Structure–function–folding relationship in a WW domain

Redesign of the Pin WW Loop 1 Substructure.

In the following, we use the shorthand (X:Y) for β-hairpin classification introduced by Thornton and coworkers (28). The first integer (X) represents the number of residues required for turn formation, and the second integer (Y) identifies the hydrogen (H)-bonding pattern in the closure of the turn. A (2:2) β-hairpin has two residues [(i + 1) and (i + 2) position] in the turn, and loop closure is accomplished by formation of two cross-strand H-bonds. A (3:5) β-hairpin has three turn residues [(i + 1) − (i + 3) position)], but only one cross-strand H-bond is formed upon loop closure.

Loop 1 of WW adopts a rare (4:6) loop conformation, with a type-II four-residue turn (sequence: Ser-16–Arg-17–Ser-18–Ser-19) incorporated into a larger, six-residue (sequence: Ser-16–Arg-17–Ser-18–Ser-19–Gly-20–Arg-21) loop (5, 9) (Fig. 1a). From an evolutionary perspective, the Pin WW loop 1 sequence could have originated from a one-residue insertion into a (3:5) type-I G1 bulge turn, with the additional residue being added N-terminal to the Gly at the bulge position (i + 3) (28, 29). Sequence alignment and structural studies indicate that the type-I G1 bulge turn is the predominant loop 1 conformation in WW domain family members at topologically equivalent positions. The longer loop 1 in Pin WW may be necessary to correctly position the functionally important Arg-17 side chain for optimal ligand binding. To investigate the structure–function–folding relationship in Pin WW, we conducted a loop 1 redesign study, starting with wild-type (WT) Pin WW, referred to as variant 1 in Fig. 1c.

The loop 1 sequence in variant 2 is identical to the sequence in FBP WW, which adopts a five-residue type-I G1 bulge turn conformation (Fig. 1b) (22). Ala, Asp, and Gly occupy the (i + 1) to (i + 3) positions, respectively. Asp exhibits the highest statistical occurrence at the (i + 2) position (30–32), whereas Gly is suited best to adopt the positive ϕ/ψ-dihedral angles at the bulge position (i + 3) in the type-I G1 bulge turn geometry (28, 32–34). The sequences corresponding to variants 3 and 4 were designed to test whether repositioning the functional Arg residue at the (i + 1) or (i + 2) position in a shorter sequence expected to adopt a type-I G1 bulge loop would yield WW domains with ligand-binding affinities comparable with variant 1. Variant 5 was obtained by deletion of the functionally important residue Arg-17 in variant 1. Variant 6 is a negative control in which the (i + 3) bulge position is occupied by a less favorable Ser residue. We predict that variant 6 will be significantly less stable than variants 2–5.

The sequences corresponding to variants 7–9 were envisioned to fold into a (2:2) type-I′ β-hairpin turn. In variant 7, Asn and Gly occupy the (i + 1) and (i + 2) positions, which are particularly well suited to adopt a type-I′ β-turn conformation (35–37). In variant 8 and 9, the (i + 1) positions are occupied by less favorable Arg and Ser residues, respectively (36, 37).

In a reverse control experiment, we replaced the type-I G1 bulge loop in a sequence-minimized version of FBP WW (variant 10) (2) with the longer loop 1 sequence of Pin WW (variant 1), yielding variant 11.

Structural Verification of Pin WW Loop 1 Redesign.

Crystallization of the isolated WW domains was unsuccessful. However, x-ray crystal structures could be obtained for variant 2 (predicted type-I G1 bulge as loop 1) at 1.9-Å resolution and variant 7 (predicted type-I′ β-turn as loop 1) at 1.5-Å resolution in the context of the full-length hPin1 rotamase. An additional Arg-14 → Ala mutation was required for crystallization of variant 2. Details on data collection/refinement statistics and electron density maps of the loop 1 substructure in variants 2 and 7 are provided as, respectively, Table 3 and Fig. 6, which are published as supporting information on the PNAS web site.

Fig. 2 depicts backbone superpositions of variants 1 and 2 (Fig. 2a) and of variants 1 and 7 (Fig. 2b). For variants 2 and 7, loop remodeling causes only local structural rearrangements within the loop 1 substructure of Pin WW. The triple-stranded β-sheet structure, the loop 2 substructure, and the conformation of N- and C-terminal extensions involved in the formation of the hydrophobic stability core are essentially superimposable with those of variant 1 (Fig. 2 a and b). A closer inspection of the loop 1 substructure in variants 2 and 7 confirms the success of our loop 1 redesign. For variant 2 (loop 1 sequence: SADGR), the dihedral angles of Ala (−60°, −28°), Asp (−97°, 12°), and Gly (94.3°, 19°) and two backbone H-bonds between Asn (donor) and Ser (acceptor) (3.1 Å) and Ser (donor) and Arg (acceptor) (2.9 Å) are consistent with a (3:5) type-I G1 bulge turn conformation (29, 33) (Figs. 1b and 2c). For variant 7 (loop 1 sequence: SNGR), the dihedral angles of Asn (53°, 55°) and Gly (70.1°, −2.2°), and two backbone H-bonds between Arg (donor) and Ser (acceptor) (3.0 Å) and Ser (donor) and Arg (acceptor) (2.8 Å) are those expected for a (2:2) type-I′ turn (29, 33) (Fig. 2d).

Structure–Stability Relationship.

All loop 1 redesigned Pin WW variants are monomeric up to at least 100 μM, as confirmed by analytical gel filtration (data not shown) and unfold cooperatively with essentially identical unfolding cooperativities (Fig. 3a and b). The unfolding transitions were analyzed assuming the validity of a two-state unfolding model (3) (Table 1).

Variants 2 and 3 with a demonstrated and a likely (3:5) type-I G1 bulge turn and with sequence-optimized (i + 2) and (i + 3) turn positions are significantly more stable than WT Pin WW (variant 1) (ΔΔG = 7.26 and 5.61 kJ·mol⁻¹, respectively; see Fig. 3a). The stability of variants 4 and 5 also exceed that of variant 1 (although variant 4 only slightly). The decrease in stability of variants 4 and 5 relative to variants 2 and 3 most likely originates from the replacement of the energetically favorable Asp at the (i + 2) position with a less favorable Arg and Ser residue, respectively (31, 32). As predicted, variant 6, in which a Ser residue replaces the energetically favorable Gly at the (i + 3) position (l-α region of the Ramachandran plot), is the least stable of the single-residue deletion variants and even less stable than variant 1. This result implies that the high stability of variants 2 and 3 cannot be entirely due to a more favorable loop closure entropy but mainly reflects the high propensity of the local sequence to fold into an energetically stable conformation (type-I G1 bulge turn). Generally, shorter loops result in more stable hairpins because of the reduced entropy penalty; however, the conformational preferences of the sequence can contribute substantially as reflected by these results.

Remodeling of Pin WW (variant 1) loop 1 with sequences preferring a (2:2) type-I′ β-turn conformation (variants 7–9) also resulted in a more stable WW domain fold (Fig. 3b). However, even though (2:2) type-I′ turns are statistically more prevalent in β-hairpins, the extent of stabilization was on average lower than that achieved by the five-residue type-I G1 bulge variants, possibly because the latter turns better match the preferred right-handed twist of the WW domain β-sheet. Variant 7 exhibits superior stability in comparison with variants 8 and 9, which might be ascribed to the energetically favorable Asn residue at the (i + 1) position in variant 7 (35, 37).

Substitution of the type-I G1 bulge turn in FBP WW (variant 10) by the longer loop 1 of WT Pin WW (variant 11) leads to a significant destabilization (7.11 kJ·mol⁻¹ at 42°C) (Fig. 3c and Table 1), which is almost identical to the gain in stability obtained at 65°C (7.26 kJ·mol⁻¹) upon engineering of variant 1 into variant 2.

We conclude that non-WT loop 1 substructures increase thermodynamic stability in a nearly independent manner relative to other substructures that are known to be important for thermodynamic stability (1–4, 16). A type-I G1 bulge turn seems to be best suited to connect the twisted β-strands 1 and 2 in the WW domain fold, providing a rationalization for the high statistical preference of this turn type for the loop 1 substructure of the WW domain family.

Structure–Folding Relationship.

The effect of loop 1 remodeling on WW domain folding kinetics was investigated by nanosecond-resolution laser T-jump relaxation. Variants 1, 3, 4, 6, 8, 10, and 11 exhibit single-exponential folding kinetics at all temperatures investigated. Variants 2, 5, 7, and 9 exhibited an additional slower, concentration-dependent relaxation phase, possibly resulting from transient, reversible oligomer formation. For the latter variants, the T-jump-induced relaxation traces were fit with a biexponential decay function, and only the rate of the fast, concentration-independent phase was considered for further analysis (2).

Arrhenius plots of the refolding rates for the redesigned type-I G1 bulge and type-I′ β-turn Pin WW variants are shown in Fig. 4a and b, respectively. A summary of the kinetic parameters is provided in Table 2. Except for the highly destabilized variant 6, all variants fold faster than the WT Pin WW domain and exhibit a pronounced anti-Arrhenius rate turnover at higher temperature. The unfolding rates, which can be computed from Tables 1 and 2, follow a standard Arrhenius-type behavior (see Fig. 7, which is published as supporting information on the PNAS web site). For Pin WW (variants 1–9), both the folding rates under isoenergetic conditions (folding rate at the denaturation midpoint T_M) and the rates of maximal folding (T_max, with T_max ≈ 7–15°C below T_M) correlate reasonably well (correlation coefficients 0.95 and 0.94, respectively) with domain stability (T_M) (see Fig. 8, which is published as supporting information on the PNAS web site). Such a strong thermodynamics–kinetics correlation is expected either for a very late transition state or in the absence of a significant folding barrier (16). Loop 1 redesign in FBP WW exhibits the expected effect. Replacing the natural (3:5) type-I G1 bulge loop in variant 10 with the longer and highly destabilizing (4:6) Pin WW loop (variant 11) significantly slows down folding, with rates comparable with Pin WW variant 1 (Table 2). That variant 6 folds slower than all of the other sequences examined implies that the expanded φ/ψ conformational space available to the Gly residue plays an important role in nucleating the WW domain fold.

Folding–Function Relationship.

Our kinetic studies demonstrate that loop 1 substructure formation is rate-limiting in WW domain folding, consistent with previous protein-engineering experiments (1, 3). A more stable loop 1 substructure in Pin WW hastens folding by up to an order of magnitude and stabilizes the domain significantly. To interrogate the influence of loop 1 remodeling on Pin WW function, the binding affinities of the Pin WW variants 1–4, 7, and 8 toward a phosphorylated peptide (sequence: YSPTpSPS), derived from the C-terminal domain of RNA-polymerase II, a natural Pin WW ligand, were determined by isothermal titration calorimetry. Fig. 5a depicts the heat evolved upon titration of a 50 μM solution of variant 1 with a 1 mM stock solution of phosphorylated ligand. Similar to previous binding studies with hYap65 WW (12), the binding isotherm of Pin WW variant 1 lacks a well defined plateau region before reaching the transition area of the titration. An immediate decrease in the amount of heat released in each injection step is evidence that the dissociation constant (K_d) of variant 1 is comparable with (or higher than) the WW domain concentration used (50 μM).

Fig. 5b depicts the integrated peak area for variants 1 and 3 after each ligand titration step, normalized to moles of ligand added and corrected for dilution effects. A nonlinear least-squares fit yielded a K_d of 35 ± 25 μM for variant 1, in good agreement with the K_d measured previously by fluorescence polarization (K_d = 34 ± 5.9 μM) (9). Notably, variant 3, which retains the functionally important Arg residue, does not exhibit any measurable heat transfer upon titration with ligand, demonstrating that variant 3, like variants 2, 4, 7, and 8 (see Fig. 9, which is published as supporting information on the PNAS web site), do not bind ligand at the WW domain concentration used (50 μM). No binding data were measured for variants 5, 6, and 9. Variant 6 is less stable and folds significantly slower than variant 1, and therefore any ligand-binding affinity exhibited by variant 6 would not contradict our hypothesis that the loop 1 substructure is optimized for functional duties at the expense of folding energetics. Variants 5 and 9 are closely related to variants 2 and 3 and 7 and 8, respectively, and it is unlikely that variants 5 and 9, which lack the functionally important Arg in loop 1, would exhibit superior binding affinity toward the phosphorylated ligand.

Variant 1, with its thermodynamically and kinetically suboptimal parameters, seems to be the best evolutionary solution for function (YSPTpSPS binding). Deletion of Arg-17 (as in variants 2 and 7), or only its minor repositioning in the shortened but energetically optimized loop 1 context (as in variants 3, 4, and 8), eliminates physiologically meaningful binding. We conclude that the loop 1 sequence and conformation in WT Pin WW (variant 1) is fine-tuned by evolution for optimal recognition and binding affinity to the phosphorylated Ser or Thr residue N-terminal to Pro, at the expense of imparting additional stability and an increased folding rate to the three-stranded β-sheet structure.