PEGylation increases the strength of a nearby NH-π Hydrogen Bond in the WW Domain

Abstract

Here we show that an NH-π interaction between a highly conserved Asn and a nearby Trp stabilizes the WW domain of the human protein Pin1. The strength of this NH-π interaction depends on the structure of the arene, with NH-π interactions involving Trp or naphthylalanine being substantially more stabilizing than those involving Tyr or Phe. Calculations suggest arene size and polarizability are key structural determinants of NH-π interaction strength. Methylation or PEGylation of the Asn side-chain amide nitrogen each strengthen the associated NH-π interaction, though likely for different reasons. We hypothesize that methylation introduces steric clashes that destabilize conformations in which the NH-π interaction is not possible, whereas PEGylation strengthens the NH-π interaction via localized desolvation of the protein surface.

Graphical Abstract

Introduction

The side chains of the amino acids Phe, Tyr, and Trp each have electron-rich faces surrounded by an electron-poor edge. This anisotropic distribution of electron density allows Phe, Tyr, and Trp to participate in favorable noncovalent interactions within proteins, including offset face-to-face or edge-to-face interactions with other arenes (π-stacking);^{1, 2} edgewise interactions with anions (anion-π);^3–5 and facewise interactions with cations (cation-π)^6–8 or with electron-deficient C–H and X–H bonds (CH-π^{9, 10} and XH-π^11–16 interactions, respectively, where X=N, O, S). Some have argued that XH-π interactions represent an additional kind of hydrogen bond.¹⁷ NH-π interactions are particularly common: a survey of 593 high-resolution protein crystal structures found 1311 instances of XH-π interactions; 40% of these were NH-π interactions.¹⁸ A recent study suggested that an NH-π interaction contributes between −0.2 to −0.9 kcal/mol to the conformational stability of the model protein GB3.¹⁹

Results and Discussion

The WW domain of the human protein Pin1 (hereafter called WW) is an extensively characterized triple-stranded antiparallel β-sheet protein consisting of only 34 residues.^20–23 The published crystal structure of WW indicates that the side-chain amide nitrogen of Asn26 is only 3.2 Å away from the plane of the Trp11 indole (Figure 1A,B),²⁴ a distance that is well within established geometric criteria for an NH-π interaction.¹⁸ Consistent with this observation, the published chemical shifts of the Asn26 side-chain amide protons (6.668 ppm for HD21 and 4.223 ppm for HD22)²² are substantially of upfield of the average values (7.336 and 7.145 ppm, respectively) reported at the Biological Magnetic Resonance Data Bank.²⁵ Presumably HD22 is more shielded because its participation in an NH-π interaction orients it directly toward the face of the Trp11 indole.

Figure 1.

(A) Space-filling and (B) stick structures of the WW domain of the human protein Pin1 (PDB: 1PIN) with Asn26 highlighted in orange and Trp11, Pro9, Ile28, and Thr29 highlighted in black. Hydrogen bonds between Asn26 side-chain HD21 and Pro9 backbone CO; between Asn26 side-chain CO and Ile28 backbone NH; between Asn26 side-chain CO and Thr29 backbone NH; and NH-π interaction between Asn side-chain HD22 and Trp11 indole are shown as dashed orange lines. (C) Structures of residue 26 within WW variants WN, WQ, and WNm, along with folding free energies (and their enthalpic vs. entropic contributions) of variants WQ and WNm relative to unmodified parent variant WN, obtained from variable temperature CD data at 50 μM protein concentration in 20 mM sodium phosphate buffer (pH 7), and calculated at the melting temperature of variant WN. (D) Calculated interaction energies for complexes of formamide with indole, benzene, naphthalene, and phenol, calculated via the MP2 method in Gaussian 16 using the aug-cc-pVDZ basis set and corrected for basis set superposition error (see supporting information for details). (E) Relationship between melting temperatures (T_m) of WW variants FN, YN, WN, and ZN and the calculated interaction energies for the corresponding formamide/arene complexes described above.

Jäger et al previously showed that WW is highly sensitive to mutation at Asn26.²⁶ Moreover, Asn is highly conserved at analogous positions within related WW family members.^{27, 28} We wondered whether a stabilizing NH-π interaction between the Asn26 side-chain amide proton HD22 and the Trp11 indole could account for the importance of Asn26 to WW folding and conformational stability. To explore this possibility, we replaced Asn26 in the parent WW sequence (called WN because of Trp11 and Asn26) with Gln (variant WQ). Gln has the same side-chain amide protons as Asn but is one methylene unit longer; we expected this increased length to move the amide protons of Gln26 away from the center of the electron-rich face of the Trp11 indole and closer to its electron-poor edge, thereby decreasing the strength of the NH-π interaction. Consistent with this expectation, variable temperature CD data reveal that the melting temperature (T_m) of variant WQ is 22.5 °C lower than that of previously characterized WN²⁹ (Figure 1C).

However, Asn26 participates in other hydrogen bonds that would presumably also be disrupted by substitution with Gln. For example, Asn26 occupies the i-position of a type I β-turn between two antiparallel β-strands in WW. The Asn26 side-chain CO hydrogen bonds with the main-chain NHs of the i+2 Ile28 and the i+3 Thr29,²⁴ resulting in two embedded 10- and 13-membered hydrogen-bound rings that comprise an Asx turn (Figure 1A,B). Hutchinson and Thornton³⁰ cite the formation of such Asx turns as a reason for the much higher propensity for Asn vs. Gln to occupy the i-position within type I β-turns. Additionally, the Asn26 side-chain amide proton HD21 is oriented away from the face of the Trp11 indole such that it can hydrogen bond with the nearby Pro9 CO (Figure 1A,B). Substitution of Asn26 with Gln would be expected to interfere with both of these hydrogen bonds in addition to the NH-π interaction and could therefore account for some of the diminished stability of WQ relative to WN.

Jäger et al previously found that replacing Asn26 with Asp strongly destabilizes WW (ΔΔG = −1.9 kcal/mol).²⁶ This result provides insight into the energetic significance of the NH-π interaction between Asn26 and Trp11 vs. the hydrogen bonds described above. Asp has an even higher propensity than Asn for occupying the i-position of a type I β-turn.³⁰ At physiological pH, Asp should be negatively charged; its side-chain carbonyl oxygen should therefore be a better hydrogen bond acceptor and form more stable Asx turns. In contrast, negatively charged Asp should be unable to form a hydrogen bond with the Pro9 CO or to participate in an NH-π interaction with Trp11. The destabilization associated with replacing Asn26 with Asp suggests that the Asx turn does not contribute strongly to WW conformational stability in the absence of the NH-π interaction with Trp11 and/or the hydrogen bond with the Pro9 CO.

To explore the energetic significance of the NH-π interaction between Asn26 HD22 and Trp11 vs. the hydrogen bond between Asn26 HD21 and Pro9 CO, we prepared variant WNm (Figure 1C), in which Asn26 has been replaced with N-methyl Asn. Presumably, the N-methyl Asn adopts the more stable trans amide stereoisomer such that the methyl group replaces the less crowded HD21 rather than HD22. Accordingly, the NH-π interaction with Trp11 should still be possible in WNm, but the hydrogen bond with the Pro9 CO should not. Variant WNm is substantially more stable than unmodified variant WN (ΔΔG = −1.05 ± 0.04 kcal/mol) due to a favorable entropic term (−TΔΔS = −2.4 ± 0.6 kcal/mol) offset by a smaller unfavorable enthalpic term (ΔΔH = 1.3 ± 0.6 kcal/mol). This observation suggests that the hydrogen bond between HD21 and the Pro9 CO does not contribute substantially to WW conformational stability. We hypothesize that N-methylation stabilizes WNm relative to WN by introducing short range steric clashes that destabilize alternative Asn26 rotamers in which the NH-π interaction with Trp11 is not possible, thereby preorganizing Asn26 in a conformation that is compatible with the NH-π interaction.

Next, we explored the energetic consequences of structural variations at position 11 while keeping Asn26 unchanged. Previous structural analysis found that 7.7% of all Trp residues within 592 high-resolution protein crystal structures were engaged in an NH-π interaction; in contrast, only 3.2% of Tyr residues, 1.9% of Phe residues, and 0.7% of His residues were similarly engaged.¹⁸ Of the 519 NH-π interactions identified in this survey, Trp accounted for 196, Tyr for 185, and Phe for 113, and His for 25.¹⁸ Based on average expression frequencies of Trp (1.2%), Tyr (3.0%), Phe (3.9%), and His (2.5%) in more than a half-million proteins from 38 proteomes,³¹ we estimate that Trp appears 3.4 times more frequently in NH-π interactions than would be expected at random. In contrast, Tyr only appears only slightly more frequently than expected (by a factor of 1.3), whereas Phe and His appear less frequently than expected (by factors of 0.6 and 0.2, respectively). Together, these observations suggest the possibility that NH-π interactions with Trp are more favorable than with Tyr, Phe, or His (note that we did not include His in the present study out of concern that its ambiguous protonation state at pH 7 might complicate analysis of our results). Previous computational studies support this hypothesis: OH-π interactions are stronger for complexes of water with indole (−3.5 kcal/mol) than with phenol (−2.3 kcal/mol) or benzene (−2.1 kcal/mol).¹⁶ Similarly, our MP2 calculations³² in Gaussian16³³ (with the aug-cc-pVDZ basis set,³⁴ implicit water,³⁵ and counterpoise correction for basis set superposition error^{36, 37}) reveal that NH-π interactions (Figure 1D) are stronger for complexes of formamide with indole (−4.9 kcal/mol) than with phenol (−4.0 kcal/mol) or benzene (−4.1 kcal/mol).

We wondered whether substituting Phe or Tyr for Trp11 would result in similar changes in WW conformational stability. To explore this possibility, we prepared WW variants FN and YN, in which Phe and Tyr, respectively, occupy position 11. We then assessed the stability of these variants using variable temperature circular dichroism experiments (Table 1): FN (T_m = 29.3 ± 0.2 °C) and YN (T_m = 28.3 ± 0.2 °C) are substantially less stable than WN (T_m = 58.0 ± 0.3), an observation that agrees qualitatively with (1) the apparent enrichment of Trp relative to Tyr and Phe in observed NH-π interactions within proteins; and (2) with the more favorable calculated NH-π interaction energies for indole vs. phenol or benzene. Trp11 occupies a somewhat buried position within WW; we wondered whether the smaller sizes of Tyr and Phe relative to Trp might account for some of the observed destabilization of YN and FN relative to WN due to cavity creation. To explore this possibility, we prepared WW variant ZN, in which we replaced Trp11 with isosteric naphthylalanine (Nap). Variant ZN (T_m = 55.0 ± 0.2 °C) is only slightly less stable than WN. Similarly, MP2 calculations suggest formamide interacts only slightly less strongly with naphthalene (−4.7 kcal/mol) than with indole. Figure 1E plots the melting temperatures of variants WN, YN, FN, and ZN vs. the calculated NH-π interaction energies for complexes of formamide with indole, benzene, phenol, and naphthalene, respectively. The strength of the resulting correlation (R² = 0.98) suggests that the calculated NH-π interaction energies account for most of the variation in the melting temperatures of WN, YN, FN, and ZN but does not rule out a contribution from variations in the size of the arene and the resulting creation of a cavity near position 11 in FN and YN but not in WN and ZN. We explore this issue in greater detail below.

Table 1.

Sequences, melting temperatures, and ΔΔG_f values of PEGylated and methylated WW variants relative to their unmodified counterparts^a

Peptide	Sequence	Tm (°C)	ΔΔGf (kcal/mol)
WN	KLPPGWEKRMSRSSGRVYYFNHITNASQFERPSG	58.0 ± 0.3
WQ	•••••W••••••••••••••Q•••••••••••••	35.5 ± 0.3
FN	•••••F••••••••••••••N•••••••••••••	29.3 ± 0.2
YN	•••••Y••••••••••••••N•••••••••••••	28.3 ± 0.2
ZN	•••••Z••••••••••••••N•••••••••••••	55.0 ± 0.2
WNp	•••••W••••••••••••••N•••••••••••••	64.5 ± 0.1	−0.58 ± 0.03
WQp	•••••W••••••••••••••Q•••••••••••••	33.7 ± 0.2	0.14 ± 0.02
FNp	•••••F••••••••••••••N•••••••••••••	34.7 ± 0.1	−0.34 ± 0.02
YNp	•••••Y••••••••••••••N•••••••••••••	35.2 ± 0.6	−0.47 ± 0.05
ZNp	•••••Z••••••••••••••N•••••••••••••	62.2 ± 0.2	−0.62 ± 0.03
WNm	•••••W••••••••••••••N•••••••••••••	70.0 ± 0.2	−1.05 ± 0.02
FNm	•••••F••••••••••••••N•••••••••••••	38.6 ± 0.2	−0.66 ± 0.02
YNm	•••••Y••••••••••••••N•••••••••••••	37.5 ± 0.4	−0.59 ± 0.03
ZNm	•••••Z••••••••••••••N•••••••••••••	64.5 ± 0.2	−0.84 ± 0.03

^aZ = naphthylalanine (Nal). The ΔΔG_f value for each PEGylated protein relative to its non-PEGylated counterpart is given ± standard error at 50 μM protein concentration in 20 mM sodium phosphate buffer (pH 7) at the melting temperature of the non-PEGylated variant. Data for WN and WNp are from ref 29.

The strength of a noncovalent interaction within a protein is highly dependent on its surroundings. For example, buried hydrogen bonds are strengthened by their non-polar microenvironment^{38, 39} presumably due to the lower effective dielectric constant of the protein interior vs. surface and to the absence of interfering water molecules. Similarly, a growing body of experimental and theoretical evidence suggests that attaching polyethylene glycol to a protein side chain (i.e., PEGylation) can strengthen nearby noncovalent interactions by localized desolvation of the area immediately surrounding the PEGylation site.^{29, 40, 41} We previously found that conjugating a four-unit PEG oligomer to the side-chain amide nitrogen of Asn26 increases the conformational stability of the PEGylated WW variant WNp by −0.58 ± 0.06 kcal/mol relative its non-PEGylated counterpart WN (Figure 2A).²⁹ Analysis of solvent radial gradient distribution functions derived from previous atomistic simulations of WN and WNp in explicit water suggested that fewer water molecules surround Trp11 in PEGylated WNp than in non-PEGylated WN (Figure 2A); this suggestion is consistent with our observations that the increased stability of WNp relative to WN comes from a favorable entropic term (−TΔΔS = −3.8 ± 0.5 kcal/mol) offset by an unfavorable enthalpic term (ΔΔH = 3.2 ± 0.5 kcal/mol)²⁹ We wondered whether the increased stability of WNp relative to WN might also reflect strengthening of the NH-π interaction between Asn26 and Trp11 due to localized PEG-based desolvation.

Figure 2.

(A) Structure of PEGylated Asn26 in variant WNp along with PEG-based stabilization of WNp relative to its non-PEGylated counterpart WN. Also shown is the simulated radial distribution function of water around Trp11 in WNp (red line) vs. WN (black line) along with an inset histogram of the number of water molecules within 3 Å of Trp11 in each variant. Adapted with permission from reference 29. Copyright © 2014 American Chemical Society. Also shown is the structure of methylated Asn26 in variant WNm along with its conformational stability relative to non-PEGylated WN. (B) Relationship between the T_m of WW variants FN, YN, WN, and ZN or the calculated NH-π interaction energies for complexes of formamide and benzene, phenol, indole, and naphthalene and the stabilization (ΔΔG) of Asn-PEGylated FNp, YNp, WNp, and ZNp (red circles) or of Asn-methylated FNm, YNm, WNm, and ZNm (gray circles) relative to their unmodified counterparts FN, YN, WN, and ZN. Solid red and black lines represent fits of these relationships via linear least-squares regression for PEGylation and methylation, respectively, with the R², F statistic, and p values shown. Dotted vertical lines represent the difference in ΔΔG values for PEGylation vs. methylation for a given WW sequence.

However, our observation that N-methylation of Asn26 is more stabilizing to WW (ΔΔG = −1.05 ± 0.04 kcal/mol) than is PEGylation (ΔΔG = −0.58 ± 0.06 kcal/mol) is surprising, because we previously observed the opposite trend for Asn-methylation vs. PEGyation at position 19: whereas N-methylated WNm is −0.46 ± 0.03 kcal/mol more stable than PEGylated WNp (Table 1), N-methylated variant 19m is 0.30 ± 0.01 kcal/mol less stable than PEGylated 19p.⁴² We hypothesized that the structural context of the N-methylation site accounts for this disparity: Asn26 in variant WN appears to be engaged in an NH-π interaction with a nearby arene (Trp11), whereas Asn19 does not.

We wondered whether the stabilizing impact of Asn26 N-methylation might therefore depend on the strength of the NH-π interaction with the arene at position 11. To explore this possibility, we prepared N-methylated variants FNm, YNm, and ZNm, in which N-methyl Asn occupies position 26, and Phe, Tyr, and Nap, respectively, occupy position 11 (Table 1). N-methylation stabilizes variants FNm, YNm, and ZNm relative to their non-methylated counterparts by −0.66 ± 0.03, −0.59 ± 0.03, and −0.84 ± 0.04 kcal/mol, respectively. This stabilization is well correlated with the melting temperatures of variants WN, FN, YN, and ZN (R² = 0.87) and with the NH-π interaction energies between formamide and indole, benzene, phenol, and naphthalene (R² = 0.94), suggesting that N-methylation of Asn26 strengthens the NH-π interaction with the arene at position 11 and that stronger NH-π interactions are more readily stabilized via methylation (Figure 2B). In contrast, PEGylation has a smaller stabilizing effect on variants FNp, YNp, and ZNp relative to their non-PEGylated counterparts than does methylation (Table 1) and the correlations between this stabilization and the melting temperatures of variants WN, FN, YN, and ZN or with the NH-π interaction energies are correspondingly weaker (R² = 0.75 and 0.68, respectively; Figure 2B).

We speculate that these observations derive from two offsetting considerations. First, as described above, we hypothesize that N-methylation restricts the conformational freedom of Asn26 by introducing short range steric clashes that destabilize alternative Asn26 rotamers in which the NH-π interaction with Trp11 is not possible. Such localized steric clashes would also be present in PEGylated WNp, but their effect would be diminished by the additional flexibility and conformational entropy of the longer PEG chain in WNp relative to the shorter methyl group in WMn. Second, we envision that the PEG chain in WNp strengthens the NH-π interaction between Asn26 and Trp11 via localized desolvation of the surface in the immediate vicinity of the PEGylation site, an effect not available the smaller methyl group. We hypothesize that the indirect effect of PEG-based desolvation on the strength of the NH-π interaction is weaker than the direct effect of methyl-based conformational restriction.

We wondered what structural features of Trp and Nap vs. Tyr and Phe facilitate the stronger NH-π interactions in WN and ZN vs. FN and YN; we hypothesized that these features might include the size, polarizability, electrostatic potential, and hydrophobicity of each arene. We used density functional theory (DFT) via the APFD method⁴³ with the 6-311+G(2d,p) basis set^{44, 45} to calculate the molar volume and molecular polarizability of indole, naphthalene, benzene, and phenol, along with the maximum electrostatic potential above the centroid of the benzene ring within each arene (see supporting information for details). We estimated the hydrophobicity of each arene using the water-octanol partition coefficients (c log P) for indole, naphthalene, phenol, and benzene recommended by Sangster.⁴⁶ We plotted the calculated NH-π interaction energies for complexes of formamide with indole, naphthalene, benzene, and phenol and the melting temperatures of WN, ZN, FN, and YN vs. arene properties described above (Figure 3). We then assessed the strength of the resulting relationships via least-squares regression. The strongest correlations for NH-π interaction energy and T_m are each with arene molar volume and molecular polarizability, suggesting that arene size and polarizability are stronger determinants of NH-π interaction strength than are hydrophobicity or electrostatic potential. Moreover, the strong correlation of NH-π interaction energy with arene molar volume suggest that the impact of arene size on the T_m of WW variants WN and ZN vs. FN and YN is not a simple function of cavity creation. Larger more polarizable arenes appear to be better NH-π hydrogen bond acceptors (Figure 3), resulting in more stable WW variants, which are in turn better stabilized by N-methylation or PEGylation at Asn26 (Figure 2B).

Figure 3.

Relationship between calculated NH-π interaction energies for complexes of formamide with indole, benzene, phenol, and naphthalene (bottom row) or experimental melting temperatures (T_m; top row) of WW variants WN, FN, YN, and ZN with calculated properties of benzene, phenol, indole, and naphthalene, including molar volume (yellow circles); molecular polarizability (blue circles); the maximum electrostatic potential above the benzene centroid within each arene (orange circles); and c log P (purple circles); Solid yellow, blue, orange, and purple lines represent fits of these relationships via linear least-squares regression, with the R², F statistic, and p values shown. Vertical dotted yellow, blue, yellow, orange, and purple lines connect NH-π interaction energy, T_m, and ΔΔG for each variant. c log P values are from reference 46; all other properties were calculated using Gaussian16 (see supporting information for details).

Here we have shown that an NH-π interaction between Asn26 and Trp11 stabilizes the WW domain. Calculated interaction energies of formamide with indole, phenol, benzene, or naphthalene account well for observed changes in the stability of WW variants upon substitution of Trp11 with Tyr, Phe, or Nap. Arene size and polarizability are the key structural features that account for the increased strength of the NH-π interactions involving Trp or Nap vs. those involving Tyr or Phe. PEGylation of Asn26 stabilizes these WW variants relative to their non-PEGylated counterparts in a way that correlates with the melting temperatures of the non-PEGylated variant and with the calculated strength of the NH-π interaction, consistent with the growing body of evidence that PEGylation can enhance the strength of a noncovalent interaction (in this case, an NH-π interaction) in the immediate vicinity of the PEGylation site via localized desolvation of the protein surface. Asn-methylation is even more stabilizing than Asn-PEGylation and its effect correlates even more closely with NH-π interaction strength. We anticipate that an increasingly detailed understanding of the impact of PEGylation on protein structure, stability, and noncovalent interactions should facilitate more efficient design of PEGylated protein drugs with enhanced pharmacokinetic properties that retain full biological function.