A cross-strand Trp–Trp pair stabilizes the hPin1 WW domain at the expense of function

Abstract

Using the human Pin1 WW domain (hPin1 WW), we show that replacement of two nearest neighbor non-hydrogen-bonded residues on adjacent β-strands with tryptophan (Trp) residues increases β-sheet thermodynamic stability by 4.8 kJ mol⁻¹ at physiological temperature. One-dimensional NMR studies confirmed that introduction of the Trp–Trp pair does not globally perturb the structure of the triple-stranded β-sheet, while circular dichroism studies suggest that the engineered cross-strand Trp–Trp pair adopts a side-chain conformation similar to that first reported for a designed “Trp-zipper” β-hairpin peptide, wherein the indole side chains stack perpendicular to each other. Even though the mutated side chains in wild-type hPin1 WW are not conserved among WW domains and compose the β-sheet surface opposite to that responsible for ligand binding, introduction of the cross-strand Trp–Trp pair effectively eliminates hPin1 WW function as assessed by the loss of binding affinity toward a natural peptide ligand. Maximizing both thermodynamic stability and the domain function of hPin1 WW by the above mentioned approach appears to be difficult, analogous to the situation with loop 1 optimization explored previously. That introduction of a non-hydrogen-bonded cross-strand Trp–Trp pair within the hPin1 WW domain eliminates function may provide a rationale for why this energetically favorable pairwise interaction has not yet been identified in WW domains or any other biologically evolved protein with known three-dimensional structure.

The two predominant protein secondary structural motifs are α-helices and β-sheets. A combination of the early availability of suitable α-helical model systems and sustained research has resulted in a detailed understanding of α-helix folding (Zimm et al. 1959; Schwarz 1965; Gruenewald et al. 1979; Chakrabartty and Baldwin 1995; Rohl and Baldwin 1998). Comparatively little is known about β-sheet structure acquisition, although the relatively recent introduction of several small β-sheet model systems into the literature has afforded detailed insights into the forces and interactions that influence β-sheet folding and stability (Minor Jr. and Kim 1994a,b; Smith et al. 1994; Munoz et al. 1997; Ramirez-Alvarado et al. 1997; Gellman 1998; Munoz et al. 1998; Searle and Ciani 2004). Reverse turn or loop formation is often rate limiting for β-sheet folding (Grantcharova et al. 2000; Kim et al. 2000; Ferguson et al. 2001; Jäger et al. 2001; Nauli et al. 2001; Du et al. 2006), although hydrophobic interactions between side chains of neighboring β-strands, interstrand hydrogen bonds, and the intrinsic propensity of the amino acids to adopt a β-strand conformation also contribute significantly to folding energetics (de Alba et al. 1997; Tatko and Waters 2002; Ciani et al. 2003; Syud et al. 2003; Santiveri et al. 2004).

Recently, Cochran and coworkers compared the stability of 19 β-hairpins differing in amino acid composition at one position. Their studies led to the unexpected observation that Trp, when positioned in a non-hydrogen-bonded (NHB) strand site, was the most stabilizing residue tested, irrespective of the composition of its nearest neighbor (Cochran et al. 2001b; Russell and Cochran 2001). The solution structure of a β-hairpin containing two NHB cross-strand Trp–Trp pairs (named “Trp-zipper” hereafter) (Cochran et al. 2001a) revealed that the indole side chains of each Trp–Trp pair interact with each other by a perpendicular stacking of the solvent-exposed indole rings (Fig. 1A), leading to dramatic increases in thermal stability of the host hairpin sequence.

Figure 1.

(A) Structural depiction of the “Trp-zipper” β-hairpin (Cochran et al. 2001a). Residues that adopt β-strand dihedral angles are color-coded in orange and the loop and N- and C-terminal residues are color-coded gray. The side chains of the four non-hydrogen-bonded Trp residues are color-coded blue, shown explicitly in stick-mode representation, and labeled in single letter code. (B) Structural depiction of the hPin1 WW domain (Ranganathan et al. 1997), color-coded as in A. The side chains of the two non-hydrogen-bonded residues Met15 and Val22 (color-coded blue), as well as neighboring residues Lys13 and Tyr24 are shown in stick-mode representation (single letter code). (C) Sequence alignment of the Pin WW variants 1–4 employed in this study.

To test whether a NHB cross-strand Trp–Trp pair can also be used to stabilize a well-defined cooperatively folded protein domain, we introduced a single NHB cross-strand Trp–Trp pair into a solvent-exposed position of an autonomously folded 34-residue three-stranded β-sheet, the WW domain from the human cis/trans-isomerase hPin1 (hPin1 WW hereafter) (Fig. 1B; Sudol 1996^; Ranganathan et al. 1997^; Jäger et al. 2001^; Macias et al. 2002). The influence of the introduction of a NHB cross-strand Trp–Trp pair on domain stability was assessed by equilibrium denaturation experiments, whereas its effect on function (binding of a natural YSPTpSPS phosphopeptide ligand from the RNA polymerase II carboxy-terminal repeat) was ascertained by isothermal titration calorimetry.

Results and Discussion

Introducing a NHB Trp–Trp cross-strand pair into the WW domain

The structure of hPin1 WW is depicted in Figure 1B. The Met15 (β-strand 1) and Val22 (β-strand 2) NHB cross-strand residues (side-chain substructures are colored blue) are those to be replaced by Trp residues in this study. These side chains do not directly contribute to ligand binding and are partially solvent-accessible (Met15: 47%, Val22: 41%). Van der Waals contacts beyond the C_β atoms exist between Met15 and Val22, Val22 and Lys13, and Met15 and Tyr24. A statistical analysis of the >200 WW domain family sequences suggests that conservation at Met15 and Val22 (or topologically equivalent positions) is low (<25%), and none of the WW-domain sequences in the database have a cross-strand Trp15–Trp22 pair or the equivalent. Met15 and Val22 in wild-type hPin1 WW (variant 1) were replaced either simultaneously or individually with Trp residues affording variants 2–4 (Fig. 1C). Analytical ultracentrifugation was employed to ascertain that variants 1–4 remain monomeric up to a protein concentration of at least 100 μM (data not shown).

Circular dichroism analysis

The far-UV CD spectrum of the wild-type WW domain (variant 1) exhibits an unusual but characteristic positive ellipticity at ∼220–240 nm with a maximum at 227 nm (Fig. 2A), attributable to contributions by the two conserved Trp residues (Trp11, Trp34), for which the domain was named. The stated origin of the far UV CD maximum is in good agreement with the sum of the far-UV CD spectrum of two variants in which one of the Trp residues (Trp11Phe, Trp34Ala) was mutated to Phe (data not shown). The maximum disappears upon thermal denaturation and serves as a probe to monitor unfolding of the WW domain.

Figure 2.

Circular dichroism analysis. (A) Far-UV CD spectra of the native and unfolded variant 1 (black and blue, respectively) and native and unfolded variant 2 (red and green, respectively). Spectra of native proteins were collected at 10°C and spectra of unfolded proteins were collected at 98°C (2-mm path length, 8 μM protein). (B) Difference spectra calculated from variant 2-1 (black), variant 3-1 (red), and variant 4-1 (blue). (C) Near-UV CD spectra (10-mm pathlength, 50 μM protein) of native and unfolded variant 1 (black and blue, respectively) and native and unfolded variant 2 (red and green, respectively). (D) Difference spectra calculated from variant 2-1 under folding conditions (black) and unfolding conditions (red).

The far-UV CD spectrum of variant 2 exhibits an even more pronounced maximum at ∼229 nm and exhibits similar fine structure in the 190–210 nm range (Fig. 2A). A difference spectrum, variants 2-1 removing the WW-fold component, exhibits intense exciton-coupled bands at 215 (minimum) and 229 nm (maximum), indicating an interaction between the indole chromophores (Fig. 2B). The difference spectrum closely resembles the far-UV CD spectrum of the Trp-zipper hairpin (Cochran et al. 2001a). Difference spectra (variants 3-1 and 4-1) with weak uninterpretable amplitude were observed for variants 3 and 4 (Fig. 2B).

The near-UV CD spectrum of variant 1 exhibits a maximum at ∼268 nm, with little fine structure (Fig. 2C). The near-UV spectrum of variant 2 resembles that of variant 1, but displays additional fine structure in the 285–310 nm range. The difference spectrum (variant 2 − 1), resembles the near-UV spectra of Cochran's Trp-zipper β-hairpin peptide, with minima at 283 and 293 nm (Cochran et al. 2001a). This near-UV CD difference spectrum essentially vanishes upon thermal unfolding, suggesting that the interactions between the NHB-indole side chains in the folded variant 2 are largely disrupted upon heat denaturation.

1-D NMR analysis

The ¹H NMR spectra exhibited by wild-type hPin1 WW domain (variant 1; Fig. 3B,D) and variant 2 (Fig. 3A,C) incorporating the NHB Trp–Trp pair exhibit chemical shift dispersion and downfield aromatic resonances expected from a folded WW domain (Kowalski et al. 2002). The superior chemical shift dispersion, especially in the downfield (Fig. 3A) and upfield (Fig. 3C) regions of variant 2 is inconsistent with significant conformational changes such as β-strand 3 dissociating from β-strands 1 and 2 as a consequence of possible changes in the twist of the first hairpin substructure of the triple-stranded β-sheet.

Figure 3.

1-D proton NMR analysis of selected hPin1 WW domains. (A) Variant 2 (33.2 μM) downfield region of the 800-MHz spectrum. (B) Variant 1 (7.6 μM) downfield region of the 800-MHz spectrum. (C) Variant 2 upfield region of the 800-MHz spectrum. (D) Variant 1 upfield region of the 800-MHz spectrum.

There is no NMR evidence for dimerization or aggregation of variant 1 (7.6 μM) or variant 2 (33.2 μM), consistent with the analytical ultracentrifugation results described above. As previously demonstrated for the structurally related WW domain from the human Yes-associated protein (hYap) (Koepf et al. 1999), the two downfield resonances in hPin1 WW variant 1 with chemical shifts of 9.90 and 10.40 ppm most likely result from the indole N-H protons of Trp34 and Trp11, respectively. Resonances with essentially identical shifts are also visible in variant 2, clearly demonstrating that the two conserved Trps exhibit very similar structural microenvironments in both domains. The two additional downfield resonances in variant 2 with chemical shifts of 9.95 and 10.20 ppm most likely result from N–H indole resonances of the 15/22 NHB cross-strand Trp residues.

Fluoresence

The fluorescence spectrum of variant 2 is only slightly higher in intensity than the spectrum obtained with variant 1 (data not shown), suggesting considerable self-quenching of the interacting solvent-accessible indole side chains and a very similar overall tertiary structure.

The 1-D NMR and CD spectra fully support the idea that the overall structures of hPin1 WW variants 1 and 2 are very similar. Furthermore, the circular dichroism spectra of variant 2 suggests a close interaction of the NHB indole side chains in a cross-strand geometry. A direct confirmation of this hypothesis by X-ray crystallography has so far been unsuccessful, owing to the unavailability of suitable crystals (J. Noel, pers. comm.).

Thermodynamic analysis

Equilibrium thermal denaturation curves for variants 1 and 2 (far-UV CD at 229 nm, 20 μM protein) are shown in Figure 4A. The solid lines represent fits of the data to a two-state unfolding model (Table 1). Normalized equilibrium unfolding curves of variants 1 and 2 are displayed in Figure 4B. The midpoint of the thermal folding transition (T_M; defined as the temperature at which ΔG = 0) increases from 59°C (variant 1) to 75°C (variant 2), and at 67°C (the average T_M of variants 1 and 2), the stability of variant 2 (ΔG = −3.7 kJ mol⁻¹) exceeds that of variant 1 by 7.3 kJ mol⁻¹ (Table 1).

Figure 4.

Summary of thermodynamic analysis. (A) Equilibrium denaturation curves of variant 1 (black) and variant 2 (red), monitored by far-UV CD at 229 nm (20 μM protein). Solid lines represent fits of the experimental data to a two-state unfolding model. (B) Normalized equilibrium denaturation curves of variant 1 (black), variant 2 (red), variant 3 (blue), and variant 4 (green), calculated from fitting the raw data to a two-state unfolding model. Solid lines have no theoretical foundation and are shown to guide the eye.

Table 1.

Summary of thermodynamic data

Variants 3 and 4, each made up of only a single Trp residue in 15/22 NHB cross-strand guest positions, were also more stable than variant 1 (Fig. 4B; Table 1), consistent with the reported higher stability of β-hairpins incorporating a single Trp residue in NHB positions (Cochran et al. 2001b). The gain in stability at physiological temperature observed in variant 3 (ΔΔG = 1.6 kJ mol⁻¹) and variant 4 (ΔΔG = 0.9 kJ mol⁻¹) is in good agreement with the thermodynamic values reported by Russell and Cochran for identical mutations in a β-hairpin model peptide (1–2 kJ mol⁻¹) obtained from thiol-disulfide exchange measurements (Cochran et al. 2001b; Russell and Cochran 2001). However, the sum of the gains in stability for variants 3 and 4 (ΔΔG = 2.5 kJ mol⁻¹) is clearly less than that the difference in free energy between variants 1 and 2 (ΔΔG = 4.8 kJ mol⁻¹) (Table 1).

As demonstrated by Cochran and coworkers for an unrelated β-hairpin, this nonadditivity in free energies between hPin1 WW variants 2, 3, and 4 might result from a global stabilization of the WW domain upon incorporation of the first NHB-Trp, which leads to a favorable preorganization of the backbone that increases the NHB-position potential of the second Trp. This interpretation differs fundamentally from more classical interpretations of nonadditivity in free energy values derived from double-mutant cycle analysis and side-chain coupling energies (Horovitz 1996). Interestingly, calculating a double-mutant cycle for variants 1–4 using the data in Table 1 would yield a nonzero coupling energy for the cross-strand Trp–Trp pair of 2.3 kJ mol⁻¹ at physiological temperature, which is comparable in magnitude to side-chain coupling energies reported for other cross-strand aromatic pairwise interactions (Tatko and Waters 2002). Whether the observed nonadditivity of free energy in hPin1 WW indeed results from an energetically favorable side-chain coupling energy, as suggested by a double- mutant cycle analysis, or from a global stabilization of the WW domain upon incorporation of the first NHB-Trp remains to be demonstrated by future studies.

Functional studies

According to the X-ray structure of hPin1 in complex with a phosphorylated peptide ligand (sequence: YSPTpSPS), derived from the natural hPin1 binding partner RNA polymerase II carboxy-terminal repeat (Verdecia et al. 2000), it can be concluded that Met15 and Val22 do not directly contribute to ligand binding. In fact the Trp residues incorporated into variant 2 are positioned at the concave side of the three-stranded β-sheet opposite to the ligand binding patch (formed by side chains protruding from the convex side of the β-sheet). The cross-strand NHB Trp–Trp pair mediated increase in the thermodynamic stability of the hPin1 WW by 4.80 kJ mol⁻¹ at physiological temperature raises the question of why this energetically favorable pairwise interaction was not selected during the evolution of hPin1 WW domain. This prompted us to investigate whether variant 2 would retain its function, as ascertained by its phosphopeptide ligand binding affinity. To assess binding capability, we titrated a fixed amount of variant 1 or 2 (50 μM protein) with 5 μL injections of a stock solution of YSPTpSPS (1 mM) and measured the heat evolved using an isothermal titration calorimeter (ITC). For wild-type hPin1 WW (variant 1), the binding isotherm shows an immediate decrease in the amount of heat evolved for each consecutive injection (Fig. 5A). The absence of a well-defined plateau region prior to the transition area of the titration curve is expected, as the affinity of hPin1 WW toward peptide YSPTpSPS is rather weak, with a dissociation constant in the micromolar range (Koepf et al. 1999; Verdecia et al. 2000; Jäger et al. 2006). The normalized peak area data calculated in kilocalories per mole of injectant are displayed in Figure 5B (solid black squares). Because of the poor signal-to-noise ratio of the binding data, no efforts were made to extract a binding constant.

Figure 5.

Determination of the WW domain peptide ligand binding affinities using isothermal titration calorimetry (ITC). (A) Raw heat responses (Cp) upon sequential titration of a solution of variant 1 (50 μM) with 5-μL aliquots of a 1 mM stock solution of the phosphorylated peptide YSPTpSPS. (B) Integrated peak areas for variant 1 (solid black squares) and variant 2 (open black squares), normalized to moles of phosphorylated peptide ligand added and corrected for dilution effects.

The normalized ITC data obtained with variant 2 (Fig. 5B, open black squares) indicate that no heat (in excess of the heat of dilution) was evolved, demonstrating that variant 2 does not bind the natural peptide ligand at physiologically relevant concentrations. In order to understand the loss of binding in variant 2 further, we also conducted ligand binding studies on the single Trp variants 3 and 4. While variant 4 (Trp at position 22) exhibited a binding affinity comparable to variant 1, variant 3 (Trp at position 15) showed substantially reduced binding affinity, too low to be biologically meaningful (data not shown). The reduced binding of variant 3 cannot be attributed to substitution of residue Met15, as a control variant with a Ala15–Trp22 cross-strand pair (lacking both Met15 and Val22 in variant 1) still binds the phosphopeptide ligand with an affinity comparable to variant 1 or 4 (data not shown).

In the absence of high-resolution structural data, we can only speculate about the possible origin of the significantly reduced binding affinity of variants 2 and 3 toward the phosphopeptide. Our 1-D NMR data are inconsistent with large-scale structural perturbations. We hypothesize that loss of function in variants 2 and 3 may result from local changes in backbone conformation or dynamics. For example, Cochran and coworkers (Russell et al. 2003) noticed that the interstrand twist in their Trp-zipper peptide is not uniformly distributed along the two β-strands of the β-hairpin, but seems to result mainly from changes in local twist on the N-terminal side of the NHB Trp residues. In hPin1 WW, Trp15 is preceded by Arg14 in β-strand 1, while Trp22 follows Arg21 in the flexible loop structure. As the side chains of Arg14, Ser16, and Arg17 are important for ligand binding in wild-type hPin1 WW (variant 1) (Verdecia et al. 2000), it is plausible that changing the backbone twist between the Arg14 and Trp15 residues in variant 3 not only changes the orientation of the guanidinium side chain of Arg14, but also affects the orientation of the side chains of residues Ser16 and Arg17 C-terminal to Trp15. In variant 4, a change of the backbone twist between Trp22 and the N-terminally preceding residue Arg21 could be compensated by small conformational adjustments of the highly flexible loop 1 substructure (in particular residues Ser18–Gly20), which might explain why this variant (and its Ala15–Trp22 analog; data not shown) exhibits wild-type-like phosphopeptide binding.

Conclusions

We have shown that the thermodynamic stability of the WW domain from the human cis-trans-isomerase hPin1 protein can be significantly increased by introducing a single cross-strand Trp–Trp pair at a solvent-accessible NHB position (variant 2). However, in this case the penalty for the higher WW domain thermodynamic stability is loss of function, as ascertained by loss of phosphopeptide binding activity at physiologically relevant WW domain and phosphopeptide ligand concentrations. It is interesting that the Trp–Trp pair studied herein was not selected for during the evolution of the WW domain family; one reason in the context of hPin1 WW may be that function is eliminated. There also may be something deleterious about a highly stabilizing NHB Trp–Trp interaction that limits (backbone) motion, which may be why this pairwise interaction has not been observed in naturally occurring β-sheets to date (Cochran et al. 2001a). Despite the demonstrated loss of function of engineered hPin1 WW variant 2, introduction of a NHB Trp–Trp pair may be of general use for increasing the stability of β-sheet structures (Cochran et al. 2001a). It may even be possible in the future to recover function, once we understand why it was lost.

Materials and Methods

Protein expression and purification

Human Pin1 WW variants 1–4 were expressed recombinantly and purified as described in detail elsewhere (Jäger et al. 2001). Protein identity was confirmed by matrix-assisted laser desorption-ionization mass spectrometry (MALDI) (data not shown).

Spectroscopic analysis

Far- and near-CD spectra were recorded on an AVIV Model 202SF dichrograph using 2-mm (far-UV CD) or 10-mm (near-UV CD) quarz cuvettes (Hellma). Measurements were made in 20 mM sodium phosphate buffer (pH 7.0). All spectra were corrected for buffer background (buffer scans without protein).

Thermodynamic analysis

Equilibrium thermal unfolding transitions of hPin1 WW variants 1–4 were performed as described in detail elsewhere (Jäger et al. 2001, 2006). Raw equilibrium denaturation data were fitted to Equation 1, which assumes the validity of a two-state reaction (Crane et al. 2000):

In Equation 1, K _eq = exp(−ΔG(T)/RT) is the equilibrium constant for folding, θ is the experimentally measured ellipticity (in units of millidegrees) and θ_N(T) and θ_D(T) the slopes of the pre- and post-transitions. Folding free energies ΔG(T) in Equation 1 were expressed as a parabolic Taylor's expansion around the midpoint of equilibrium folding (T = T_M) (Equation 2; Crane et al. 2000):

Equilibrium unfolding transitions were normalized to the fraction of denatured protein (FD):

Hydrodynamic analysis

The solution molecular weights of the wild-type Pin WW domain and variants thereof were determined by sedimentation equilibrium measurements carried out as described in detail elsewhere (Deechongkit and Kelly 2002).

NMR analysis

1-D NMR spectra were recorded at 10°C for hPin1 WW variant 1 (7.6 μM) and variant 2 (33.2 μM) in 20 mM sodium phosphate (pH 7.0), 10% D₂O on a Bruker DRX 800 MHz instrument equipped with Bruker TCI cryogenic triple resonance employing a z-axis gradient.

Functional studies

Ligand binding affinities were measured by isothermal titration calorimetry (MicroCal Inc.) at 25°C as described in detail elsewhere (Koepf et al. 1999; Jäger et al. 2006). Two independent titration experiments were conducted and averaged data are reported.