx₄ (devoid of sclx₄) as the search model. The coordinates for sclx₆ and sclx₈ were built in JLigand (Lebedev et al., 2012). High mosaic spread (0.3–0.9) in the PAF–sclx₈ dataset made it difficult to obtain better R values. Truncating the images with high mosaicity did not help in this respect. Iterative cycles of manual model building in COOT (Emsley et al., 2010) and refinement in BUSTER (Smart et al., 2012) were carried out until no further improvements in R_free and electron density were observed. The final structures were validated with MolProbity (Chen et al., 2010) and deposited in the Protein Data Bank as PAF–sclx₄ (PDB reference 6ha4), PAF–sclx₆ (PDB reference 6hah) and PAF–sclx₈ (PDB reference 6haj).

2.5. Accessible surface area calculations

The effect of sclx₄, sclx₆ and sclx₈ on the accessible surface area (ASA) of PAF residues in the crystal packing environments was determined in AreaIMol as described previously (Alex et al., 2018).

2.6. NMR spectroscopy

The sample conditions were 0.3 or 0.5 mM ¹⁵N-PAF in 10 mM sodium phosphate buffer at pH 6.0. NMR titrations were performed at 298 K using 0.5–1 µl aliquots of 50 mM stocks of sclx₄, sclx₆ or sclx₈. ¹H-¹⁵N HSQC spectra were acquired with spectral widths of 12 p.p.m. (¹H) and 19 p.p.m. (¹⁵N) using two scans and 128 increments on a Bruker Avance-II-500 NMR spectrometer. Ligand-induced chemical-shift perturbations were analysed in CCPN (Delaglio et al., 1995).

2.7. Isothermal titration calorimetry and data fitting

PAF samples were dissolved in 10 mM sodium phosphate pH 6.0. The same buffer was used to dilute stocks of sclx₄ (7.1 mM, PAF 0.5 mM), sclx₆ (3.6 mM, PAF 0.5 mM) and sclx₈ (2.5 mM, PAF 0.3 mM) to the required concentration. Samples were degassed prior to the titration. Measurements were made at 25°C using a Microcal ITC-200 instrument. Titrations were performed in duplicate with similar trends between each replicate. A single replicate from each calixarene was used for model fitting. Separate titrations of each calixarene into buffer confirmed that the heats of dilution were small, exothermic and approximately constant.

NITPIC (Keller et al., 2012) was used for baseline correction and integration of the thermograms. Pytc (Duvvuri et al., 2018) was used to perform model fitting and parameter estimation. The system of equations relating the independent variables of the model (total concentrations) to the experimental observations (heat generated during injections) for the single-site and bidentate-ligand models are as follows.

Single-site model,

where [P_T]_i is the total cell concentration of protein at the ith injection (independent variable), [L_T]_i is the total cell concentration of ligand at the ith injection (independent variable), K₁ is the equilibrium association constant (fit parameter), ΔH is the enthalpy (fit parameter) associated with K, V_cell is the volume of the cell, v_i is the volume of the ith injection, q_i is the heat generated from the ith injection (dependent variable) and q_dil is the heat of dilution (fit parameter, assumed to be constant)

Bidentate-ligand model,

where K₁ and K₂ are the microscopic equilibrium association constants (fit parameters), ΔH₁ and ΔH₂ are the enthalpies (fit parameters) associated with K₁ and K₂, respectively

The expressions for mass balance of the protein and ligand can be represented by equations (1) or (4). Equation (2) or (5) can be used to define the equilibrium constants. For the bidentate ligand model, equation (5) was solved numerically (the Levenberg–Marquardt algorithm) to yield the free-protein ([P]_i) and free-ligand ([L]_i) concentrations. The free concentrations were used to compute the concentrations of the other states via the equilibrium equations. The heat generated from a given injection was determined using either equations (3) or (6). Parameters were constrained to physically reasonable bounds (e.g. K₁ and K₂ values between 10² and 10¹⁰ M⁻¹) and best-fits were obtained by maximum likelihood starting from a range of initial estimates. Parameter errors and correlations were estimated using a Bayesian approach (Markov chain Monte Carlo simulations). The error for each integrated heat was determined using NITPIC (Keller et al., 2012).

3. Results and discussion

3.1. PAF–sclx_n co-crystallization

Pure PAF proved to be recalcitrant to crystallization. A sparse-matrix screen yielded spherulites or needle-like crystals only (see experimental). In contrast, PAF–sclx₄ mixtures were crystallized readily from solutions containing PEG and sodium acetate. PAF–sclx₄, PAF–sclx₆ and PAF–sclx₈ co-crystals were obtained at 28–30% PEG 3350 and 50 mM sodium acetate pH 5.6 (Fig. S1 and Table S1 of the supporting information).

3.2. Data collection and model building

Datasets extending to 1.30, 1.45 and 1.50 Å resolution were collected from monoclinic (P12₁1) PAF–sclx₄, PAF–sclx₆ and hexagonal (P6₁) PAF–sclx₈ co-crystals, respectively (Table S1). The PAF–sclx₄ structure was determined using the NMR coordinates (PDB reference 2mhv; Fizil et al., 2015) as the search model. To obtain a satisfactory solution it was necessary to delete two loops and replace all six cysteines with alanines. After several rounds of model building and refinement a complete PAF structure was obtained. This model was used to solve the PAF–sclx₆ and PAF–sclx₈ structures. The PAF fold and the three disulfide bridges in the X-ray structures were consistent with the NMR model (Batta et al., 2009; Fizil et al., 2015, 2018). Interestingly, the fold was altered slightly in response to sclx_n binding (Fig. S2). Superposition of the three structures revealed a C^α r.m.s.d. of 0.54 Å (PAF–sclx₆) and 0.78 Å (PAF–sclx₈) relative to PAF–sclx₄, with the largest differences at loops 2, 3 and 4. The calculated energies of the disulfide bonds (Schmidt et al., 2006) were approximately threefold lower in the X-ray structures compared with the NMR structure (Table S2).

In contrast to the PAF–sclx_n crystals, the spherulites and needles of pure PAF failed to provide a usable dataset. The needles did not diffract and the spherulites yielded a 3.0 Å resolution dataset which proved difficult to index and integrate. The difficulty in obtaining suitable crystals of pure PAF suggests that the calixarene facilitates protein assembly and crystallization (Alex et al., 2018; Doolan et al., 2018; McGovern et al., 2012, 2014, 2015; Rennie et al., 2017, 2018).

3.3. Different calixarene, similar binding site

The asymmetric unit of the PAF–sclx_n complexes comprised one (in the case of PAF–sclx₄ and PAF–sclx₆) or two (PAF–sclx₈) molecules of PAF. Each structure contained one calixarene, as shown by the 2F_o—F_c electron-density maps (Figs. 2 and S1). Additional electron density adjacent to sclx₆ and sclx₈ was modelled as a PEG fragment equivalent to tetraethylene glycol (EG4) and heptaethylene glycol (EG7), respectively (Figs. 2 and 3). Sclx₄, locked in the cone conformation, encapsulates the side chain of a single lysine (Lys30), as observed previously in different protein-clx₄ complexes (Alex et al., 2018; Doolan et al., 2018; McGovern et al., 2012, 2014, 2015). The larger flexible sclx₆ and sclx₈ adopted distinct conformations and bound at least two lysines. Sclx₆ was in the double partial-cone conformation (Atwood et al., 1992; Dalgarno et al., 2003), with three sulfonates pointed upwards and three pointed downwards [Figs. 1(b) and 2(b)]. Sclx₈ adopted the double cone conformation (Liu et al., 2009; Perret et al., 2006; Smith et al., 2006), with each half of the molecule acting like a calix[4]arene to bind one PAF molecule, thus mediating a crystallographic dimer [Fig. 2(c)].

Figure 2 Binding-site interactions in PAF–sclxn. (a) sclx4, (b) sclx6 and (c) sclx8 binding to PAF at Lys30. Note the altered conformations of Lys30 and Phe31 in each structure, while Pro29 provides a rigid hydrophobic surface for face-to-face interaction with sclx6 and sclx8. In PAF–sclx8, two protein chains interact with the calixarene. PEG fragments equivalent to tetraethylene glycol and heptaethylene glycol were bound to sclx6 and sclx8, respectively.

Figure 3 Protein–PEG–calixarene interfaces. The protein–calixarene interfaces are completed by a PEG fragment in (a) PAF–sclx6 and (b) PAF–sclx8. Lys9 N^ζ simultaneously forms ion–dipole bonds to the PEG (crown-ether-like complex) and a salt bridge to one sulfonate. CH—π and lone-pair—π bonds also occur between PEG and the calixarene phenolic rings.

All three calixarenes bound to Lys30, while interacting also with neighbouring residues as well as other proteins (symmetry mates) in the crystal packing. Depending on the ligand size/conformation, the noncovalent contacts varied in their type and multiplicity. The PAF–sclx₄ complex [Fig. 2(a)] was similar to cytochrome c–sclx₄ (McGovern et al., 2012), involving a salt bridge and CH—π/cation—π bonds with the encapsulated lysine. Hydrogen bonds to the backbone amide NHs of Lys30, Phe31 and Asp32 were evident and the aromatic ring of Phe31 was in van der Waals contact with an sclx₄ methylene bridge. Considering symmetry mates [Fig. 4(a)], sclx₄ formed substantial interfaces (>150 Ų) with three proteins. Interestingly, a salt bridge was formed with the N^α of Ala1. Salt bridges also occurred with Lys2, Lys17, Lys22 and Lys35, emphasizing a substantial charge–charge component to complexation. In total, the protein–sclx₄ interfaces buried ∼660 Ų of protein.

Figure 4 Calixarenes as molecular glues. The crystal packing is dominated by PAF–sclxn interactions in (a) PAF–sclx4, (b) PAF–sclx6 and (c) PAF–sclx8. This observation suggests that the calixarene acts as a molecular glue in protein assembly. Proteins, calixarenes and unit-cell axes are depicted in grey, green and blue, respectively. The PEG fragments are depicted as sticks.

Sclx₆ (1.5 times larger than sclx₄) also completely encaged Lys30 [Fig. 2(b)]. However, one wall of the calixarene cage was composed of three phenolic groups. The phenolic oxygens were in van der Waals contact with the C^β, C^γ and C^δ of Lys30, indicative of CH⋯O hydrogen bonding and the Lys30 N^α was hydrogen bonded to a phenolic OH (rather than to a sulfonate). Other differences, with respect to sclx₄, were water-mediated salt bridges between Lys30 N^ζ and two sulfonates and a weak π–π interaction with Phe31 [Fig. 2(b)]. The adjacent residue Pro29 was also important for calixarene binding (vide infra). In terms of crystal packing [Fig. 4(b)], the larger sclx₆ was nestled between five proteins and formed numerous salt bridges (with Lys6, Lys9, Lys11, Lys27, Lys38, Lys42). The resulting protein–ligand contacts mask ∼970 Ų of protein surface. Compared with sclx₄, the more extensive interactions exhibited by sclx₆ may explain why four times less ligand was required to achieve crystal growth (see experimental and Table S1).

The interactions of sclx₈ with PAF were similar to those observed with sclx₆, though less extensive. At twice the size of sclx₄ it might be expected that sclx₈ would mask a larger protein surface; however, sclx₈ formed a PAF dimer [Figs. 2(c) and 4(c)] resulting in a total protein surface coverage of ∼950 Ų. The double-cone conformation (compared with the `pleated loop', Rennie et al., 2018) adopted by sclx₈ minimized its contact with protein surfaces. Salt-bridge interactions involved up to three lysines from each monomer. Here, again a hydrogen bond was formed between the Lys30 N^α and a phenolic OH. In one of the protein chains Phe31 formed an edge-to-face interaction with an sclx₈ phenolic ring. In protein chain B, Phe31 was disordered [Fig. 2(c)].

In complex with PAF, sclx₄, sclx₆ and sclx₈ contributed an additional surface of ∼550, ∼850 and ∼1290 Ų to the protein, respectively (calculated for a single protein). The exposed calixarene surface is a relatively homogenous `mask' that is conducive to forming noncovalent bridges with other proteins. Apparently, the calixarene acts as molecular glue (Fig. 4) by providing a patch that mediates protein assembly (subsequently driving protein crystallization) in a special case of the `patchy particle model' (Alex et al., 2018; Fusco et al., 2014; James et al., 2015; Staneva & Frenkel, 2015; Derewenda & Godzik, 2017).

The presence of PEG fragments (EG4 and EG7) markedly distinguished the PAF–sclx₆ and PAF–sclx₈ complexes (Fig. 3). The PEG–calixarene interaction involved lone-pair–π (Jain et al., 2009) and CH–π bonds, while the PEG–protein contacts included hydrogen bonds between the oxygen lone pairs and Lys9 (Lys9 N^ζ⋯O—PEG = 3.0–3.3 Å). This crown-ether like Lys9–PEG interaction resembles the binding of lysine to 18-crown-6 (PDB entry 3wur; Lee et al., 2014). A heptaethylene glycol fragment has been observed bound to an antibody (PDB entry 2ajs; Zhu et al., 2006), where it adopted a crown-ether like conformation, compared with the extended conformation in PAF–sclx₈. In addition, a crystal structure of an SH3 domain (PDB entry 5xg9; Gautam et al., 2017) revealed various PEG fragments at protein–protein interfaces. These examples suggest that the role of PEG is as an interface `filler' and possibly the PEG fragments (Fig. 3) contribute towards calixarene conformation selection/stability.

3.4. Selectivity of PAF–sclx_n complexation, why Lys30?

Considering that PAF contains 13 lysines the question arises as to why Lys30 was selected by sclx_n. ASA calculations were used to probe the selectivity of sclx_n for the Pro29-Lys30-Phe31 patch over other possible binding sites (Fig. 4). The calculations accounted for contributions from symmetry mates in the crystal packing (Alex et al., 2018). The effect of ligand binding on the ASA of all Lys, Pro, Phe and Tyr residues is plotted in Fig. 5. At least half of the lysines, including Lys30, are highly exposed (ASA ≥ 125 Ų) in each structure in the absence of sclx_n. This observation suggests that steric accessibility (McGovern et al., 2014) was not the determining factor in sclx_n selectivity. For example, Lys2 (>150 Ų) was significantly masked (ΔASA ≥ 15%) by binding with sclx₄ only. Perhaps a salt-bridge interaction with Asp46 reduced the availability of Lys2 in the other complexes. In contrast, Lys30 was strongly affected by all three calixarenes (ΔASA up to 80%). Adjacent residue Lys27 was also strongly affected in the complexes with sclx₆ and sclx₈. The differences in the degree of masking can be attributed to the calixarene sizes (small, sclx₄) and conformations (`double cone', sclx₈). However, sclx₈ had more in common with sclx₆ than sclx₄. For example, Lys9, Lys11 and Lys38 were 30–50% buried by sclx₆ or sclx₈, while sclx₄ had no effect on these residues. Overall, calixarene binding resulted in significant masking of five (sclx₄), eight (sclx₆) and six (sclx₈) lysines.

Figure 5 ASA plots. Accessibility of Lys, Pro, Phe and Tyr residues in ligand-free (black) and ligand-bound (grey) PAF. The PAF–sclx8 data correspond to chain A.

PAF has five aromatic residues, Phe25, Phe31, Tyr3, Tyr16 and Tyr48 (Fig. 5); the latter is highly solvent exposed (∼200 Ų) and might be expected to interact with sclx_n. However, only minor contributions were evident (Fig. S3). Phe31 was the dominant aromatic residue for sclx_n complexation. The adjacent Lys30, Lys34 and Lys35 may facilitate (via charge–charge interactions) calixarene binding here, compared with Tyr48, which is proximal to Lys2 only. The contribution of Pro29 merits special attention as it completes the binding site for both sclx₆ and sclx₈ via face-to-face hydrophobic stacks with a phenolic ring [Figs. 2(b) and 2(c)]. These interactions are reminiscent of polyphenol binding to proline-rich proteins (Baxter et al., 1997; Charlton et al., 2002; Quideau et al., 2011). The rigid pyrrolidine ring appears to provide a stable platform for binding the `floppy' sclx₆ or sclx₈. Thus, it is perhaps unsurprising that the only proline residue in PAF was involved at the binding site.

As such, it appears to be the combination of the Pro29-Lys30-Phe31 motif and adjacent lysines (charge–charge interactions) that stabilize sclx_n binding and impart selectivity. This region has been implicated in PAF function, with decreased antifungal activity when Phe31, Lys35 or Lys38 were mutated to Asn or Ala (Batta et al., 2009; Sonderegger et al., 2016; Garrigues et al., 2017). The selectivity of the anionic calixarenes for this site suggests that it may be involved in cell membrane binding and permeation as required for antifungal activity.

3.5. NMR characterization and comparison with the solid state

PAF–calixarene binding in solution was assessed by NMR spectroscopy. Titrations were performed by the addition of microlitre aliquots of sclx_n to ¹⁵N-labelled PAF, which was monitored by ¹H-¹⁵N HSQC spectroscopy (Fizil et al., 2018; McGovern et al., 2012). The overlaid spectra (Fig. 6) revealed increasing chemical-shift perturbations (Δδ) as a function of sclx₄ or sclx₆ concentration, indicative of fast to intermediate exchange between the ligand-free and ligand-bound states. Some biphasic shifts were evident for sclx₆. Severe broadening effects were observed with ≥0.3 eq sclx₈, indicative of a slow-exchange process and suggesting the possibility of ligand-mediated oligomerization (Doolan et al., 2018; Fonseca-Ornelas et al., 2017; Mallon et al., 2016; Rennie et al., 2017, 2018).

Figure 6 NMR characterization of PAF–sclxn­ complexation. (a) Region from overlaid ^1H-^15N HSQC spectra of pure PAF (black contours) and in the presence of 0.1–0.6 mM ligand (coloured scale). Biphasic shifts occurred for resonances Lys30, Lys34 and Cys36 in the presence of sclx6. Resonances Lys11, Cys28, Lys30, Lys34 and Cys36 were broadened at 0.3 mM sclx8, while resonances Thr8, Lys11, Asp32 and Thr37 were broadened beyond detection at 0.6 mM sclx8. (b) Plots of chemical-shift perturbations measured for PAF backbone amides in the presence of 0.6 mM sclx4, sclx6 or sclx8. Blanks correspond to Pro30 and undetectable resonances (due to broadening).

The Δδ plot (Fig. 6) shows a clear selectivity for sclx₄ binding to Lys30 and neighbouring residues 31–36. In the crystal structure, all of these residues occurred in the vicinity of sclx₄. Significant Δδ were observed also for the C-terminal Val52 and Cys54, which are further from the crystallographic binding site. However, both of these residues are adjacent to Pro29, and Cys54 is hydrogen bonded to Lys34, suggesting a mechanism for how these resonances sense ligand binding. In the presence of sclx₆, the Δδ plot again shows a preference for binding around Lys30 as well as effects at the C-terminus (Val52 N^α is hydrogen bonded to sclx₆). However, compared with sclx₄, the shifts are 2–4 times larger and other segments of the primary structure (residues 6–13 and 42–45) were also affected. These two regions correspond to additional sclx₆ binding sites evident in the crystal packing. Therefore, the NMR data suggests that the PAF–sclx₆ interaction fluctuates, with the calixarene exploring different patches on the protein surface, as observed previously for cytochrome c–sclx₄ complexes (Doolan et al., 2018; McGovern et al., 2012). Judging from the magnitude of the shifts, binding to Lys30 is preferred while a weaker interaction occurred at a patch involving Lys6 and Lys42.

The titrations with sclx₈ resulted in different effects. In addition to pronounced perturbations of Lys30 and neighbours, substantial broadening effects occurred. Cys28, Lys30, Lys34 and Cys36 broadened at 0.3 mM, and Thr8, Lys11, Asp32 and Thr37 broadened beyond detection at 0.6 mM sclx₈. These eight residues are located at the crystallographically defined binding site. Thus, the broadening effects may be indicative of PAF dimerization, consistent with the sclx₈-mediated dimer in the crystal structure [Fig. 2(c)]. Previously, we observed a complete loss of the HSQC spectrum of cytochrome c in complex with pclx₆, which also yielded a dimer in the solid state (Rennie et al., 2017).

3.6. Thermodynamics of PAF–sclx_n complexation

Isothermal titration calorimetry was used to characterize the PAF–sclx_n binding affinities and stoichiometries (Fig. 7). The data were fitted to a single-site or a bidentate-ligand model. The latter model describes a bidentate ligand that can bind two protein molecules and was necessary to describe the obviously biphasic data for sclx₈. The choice of this model is supported by the observation of a PAF–sclx₈–PAF dimer in the crystal structure, and by the spectral broadening in the NMR experiments. All of the fit parameters were well determined by the data (Table 1), with parameter errors assessed by Bayesian methods (Patil et al., 2010).

Table 1
Thermodynamics of PAF–sclx_n complexation determined by ITC

Fit values are median (2.5% quantile, 97.5% quantile) from the Markov chain Monte Carlo method. In the case of sclx₆, the fit parameters for both models are shown.

[Ligand]† (µM)	[PAF]‡ (µM)	Kd (µM)	ΔH (kJ mol^−1)	TΔS (kJ mol^−1)
PAF–sclx4 (single-site model)
7143 (1248)	500 (412)	107.0 (0.0, 0.0)	−16.9 (0.1, 0.1)	−5.6 (0.2, 0.2)
PAF–sclx6 (single-site model)
3623 (633)	500 (412)	15.4 (0.0, 0.0)	−28.2 (0.2, 0.1)	−0.7 (0.2, 0.2)
PAF–sclx6 (bidentate-ligand model)
3623 (633)	500 (412)	47.8 (0.0, 0.0)	−9.2 (0.1, 0.2)	−15.4 (0.4, 0.3)
		45.8 (3.5, 4.1)	−20.1 (0.4, 0.4)	−4.6 (0.6, 0.6)
PAF–sclx8 (bidentate-ligand model)
2500 (437)	300 (247)	10.6 (1.3, 1.4)	−3.5 (0.3, 0.3)	−24.8 (0.3, 0.3)
		33.5 (4.8, 6.5)	−36.0 (1.7, 1.4)	10.5 (1.8, 2.1)

†Calixarene concentrations in the syringe; the final concentrations are indicated in parentheses.
‡Protein concentrations in the cell; the final concentrations are indicated in parentheses.

Figure 7 ITC analysis of PAF–sclxn complexation. Top panels show the baseline-corrected thermograms for injections of sclx4, sclx6 or sclx8 into PAF. Bottom panels are the observed heats (data points) and the fits (solid line) for single-site (sclx4) and bidentate-ligand (sclx8) models.

The isotherms for sclx₄ injected into PAF were fitted to a single-site binding model with K_d ∼110 µM. In contrast, the isotherms for sclx₈ were biphasic (Brautigam, 2015) and fitted to a bidentate ligand model with K_d values of ∼10 and ∼30 µM, for binding the first and second molecule of PAF, respectively. The isotherms for sclx₆ were intermediate between sclx₄ and sclx₈, suggesting that this ligand may exhibit weak bidentate binding. A satisfactory fit for this data was not obtained with either model. The ITC data demonstrate an increasing affinity for PAF as the calixarene size increases and a switch in binding mode from the small, rigid sclx₄ (single site) to the large, flexible sclx₈ (bidentate).

4. Conclusions

Using a combination of X-ray crystallography and NMR spectroscopy it was demonstrated that the sclx_n series binds selectively to the highly cationic PAF. Despite the varying size and conformational flexibility, sclx₄, sclx₆ and sclx₈ bound similarly the Pro29-Lys30-Phe31 motif in loop 3. The selectivity of the anionic calixarenes for this motif, and the role of loop 3 in antifungal activity, suggests that this region may be required for membrane binding. In addition to charge–charge interactions (showed by numerous lysine-to-sulfonate salt bridges), other noncovalent bonds including CH–π and π–π (via Pro29 and Phe31, respectively) participated in ligand stabilization. The presence of PEG fragments at the protein–sclx₆ and protein–sclx₈ interfaces suggests that PEG acts as a `filler' to complete the binding site, potentially reinforcing the calixarene conformation.

The structures of all three PAF–sclx_n co-crystals highlight the potential of calixarenes as a `sticky patch' on the protein surface that facilitates assembly and crystallization. In the case of the sclx₄ and sclx₆ co-crystals (P12₁1), it is evident that the calixarene is a dominant contributor to the crystal packing (Fig. 4). Similarly in the sclx₈ structure (P6₁), the packing involves substantial protein–calixarene contacts, and the structure is interesting as sclx₈ mediates a PAF dimer. Previously, we found that sclx₈ mediates a tetramer of cytochrome c (Rennie et al., 2018). Generally, it seems that calixarene-mediated protein crystallization may be a special case of the patchy particle model for protein assembly (Alex et al., 2018; Fusco et al., 2014; James et al., 2015; Staneva & Frenkel, 2015; Derewenda & Godzik, 2017). Considering that PAF alone did not yield diffraction-quality crystals, we conclude that co-crystallization with sclx_n was beneficial. Anionic calixarenes may generally facilitate crystallization and structure determination of small cationic proteins.

The binding surfaces observed in the NMR experiments were consistent with the X-ray data. However, the NMR effects were more pronounced with increasing calixarene size, suggesting that the larger calixarenes mask a greater portion of the protein surface and/or lead to assembly in solution. Similarly, the ITC experiments revealed tighter affinities and more complex effects with increasing calixarene size. In particular, sclx₈ behaved as a bidentate ligand that facilitated PAF dimerization. These data add to the growing evidence of calixarene-mediated protein assembly in solution (Doolan et al., 2018; Rennie et al., 2017, 2018). In terms of the biological relevance of these data it is noted that defensin oligomerization (upon phospholipid binding) has implications for antifungal activity (Poon et al., 2014; Järvå et al., 2018). Perhaps calixarenes can be used to modulate the activity of PAF and related proteins.