Structural and thermodynamic consequences of burial of an artificial ion pair in the hydrophobic interior of a protein Aaron C. Robinsona, Carlos A. Castañedaa, Jamie L. Schlessmana,b, and Bertrand García-Moreno E.a,1 a

Department of Biophysics, Johns Hopkins University, Baltimore, MD 21218; and bChemistry Department, United States Naval Academy, Annapolis, MD 21402

An artificial charge pair buried in the hydrophobic core of staphylococcal nuclease was engineered by making the V23E and L36K substitutions. Buried individually, Glu-23 and Lys-36 both titrate with pKa values near 7. When buried together their pKa values appear to be normal. The ionizable moieties of the buried Glu–Lys pair are 2.6 Å apart. The interaction between them at pH 7 is worth 5 kcal/mol. Despite this strong interaction, the buried Glu–Lys pair destabilizes the protein significantly because the apparent Coulomb interaction is sufficient to offset the dehydration of only one of the two buried charges. Save for minor reorganization of dipoles and water penetration consistent with the relatively high dielectric constant reported by the buried ion pair, there is no evidence that the presence of two charges in the hydrophobic interior of the protein induces any significant structural reorganization. The successful engineering of an artificial ion pair in a highly hydrophobic environment suggests that buried Glu–Lys pairs in dehydrated environments can be charged and that it is possible to engineer charge clusters that loosely resemble catalytic sites in a scaffold protein with high thermodynamic stability, without the need for specialized structural adaptations.

|

|

electrostatics internal charge low-barrier hydrogen bond dielectric constant enzyme design

|

P

|

aired ionizable groups buried in hydrophobic environments inside proteins are an uncommon but essential structural motif necessary for H+ transport (1), e− transfer (2), catalysis (3– 5), and other important biochemical processes. Despite their importance, the properties of these buried ionizable pairs are poorly understood because they are difficult to study experimentally in the proteins where they play functional roles. When the pair consists of a basic and an acidic group it is usually assumed to be charged, but this is often just speculation without direct experimental support (6, 7). In principle, a Coulomb interaction between two ionizable groups of opposite polarity buried in a dehydrated environment inside a protein can be very strong, but this has not been established experimentally. Simulations suggest that buried ion pairs are always destabilizing (8); however, from an experimental perspective it is not known if the favorable Coulomb interaction would be sufficient to compensate for the unfavorable dehydration of the two buried charges or if the net effect of the pair on the thermodynamic stability of a protein would be stabilizing or not. These issues were examined in detail with an artificial ion pair buried in the hydrophobic interior of staphylococcal nuclease (SNase). The buried ion pair was engineered by introducing the V23E and L36K substitutions in a highly stable form of SNase. The pKa values of Glu-23 and Lys-36 introduced individually are anomalous (pKa of 7.1 for Glu-23, ref. 9; and 7.2 for Lys-36, ref. 10), consistent with the groups existing in hydrophobic environments that are neither as polar nor as polarizable as water (11–13). Structural modeling indicated that the ionizable moieties of Glu-23 and Lys-36 could form a short-range interaction when buried simultaneously. Their similar pKa values were expected to facilitate H+ transfer between the acidic and basic moieties. The www.pnas.org/cgi/doi/10.1073/pnas.1402900111

main goals of this study were to determine the charge state of an internal pair of Glu–Lys residues, to examine how such a pair could affect its microenvironment, to measure consequences on thermodynamic stability of the protein, and by comparing the pair against the single-site substitutions, to evaluate the relative magnitudes of Coulomb interactions and dehydration energies. According to primitive, macroscopic electrostatics models, the transfer of an ion pair from water into a cavity with low dielectric constant is always unfavorable (Fig. 1). In these models the favorable Coulomb interaction between the charges in the buried pair is never strong enough to offset the unfavorable self-energy experienced by the two buried charges. In some microscopic models, Coulomb interactions can be stronger than dehydration effects, but calculations with these methods are still suspect as they cannot reproduce self-energies reliably (14, 15) and tend to overestimate the magnitude of Coulomb interactions (14, 16– 18). The experiments that were performed test assumptions made in most continuum methods, including the idea that a pair of ionizable groups interacting at close range in a dehydrated environment can be treated classically as point charges without invoking some level of quantum mechanical treatment to account for polarizability and for the properties of a proton shared between acidic and basic moieties interacting over a short distance in a highly dehydrated environment (19, 20). Improved computational methods for structure-based electrostatics calculations are necessary for progress in molecular understanding of catalysis and other biochemical processes governed by acid–base chemistry. The physical insight gained Significance Charges buried in hydrophobic environments in proteins play essential roles in energy transduction. We engineered an artificial ion pair in the hydrophobic core of a protein to demonstrate that buried ion pairs can be charged and stabilized, in this instance, by a strong Coulomb interaction worth 5 kcal/mol. Despite this interaction, the buried charge pair destabilized the folded protein because the Coulomb interaction recovered the energetic penalty for dehydrating only one of the two buried charges. Our results suggest how artificial active sites can be engineered in stable proteins without the need to design or evolve specialized structural adaptations to stabilize the buried charges. Minor structural reorganization is sufficient to mitigate the deleterious consequences of charges buried in hydrophobic environments. Author contributions: A.C.R. and B.G.-M.E. designed research; A.C.R., C.A.C., and J.L.S. performed research; A.C.R., C.A.C., and J.L.S. analyzed data; and A.C.R. and B.G.-M.E. wrote the paper. The authors declare no conflict of interest. *This Direct Submission article had a prearranged editor. Data deposition: The atomic coordinates and structure factors have been deposited in the Protein Data Bank, www.pdb.org (PDB ID codes 3NHH and 3QOL). 1

To whom correspondence should be addressed. Email: [email protected].

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. 1073/pnas.1402900111/-/DCSupplemental.

PNAS | August 12, 2014 | vol. 111 | no. 32 | 11685–11690

BIOPHYSICS AND COMPUTATIONAL BIOLOGY

Edited* by Arieh Warshel, University of Southern California, Los Angeles, CA, and approved June 12, 2014 (received for review February 21, 2014)

paired. One is buried deeply at a distance 2.8 and 2.5 Å from the Nζ and Oe1 of Lys-36 and Glu-23, respectively. The second water molecule is 3.1 Å from the Oe2 of Glu-23 (Fig. 3D). Minor rearrangements of the backbone carbonyl oxygen of Gly-20 and of the side chain hydroxyl group of Thr-62 allow these two groups to interact with the Lys-36 Nζ and Glu-23 Oe2, respectively. Structural consequences of the V23E/L36K double substitution were also examined with circular dichroism, Trp fluorescence, and NMR spectroscopy, none of which showed any indication that the V23E/L36K double substitution had any detectable effect on the conformation of the protein (Table 1 and Figs. S1–S4).

Fig. 1. Free energy for transfer of a charged pair from water into a dielectric cavity of variable dielectric constant eprot, calculated with a primitive continuum electrostatics model and dissected into contributions from dehydration and Coulomb energies. Dehydration energy (red) calculated with a Born formalism to describe the energy for the transfer of two charges of opposite sign, each of radius rion = 1.75 Å, from water into the center of the spherical dielectric cavity of radius rcavity = 20 Å. Coulomb interaction (blue) between the two charges separated by a distance rij = 3.5 Å, calculated with Coulomb’s law relative to the Coulomb energy in water (eprot = 80). Total transfer free energy (black) for moving the ion pair from water (ew = 80) into the center of the cavity, is calculated as a function of eprot, the dielectric constant of the cavity.

from the present study will help guide the development of more accurate structure-based calculations. The thermodynamic and structural data from this study will also constitute benchmarks useful for stringent testing and validation of structure-based calculations. Our results suggest a strategy for the evolution or engineering of enzymatic active sites capable of performing acid– base chemistry and proton pumping focused more on the use of a protein scaffold with high thermodynamic stability than on the manipulation of the microenvironments and chemical details needed to stabilize buried charge clusters. Results Crystal Structures. The crystal structures of the highly stable variant of SNase, Δ+PHS, used as background for this study and the L36K (solved previously in ref. 21), V23E, and V23E/L36K variants thereof are very similar. This demonstrates that the L36K and V23E substitutions had no detectable effect on the conformation of the protein (Fig. 2 and Table S1). The side chain of Lys-36 in the L36K variant (obtained at pH 8) is extended and points into a microcavity in the hydrophobic core (22). It is completely solvent inaccessible and surrounded by nonpolar atoms save for the carbonyl oxygen of Gly-20 3.5 Å away (Fig. 3B). In the V23E variant (obtained at pH 6) the side chain of Glu-23 is also extended and completely solvent inaccessible (Fig. 3C). One crystallographic water molecule occupies a small recess extending inwards between β-1 and α-1; an analogous water molecule was observed previously in the structure of the Δ+PHS protein (16). Other than this semiburied water molecule, there are no polar atoms within 4.5 Å of either carboxyl oxygen of the Glu side chain, leaving several potential hydrogen bonds unsatisfied. The structure of the V23E/L36K double variant (obtained at pH 8) is also very similar to that of Δ+PHS SNase. The conformation of the side chain of Lys-36 is identical in the presence and absence of Glu-23 (Fig. 3 B and D). The conformation of the side chain of Glu-23 is affected slightly by the presence of Lys-36, bringing Oe1 and Oe2 to within 3.2 and 2.6 Å, respectively, of the Nζ of Lys-36. Overall, the microenvironment surrounding the ionizable moieties of the pair was more polar than for the ionizable moieties of the ionizable residues buried individually. Two crystallographic water molecules interact with the ionizable groups when they are 11686 | www.pnas.org/cgi/doi/10.1073/pnas.1402900111

pKa of Glu-23. Glu-23 in the V23E variant has a pKa of 7.2 ± 0.2 (9). An attempt was made to use NMR spectroscopy (16) both to confirm this pKa by more direct measurement and to monitor changes in the chemical environment of Glu-23 upon ionization. In a correlation (CBCGCO) (16) spectrum of the V23E variant, an unusual resonance was observed that is not normally present in the spectrum of Δ+PHS SNase (Fig. 4). The peak was in an unusual position for an Asp–Glu side chain, exhibiting an upfield chemical shift to 174 ppm in its Cδ resonance (Fig. 4), far from the typical Cδ resonances of surface Glu residues, which range from 178 to 185 ppm depending on whether the group is neutral or charged, respectively (16, 23). The Cγ resonance was ∼31.8 ppm, near the upfield limit of the normal range for Glu residues from 33 to 39 ppm. The new resonance was assigned to Glu-23. Intermediate exchange at pH > 6 precluded the monitoring of this resonance in the range of pH where Glu-23 titrated. The resonance did not reappear once Glu-23 was fully charged at high pH. The origin of this exchange behavior for this residue is not known. A similar behavior has been observed for other internal carboxyl groups in SNase variants (16, 24). Although the pKa value for Glu-23 could not be measured directly by NMR spectroscopy, both its upfield chemical shift and exchange behavior upon titration are consistent with the Glu side chain being internal and having a pKa value >6, in agreement with the previously determined apparent pKa of 7.2. An attempt was also made to measure the pKa of Glu-23 in the V23E/L36K double variant. The unusual resonance with the large upfield chemical shift observed for Glu-23 in the V23E variant was not present in the spectrum for the V23E/L36K variant. A single new peak was observed in the region of the CBCGCO spectra normally populated by the surface residues Glu-101 and Glu-122 (Fig. 4). The Cγ and Cδ resonances were assigned chemical shifts of 36.4 and 179.8 ppm, respectively, within the range normally observed for surface Glu residues (16, 25, 26). The new resonance was presumed to correspond to Glu-23 in the presence of Lys-36. Despite some moderate line broadening at neutral pH, the Cδ resonance of Glu-23 could be monitored at all pH values measured (Fig. 5A, Inset). The chemical shift was largely pH independent, undergoing a minor shift of ∼0.8 ppm between pH 6.5 and 4.3, considerably smaller than the normal

Fig. 2. Stereoview of crystal structures of the V23E/L36K (white; PDB ID code 3NHH), V23E (pink; PDB ID code 3QOL), and L36K (blue; PDB ID code 3EJI) (21) variants of the Δ+PHS form of SNase, overlaid on the structure of Δ+PHS used as reference (green; PDB ID code 3BDC) (16). Side chain atoms for Glu-23 and Lys-36 in the Δ+PHS/V23E/L36K variant are shown as balls-and-sticks.

Robinson et al.

residues suggested that the carboxylic groups had pKa values comparable to those measured previously in the highly stable Δ+PHS form of SNase (Fig. S6). The data confirmed that the chemical environments of the surface carboxylic groups were not perturbed by the presence of the internal Glu-23 or Lys-36. The sole exception was Asp-21, which normally titrates with an elevated pKa of ∼6.5 in Δ+PHS (16), but had a pKa of 6.3 ± 0.1 in the V23E variant and of 5.9 ± 0.1 in the V23E/L36K variant (Fig. 5). Thermodynamic Stability. The thermodynamic stability ðΔG8H2 O Þ

of the V23E/L36K variant was determined at pH values between 4.5 and 10.0 in increments of 0.4–0.5 pH units (Table 1 and Table S2). Usually the anomalous pKa values of internal ionizable groups can be measured by fitting Eq. 1 to the difference in stability, ΔΔG8H2 O , between the reference protein (Δ+PHS) and the variant protein (Δ+PHS/V23E/L36K):

Fig. 3. (A) Microenvironment of the side chains of Val-23 and Leu-36 in the Δ+PHS protein. Hydrophobic side chains in the main hydrophobic core are shown as balls-and-sticks in yellow. Polar side chains are colored white. The blue mesh indicates the molecular surface of a small, naturally occurring internal microcavity as well as the volume occupied by internal water molecules. Microenvironment of (B) Lys-36 in the L36K variant, (C) Glu-23 in the V23E variant, and (D) Glu-23 and Lys-36 in the V23E/L36K variant. Glu-23 and Lys-36 side chains are colored light blue; the alternate conformation for Thr62 in Δ+PHS/V23E/L36K is colored gray. Water molecules are represented as blue-green spheres. Hydrogen bonds are shown as dashed lines.

chemical shift change of the Cδ of a titrating Glu residue, which ranges from 3 to 5 ppm (23). Small shifts of the order of ∼1 ppm usually result from the titration of another ionizable group coupled to the residue being monitored (16). For Glu-23 in the V23E/L36K variant, this secondary titration likely reflects the titration of the nearby Asp-21. The NMR experiment is consistent with Glu-23 in the V23E/L36K variant being charged over the entire range of pH studied. The pKa values of surface Glu and Asp residues in the V23E/ L36K variant were also monitored to determine if they were affected by Glu-23 directly through a Coulomb interaction or indirectly through conformational reorganization. It was impossible to measure complete titration curves for surface Asp and Glu residues because the protein begins to acid unfold in the pH range 4.5–4.0 (Fig. S5), the pH range where surface Asp and Glu residues begin to titrate. However, visual inspection of the pH dependence of the Cγ/Cδ resonances of the other Asp and Glu

X

1 + ez2:303ðpH−pKa;i Þ : N 1 + ez2:303ðpH−pKa;i Þ U

RT ln

[1]

In this equation ΔΔG8mut refers to the pH-independent free energy for substitution of both ionizable residues at a hypothetical pH where both groups are neutral. The right-most term reflects the coupling of the pH-dependent shift in pKa values to the transition between native (N) and denatured (U) states. pKa values in the N and U states can be extracted from the characteristic shape of the pH dependence of ΔΔG8H2 O (11, 12) (Fig. 6). In chemical denaturation experiments, the unfolding transition of the Glu-23–Lys-36 double variant exhibited apparent two-state behavior at all pH values, with a ΔG8H2 O near 2.6 kcal/mol between pH 6 and 9 and lower values outside this range (Fig. 6A). The ΔΔG8H2 O , calculated as ΔG8H2 O  Δ + PHS=V23E=L36K − ΔG8H2 O;Δ + PHS , was largely pH independent. The minor pH dependence below pH 6 is within error of the ΔΔG measurement and could reflect the minor shift in pKa of Asp-21. The most striking feature of the ΔΔG8H2 O vs. pH profile for the V23E/L36K double variant is the absence of the characteristic pH sensitivity expected from a protein that has ionizable groups with anomalous pKa values (Fig. 6B). The net coupling energy (ΔΔGint) between Glu-23 and Lys-36 was estimated with a double mutant cycle (27): ΔΔGint = ΔΔG8H2 O;V23E=L36K − ΔΔG8H2 O;V23E − ΔΔG8H2 O;L36K : [2] ΔΔGint was found to be stabilizing by 2.5–5 kcal/mol in a pHdependent manner (Fig. 6B). At pH 7, ΔΔGint is 5 kcal/mol. As the pH approaches the pH corresponding to the model pKa of 4.4 for Glu in water (10.4 for Lys), ΔΔGint reflects the energy of offsetting the shift in pKa of Lys-36 (Glu-23) alone. Although the ΔΔGint values cannot be interpreted as arising solely from a Coulomb effect, its magnitude demonstrates that a strong stabilizing interaction was established between Glu-23 and Lys-36

Table 1. Thermodynamic parameters from pH and chemical denaturation experiments pHmid Fl ΔG8H2 O ,

Variant Δ+PHS* Δ+PHSV23E† Δ+PHSL36K‡ Δ+PHSV23E/L36K ,†

kcal·mol

11.9 4.9 4.7 2.6

± ± ± ±

0.1 0.2 0.1 0.4

−1

−1

m-value, kcal·mol ·M — 6.0 ± 0.2 6.4 ± 0.1 7.2 ± 0.6

−1

ΔΔG8H2 O ,

kcal·mol

— −7.0 ± 0.2 −7.2 ± 0.2 −9.3 ± 0.4

−1

pHmid CD

Acid

Base

Acid

Base

2.1 ± 0.1 — 4.1 ± 0.1 4.3 ± 0.1

— — — 10.0 ± 0.1

2.1 ± 0.1 — 4.1 ± 0.1 4.3 ± 0.1

— — — 10.4 ± 0.1

Fl, fluorescence. *Data from ref. 46. † Data from ref. 9. ‡ Data from ref. 10.

Robinson et al.

PNAS | August 12, 2014 | vol. 111 | no. 32 | 11687

BIOPHYSICS AND COMPUTATIONAL BIOLOGY

ΔΔG8H2 O ðpHÞ = ΔΔG8mut −

substitution destabilized the protein by 7.0 kcal/mol and the L36K substitution by 7.2 kcal/mol. Therefore, if the pKa values of Glu-23 and Lys-36 in the double variant were the same as in the single variants at pH 7, the double variant would be destabilized by 14.2 kcal/mol. Instead, at pH 7, the V23E/L36K double substitution destabilized the protein by only 9.3 kcal/mol, fully consistent with the interpretation that Glu-23 and Lys-36, when paired, have normal pKa values and are charged at pH 7 and at all other pH values studied. The ionizable moieties of Glu-23 and Lys-36 are 2.6 Å apart, suggesting that the stabilizing interaction involves an H-bond between neutral Glu and Lys, or perhaps more likely given the slight reorganization of the microenvironment around the pair, a Coulomb interaction between these two charged moieties. Fig. 4. CBCGCO spectrum of the V23E variant at pH 5.28 (red) and of the V23E/L36K variant at pH 5.18 (blue). Gray, dashed boxes identify typical chemical shifts for water-exposed Glu and Asp resonances in the Δ+PHS protein. The resonances belonging to Glu-23 in the V23E and V23E/L36K variants are labeled in red and blue, respectively. Resonances for Asp-19 and -21, which reside near Glu-23 in both the primary and tertiary structure, are labeled.

when these two residues were buried together. Note that in the crystal structure, the geometry of the Glu–Lys pair is not ideal for the establishment of a hydrogen bond (Fig. 2). Discussion A systematic study of ionizable groups in the hydrophobic interior of SNase showed that internal Glu, Asp, Lys, or Arg destabilize the native state by 1.2–12 kcal/mol at pH 7 (9, 10, 28, 29). Most of these Glu, Asp, and Lys residues titrate with anomalous pKa values shifted by as much as 6 pH units (9, 10). The pKa values of these buried ionizable groups describe quantitatively the range of self-energies sampled by ionizable groups in different microenvironments in the protein interior. If the presence of even a single ionizable group buried in the hydrophobic interior of a protein can be so highly destabilizing, it is not obvious how a protein requiring more than one internal ionizable group for functional purposes can have a stable folded state, unless the dehydration experienced by buried ionizable groups can be at least partly compensated through favorable Coulomb or hydrogen-bonding interactions between them or through structural reorganization to solvate the buried ion pair. One of the goals of this study was to move beyond our previous studies of self-energies, to examine experimentally the relative magnitudes of Coulomb and self-energies in the hydrophobic interior of a protein (9, 10). Charged State of the Buried Glu–Lys Pair. The pKa value of Glu-23 in the presence of Lys-36 could not be measured directly with NMR spectroscopy, but the lack of pH sensitivity in the ΔΔG vs. pH curve for the V23E/L36K double variant suggests that either the pKa values of Glu-23 and Lys-36 are ≤4.4 and ≥10.4, respectively, (i.e., the Glu–Lys pair is charged) or the less likely interpretation that the pKa values Glu-23 and Lys-36 are ≥10.4 and ≤4.4, respectively (i.e., the Glu–Lys pair is neutral) and that they destabilize the protein by equal and cancelling amounts because of opposite pH dependencies. The V23E or L36K single substitutions were highly destabilizing throughout the entire pH range studied. In contrast the V23E/L36K double variant was much less destabilizing than expected from the behavior of the individual V23E or L36K substitutions. At pH 3.9 and 10.4, where Glu and Lys, respectively, are mostly neutral in water, the V23E and L36K substitutions, when made individually, destabilized the protein by 4.4 and 3.4 kcal/mol, respectively (9, 10). If the substitutions were additive, the nonelectrostatic contribution to destabilization of the V23E/L36K double variant would be 7.8 kcal/mol. At other pH values the stability of the proteins with either V23E or L36K substitutions is affected significantly by a pH-dependent contribution originating from shifts in the pKa values of the buried ionizable groups (Fig. 6). At pH 7 the V23E 11688 | www.pnas.org/cgi/doi/10.1073/pnas.1402900111

Interplay Between Coulomb and Self-Energies. At pH 7 the V23E/ L36K double variant is 4.9 kcal/mol more stable than it should be if the effects of the V23E and L36K single variants were additive (Fig. 6). This difference is greater than the cost of ionizing either of the two groups in the protein interior at this pH (2.6 kcal/mol for the V23E substitution and 3.8 kcal/mol for L36K). It appears that once a charge has been created in the hydrophobic environment of the protein, the cost of creating a second charge to establish a favorable ion pair is small. Both the crystallography and NMR spectroscopy data suggest that any reorganization resulting from making both the V23E and L36K substitutions is very minor; therefore, the total cost of ionizing the internal pair was estimated to be on the order of 1.5 kcal/mol (ΔΔG8H2 O;Δ + PHS=V23E=L36K;pH  7 − ΔΔG8H2 O;Δ+PHS=V23E; pH 3:9 −ΔΔG8H2 O;Δ+PHS=L36K;pH 10:4). This suggests that the interaction between Glu-23 and Lys-36 is strong, close to 5 kcal/mol, but not sufficiently strong to compensate for the dehydration of both Glu-23 and Lys-36. No Evidence for a Low-Barrier Hydrogen Bond. Interpretation of the equilibrium thermodynamic data in terms of the presence of a charged pair in the interior of the protein is complicated by the possibility of an unusual hydrogen-bonding arrangement in which

Fig. 5. (A) Titration of Glu-23 in the V23E (black) and V23E/L36K (red) variants monitored by Cδ chemical shift. (Inset) Multidimensional NMR spectroscopy spectra from pH 4.4 (red) to pH 9.3 (violet) 0.3–0.4 pH unit increments for the Δ+PHS/V23E/L36K variant. (B) Titration of Asp-21 in the V23E (black) and V23E/L36K (red) variants, monitored by Cγ chemical shift.

Robinson et al.

Fig. 6. (A) pH dependence of ΔG8H2 O for Δ+PHS (black circles), V23E (9) (red circles), L36K (10) (blue circles), and V23E/L36K (green circles) variants of SNase. Solid lines are meant only to guide the eye. The solid green line refers to the hypothetical variant in which both Glu-23 and Lys-36 are present, but with the pKa values they have in the single-site variants. (B) ΔΔG8H2 O for the V23E/L36K variant (green circle) calculated as the difference in ΔG8H2 O of Δ+PHS/V23E/L36K and Δ+PHS. The solid line refers to the fit of Eq. 1, which assumes that a single titratable group is responsible for the pH dependence observed. Simulations of ΔΔG8H2 O assuming a Glu with a pKa of 7.1 or a Lys with a pKa of 7.2 are shown as dashed red and blue lines, respectively. The dashed green line refers to the simulation for the additive situation in which a Glu has an elevated pKa of 7.1 and a Lys a depressed pKa of 7.2 (Eq. 2).

the charged and the neutral form of the pair are in barrierless exchange. Glu-23 and Lys-36 meet all of the conditions invoked for low-barrier hydrogen bonds (LBHBs): in the absence of each other they have matched pKa values, they are close (2.6 Å) and protected from bulk water by the relatively hydrophobic environment of the protein (30, 31). An LBHB is a system described by a double-well configuration (B···HA and B+H···A−) wherein the height of the barrier between the wells is sufficiently low to allow the proton to move freely between the hydrogen bond donor and acceptor (A···H···B). LBHBs are thought to be sufficient to rationalize catalysis by virtue of being able to contribute as many as 20 kcal/mol to the stabilization of transition states (30–32), an interpretation that is not without controversy (33). In the present case the NMR spectroscopy experiments that were performed offer no insight into the state of the Glu-23–Lys-36 pair. However, the measured coupling free energy from the interaction between the ionizable residues in the pair is of the order of 5 kcal/mol at pH 7, far less than the 20 kcal/mol invoked for LBHBs (30, 31). One reason the coupling energy is lower than expected is that Glu-23 and Lys-36 in the single variants are buried in what is normally a highly hydrophobic region of SNase, but when buried as a charged pair, their microenvironment become considerably more polar. It is possible that a truly hydrophobic microenvironment sufficient to achieve the conditions invoked for LBHBs cannot be preserved in the presence of a buried charged pair. The other reason Glu-23 and Lys-36 in SNase cannot establish an LBHB is that the geometry is unfavorable.

Preorganized vs. Reorganization. The energy required to ionize an individual basic or acidic residue buried in the hydrophobic interior of SNase, calculated from the measured pKa values of ionizable groups buried at 25 internal positions, ranges from 1.5 to 6.9 kcal/mol (9, 10). It has been suggested that naturally occurring, internal ionizable groups are stabilized by preorganized environments in which dipoles and other charges offset the unfavorable dehydration energy by behaving as materials with relatively high dielectric constants (3–5). The successful introduction of the Glu-23–Lys-36 pair into the hydrophobic interior of SNase demonstrates that not one, but two ionizable groups can be introduced into the interior of a protein without requiring preorganization of the site. The Glu-23–Lys-36 pair appears to be stabilized by unavoidable contacts with internal protein dipoles and internal water molecules that are more a consequence of reorganization than of preorganization. The slight differences between the crystal structures of the Glu-23–Lys-36 variant and those of the single variants or the reference Δ+PHS protein provide an indication of the subtle structural reorganization that facilitates favorable interactions between the buried charge pair and permanent dipoles of the protein. These small structural changes lead to favorable interactions that, collectively, lower the overall energy of the short ionic bond between Glu-23 and Lys-36. Ion pairs in hydrophobic environments have been studied previously. The interaction between the buried His-31–Asp-70 pair in T4 lysozyme was also found to be stabilizing (34), whereas in the Arc repressor the interactions within the buried Arg–Glu– Arg triad were destabilizing (35). The discrepancy between the properties of different buried charge pairs and clusters could be explained in part by the methods used to measure the Coulomb contribution of the ion pairs (36). Alternatively, these discrepancies might reflect the fact that although internal ion pairs may destabilize a protein overall, the net Coulomb interaction might be stabilizing relative to either a charge–dipole or dipole–dipole interaction. Note, however, that although the ion pairs in the T4 lysozyme and in the Arc repressor are referred to as being buried, they are both partially solvent exposed, and at least one of the ionizable groups had some contact with bulk water. The properties of ion pairs that are completely buried and secluded from bulk water can, in principle, be quite different from those of interfacial ion pairs that contact bulk water. Warshel has already noted that the polarity of the microenvironment surrounding the buried ion pairs can be the dominant factor that determines if the buried ion pair has a net stabilizing or destabilizing influence on the protein, and that alone might explain differences between the different internal ion pairs (37). Implications. The buried Glu–Lys pair allowed a quantitative

Origins of the High Apparent Dielectric Constant. When Glu-23 and

Lys-36 are introduced alone the ionizable moieties are buried in highly hydrophobic microenvironments. When in the neutral state, neither group alters its microenvironment. When introduced together, Glu-23 and Lys-36 deform their environment Robinson et al.

description of the balance between self-energy and Coulomb interactions experienced by interacting charges buried in a hydrophobic environment in a protein. These data constitute stringent benchmarks useful to test the ability of structure-based calculations to reproduce pKa values of internal ionizable groups for PNAS | August 12, 2014 | vol. 111 | no. 32 | 11689

BIOPHYSICS AND COMPUTATIONAL BIOLOGY

in a subtle manner, such that the ionizable moieties end up in a relatively polar environment that behaves as having dielectric properties comparable to those of a material with a relatively high dielectric constant. This deformation is evident in the crystal structure of the V23E/L36K double variant: (i) the backbone carbonyl of Gly-20 moved slightly to hydrogen bond with the Nζ of Lys-36, (ii) the side chain of Thr-62 exists in multiple rotameric states allowing it to hydrogen bond to Glu-23 or Gly-20, and (iii) there are two deeply buried water molecules that hydrate the Glu–Lys pair, one of which has a potential to form a single-file water chain linking the internal paired groups with bulk water. This “water wire” could constitute a conduit whereby H+ can move between the internal pair and bulk water. Overall, the data suggest that although the Glu-23–Lys-36 pair is deeply buried in the protein, it is well solvated by internal water molecules and by the reorganized protein microenvironment. The overall effect is to stabilize the Glu-23–Lys-36 pair in the charged state.

the right physical reasons, with the magnitudes of Coulomb and self-energy contributions balanced correctly. Warshel and a colleague have suggested that it is physically inappropriate to calculate these two different types of energy terms in continuum electrostatics models using a single value for the protein dielectric constant (14). The data from this study will allow an unprecedented, direct test of this idea. The factors that govern the properties of the artificial, buried Glu–Lys pair are the same ones that govern the properties of ionizable groups in the active sites of enzymes (38–40). Therefore, the successful engineering of a Glu–Lys pair in the hydrophobic interior of a highly stable form of SNase is also significant because it constitutes an empirical demonstration that buried ion pairs and charge clusters resembling enzymatic active sites can be engineered in the hydrophobic interior of a scaffold protein. It demonstrates that in a protein with an inherently high thermodynamic stability, it is possible to bury multiple ionizable residues without the need for specialized structural adaptations to accommodate them. It appears that the inherent conformational flexibility of proteins, water penetration into the core, the unavoidable presence of backbone polar atoms, and favorable Coulomb interactions are sufficient to stabilize a charge cluster buried in the relatively dry protein interior. This study provides

a clear demonstration that one of the most important factors that has to be considered in rational design or evolution of novel enzymatic active sites is the need for a high thermodynamic stability sufficient to allow the protein to withstand the unavoidable, destabilizing consequences associated with the presence of ionizable residues in hydrophobic environments (41).

1. von Ballmoos C, Wiedenmann A, Dimroth P (2009) Essentials for ATP synthesis by F1F0 ATP synthases. Annu Rev Biochem 78:649–672. 2. Iwata S, et al. (1998) Complete structure of the 11-subunit bovine mitochondrial cytochrome bc1 complex. Science 281(5373):64–71. 3. Harris TK, Turner GJ (2002) Structural basis of perturbed pKa values of catalytic groups in enzyme active sites. IUBMB Life 53(2):85–98. 4. Warshel A (1978) Energetics of enzyme catalysis. Proc Natl Acad Sci USA 75(11): 5250–5254. 5. Cannon WR, Benkovic SJ (1998) Solvation, reorganization energy, and biological catalysis. J Biol Chem 273(41):26257–26260. 6. Pisliakov AV, Sharma PK, Chu ZT, Haranczyk M, Warshel A (2008) Electrostatic basis for the unidirectionality of the primary proton transfer in cytochrome c oxidase. Proc Natl Acad Sci USA 105(22):7726–7731. 7. Rivalta I, et al. (2011) Structural-functional role of chloride in photosystem II. Biochemistry 50(29):6312–6315. 8. Hendsch ZS, Tidor B (1994) Do salt bridges stabilize proteins? A continuum electrostatic analysis. Protein Sci 3(2):211–226. 9. Isom DG, Castañeda CA, Cannon BR, Velu PD, García-Moreno E B (2010) Charges in the hydrophobic interior of proteins. Proc Natl Acad Sci USA 107(37):16096–16100. 10. Isom DG, Castañeda CA, Cannon BR, García-Moreno E B (2011) Large shifts in pKa values of lysine residues buried inside a protein. Proc Natl Acad Sci USA 108(13): 5260–5265. 11. Dwyer JJ, et al. (2000) High apparent dielectric constants in the interior of a protein reflect water penetration. Biophys J 79(3):1610–1620. 12. García-Moreno E B, et al. (1997) Experimental measurement of the effective dielectric in the hydrophobic core of a protein. Biophys Chem 64(1-3):211–224. 13. Isom DG, Cannon BR, Castañeda CA, Robinson A, García-Moreno E B (2008) High tolerance for ionizable residues in the hydrophobic interior of proteins. Proc Natl Acad Sci USA 105(46):17784–17788. 14. Schutz CN, Warshel A (2001) What are the dielectric “constants” of proteins and how to validate electrostatic models? Proteins 44(4):400–417. 15. Harms MJ, et al. (2009) The pK(a) values of acidic and basic residues buried at the same internal location in a protein are governed by different factors. J Mol Biol 389(1):34–47. 16. Castañeda CA, et al. (2009) Molecular determinants of the pKa values of Asp and Glu residues in staphylococcal nuclease. Proteins 77(3):570–588. 17. Antosiewicz J, McCammon JA, Gilson MK (1996) The determinants of pKas in proteins. Biochemistry 35(24):7819–7833. 18. Antosiewicz J, McCammon JA, Gilson MK (1994) Prediction of pH-dependent properties of proteins. J Mol Biol 238(3):415–436. 19. Kamerlin SCL, Haranczyk M, Warshel A (2009) Progress in ab initio QM/MM freeenergy simulations of electrostatic energies in proteins: Accelerated QM/MM studies of pKa, redox reactions and solvation free energies. J Phys Chem B 113(5):1253–1272. 20. Nagy PI, Erhardt PW (2010) Theoretical studies of salt-bridge formation by amino acid side chains in low and medium polarity environments. J Phys Chem B 114(49):16436–16442. 21. Chimenti MS, et al. (2012) Structural reorganization triggered by charging of Lys residues in the hydrophobic interior of a protein. Structure 20(6):1071–1085. 22. Wynn R, Harkins PC, Richards FM, Fox RO (1996) Mobile unnatural amino acid side chains in the core of staphylococcal nuclease. Protein Sci 5(6):1026–1031. 23. Quirt AR, et al. (1974) Carbon-13 nuclear magnetic resonance titration shifts in amino acids. J Am Chem Soc 96(2):570–574. 24. Chimenti MS, Castañeda CA, Majumdar A, García-Moreno E B (2011) Structural origins of high apparent dielectric constants experienced by ionizable groups in the hydrophobic core of a protein. J Mol Biol 405(2):361–377.

25. Joshi MD, Hedberg A, McIntosh LP (1997) Complete measurement of the pKa values of the carboxyl and imidazole groups in Bacillus circulans xylanase. Protein Sci 6(12): 2667–2670. 26. Oda Y, et al. (1994) Individual ionization constants of all the carboxyl groups in ribonuclease HI from Escherichia coli determined by NMR. Biochemistry 33(17): 5275–5284. 27. Carter PJ, Winter G, Wilkinson AJ, Fersht AR (1984) The use of double mutants to detect structural changes in the active site of the tyrosyl-tRNA synthetase (Bacillus stearothermophilus). Cell 38(3):835–840. 28. Harms MJ, Schlessman JL, Sue GR, García-Moreno E B (2011) Arginine residues at internal positions in a protein are always charged. Proc Natl Acad Sci USA 108(47): 18954–18959. 29. Cannon B (2008) Thermodynamic Consequences of Substitution of Internal Positions in Proteins with Polar and Ionizable Residues (Johns Hopkins Univ, Baltimore). 30. Cleland WW, Frey PA, Gerlt JA (1998) The low barrier hydrogen bond in enzymatic catalysis. J Biol Chem 273(40):25529–25532. 31. Cleland WW, Kreevoy MM (1994) Low-barrier hydrogen bonds and enzymic catalysis. Science 264(5167):1887–1890. 32. Gerlt JA, Kreevoy MM, Cleland W, Frey PA (1997) Understanding enzymic catalysis: The importance of short, strong hydrogen bonds. Chem Biol 4(4):259–267. 33. Warshel A, Papazyan A, Kollman PA, Cleland WW (1995) On low-barrier hydrogen bonds and enzyme catalysis. Science 269(5220):102–106. 34. Anderson DE, Becktel WJ, Dahlquist FW (1990) pH-induced denaturation of proteins: A single salt bridge contributes 3-5 kcal/mol to the free energy of folding of T4 lysozyme. Biochemistry 29(9):2403–2408. 35. Waldburger CD, Schildbach JF, Sauer RT (1995) Are buried salt bridges important for protein stability and conformational specificity? Nat Struct Biol 2(2):122–128. 36. Bosshard HR, Marti DN, Jelesarov I (2004) Protein stabilization by salt bridges: Concepts, experimental approaches and clarification of some misunderstandings. J Mol Recognit 17(1):1–16. 37. Warshel A (1987) What about protein polarity? Nature 330(6143):15–16. 38. Gul S, et al. (2008) Generation of nucleophilic character in the Cys25/His159 ion pair of papain involves Trp177 but not Asp158. Biochemistry 47(7):2025–2035. 39. Mössner E, Iwai H, Glockshuber R (2000) Influence of the pK(a) value of the buried, active-site cysteine on the redox properties of thioredoxin-like oxidoreductases. FEBS Lett 477(1-2):21–26. 40. Schwans JP, Sunden F, Gonzalez A, Tsai Y, Herschlag D (2013) Uncovering the determinants of a highly perturbed tyrosine pKa in the active site of ketosteroid isomerase. Biochemistry 52(44):7840–7855. 41. Merski M, Shoichet BK (2012) Engineering a model protein cavity to catalyze the Kemp elimination. Proc Natl Acad Sci USA 109(40):16179–16183. 42. Bailey S; Collaborative Computational Project, Number 4 (1994) The CCP4 suite: Programs for protein crystallography. Acta Crystallogr D Biol Crystallogr 50(Pt 5): 760–763. 43. Emsley P, Cowtan K (2004) Coot: Model-building tools for molecular graphics. Acta Crystallogr D Biol Crystallogr 60(Pt 12 Pt 1):2126–2132. 44. Delaglio F, et al. (1995) NMRPipe: A multidimensional spectral processing system based on UNIX pipes. J Biomol NMR 6(3):277–293. 45. Goddard T, Kneller D (2008) Sparky 3 (Univ of California, San Francisco). 46. Bell-Upp P, et al. (2011) Thermodynamic principles for the engineering of pH-driven conformational switches and acid insensitive proteins. Biophys Chem 159(1):217–226.

11690 | www.pnas.org/cgi/doi/10.1073/pnas.1402900111

Methods All variants were engineered using a highly stable variant of SNase (Δ+PHS) using protein purification and molecular cloning techniques described previously (12). Crystal structures of Δ+PHS/V23E and Δ+PHS/V23E/L36K were solved by molecular replacement using Δ+PHS (Protein Data Bank ID code 3BDC) (16) as a search model and iterative model building was performed using Refmac5 (42) and Coot (43). NMR data were collected on a Bruker Avance II-600 equipped with a cryoprobe, and spectra were processed and analyzed with NMRPipe (44) and Sparky (45), respectively. Trp fluorescence and CD222nm data were collected with an Aviv Automatic Titrating Fluorometer 105 and an Aviv CD spectrometer model 215, respectively, as described previously (9). For detailed methods, see SI Methods. ACKNOWLEDGMENTS. NMR data were collected with the help of Dr. Ananya Majumdar in the NMR Center at Johns Hopkins University. This work was supported by National Institutes of Health Grant GM-065197 (to B.G.-M.E.).

Robinson et al.